![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to gcc.info-6 CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GCC 2.x] / gcc|
Return to gcc.info-6 CVS log | Up to [GCC 2.x] / gcc |
1.1 root 1: This is Info file gcc.info, produced by Makeinfo-1.43 from the input
2: file gcc.texi.
3:
4: This file documents the use and the internals of the GNU compiler.
5:
6: Copyright (C) 1988, 1989, 1992 Free Software Foundation, Inc.
7:
8: Permission is granted to make and distribute verbatim copies of
9: this manual provided the copyright notice and this permission notice
10: are preserved on all copies.
11:
12: Permission is granted to copy and distribute modified versions of
13: this manual under the conditions for verbatim copying, provided also
14: that the section entitled "GNU General Public License" is included
15: exactly as in the original, and provided that the entire resulting
16: derived work is distributed under the terms of a permission notice
17: identical to this one.
18:
19: Permission is granted to copy and distribute translations of this
20: manual into another language, under the above conditions for modified
21: versions, except that the section entitled "GNU General Public
22: License" and this permission notice may be included in translations
23: approved by the Free Software Foundation instead of in the original
24: English.
25:
26:
27: File: gcc.info, Node: Non-bugs, Prev: Bug Reporting, Up: Bugs
28:
29: Certain Changes We Don't Want to Make
30: =====================================
31:
32: This section lists changes that people frequently request, but which
33: we do not make because we think GNU CC is better without them.
34:
35: * Checking the number and type of arguments to a function which has
36: an old-fashioned definition and no prototype.
37:
38: Such a feature would work only occasionally--only for calls that
39: appear in the same file as the called function, following the
40: definition. The only way to check all calls reliably is to add a
41: prototype for the function. But adding a prototype will
42: eliminate the need for this feature. So the feature is not
43: worthwhile.
44:
45: * Warning about using an expression whose type is signed as a shift
46: count.
47:
48: Shift count operands are probably signed more often than unsigned.
49: Warning about this would cause far more annoyance than good.
50:
51: * Warning about assigning a signed value to an unsigned variable.
52:
53: Such assignments must be very common; warning about them would
54: cause more annoyance than good.
55:
56: * Making bitfields unsigned by default on particular machines where
57: "the ABI standard" says to do so.
58:
59: The ANSI C standard leaves it up to the implementation whether a
60: bitfield declared plain `int' is signed or not. This in effect
61: creates two alternative dialects of C.
62:
63: The GNU C compiler supports both dialects; you can specify the
64: dialect you want with the option `-fsigned-bitfields' or
65: `-funsigned-bitfields'. However, this leaves open the question
66: of which dialect to use by default.
67:
68: Currently, the preferred dialect makes plain bitfields signed,
69: because this is simplest. Since `int' is the same as `signed
70: int' in every other context, it is cleanest for them to be the
71: same in bitfields as well.
72:
73: Some computer manufacturers have published Application Binary
74: Interface standards which specify that plain bitfields should be
75: unsigned. It is a mistake, however, to say anything about this
76: issue in an ABI. This is because the handling of plain bitfields
77: distinguishes two dialects of C. Both dialects are meaningful on
78: every type of machine. Whether a particular object file was
79: compiled using signed bitfields or unsigned is of no concern to
80: functions in any other object file, even if they access the same
81: bitfields in the same data structures.
82:
83: A given program is written in one or the other of these two
84: dialects. The program stands a chance to work on most any
85: machine if it is compiled with the proper dialect. It is
86: unlikely to work at all if compiled with the wrong dialect.
87:
88: Many users appreciate the GNU C compiler because it provides an
89: environment that is uniform across machines. These users would be
90: inconvenienced if the compiler treated plain bitfields
91: differently on certain machines.
92:
93: Occasionally users write programs intended only for a particular
94: machine type. On these occasions, the users would benefit if the
95: GNU C compiler were to support by default the same dialect as the
96: other compilers on that machine. But such applications are rare.
97: And users writing a program to run on more than one type of
98: machine cannot possibly benefit from this kind of compatibility.
99:
100: This is why GNU CC does and will treat plain bitfields in the same
101: fashion on all types of machines (by default).
102:
103: (Of course, users strongly concerned about portability should
104: indicate explicitly in each bitfield whether it is signed or not.)
105:
106: * Undefining `__STDC__' when `-ansi' is not used.
107:
108: Currently, GNU CC defines `__STDC__' as long as you don't use
109: `-traditional'. This provides good results in practice.
110:
111: Programmers normally use conditionals on `__STDC__' to ask whether
112: it is safe to use certain features of ANSI C, such as function
113: prototypes or ANSI token concatenation. Since plain `gcc'
114: supports all the features of ANSI C, the correct answer to these
115: questions is "yes".
116:
117: Some users try to use `__STDC__' to check for the availability of
118: certain library facilities. This is actually incorrect usage in
119: an ANSI C program, because the ANSI C standard says that a
120: conforming freestanding implementation should define `__STDC__'
121: even though it does not have the library facilities. `gcc -ansi
122: -pedantic' is a conforming freestanding implementation, and it is
123: therefore required to define `__STDC__', even though it does not
124: come with an ANSI C library.
125:
126: Sometimes people say that defining `__STDC__' in a compiler that
127: does not completely conform to the ANSI C standard somehow
128: violates the standard. This is illogical. The standard is a
129: standard for compilers that are supposed to conform. It says
130: nothing about what any other compilers should do. Whatever the
131: ANSI C standard says is relevant to the design of plain `gcc'
132: without `-ansi' only for pragmatic reasons, not as a requirement.
133:
134: * Undefining `__STDC__' in C++.
135:
136: Programs written to compile with C++-to-C translators get the
137: value of `__STDC__' that goes with the C compiler that is
138: subsequently used. These programs must test `__STDC__' to
139: determine what kind of C preprocessor that compiler uses: whether
140: they should concatenate tokens in the ANSI C fashion or in the
141: traditional fashion.
142:
143: These programs work properly with GNU C++ if `__STDC__' is
144: defined. They would not work otherwise.
145:
146: In addition, many header files are written to provide prototypes
147: in ANSI C but not in traditional C. Many of these header files
148: can work without change in C++ provided `__STDC__' is defined.
149: If `__STDC__' is not defined, they will all fail, and will all
150: need to be changed to test explicitly for C++ as well.
151:
152:
153: File: gcc.info, Node: VMS, Next: Portability, Prev: Bugs, Up: Top
154:
155: Using GNU CC on VMS
156: *******************
157:
158: * Menu:
159:
160: * Include Files and VMS:: Where the preprocessor looks for the include files.
161: * Global Declarations:: How to do globaldef, globalref and globalvalue with
162: GNU CC.
163: * VMS Misc:: Misc information.
164:
165:
166: File: gcc.info, Node: Include Files and VMS, Next: Global Declarations, Prev: VMS, Up: VMS
167:
168: Include Files and VMS
169: =====================
170:
171: Due to the differences between the filesystems of Unix and VMS, GNU
172: CC attempts to translate file names in `#include' into names that VMS
173: will understand. The basic strategy is to prepend a prefix to the
174: specification of the include file, convert the whole filename to a VMS
175: filename, and then try to open the file. GNU CC tries various prefixes
176: one by one until one of them succeeds:
177:
178: 1. The first prefix is the `GNU_CC_INCLUDE:' logical name: this is
179: where GNU C header files are traditionally stored. If you wish
180: to store header files in non-standard locations, then you can
181: assign the logical `GNU_CC_INCLUDE' to be a search list, where
182: each element of the list is suitable for use with a rooted
183: logical.
184:
185: 2. The next prefix tried is `SYS$SYSROOT:[SYSLIB.]'. This is where
186: VAX-C header files are traditionally stored.
187:
188: 3. If the include file specification by itself is a valid VMS
189: filename, the preprocessor then uses this name with no prefix in
190: an attempt to open the include file.
191:
192: 4. If the file specification is not a valid VMS filename (i.e. does
193: not contain a device or a directory specifier, and contains a `/'
194: character), the preprocessor tries to convert it from Unix syntax
195: to VMS syntax.
196:
197: Conversion works like this: the first directory name becomes a
198: device, and the rest of the directories are converted into
199: VMS-format directory names. For example, `X11/foobar.h' is
200: translated to `X11:[000000]foobar.h' or `X11:foobar.h', whichever
201: one can be opened. This strategy allows you to assign a logical
202: name to point to the actual location of the header files.
203:
204: 5. If none of these strategies succeeds, the `#include' fails.
205:
206: Include directives of the form:
207:
208: #include foobar
209:
210: are a common source of incompatibility between VAX-C and GNU CC. VAX-C
211: treats this much like a standard `#include <foobar.h>' directive.
212: That is incompatible with the ANSI C behavior implemented by GNU CC: to
213: expand the name `foobar' as a macro. Macro expansion should
214: eventually yield one of the two standard formats for `#include':
215:
216: #include "FILE"
217: #include <FILE>
218:
219: If you have this problem, the best solution is to modify the source
220: to convert the `#include' directives to one of the two standard forms.
221: That will work with either compiler. If you want a quick and dirty
222: fix, define the file names as macros with the proper expansion, like
223: this:
224:
225: #define stdio <stdio.h>
226:
227: This will work, as long as the name doesn't conflict with anything else
228: in the program.
229:
230: Another source of incompatibility is that VAX-C assumes that:
231:
232: #include "foobar"
233:
234: is actually asking for the file `foobar.h'. GNU CC does not make this
235: assumption, and instead takes what you ask for literally; it tries to
236: read the file `foobar'. The best way to avoid this problem is to
237: always specify the desired file extension in your include directives.
238:
239: GNU CC for VMS is distributed with a set of include files that is
240: sufficient to compile most general purpose programs. Even though the
241: GNU CC distribution does not contain header files to define constants
242: and structures for some VMS system-specific functions, there is no
243: reason why you cannot use GNU CC with any of these functions. You
244: first may have to generate or create header files, either by using the
245: public domain utility `UNSDL' (which can be found on a DECUS tape), or
246: by extracting the relevant modules from one of the system macro
247: libraries, and using an editor to construct a C header file.
248:
249:
250: File: gcc.info, Node: Global Declarations, Next: VMS Misc, Prev: Include Files and VMS, Up: VMS
251:
252: Global Declarations and VMS
253: ===========================
254:
255: GNU CC does not provide the `globalref', `globaldef' and
256: `globalvalue' keywords of VAX-C. You can get the same effect with an
257: obscure feature of GAS, the GNU assembler. (This requires GAS version
258: 1.39 or later.) The following macros allow you to use this feature in
259: a fairly natural way:
260:
261: #ifdef __GNUC__
262: #define GLOBALREF(NAME) \
263: NAME asm("_$$PsectAttributes_GLOBALSYMBOL$$" #NAME )
264: #define GLOBALDEF(NAME,VALUE) \
265: NAME asm("_$$PsectAttributes_GLOBALSYMBOL$$" #NAME ) = VALUE
266: #define GLOBALVALUEREF(NAME) \
267: const NAME [1] asm("_$$PsectAttributes_GLOBALVALUE$$" #NAME )
268: #define GLOBALVALUEDEF(NAME,VALUE) \
269: const NAME [1] asm("_$$PsectAttributes_GLOBALVALUE$$" #NAME ) = {VALUE}
270: #else
271: #define GLOBALREF(NAME) globalref NAME
272: #define GLOBALDEF(NAME,VALUE) globaldef NAME = VALUE
273: #define GLOBALVALUEDEF(NAME,VALUE) globalvalue NAME = VALUE
274: #define GLOBALVALUEREF(NAME) globalvalue NAME
275: #endif
276:
277: (The `_$$PsectAttributes_GLOBALSYMBOL' prefix at the start of the name
278: is removed by the assembler, after it has modified the attributes of
279: the symbol). These macros are provided in the VMS binaries
280: distribution in a header file `GNU_HACKS.H'. An example of the usage
281: is:
282:
283: int GLOBALREF (ijk);
284: int GLOBALDEF (jkl, 0);
285:
286: The macros `GLOBALREF' and `GLOBALDEF' cannot be used
287: straightforwardly for arrays, since there is no way to insert the array
288: dimension into the declaration at the right place. However, you can
289: declare an array with these macros if you first define a typedef for
290: the array type, like this:
291:
292: typedef int intvector[10];
293: intvector GLOBALREF (foo);
294:
295: Array and structure initializers will also break the macros; you can
296: define the initializer to be a macro of its own, or you can expand the
297: `GLOBALDEF' macro by hand. You may find a case where you wish to use
298: the `GLOBALDEF' macro with a large array, but you are not interested
299: in explicitly initializing each element of the array. In such cases
300: you can use an initializer like: `{0,}', which will initialize the
301: entire array to `0'.
302:
303: A shortcoming of this implementation is that a variable declared
304: with `GLOBALVALUEREF' or `GLOBALVALUEDEF' is always an array. For
305: example, the declaration:
306:
307: int GLOBALVALUEREF(ijk);
308:
309: declares the variable `ijk' as an array of type `int [1]'. This is
310: done because a globalvalue is actually a constant; its "value" is what
311: the linker would normally consider an address. That is not how an
312: integer value works in C, but it is how an array works. So treating
313: the symbol as an array name gives consistent results--with the
314: exception that the value seems to have the wrong type. *Don't try to
315: access an element of the array.* It doesn't have any elements. The
316: array "address" may not be the address of actual storage.
317:
318: The fact that the symbol is an array may lead to warnings where the
319: variable is used. Insert type casts to avoid the warnings. Here is an
320: example; it takes advantage of the ANSI C feature allowing macros that
321: expand to use the same name as the macro itself.
322:
323: int GLOBALVALUEREF (ss$_normal);
324: int GLOBALVALUEDEF (xyzzy,123);
325: #ifdef __GNUC__
326: #define ss$_normal ((int) ss$_normal)
327: #define xyzzy ((int) xyzzy)
328: #endif
329:
330: Don't use `globaldef' or `globalref' with a variable whose type is
331: an enumeration type; this is not implemented. Instead, make the
332: variable an integer, and use a `globalvaluedef' for each of the
333: enumeration values. An example of this would be:
334:
335: #ifdef __GNUC__
336: int GLOBALDEF (color, 0);
337: int GLOBALVALUEDEF (RED, 0);
338: int GLOBALVALUEDEF (BLUE, 1);
339: int GLOBALVALUEDEF (GREEN, 3);
340: #else
341: enum globaldef color {RED, BLUE, GREEN = 3};
342: #endif
343:
344:
345: File: gcc.info, Node: VMS Misc, Prev: Global Declarations, Up: VMS
346:
347: Other VMS Issues
348: ================
349:
350: GNU CC automatically arranges for `main' to return 1 by default if
351: you fail to specify an explicit return value. This will be interpreted
352: by VMS as a status code indicating a normal successful completion.
353: Version 1 of GNU CC did not provide this default.
354:
355: GNU CC on VMS works only with the GNU assembler, GAS. You need
356: version 1.37 or later of GAS in order to produce value debugging
357: information for the VMS debugger. Use the ordinary VMS linker with
358: the object files produced by GAS.
359:
360: Under previous versions of GNU CC, the generated code would
361: occasionally give strange results when linked to the sharable
362: `VAXCRTL' library. Now this should work.
363:
364: A caveat for use of `const' global variables: the `const' modifier
365: must be specified in every external declaration of the variable in all
366: of the source files that use that variable. Otherwise the linker will
367: issue warnings about conflicting attributes for the variable. Your
368: program will still work despite the warnings, but the variable will be
369: placed in writable storage.
370:
371: The VMS linker does not distinguish between upper and lower case
372: letters in function and variable names. However, usual practice in C
373: is to distinguish case. Normally GNU CC (by means of the assembler
374: GAS) implements usual C behavior by augmenting each name that is not
375: all lower-case. A name is augmented by truncating it to at most 23
376: characters and then adding more characters at the end which encode the
377: case pattern the rest.
378:
379: Name augmentation yields bad results for programs that use
380: precompiled libraries (such as Xlib) which were generated by another
381: compiler. You can use the compiler option `/NOCASE_HACK' to inhibit
382: augmentation; it makes external C functions and variables
383: case-independent as is usual on VMS. Alternatively, you could write
384: all references to the functions and variables in such libraries using
385: lower case; this will work on VMS, but is not portable to other
386: systems.
387:
388: Function and variable names are handled somewhat differently with
389: GNU C++. The GNU C++ compiler performs "name mangling" on function
390: names, which means that it adds information to the function name to
391: describe the data types of the arguments that the function takes. One
392: result of this is that the name of a function can become very long.
393: Since the VMS linker only recognizes the first 31 characters in a name,
394: special action is taken to ensure that each function and variable has a
395: unique name that can be represented in 31 characters.
396:
397: If the name (plus a name augmentation, if required) is less than 32
398: characters in length, then no special action is performed. If the name
399: is longer than 31 characters, the assembler (GAS) will generate a hash
400: string based upon the function name, truncate the function name to 23
401: characters, and append the hash string to the truncated name. If the
402: `/VERBOSE' compiler option is used, the assembler will print both the
403: full and truncated names of each symbol that is truncated.
404:
405: The `/NOCASE_HACK' compiler option should not be used when you are
406: compiling programs that use libg++. libg++ has several instances of
407: objects (i.e. `Filebuf' and `filebuf') which become indistinguishable
408: in a case-insensitive environment. This leads to cases where you need
409: to inhibit augmentation selectively (if you were using libg++ and Xlib
410: in the same program, for example). There is no special feature for
411: doing this, but you can get the result by defining a macro for each
412: mixed case symbol for which you wish to inhibit augmentation. The
413: macro should expand into the lower case equivalent of itself. For
414: example:
415:
416: #define StuDlyCapS studlycaps
417:
418: These macro definitions can be placed in a header file to minimize
419: the number of changes to your source code.
420:
421:
422: File: gcc.info, Node: Portability, Next: Interface, Prev: VMS, Up: Top
423:
424: GNU CC and Portability
425: **********************
426:
427: The main goal of GNU CC was to make a good, fast compiler for
428: machines in the class that the GNU system aims to run on: 32-bit
429: machines that address 8-bit bytes and have several general registers.
430: Elegance, theoretical power and simplicity are only secondary.
431:
432: GNU CC gets most of the information about the target machine from a
433: machine description which gives an algebraic formula for each of the
434: machine's instructions. This is a very clean way to describe the
435: target. But when the compiler needs information that is difficult to
436: express in this fashion, I have not hesitated to define an ad-hoc
437: parameter to the machine description. The purpose of portability is
438: to reduce the total work needed on the compiler; it was not of
439: interest for its own sake.
440:
441: GNU CC does not contain machine dependent code, but it does contain
442: code that depends on machine parameters such as endianness (whether
443: the most significant byte has the highest or lowest address of the
444: bytes in a word) and the availability of autoincrement addressing. In
445: the RTL-generation pass, it is often necessary to have multiple
446: strategies for generating code for a particular kind of syntax tree,
447: strategies that are usable for different combinations of parameters.
448: Often I have not tried to address all possible cases, but only the
449: common ones or only the ones that I have encountered. As a result, a
450: new target may require additional strategies. You will know if this
451: happens because the compiler will call `abort'. Fortunately, the new
452: strategies can be added in a machine-independent fashion, and will
453: affect only the target machines that need them.
454:
455:
456: File: gcc.info, Node: Interface, Next: Passes, Prev: Portability, Up: Top
457:
458: Interfacing to GNU CC Output
459: ****************************
460:
461: GNU CC is normally configured to use the same function calling
462: convention normally in use on the target system. This is done with the
463: machine-description macros described (*note Machine Macros::.).
464:
465: However, returning of structure and union values is done
466: differently on some target machines. As a result, functions compiled
467: with PCC returning such types cannot be called from code compiled with
468: GNU CC, and vice versa. This does not cause trouble often because few
469: Unix library routines return structures or unions.
470:
471: GNU CC code returns structures and unions that are 1, 2, 4 or 8
472: bytes long in the same registers used for `int' or `double' return
473: values. (GNU CC typically allocates variables of such types in
474: registers also.) Structures and unions of other sizes are returned by
475: storing them into an address passed by the caller (usually in a
476: register). The machine-description macros `STRUCT_VALUE' and
477: `STRUCT_INCOMING_VALUE' tell GNU CC where to pass this address.
478:
479: By contrast, PCC on most target machines returns structures and
480: unions of any size by copying the data into an area of static storage,
481: and then returning the address of that storage as if it were a pointer
482: value. The caller must copy the data from that memory area to the
483: place where the value is wanted. This is slower than the method used
484: by GNU CC, and fails to be reentrant.
485:
486: On some target machines, such as RISC machines and the 80386, the
487: standard system convention is to pass to the subroutine the address of
488: where to return the value. On these machines, GNU CC has been
489: configured to be compatible with the standard compiler, when this
490: method is used. It may not be compatible for structures of 1, 2, 4 or
491: 8 bytes.
492:
493: GNU CC uses the system's standard convention for passing arguments.
494: On some machines, the first few arguments are passed in registers; in
495: others, all are passed on the stack. It would be possible to use
496: registers for argument passing on any machine, and this would probably
497: result in a significant speedup. But the result would be complete
498: incompatibility with code that follows the standard convention. So
499: this change is practical only if you are switching to GNU CC as the
500: sole C compiler for the system. We may implement register argument
501: passing on certain machines once we have a complete GNU system so that
502: we can compile the libraries with GNU CC.
503:
504: On some machines (particularly the Sparc), certain types of
505: arguments are passed "by invisible reference". This means that the
506: value is stored in memory, and the address of the memory location is
507: passed to the subroutine.
508:
509: If you use `longjmp', beware of automatic variables. ANSI C says
510: that automatic variables that are not declared `volatile' have
511: undefined values after a `longjmp'. And this is all GNU CC promises
512: to do, because it is very difficult to restore register variables
513: correctly, and one of GNU CC's features is that it can put variables
514: in registers without your asking it to.
515:
516: If you want a variable to be unaltered by `longjmp', and you don't
517: want to write `volatile' because old C compilers don't accept it, just
518: take the address of the variable. If a variable's address is ever
519: taken, even if just to compute it and ignore it, then the variable
520: cannot go in a register:
521:
522: {
523: int careful;
524: &careful;
525: ...
526: }
527:
528: Code compiled with GNU CC may call certain library routines. Most
529: of them handle arithmetic for which there are no instructions. This
530: includes multiply and divide on some machines, and floating point
531: operations on any machine for which floating point support is disabled
532: with `-msoft-float'. Some standard parts of the C library, such as
533: `bcopy' or `memcpy', are also called automatically. The usual
534: function call interface is used for calling the library routines.
535:
536: These library routines should be defined in the library `libgcc.a',
537: which GNU CC automatically searches whenever it links a program. On
538: machines that have multiply and divide instructions, if hardware
539: floating point is in use, normally `libgcc.a' is not needed, but it is
540: searched just in case.
541:
542: Each arithmetic function is defined in `libgcc1.c' to use the
543: corresponding C arithmetic operator. As long as the file is compiled
544: with another C compiler, which supports all the C arithmetic operators,
545: this file will work portably. However, `libgcc1.c' does not work if
546: compiled with GNU CC, because each arithmetic function would compile
547: into a call to itself!
548:
549:
550: File: gcc.info, Node: Passes, Next: RTL, Prev: Interface, Up: Top
551:
552: Passes and Files of the Compiler
553: ********************************
554:
555: The overall control structure of the compiler is in `toplev.c'.
556: This file is responsible for initialization, decoding arguments,
557: opening and closing files, and sequencing the passes.
558:
559: The parsing pass is invoked only once, to parse the entire input.
560: The RTL intermediate code for a function is generated as the function
561: is parsed, a statement at a time. Each statement is read in as a
562: syntax tree and then converted to RTL; then the storage for the tree
563: for the statement is reclaimed. Storage for types (and the
564: expressions for their sizes), declarations, and a representation of
565: the binding contours and how they nest, remain until the function is
566: finished being compiled; these are all needed to output the debugging
567: information.
568:
569: Each time the parsing pass reads a complete function definition or
570: top-level declaration, it calls the function `rest_of_compilation' or
571: `rest_of_decl_compilation' in `toplev.c', which are responsible for
572: all further processing necessary, ending with output of the assembler
573: language. All other compiler passes run, in sequence, within
574: `rest_of_compilation'. When that function returns from compiling a
575: function definition, the storage used for that function definition's
576: compilation is entirely freed, unless it is an inline function (*note
577: Inline::.).
578:
579: Here is a list of all the passes of the compiler and their source
580: files. Also included is a description of where debugging dumps can be
581: requested with `-d' options.
582:
583: * Parsing. This pass reads the entire text of a function
584: definition, constructing partial syntax trees. This and RTL
585: generation are no longer truly separate passes (formerly they
586: were), but it is easier to think of them as separate.
587:
588: The tree representation does not entirely follow C syntax,
589: because it is intended to support other languages as well.
590:
591: Language-specific data type analysis is also done in this pass,
592: and every tree node that represents an expression has a data type
593: attached. Variables are represented as declaration nodes.
594:
595: Constant folding and some arithmetic simplifications are also done
596: during this pass.
597:
598: The language-independent source files for parsing are
599: `stor-layout.c', `fold-const.c', and `tree.c'. There are also
600: header files `tree.h' and `tree.def' which define the format of
601: the tree representation.
602:
603: The source files for parsing C are `c-parse.y', `c-decl.c',
604: `c-typeck.c', `c-convert.c', `c-lang.c', and `c-aux-info.c' along
605: with header files `c-lex.h', and `c-tree.h'.
606:
607: The source files for parsing C++ are `cp-parse.y', `cp-class.c',
608: `cp-cvt.c',
609: `cp-decl.c', `cp-decl.c', `cp-decl2.c', `cp-dem.c',
610: `cp-except.c',
611: `cp-expr.c', `cp-init.c', `cp-lex.c', `cp-method.c',
612: `cp-ptree.c',
613: `cp-search.c', `cp-tree.c', `cp-type2.c', and `cp-typeck.c',
614: along with header files `cp-tree.def', `cp-tree.h', and
615: `cp-decl.h'.
616:
617: The special source files for parsing Objective C are
618: `objc-parse.y', `objc-actions.c', `objc-tree.def', and
619: `objc-actions.h'. Certain C-specific files are used for this as
620: well.
621:
622: The file `c-common.c' is also used for all of the above languages.
623:
624: * RTL generation. This is the conversion of syntax tree into RTL
625: code. It is actually done statement-by-statement during parsing,
626: but for most purposes it can be thought of as a separate pass.
627:
628: This is where the bulk of target-parameter-dependent code is
629: found, since often it is necessary for strategies to apply only
630: when certain standard kinds of instructions are available. The
631: purpose of named instruction patterns is to provide this
632: information to the RTL generation pass.
633:
634: Optimization is done in this pass for `if'-conditions that are
635: comparisons, boolean operations or conditional expressions. Tail
636: recursion is detected at this time also. Decisions are made
637: about how best to arrange loops and how to output `switch'
638: statements.
639:
640: The source files for RTL generation include `stmt.c',
641: `function.c', `expr.c', `calls.c', `explow.c', `expmed.c',
642: `optabs.c' and `emit-rtl.c'. Also, the file `insn-emit.c',
643: generated from the machine description by the program `genemit',
644: is used in this pass. The header file `expr.h' is used for
645: communication within this pass.
646:
647: The header files `insn-flags.h' and `insn-codes.h', generated
648: from the machine description by the programs `genflags' and
649: `gencodes', tell this pass which standard names are available for
650: use and which patterns correspond to them.
651:
652: Aside from debugging information output, none of the following
653: passes refers to the tree structure representation of the
654: function (only part of which is saved).
655:
656: The decision of whether the function can and should be expanded
657: inline in its subsequent callers is made at the end of rtl
658: generation. The function must meet certain criteria, currently
659: related to the size of the function and the types and number of
660: parameters it has. Note that this function may contain loops,
661: recursive calls to itself (tail-recursive functions can be
662: inlined!), gotos, in short, all constructs supported by GNU CC.
663: The file `integrate.c' contains the code to save a function's rtl
664: for later inlining and to inline that rtl when the function is
665: called. The header file `integrate.h' is also used for this
666: purpose.
667:
668: The option `-dr' causes a debugging dump of the RTL code after
669: this pass. This dump file's name is made by appending `.rtl' to
670: the input file name.
671:
672: * Jump optimization. This pass simplifies jumps to the following
673: instruction, jumps across jumps, and jumps to jumps. It deletes
674: unreferenced labels and unreachable code, except that unreachable
675: code that contains a loop is not recognized as unreachable in
676: this pass. (Such loops are deleted later in the basic block
677: analysis.) It also converts some code originally written with
678: jumps into sequences of instructions that directly set values
679: from the results of comparisons, if the machine has such
680: instructions.
681:
682: Jump optimization is performed two or three times. The first
683: time is immediately following RTL generation. The second time is
684: after CSE, but only if CSE says repeated jump optimization is
685: needed. The last time is right before the final pass. That
686: time, cross-jumping and deletion of no-op move instructions are
687: done together with the optimizations described above.
688:
689: The source file of this pass is `jump.c'.
690:
691: The option `-dj' causes a debugging dump of the RTL code after
692: this pass is run for the first time. This dump file's name is
693: made by appending `.jump' to the input file name.
694:
695: * Register scan. This pass finds the first and last use of each
696: register, as a guide for common subexpression elimination. Its
697: source is in `regclass.c'.
698:
699: * Jump threading. This pass detects a condition jump that branches
700: to an identical or inverse test. Such jumps can be `threaded'
701: through the second conditional test. The source code for this
702: pass is in `jump.c'. This optimization is only performed if
703: `-fthread-jumps' is enabled.
704:
705: * Common subexpression elimination. This pass also does constant
706: propagation. Its source file is `cse.c'. If constant
707: propagation causes conditional jumps to become unconditional or to
708: become no-ops, jump optimization is run again when CSE is
709: finished.
710:
711: The option `-ds' causes a debugging dump of the RTL code after
712: this pass. This dump file's name is made by appending `.cse' to
713: the input file name.
714:
715: * Loop optimization. This pass moves constant expressions out of
716: loops, and optionally does strength-reduction and loop unrolling
717: as well. Its source files are `loop.c' and `unroll.c', plus the
718: header `loop.h' used for communication between them. Loop
719: unrolling uses some functions in `integrate.c' and the header
720: `integrate.h'.
721:
722: The option `-dL' causes a debugging dump of the RTL code after
723: this pass. This dump file's name is made by appending `.loop' to
724: the input file name.
725:
726: * If `-frerun-cse-after-loop' was enabled, a second common
727: subexpression elimination pass is performed after the loop
728: optimization pass. Jump threading is also done again at this
729: time if it was specified.
730:
731: The option `-dt' causes a debugging dump of the RTL code after
732: this pass. This dump file's name is made by appending `.cse2' to
733: the input file name.
734:
735: * Stupid register allocation is performed at this point in a
736: nonoptimizing compilation. It does a little data flow analysis as
737: well. When stupid register allocation is in use, the next pass
738: executed is the reloading pass; the others in between are skipped.
739: The source file is `stupid.c'.
740:
741: * Data flow analysis (`flow.c'). This pass divides the program
742: into basic blocks (and in the process deletes unreachable loops);
743: then it computes which pseudo-registers are live at each point in
744: the program, and makes the first instruction that uses a value
745: point at the instruction that computed the value.
746:
747: This pass also deletes computations whose results are never used,
748: and combines memory references with add or subtract instructions
749: to make autoincrement or autodecrement addressing.
750:
751: The option `-df' causes a debugging dump of the RTL code after
752: this pass. This dump file's name is made by appending `.flow' to
753: the input file name. If stupid register allocation is in use,
754: this dump file reflects the full results of such allocation.
755:
756: * Instruction combination (`combine.c'). This pass attempts to
757: combine groups of two or three instructions that are related by
758: data flow into single instructions. It combines the RTL
759: expressions for the instructions by substitution, simplifies the
760: result using algebra, and then attempts to match the result
761: against the machine description.
762:
763: The option `-dc' causes a debugging dump of the RTL code after
764: this pass. This dump file's name is made by appending `.combine'
765: to the input file name.
766:
767: * Instruction scheduling (`sched.c'). This pass looks for
768: instructions whose output will not be available by the time that
769: it is used in subsequent instructions. (Memory loads and
770: floating point instructions often have this behavior on RISC
771: machines). It re-orders instructions within a basic block to try
772: to separate the definition and use of items that otherwise would
773: cause pipeline stalls.
774:
775: Instruction scheduling is performed twice. The first time is
776: immediately after instruction combination and the second is
777: immediately after reload.
778:
779: The option `-dS' causes a debugging dump of the RTL code after
780: this pass is run for the first time. The dump file's name is
781: made by appending `.sched' to the input file name.
782:
783: * Register class preferencing. The RTL code is scanned to find out
784: which register class is best for each pseudo register. The source
785: file is `regclass.c'.
786:
787: * Local register allocation (`local-alloc.c'). This pass allocates
788: hard registers to pseudo registers that are used only within one
789: basic block. Because the basic block is linear, it can use fast
790: and powerful techniques to do a very good job.
791:
792: The option `-dl' causes a debugging dump of the RTL code after
793: this pass. This dump file's name is made by appending `.lreg' to
794: the input file name.
795:
796: * Global register allocation (`global-alloc.c'). This pass
797: allocates hard registers for the remaining pseudo registers (those
798: whose life spans are not contained in one basic block).
799:
800: * Reloading. This pass renumbers pseudo registers with the hardware
801: registers numbers they were allocated. Pseudo registers that did
802: not get hard registers are replaced with stack slots. Then it
803: finds instructions that are invalid because a value has failed to
804: end up in a register, or has ended up in a register of the wrong
805: kind. It fixes up these instructions by reloading the
806: problematical values temporarily into registers. Additional
807: instructions are generated to do the copying.
808:
809: The reload pass also optionally eliminates the frame pointer and
810: inserts instructions to save and restore call-clobbered registers
811: around calls.
812:
813: Source files are `reload.c' and `reload1.c', plus the header
814: `reload.h' used for communication between them.
815:
816: The option `-dg' causes a debugging dump of the RTL code after
817: this pass. This dump file's name is made by appending `.greg' to
818: the input file name.
819:
820: * Instruction scheduling is repeated here to try to avoid pipeline
821: stalls due to memory loads generated for spilled pseudo registers.
822:
823: The option `-dR' causes a debugging dump of the RTL code after
824: this pass. This dump file's name is made by appending `.sched2'
825: to the input file name.
826:
827: * Jump optimization is repeated, this time including cross-jumping
828: and deletion of no-op move instructions.
829:
830: The option `-dJ' causes a debugging dump of the RTL code after
831: this pass. This dump file's name is made by appending `.jump2'
832: to the input file name.
833:
834: * Delayed branch scheduling. This optional pass attempts to find
835: instructions that can go into the delay slots of other
836: instructions, usually jumps and calls. The source file name is
837: `reorg.c'.
838:
839: The option `-dd' causes a debugging dump of the RTL code after
840: this pass. This dump file's name is made by appending `.dbr' to
841: the input file name.
842:
843: * Conversion from usage of some hard registers to usage of a
844: register stack may be done at this point. Currently, this is
845: supported only for the floating-point registers of the Intel
846: 80387 coprocessor. The source file name is `reg-stack.c'.
847:
848: The options `-dk' causes a debugging dump of the RTL code after
849: this pass. This dump file's name is made by appending `.stack'
850: to the input file name.
851:
852: * Final. This pass outputs the assembler code for the function.
853: It is also responsible for identifying spurious test and compare
854: instructions. Machine-specific peephole optimizations are
855: performed at the same time. The function entry and exit
856: sequences are generated directly as assembler code in this pass;
857: they never exist as RTL.
858:
859: The source files are `final.c' plus `insn-output.c'; the latter
860: is generated automatically from the machine description by the
861: tool `genoutput'. The header file `conditions.h' is used for
862: communication between these files.
863:
864: * Debugging information output. This is run after final because it
865: must output the stack slot offsets for pseudo registers that did
866: not get hard registers. Source files are `dbxout.c' for DBX
867: symbol table format, `sdbout.c' for SDB symbol table format, and
868: `dwarfout.c' for DWARF symbol table format.
869:
870: Some additional files are used by all or many passes:
871:
872: * Every pass uses `machmode.def' and `machmode.h' which define the
873: machine modes.
874:
875: * Several passes use `real.h', which defines the default
876: representation of floating point constants and how to operate on
877: them.
878:
879: * All the passes that work with RTL use the header files `rtl.h'
880: and `rtl.def', and subroutines in file `rtl.c'. The tools `gen*'
881: also use these files to read and work with the machine
882: description RTL.
883:
884: * Several passes refer to the header file `insn-config.h' which
885: contains a few parameters (C macro definitions) generated
886: automatically from the machine description RTL by the tool
887: `genconfig'.
888:
889: * Several passes use the instruction recognizer, which consists of
890: `recog.c' and `recog.h', plus the files `insn-recog.c' and
891: `insn-extract.c' that are generated automatically from the
892: machine description by the tools `genrecog' and `genextract'.
893:
894: * Several passes use the header files `regs.h' which defines the
895: information recorded about pseudo register usage, and
896: `basic-block.h' which defines the information recorded about
897: basic blocks.
898:
899: * `hard-reg-set.h' defines the type `HARD_REG_SET', a bit-vector
900: with a bit for each hard register, and some macros to manipulate
901: it. This type is just `int' if the machine has few enough hard
902: registers; otherwise it is an array of `int' and some of the
903: macros expand into loops.
904:
905: * Several passes use instruction attributes. A definition of the
906: attributes defined for a particular machine is in file
907: `insn-attr.h', which is generated from the machine description by
908: the program `genattr'. The file `insn-attrtab.c' contains
909: subroutines to obtain the attribute values for insns. It is
910: generated from the machine description by the program
911: `genattrtab'.
912:
913:
914: File: gcc.info, Node: RTL, Next: Machine Desc, Prev: Passes, Up: Top
915:
916: RTL Representation
917: ******************
918:
919: Most of the work of the compiler is done on an intermediate
920: representation called register transfer language. In this language,
921: the instructions to be output are described, pretty much one by one,
922: in an algebraic form that describes what the instruction does.
923:
924: RTL is inspired by Lisp lists. It has both an internal form, made
925: up of structures that point at other structures, and a textual form
926: that is used in the machine description and in printed debugging
927: dumps. The textual form uses nested parentheses to indicate the
928: pointers in the internal form.
929:
930: * Menu:
931:
932: * RTL Objects:: Expressions vs vectors vs strings vs integers.
933: * Accessors:: Macros to access expression operands or vector elts.
934: * Flags:: Other flags in an RTL expression.
935: * Machine Modes:: Describing the size and format of a datum.
936: * Constants:: Expressions with constant values.
937: * Regs and Memory:: Expressions representing register contents or memory.
938: * Arithmetic:: Expressions representing arithmetic on other expressions.
939: * Comparisons:: Expressions representing comparison of expressions.
940: * Bit Fields:: Expressions representing bit-fields in memory or reg.
941: * Conversions:: Extending, truncating, floating or fixing.
942: * RTL Declarations:: Declaring volatility, constancy, etc.
943: * Side Effects:: Expressions for storing in registers, etc.
944: * Incdec:: Embedded side-effects for autoincrement addressing.
945: * Assembler:: Representing `asm' with operands.
946: * Insns:: Expression types for entire insns.
947: * Calls:: RTL representation of function call insns.
948: * Sharing:: Some expressions are unique; others *must* be copied.
949:
950:
951: File: gcc.info, Node: RTL Objects, Next: Accessors, Prev: RTL, Up: RTL
952:
953: RTL Object Types
954: ================
955:
956: RTL uses four kinds of objects: expressions, integers, strings and
957: vectors. Expressions are the most important ones. An RTL expression
958: ("RTX", for short) is a C structure, but it is usually referred to
959: with a pointer; a type that is given the typedef name `rtx'.
960:
961: An integer is simply an `int'; their written form uses decimal
962: digits.
963:
964: A string is a sequence of characters. In core it is represented as
965: a `char *' in usual C fashion, and it is written in C syntax as well.
966: However, strings in RTL may never be null. If you write an empty
967: string in a machine description, it is represented in core as a null
968: pointer rather than as a pointer to a null character. In certain
969: contexts, these null pointers instead of strings are valid. Within
970: RTL code, strings are most commonly found inside `symbol_ref'
971: expressions, but they appear in other contexts in the RTL expressions
972: that make up machine descriptions.
973:
974: A vector contains an arbitrary number of pointers to expressions.
975: The number of elements in the vector is explicitly present in the
976: vector. The written form of a vector consists of square brackets
977: (`[...]') surrounding the elements, in sequence and with whitespace
978: separating them. Vectors of length zero are not created; null
979: pointers are used instead.
980:
981: Expressions are classified by "expression codes" (also called RTX
982: codes). The expression code is a name defined in `rtl.def', which is
983: also (in upper case) a C enumeration constant. The possible expression
984: codes and their meanings are machine-independent. The code of an RTX
985: can be extracted with the macro `GET_CODE (X)' and altered with
986: `PUT_CODE (X, NEWCODE)'.
987:
988: The expression code determines how many operands the expression
989: contains, and what kinds of objects they are. In RTL, unlike Lisp,
990: you cannot tell by looking at an operand what kind of object it is.
991: Instead, you must know from its context--from the expression code of
992: the containing expression. For example, in an expression of code
993: `subreg', the first operand is to be regarded as an expression and the
994: second operand as an integer. In an expression of code `plus', there
995: are two operands, both of which are to be regarded as expressions. In
996: a `symbol_ref' expression, there is one operand, which is to be
997: regarded as a string.
998:
999: Expressions are written as parentheses containing the name of the
1000: expression type, its flags and machine mode if any, and then the
1001: operands of the expression (separated by spaces).
1002:
1003: Expression code names in the `md' file are written in lower case,
1004: but when they appear in C code they are written in upper case. In this
1005: manual, they are shown as follows: `const_int'.
1006:
1007: In a few contexts a null pointer is valid where an expression is
1008: normally wanted. The written form of this is `(nil)'.
1009:
1010:
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.