1 This is doc/gccint.info, produced by makeinfo version 4.8 from
2 /home/jingyu/projects/gcc/android-toolchain/gcc-4.4.0/gcc/doc/gccint.texi.
4 Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
5 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 Free
6 Software Foundation, Inc.
8 Permission is granted to copy, distribute and/or modify this document
9 under the terms of the GNU Free Documentation License, Version 1.2 or
10 any later version published by the Free Software Foundation; with the
11 Invariant Sections being "Funding Free Software", the Front-Cover Texts
12 being (a) (see below), and with the Back-Cover Texts being (b) (see
13 below). A copy of the license is included in the section entitled "GNU
14 Free Documentation License".
16 (a) The FSF's Front-Cover Text is:
20 (b) The FSF's Back-Cover Text is:
22 You have freedom to copy and modify this GNU Manual, like GNU
23 software. Copies published by the Free Software Foundation raise
24 funds for GNU development.
26 INFO-DIR-SECTION Software development
28 * gccint: (gccint). Internals of the GNU Compiler Collection.
30 This file documents the internals of the GNU compilers.
32 Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
33 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 Free
34 Software Foundation, Inc.
36 Permission is granted to copy, distribute and/or modify this document
37 under the terms of the GNU Free Documentation License, Version 1.2 or
38 any later version published by the Free Software Foundation; with the
39 Invariant Sections being "Funding Free Software", the Front-Cover Texts
40 being (a) (see below), and with the Back-Cover Texts being (b) (see
41 below). A copy of the license is included in the section entitled "GNU
42 Free Documentation License".
44 (a) The FSF's Front-Cover Text is:
48 (b) The FSF's Back-Cover Text is:
50 You have freedom to copy and modify this GNU Manual, like GNU
51 software. Copies published by the Free Software Foundation raise
52 funds for GNU development.
56 File: gccint.info, Node: Top, Next: Contributing, Up: (DIR)
61 This manual documents the internals of the GNU compilers, including how
62 to port them to new targets and some information about how to write
63 front ends for new languages. It corresponds to the compilers
64 (GCC) version 4.4.0. The use of the GNU compilers is documented in a
65 separate manual. *Note Introduction: (gcc)Top.
67 This manual is mainly a reference manual rather than a tutorial. It
68 discusses how to contribute to GCC (*note Contributing::), the
69 characteristics of the machines supported by GCC as hosts and targets
70 (*note Portability::), how GCC relates to the ABIs on such systems
71 (*note Interface::), and the characteristics of the languages for which
72 GCC front ends are written (*note Languages::). It then describes the
73 GCC source tree structure and build system, some of the interfaces to
74 GCC front ends, and how support for a target system is implemented in
77 Additional tutorial information is linked to from
78 `http://gcc.gnu.org/readings.html'.
82 * Contributing:: How to contribute to testing and developing GCC.
83 * Portability:: Goals of GCC's portability features.
84 * Interface:: Function-call interface of GCC output.
85 * Libgcc:: Low-level runtime library used by GCC.
86 * Languages:: Languages for which GCC front ends are written.
87 * Source Tree:: GCC source tree structure and build system.
88 * Options:: Option specification files.
89 * Passes:: Order of passes, what they do, and what each file is for.
90 * Trees:: The source representation used by the C and C++ front ends.
91 * RTL:: The intermediate representation that most passes work on.
92 * GENERIC:: Language-independent representation generated by Front Ends
93 * GIMPLE:: Tuple representation used by Tree SSA optimizers
94 * Tree SSA:: Analysis and optimization of GIMPLE
95 * Control Flow:: Maintaining and manipulating the control flow graph.
96 * Loop Analysis and Representation:: Analysis and representation of loops
97 * Machine Desc:: How to write machine description instruction patterns.
98 * Target Macros:: How to write the machine description C macros and functions.
99 * Host Config:: Writing the `xm-MACHINE.h' file.
100 * Fragments:: Writing the `t-TARGET' and `x-HOST' files.
101 * Collect2:: How `collect2' works; how it finds `ld'.
102 * Header Dirs:: Understanding the standard header file directories.
103 * Type Information:: GCC's memory management; generating type information.
104 * Plugins:: Extending the compiler with plugins.
106 * Funding:: How to help assure funding for free software.
107 * GNU Project:: The GNU Project and GNU/Linux.
109 * Copying:: GNU General Public License says
110 how you can copy and share GCC.
111 * GNU Free Documentation License:: How you can copy and share this manual.
112 * Contributors:: People who have contributed to GCC.
114 * Option Index:: Index to command line options.
115 * Concept Index:: Index of concepts and symbol names.
118 File: gccint.info, Node: Contributing, Next: Portability, Prev: Top, Up: Top
120 1 Contributing to GCC Development
121 *********************************
123 If you would like to help pretest GCC releases to assure they work well,
124 current development sources are available by SVN (see
125 `http://gcc.gnu.org/svn.html'). Source and binary snapshots are also
126 available for FTP; see `http://gcc.gnu.org/snapshots.html'.
128 If you would like to work on improvements to GCC, please read the
129 advice at these URLs:
131 `http://gcc.gnu.org/contribute.html'
132 `http://gcc.gnu.org/contributewhy.html'
134 for information on how to make useful contributions and avoid
135 duplication of effort. Suggested projects are listed at
136 `http://gcc.gnu.org/projects/'.
139 File: gccint.info, Node: Portability, Next: Interface, Prev: Contributing, Up: Top
141 2 GCC and Portability
142 *********************
144 GCC itself aims to be portable to any machine where `int' is at least a
145 32-bit type. It aims to target machines with a flat (non-segmented)
146 byte addressed data address space (the code address space can be
147 separate). Target ABIs may have 8, 16, 32 or 64-bit `int' type. `char'
148 can be wider than 8 bits.
150 GCC gets most of the information about the target machine from a
151 machine description which gives an algebraic formula for each of the
152 machine's instructions. This is a very clean way to describe the
153 target. But when the compiler needs information that is difficult to
154 express in this fashion, ad-hoc parameters have been defined for
155 machine descriptions. The purpose of portability is to reduce the
156 total work needed on the compiler; it was not of interest for its own
159 GCC does not contain machine dependent code, but it does contain code
160 that depends on machine parameters such as endianness (whether the most
161 significant byte has the highest or lowest address of the bytes in a
162 word) and the availability of autoincrement addressing. In the
163 RTL-generation pass, it is often necessary to have multiple strategies
164 for generating code for a particular kind of syntax tree, strategies
165 that are usable for different combinations of parameters. Often, not
166 all possible cases have been addressed, but only the common ones or
167 only the ones that have been encountered. As a result, a new target
168 may require additional strategies. You will know if this happens
169 because the compiler will call `abort'. Fortunately, the new
170 strategies can be added in a machine-independent fashion, and will
171 affect only the target machines that need them.
174 File: gccint.info, Node: Interface, Next: Libgcc, Prev: Portability, Up: Top
176 3 Interfacing to GCC Output
177 ***************************
179 GCC is normally configured to use the same function calling convention
180 normally in use on the target system. This is done with the
181 machine-description macros described (*note Target Macros::).
183 However, returning of structure and union values is done differently on
184 some target machines. As a result, functions compiled with PCC
185 returning such types cannot be called from code compiled with GCC, and
186 vice versa. This does not cause trouble often because few Unix library
187 routines return structures or unions.
189 GCC code returns structures and unions that are 1, 2, 4 or 8 bytes
190 long in the same registers used for `int' or `double' return values.
191 (GCC typically allocates variables of such types in registers also.)
192 Structures and unions of other sizes are returned by storing them into
193 an address passed by the caller (usually in a register). The target
194 hook `TARGET_STRUCT_VALUE_RTX' tells GCC where to pass this address.
196 By contrast, PCC on most target machines returns structures and unions
197 of any size by copying the data into an area of static storage, and then
198 returning the address of that storage as if it were a pointer value.
199 The caller must copy the data from that memory area to the place where
200 the value is wanted. This is slower than the method used by GCC, and
201 fails to be reentrant.
203 On some target machines, such as RISC machines and the 80386, the
204 standard system convention is to pass to the subroutine the address of
205 where to return the value. On these machines, GCC has been configured
206 to be compatible with the standard compiler, when this method is used.
207 It may not be compatible for structures of 1, 2, 4 or 8 bytes.
209 GCC uses the system's standard convention for passing arguments. On
210 some machines, the first few arguments are passed in registers; in
211 others, all are passed on the stack. It would be possible to use
212 registers for argument passing on any machine, and this would probably
213 result in a significant speedup. But the result would be complete
214 incompatibility with code that follows the standard convention. So this
215 change is practical only if you are switching to GCC as the sole C
216 compiler for the system. We may implement register argument passing on
217 certain machines once we have a complete GNU system so that we can
218 compile the libraries with GCC.
220 On some machines (particularly the SPARC), certain types of arguments
221 are passed "by invisible reference". This means that the value is
222 stored in memory, and the address of the memory location is passed to
225 If you use `longjmp', beware of automatic variables. ISO C says that
226 automatic variables that are not declared `volatile' have undefined
227 values after a `longjmp'. And this is all GCC promises to do, because
228 it is very difficult to restore register variables correctly, and one
229 of GCC's features is that it can put variables in registers without
233 File: gccint.info, Node: Libgcc, Next: Languages, Prev: Interface, Up: Top
235 4 The GCC low-level runtime library
236 ***********************************
238 GCC provides a low-level runtime library, `libgcc.a' or `libgcc_s.so.1'
239 on some platforms. GCC generates calls to routines in this library
240 automatically, whenever it needs to perform some operation that is too
241 complicated to emit inline code for.
243 Most of the routines in `libgcc' handle arithmetic operations that the
244 target processor cannot perform directly. This includes integer
245 multiply and divide on some machines, and all floating-point and
246 fixed-point operations on other machines. `libgcc' also includes
247 routines for exception handling, and a handful of miscellaneous
250 Some of these routines can be defined in mostly machine-independent C.
251 Others must be hand-written in assembly language for each processor
254 GCC will also generate calls to C library routines, such as `memcpy'
255 and `memset', in some cases. The set of routines that GCC may possibly
256 use is documented in *Note Other Builtins: (gcc)Other Builtins.
258 These routines take arguments and return values of a specific machine
259 mode, not a specific C type. *Note Machine Modes::, for an explanation
260 of this concept. For illustrative purposes, in this chapter the
261 floating point type `float' is assumed to correspond to `SFmode';
262 `double' to `DFmode'; and `long double' to both `TFmode' and `XFmode'.
263 Similarly, the integer types `int' and `unsigned int' correspond to
264 `SImode'; `long' and `unsigned long' to `DImode'; and `long long' and
265 `unsigned long long' to `TImode'.
269 * Integer library routines::
270 * Soft float library routines::
271 * Decimal float library routines::
272 * Fixed-point fractional library routines::
273 * Exception handling routines::
274 * Miscellaneous routines::
277 File: gccint.info, Node: Integer library routines, Next: Soft float library routines, Up: Libgcc
279 4.1 Routines for integer arithmetic
280 ===================================
282 The integer arithmetic routines are used on platforms that don't provide
283 hardware support for arithmetic operations on some modes.
285 4.1.1 Arithmetic functions
286 --------------------------
288 -- Runtime Function: int __ashlsi3 (int A, int B)
289 -- Runtime Function: long __ashldi3 (long A, int B)
290 -- Runtime Function: long long __ashlti3 (long long A, int B)
291 These functions return the result of shifting A left by B bits.
293 -- Runtime Function: int __ashrsi3 (int A, int B)
294 -- Runtime Function: long __ashrdi3 (long A, int B)
295 -- Runtime Function: long long __ashrti3 (long long A, int B)
296 These functions return the result of arithmetically shifting A
299 -- Runtime Function: int __divsi3 (int A, int B)
300 -- Runtime Function: long __divdi3 (long A, long B)
301 -- Runtime Function: long long __divti3 (long long A, long long B)
302 These functions return the quotient of the signed division of A and
305 -- Runtime Function: int __lshrsi3 (int A, int B)
306 -- Runtime Function: long __lshrdi3 (long A, int B)
307 -- Runtime Function: long long __lshrti3 (long long A, int B)
308 These functions return the result of logically shifting A right by
311 -- Runtime Function: int __modsi3 (int A, int B)
312 -- Runtime Function: long __moddi3 (long A, long B)
313 -- Runtime Function: long long __modti3 (long long A, long long B)
314 These functions return the remainder of the signed division of A
317 -- Runtime Function: int __mulsi3 (int A, int B)
318 -- Runtime Function: long __muldi3 (long A, long B)
319 -- Runtime Function: long long __multi3 (long long A, long long B)
320 These functions return the product of A and B.
322 -- Runtime Function: long __negdi2 (long A)
323 -- Runtime Function: long long __negti2 (long long A)
324 These functions return the negation of A.
326 -- Runtime Function: unsigned int __udivsi3 (unsigned int A, unsigned
328 -- Runtime Function: unsigned long __udivdi3 (unsigned long A,
330 -- Runtime Function: unsigned long long __udivti3 (unsigned long long
331 A, unsigned long long B)
332 These functions return the quotient of the unsigned division of A
335 -- Runtime Function: unsigned long __udivmoddi3 (unsigned long A,
336 unsigned long B, unsigned long *C)
337 -- Runtime Function: unsigned long long __udivti3 (unsigned long long
338 A, unsigned long long B, unsigned long long *C)
339 These functions calculate both the quotient and remainder of the
340 unsigned division of A and B. The return value is the quotient,
341 and the remainder is placed in variable pointed to by C.
343 -- Runtime Function: unsigned int __umodsi3 (unsigned int A, unsigned
345 -- Runtime Function: unsigned long __umoddi3 (unsigned long A,
347 -- Runtime Function: unsigned long long __umodti3 (unsigned long long
348 A, unsigned long long B)
349 These functions return the remainder of the unsigned division of A
352 4.1.2 Comparison functions
353 --------------------------
355 The following functions implement integral comparisons. These functions
356 implement a low-level compare, upon which the higher level comparison
357 operators (such as less than and greater than or equal to) can be
358 constructed. The returned values lie in the range zero to two, to allow
359 the high-level operators to be implemented by testing the returned
360 result using either signed or unsigned comparison.
362 -- Runtime Function: int __cmpdi2 (long A, long B)
363 -- Runtime Function: int __cmpti2 (long long A, long long B)
364 These functions perform a signed comparison of A and B. If A is
365 less than B, they return 0; if A is greater than B, they return 2;
366 and if A and B are equal they return 1.
368 -- Runtime Function: int __ucmpdi2 (unsigned long A, unsigned long B)
369 -- Runtime Function: int __ucmpti2 (unsigned long long A, unsigned
371 These functions perform an unsigned comparison of A and B. If A
372 is less than B, they return 0; if A is greater than B, they return
373 2; and if A and B are equal they return 1.
375 4.1.3 Trapping arithmetic functions
376 -----------------------------------
378 The following functions implement trapping arithmetic. These functions
379 call the libc function `abort' upon signed arithmetic overflow.
381 -- Runtime Function: int __absvsi2 (int A)
382 -- Runtime Function: long __absvdi2 (long A)
383 These functions return the absolute value of A.
385 -- Runtime Function: int __addvsi3 (int A, int B)
386 -- Runtime Function: long __addvdi3 (long A, long B)
387 These functions return the sum of A and B; that is `A + B'.
389 -- Runtime Function: int __mulvsi3 (int A, int B)
390 -- Runtime Function: long __mulvdi3 (long A, long B)
391 The functions return the product of A and B; that is `A * B'.
393 -- Runtime Function: int __negvsi2 (int A)
394 -- Runtime Function: long __negvdi2 (long A)
395 These functions return the negation of A; that is `-A'.
397 -- Runtime Function: int __subvsi3 (int A, int B)
398 -- Runtime Function: long __subvdi3 (long A, long B)
399 These functions return the difference between B and A; that is `A
405 -- Runtime Function: int __clzsi2 (int A)
406 -- Runtime Function: int __clzdi2 (long A)
407 -- Runtime Function: int __clzti2 (long long A)
408 These functions return the number of leading 0-bits in A, starting
409 at the most significant bit position. If A is zero, the result is
412 -- Runtime Function: int __ctzsi2 (int A)
413 -- Runtime Function: int __ctzdi2 (long A)
414 -- Runtime Function: int __ctzti2 (long long A)
415 These functions return the number of trailing 0-bits in A, starting
416 at the least significant bit position. If A is zero, the result is
419 -- Runtime Function: int __ffsdi2 (long A)
420 -- Runtime Function: int __ffsti2 (long long A)
421 These functions return the index of the least significant 1-bit in
422 A, or the value zero if A is zero. The least significant bit is
425 -- Runtime Function: int __paritysi2 (int A)
426 -- Runtime Function: int __paritydi2 (long A)
427 -- Runtime Function: int __parityti2 (long long A)
428 These functions return the value zero if the number of bits set in
429 A is even, and the value one otherwise.
431 -- Runtime Function: int __popcountsi2 (int A)
432 -- Runtime Function: int __popcountdi2 (long A)
433 -- Runtime Function: int __popcountti2 (long long A)
434 These functions return the number of bits set in A.
436 -- Runtime Function: int32_t __bswapsi2 (int32_t A)
437 -- Runtime Function: int64_t __bswapdi2 (int64_t A)
438 These functions return the A byteswapped.
441 File: gccint.info, Node: Soft float library routines, Next: Decimal float library routines, Prev: Integer library routines, Up: Libgcc
443 4.2 Routines for floating point emulation
444 =========================================
446 The software floating point library is used on machines which do not
447 have hardware support for floating point. It is also used whenever
448 `-msoft-float' is used to disable generation of floating point
449 instructions. (Not all targets support this switch.)
451 For compatibility with other compilers, the floating point emulation
452 routines can be renamed with the `DECLARE_LIBRARY_RENAMES' macro (*note
453 Library Calls::). In this section, the default names are used.
455 Presently the library does not support `XFmode', which is used for
456 `long double' on some architectures.
458 4.2.1 Arithmetic functions
459 --------------------------
461 -- Runtime Function: float __addsf3 (float A, float B)
462 -- Runtime Function: double __adddf3 (double A, double B)
463 -- Runtime Function: long double __addtf3 (long double A, long double
465 -- Runtime Function: long double __addxf3 (long double A, long double
467 These functions return the sum of A and B.
469 -- Runtime Function: float __subsf3 (float A, float B)
470 -- Runtime Function: double __subdf3 (double A, double B)
471 -- Runtime Function: long double __subtf3 (long double A, long double
473 -- Runtime Function: long double __subxf3 (long double A, long double
475 These functions return the difference between B and A; that is,
478 -- Runtime Function: float __mulsf3 (float A, float B)
479 -- Runtime Function: double __muldf3 (double A, double B)
480 -- Runtime Function: long double __multf3 (long double A, long double
482 -- Runtime Function: long double __mulxf3 (long double A, long double
484 These functions return the product of A and B.
486 -- Runtime Function: float __divsf3 (float A, float B)
487 -- Runtime Function: double __divdf3 (double A, double B)
488 -- Runtime Function: long double __divtf3 (long double A, long double
490 -- Runtime Function: long double __divxf3 (long double A, long double
492 These functions return the quotient of A and B; that is, A / B.
494 -- Runtime Function: float __negsf2 (float A)
495 -- Runtime Function: double __negdf2 (double A)
496 -- Runtime Function: long double __negtf2 (long double A)
497 -- Runtime Function: long double __negxf2 (long double A)
498 These functions return the negation of A. They simply flip the
499 sign bit, so they can produce negative zero and negative NaN.
501 4.2.2 Conversion functions
502 --------------------------
504 -- Runtime Function: double __extendsfdf2 (float A)
505 -- Runtime Function: long double __extendsftf2 (float A)
506 -- Runtime Function: long double __extendsfxf2 (float A)
507 -- Runtime Function: long double __extenddftf2 (double A)
508 -- Runtime Function: long double __extenddfxf2 (double A)
509 These functions extend A to the wider mode of their return type.
511 -- Runtime Function: double __truncxfdf2 (long double A)
512 -- Runtime Function: double __trunctfdf2 (long double A)
513 -- Runtime Function: float __truncxfsf2 (long double A)
514 -- Runtime Function: float __trunctfsf2 (long double A)
515 -- Runtime Function: float __truncdfsf2 (double A)
516 These functions truncate A to the narrower mode of their return
517 type, rounding toward zero.
519 -- Runtime Function: int __fixsfsi (float A)
520 -- Runtime Function: int __fixdfsi (double A)
521 -- Runtime Function: int __fixtfsi (long double A)
522 -- Runtime Function: int __fixxfsi (long double A)
523 These functions convert A to a signed integer, rounding toward
526 -- Runtime Function: long __fixsfdi (float A)
527 -- Runtime Function: long __fixdfdi (double A)
528 -- Runtime Function: long __fixtfdi (long double A)
529 -- Runtime Function: long __fixxfdi (long double A)
530 These functions convert A to a signed long, rounding toward zero.
532 -- Runtime Function: long long __fixsfti (float A)
533 -- Runtime Function: long long __fixdfti (double A)
534 -- Runtime Function: long long __fixtfti (long double A)
535 -- Runtime Function: long long __fixxfti (long double A)
536 These functions convert A to a signed long long, rounding toward
539 -- Runtime Function: unsigned int __fixunssfsi (float A)
540 -- Runtime Function: unsigned int __fixunsdfsi (double A)
541 -- Runtime Function: unsigned int __fixunstfsi (long double A)
542 -- Runtime Function: unsigned int __fixunsxfsi (long double A)
543 These functions convert A to an unsigned integer, rounding toward
544 zero. Negative values all become zero.
546 -- Runtime Function: unsigned long __fixunssfdi (float A)
547 -- Runtime Function: unsigned long __fixunsdfdi (double A)
548 -- Runtime Function: unsigned long __fixunstfdi (long double A)
549 -- Runtime Function: unsigned long __fixunsxfdi (long double A)
550 These functions convert A to an unsigned long, rounding toward
551 zero. Negative values all become zero.
553 -- Runtime Function: unsigned long long __fixunssfti (float A)
554 -- Runtime Function: unsigned long long __fixunsdfti (double A)
555 -- Runtime Function: unsigned long long __fixunstfti (long double A)
556 -- Runtime Function: unsigned long long __fixunsxfti (long double A)
557 These functions convert A to an unsigned long long, rounding
558 toward zero. Negative values all become zero.
560 -- Runtime Function: float __floatsisf (int I)
561 -- Runtime Function: double __floatsidf (int I)
562 -- Runtime Function: long double __floatsitf (int I)
563 -- Runtime Function: long double __floatsixf (int I)
564 These functions convert I, a signed integer, to floating point.
566 -- Runtime Function: float __floatdisf (long I)
567 -- Runtime Function: double __floatdidf (long I)
568 -- Runtime Function: long double __floatditf (long I)
569 -- Runtime Function: long double __floatdixf (long I)
570 These functions convert I, a signed long, to floating point.
572 -- Runtime Function: float __floattisf (long long I)
573 -- Runtime Function: double __floattidf (long long I)
574 -- Runtime Function: long double __floattitf (long long I)
575 -- Runtime Function: long double __floattixf (long long I)
576 These functions convert I, a signed long long, to floating point.
578 -- Runtime Function: float __floatunsisf (unsigned int I)
579 -- Runtime Function: double __floatunsidf (unsigned int I)
580 -- Runtime Function: long double __floatunsitf (unsigned int I)
581 -- Runtime Function: long double __floatunsixf (unsigned int I)
582 These functions convert I, an unsigned integer, to floating point.
584 -- Runtime Function: float __floatundisf (unsigned long I)
585 -- Runtime Function: double __floatundidf (unsigned long I)
586 -- Runtime Function: long double __floatunditf (unsigned long I)
587 -- Runtime Function: long double __floatundixf (unsigned long I)
588 These functions convert I, an unsigned long, to floating point.
590 -- Runtime Function: float __floatuntisf (unsigned long long I)
591 -- Runtime Function: double __floatuntidf (unsigned long long I)
592 -- Runtime Function: long double __floatuntitf (unsigned long long I)
593 -- Runtime Function: long double __floatuntixf (unsigned long long I)
594 These functions convert I, an unsigned long long, to floating
597 4.2.3 Comparison functions
598 --------------------------
600 There are two sets of basic comparison functions.
602 -- Runtime Function: int __cmpsf2 (float A, float B)
603 -- Runtime Function: int __cmpdf2 (double A, double B)
604 -- Runtime Function: int __cmptf2 (long double A, long double B)
605 These functions calculate a <=> b. That is, if A is less than B,
606 they return -1; if A is greater than B, they return 1; and if A
607 and B are equal they return 0. If either argument is NaN they
608 return 1, but you should not rely on this; if NaN is a
609 possibility, use one of the higher-level comparison functions.
611 -- Runtime Function: int __unordsf2 (float A, float B)
612 -- Runtime Function: int __unorddf2 (double A, double B)
613 -- Runtime Function: int __unordtf2 (long double A, long double B)
614 These functions return a nonzero value if either argument is NaN,
617 There is also a complete group of higher level functions which
618 correspond directly to comparison operators. They implement the ISO C
619 semantics for floating-point comparisons, taking NaN into account. Pay
620 careful attention to the return values defined for each set. Under the
621 hood, all of these routines are implemented as
623 if (__unordXf2 (a, b))
625 return __cmpXf2 (a, b);
627 where E is a constant chosen to give the proper behavior for NaN.
628 Thus, the meaning of the return value is different for each set. Do
629 not rely on this implementation; only the semantics documented below
632 -- Runtime Function: int __eqsf2 (float A, float B)
633 -- Runtime Function: int __eqdf2 (double A, double B)
634 -- Runtime Function: int __eqtf2 (long double A, long double B)
635 These functions return zero if neither argument is NaN, and A and
638 -- Runtime Function: int __nesf2 (float A, float B)
639 -- Runtime Function: int __nedf2 (double A, double B)
640 -- Runtime Function: int __netf2 (long double A, long double B)
641 These functions return a nonzero value if either argument is NaN,
642 or if A and B are unequal.
644 -- Runtime Function: int __gesf2 (float A, float B)
645 -- Runtime Function: int __gedf2 (double A, double B)
646 -- Runtime Function: int __getf2 (long double A, long double B)
647 These functions return a value greater than or equal to zero if
648 neither argument is NaN, and A is greater than or equal to B.
650 -- Runtime Function: int __ltsf2 (float A, float B)
651 -- Runtime Function: int __ltdf2 (double A, double B)
652 -- Runtime Function: int __lttf2 (long double A, long double B)
653 These functions return a value less than zero if neither argument
654 is NaN, and A is strictly less than B.
656 -- Runtime Function: int __lesf2 (float A, float B)
657 -- Runtime Function: int __ledf2 (double A, double B)
658 -- Runtime Function: int __letf2 (long double A, long double B)
659 These functions return a value less than or equal to zero if
660 neither argument is NaN, and A is less than or equal to B.
662 -- Runtime Function: int __gtsf2 (float A, float B)
663 -- Runtime Function: int __gtdf2 (double A, double B)
664 -- Runtime Function: int __gttf2 (long double A, long double B)
665 These functions return a value greater than zero if neither
666 argument is NaN, and A is strictly greater than B.
668 4.2.4 Other floating-point functions
669 ------------------------------------
671 -- Runtime Function: float __powisf2 (float A, int B)
672 -- Runtime Function: double __powidf2 (double A, int B)
673 -- Runtime Function: long double __powitf2 (long double A, int B)
674 -- Runtime Function: long double __powixf2 (long double A, int B)
675 These functions convert raise A to the power B.
677 -- Runtime Function: complex float __mulsc3 (float A, float B, float
679 -- Runtime Function: complex double __muldc3 (double A, double B,
681 -- Runtime Function: complex long double __multc3 (long double A, long
682 double B, long double C, long double D)
683 -- Runtime Function: complex long double __mulxc3 (long double A, long
684 double B, long double C, long double D)
685 These functions return the product of A + iB and C + iD, following
686 the rules of C99 Annex G.
688 -- Runtime Function: complex float __divsc3 (float A, float B, float
690 -- Runtime Function: complex double __divdc3 (double A, double B,
692 -- Runtime Function: complex long double __divtc3 (long double A, long
693 double B, long double C, long double D)
694 -- Runtime Function: complex long double __divxc3 (long double A, long
695 double B, long double C, long double D)
696 These functions return the quotient of A + iB and C + iD (i.e., (A
697 + iB) / (C + iD)), following the rules of C99 Annex G.
700 File: gccint.info, Node: Decimal float library routines, Next: Fixed-point fractional library routines, Prev: Soft float library routines, Up: Libgcc
702 4.3 Routines for decimal floating point emulation
703 =================================================
705 The software decimal floating point library implements IEEE 754-2008
706 decimal floating point arithmetic and is only activated on selected
709 The software decimal floating point library supports either DPD
710 (Densely Packed Decimal) or BID (Binary Integer Decimal) encoding as
711 selected at configure time.
713 4.3.1 Arithmetic functions
714 --------------------------
716 -- Runtime Function: _Decimal32 __dpd_addsd3 (_Decimal32 A, _Decimal32
718 -- Runtime Function: _Decimal32 __bid_addsd3 (_Decimal32 A, _Decimal32
720 -- Runtime Function: _Decimal64 __dpd_adddd3 (_Decimal64 A, _Decimal64
722 -- Runtime Function: _Decimal64 __bid_adddd3 (_Decimal64 A, _Decimal64
724 -- Runtime Function: _Decimal128 __dpd_addtd3 (_Decimal128 A,
726 -- Runtime Function: _Decimal128 __bid_addtd3 (_Decimal128 A,
728 These functions return the sum of A and B.
730 -- Runtime Function: _Decimal32 __dpd_subsd3 (_Decimal32 A, _Decimal32
732 -- Runtime Function: _Decimal32 __bid_subsd3 (_Decimal32 A, _Decimal32
734 -- Runtime Function: _Decimal64 __dpd_subdd3 (_Decimal64 A, _Decimal64
736 -- Runtime Function: _Decimal64 __bid_subdd3 (_Decimal64 A, _Decimal64
738 -- Runtime Function: _Decimal128 __dpd_subtd3 (_Decimal128 A,
740 -- Runtime Function: _Decimal128 __bid_subtd3 (_Decimal128 A,
742 These functions return the difference between B and A; that is,
745 -- Runtime Function: _Decimal32 __dpd_mulsd3 (_Decimal32 A, _Decimal32
747 -- Runtime Function: _Decimal32 __bid_mulsd3 (_Decimal32 A, _Decimal32
749 -- Runtime Function: _Decimal64 __dpd_muldd3 (_Decimal64 A, _Decimal64
751 -- Runtime Function: _Decimal64 __bid_muldd3 (_Decimal64 A, _Decimal64
753 -- Runtime Function: _Decimal128 __dpd_multd3 (_Decimal128 A,
755 -- Runtime Function: _Decimal128 __bid_multd3 (_Decimal128 A,
757 These functions return the product of A and B.
759 -- Runtime Function: _Decimal32 __dpd_divsd3 (_Decimal32 A, _Decimal32
761 -- Runtime Function: _Decimal32 __bid_divsd3 (_Decimal32 A, _Decimal32
763 -- Runtime Function: _Decimal64 __dpd_divdd3 (_Decimal64 A, _Decimal64
765 -- Runtime Function: _Decimal64 __bid_divdd3 (_Decimal64 A, _Decimal64
767 -- Runtime Function: _Decimal128 __dpd_divtd3 (_Decimal128 A,
769 -- Runtime Function: _Decimal128 __bid_divtd3 (_Decimal128 A,
771 These functions return the quotient of A and B; that is, A / B.
773 -- Runtime Function: _Decimal32 __dpd_negsd2 (_Decimal32 A)
774 -- Runtime Function: _Decimal32 __bid_negsd2 (_Decimal32 A)
775 -- Runtime Function: _Decimal64 __dpd_negdd2 (_Decimal64 A)
776 -- Runtime Function: _Decimal64 __bid_negdd2 (_Decimal64 A)
777 -- Runtime Function: _Decimal128 __dpd_negtd2 (_Decimal128 A)
778 -- Runtime Function: _Decimal128 __bid_negtd2 (_Decimal128 A)
779 These functions return the negation of A. They simply flip the
780 sign bit, so they can produce negative zero and negative NaN.
782 4.3.2 Conversion functions
783 --------------------------
785 -- Runtime Function: _Decimal64 __dpd_extendsddd2 (_Decimal32 A)
786 -- Runtime Function: _Decimal64 __bid_extendsddd2 (_Decimal32 A)
787 -- Runtime Function: _Decimal128 __dpd_extendsdtd2 (_Decimal32 A)
788 -- Runtime Function: _Decimal128 __bid_extendsdtd2 (_Decimal32 A)
789 -- Runtime Function: _Decimal128 __dpd_extendddtd2 (_Decimal64 A)
790 -- Runtime Function: _Decimal128 __bid_extendddtd2 (_Decimal64 A)
791 -- Runtime Function: _Decimal32 __dpd_truncddsd2 (_Decimal64 A)
792 -- Runtime Function: _Decimal32 __bid_truncddsd2 (_Decimal64 A)
793 -- Runtime Function: _Decimal32 __dpd_trunctdsd2 (_Decimal128 A)
794 -- Runtime Function: _Decimal32 __bid_trunctdsd2 (_Decimal128 A)
795 -- Runtime Function: _Decimal64 __dpd_trunctddd2 (_Decimal128 A)
796 -- Runtime Function: _Decimal64 __bid_trunctddd2 (_Decimal128 A)
797 These functions convert the value A from one decimal floating type
800 -- Runtime Function: _Decimal64 __dpd_extendsfdd (float A)
801 -- Runtime Function: _Decimal64 __bid_extendsfdd (float A)
802 -- Runtime Function: _Decimal128 __dpd_extendsftd (float A)
803 -- Runtime Function: _Decimal128 __bid_extendsftd (float A)
804 -- Runtime Function: _Decimal128 __dpd_extenddftd (double A)
805 -- Runtime Function: _Decimal128 __bid_extenddftd (double A)
806 -- Runtime Function: _Decimal128 __dpd_extendxftd (long double A)
807 -- Runtime Function: _Decimal128 __bid_extendxftd (long double A)
808 -- Runtime Function: _Decimal32 __dpd_truncdfsd (double A)
809 -- Runtime Function: _Decimal32 __bid_truncdfsd (double A)
810 -- Runtime Function: _Decimal32 __dpd_truncxfsd (long double A)
811 -- Runtime Function: _Decimal32 __bid_truncxfsd (long double A)
812 -- Runtime Function: _Decimal32 __dpd_trunctfsd (long double A)
813 -- Runtime Function: _Decimal32 __bid_trunctfsd (long double A)
814 -- Runtime Function: _Decimal64 __dpd_truncxfdd (long double A)
815 -- Runtime Function: _Decimal64 __bid_truncxfdd (long double A)
816 -- Runtime Function: _Decimal64 __dpd_trunctfdd (long double A)
817 -- Runtime Function: _Decimal64 __bid_trunctfdd (long double A)
818 These functions convert the value of A from a binary floating type
819 to a decimal floating type of a different size.
821 -- Runtime Function: float __dpd_truncddsf (_Decimal64 A)
822 -- Runtime Function: float __bid_truncddsf (_Decimal64 A)
823 -- Runtime Function: float __dpd_trunctdsf (_Decimal128 A)
824 -- Runtime Function: float __bid_trunctdsf (_Decimal128 A)
825 -- Runtime Function: double __dpd_extendsddf (_Decimal32 A)
826 -- Runtime Function: double __bid_extendsddf (_Decimal32 A)
827 -- Runtime Function: double __dpd_trunctddf (_Decimal128 A)
828 -- Runtime Function: double __bid_trunctddf (_Decimal128 A)
829 -- Runtime Function: long double __dpd_extendsdxf (_Decimal32 A)
830 -- Runtime Function: long double __bid_extendsdxf (_Decimal32 A)
831 -- Runtime Function: long double __dpd_extendddxf (_Decimal64 A)
832 -- Runtime Function: long double __bid_extendddxf (_Decimal64 A)
833 -- Runtime Function: long double __dpd_trunctdxf (_Decimal128 A)
834 -- Runtime Function: long double __bid_trunctdxf (_Decimal128 A)
835 -- Runtime Function: long double __dpd_extendsdtf (_Decimal32 A)
836 -- Runtime Function: long double __bid_extendsdtf (_Decimal32 A)
837 -- Runtime Function: long double __dpd_extendddtf (_Decimal64 A)
838 -- Runtime Function: long double __bid_extendddtf (_Decimal64 A)
839 These functions convert the value of A from a decimal floating type
840 to a binary floating type of a different size.
842 -- Runtime Function: _Decimal32 __dpd_extendsfsd (float A)
843 -- Runtime Function: _Decimal32 __bid_extendsfsd (float A)
844 -- Runtime Function: _Decimal64 __dpd_extenddfdd (double A)
845 -- Runtime Function: _Decimal64 __bid_extenddfdd (double A)
846 -- Runtime Function: _Decimal128 __dpd_extendtftd (long double A)
847 -- Runtime Function: _Decimal128 __bid_extendtftd (long double A)
848 -- Runtime Function: float __dpd_truncsdsf (_Decimal32 A)
849 -- Runtime Function: float __bid_truncsdsf (_Decimal32 A)
850 -- Runtime Function: double __dpd_truncdddf (_Decimal64 A)
851 -- Runtime Function: double __bid_truncdddf (_Decimal64 A)
852 -- Runtime Function: long double __dpd_trunctdtf (_Decimal128 A)
853 -- Runtime Function: long double __bid_trunctdtf (_Decimal128 A)
854 These functions convert the value of A between decimal and binary
855 floating types of the same size.
857 -- Runtime Function: int __dpd_fixsdsi (_Decimal32 A)
858 -- Runtime Function: int __bid_fixsdsi (_Decimal32 A)
859 -- Runtime Function: int __dpd_fixddsi (_Decimal64 A)
860 -- Runtime Function: int __bid_fixddsi (_Decimal64 A)
861 -- Runtime Function: int __dpd_fixtdsi (_Decimal128 A)
862 -- Runtime Function: int __bid_fixtdsi (_Decimal128 A)
863 These functions convert A to a signed integer.
865 -- Runtime Function: long __dpd_fixsddi (_Decimal32 A)
866 -- Runtime Function: long __bid_fixsddi (_Decimal32 A)
867 -- Runtime Function: long __dpd_fixdddi (_Decimal64 A)
868 -- Runtime Function: long __bid_fixdddi (_Decimal64 A)
869 -- Runtime Function: long __dpd_fixtddi (_Decimal128 A)
870 -- Runtime Function: long __bid_fixtddi (_Decimal128 A)
871 These functions convert A to a signed long.
873 -- Runtime Function: unsigned int __dpd_fixunssdsi (_Decimal32 A)
874 -- Runtime Function: unsigned int __bid_fixunssdsi (_Decimal32 A)
875 -- Runtime Function: unsigned int __dpd_fixunsddsi (_Decimal64 A)
876 -- Runtime Function: unsigned int __bid_fixunsddsi (_Decimal64 A)
877 -- Runtime Function: unsigned int __dpd_fixunstdsi (_Decimal128 A)
878 -- Runtime Function: unsigned int __bid_fixunstdsi (_Decimal128 A)
879 These functions convert A to an unsigned integer. Negative values
882 -- Runtime Function: unsigned long __dpd_fixunssddi (_Decimal32 A)
883 -- Runtime Function: unsigned long __bid_fixunssddi (_Decimal32 A)
884 -- Runtime Function: unsigned long __dpd_fixunsdddi (_Decimal64 A)
885 -- Runtime Function: unsigned long __bid_fixunsdddi (_Decimal64 A)
886 -- Runtime Function: unsigned long __dpd_fixunstddi (_Decimal128 A)
887 -- Runtime Function: unsigned long __bid_fixunstddi (_Decimal128 A)
888 These functions convert A to an unsigned long. Negative values
891 -- Runtime Function: _Decimal32 __dpd_floatsisd (int I)
892 -- Runtime Function: _Decimal32 __bid_floatsisd (int I)
893 -- Runtime Function: _Decimal64 __dpd_floatsidd (int I)
894 -- Runtime Function: _Decimal64 __bid_floatsidd (int I)
895 -- Runtime Function: _Decimal128 __dpd_floatsitd (int I)
896 -- Runtime Function: _Decimal128 __bid_floatsitd (int I)
897 These functions convert I, a signed integer, to decimal floating
900 -- Runtime Function: _Decimal32 __dpd_floatdisd (long I)
901 -- Runtime Function: _Decimal32 __bid_floatdisd (long I)
902 -- Runtime Function: _Decimal64 __dpd_floatdidd (long I)
903 -- Runtime Function: _Decimal64 __bid_floatdidd (long I)
904 -- Runtime Function: _Decimal128 __dpd_floatditd (long I)
905 -- Runtime Function: _Decimal128 __bid_floatditd (long I)
906 These functions convert I, a signed long, to decimal floating
909 -- Runtime Function: _Decimal32 __dpd_floatunssisd (unsigned int I)
910 -- Runtime Function: _Decimal32 __bid_floatunssisd (unsigned int I)
911 -- Runtime Function: _Decimal64 __dpd_floatunssidd (unsigned int I)
912 -- Runtime Function: _Decimal64 __bid_floatunssidd (unsigned int I)
913 -- Runtime Function: _Decimal128 __dpd_floatunssitd (unsigned int I)
914 -- Runtime Function: _Decimal128 __bid_floatunssitd (unsigned int I)
915 These functions convert I, an unsigned integer, to decimal
918 -- Runtime Function: _Decimal32 __dpd_floatunsdisd (unsigned long I)
919 -- Runtime Function: _Decimal32 __bid_floatunsdisd (unsigned long I)
920 -- Runtime Function: _Decimal64 __dpd_floatunsdidd (unsigned long I)
921 -- Runtime Function: _Decimal64 __bid_floatunsdidd (unsigned long I)
922 -- Runtime Function: _Decimal128 __dpd_floatunsditd (unsigned long I)
923 -- Runtime Function: _Decimal128 __bid_floatunsditd (unsigned long I)
924 These functions convert I, an unsigned long, to decimal floating
927 4.3.3 Comparison functions
928 --------------------------
930 -- Runtime Function: int __dpd_unordsd2 (_Decimal32 A, _Decimal32 B)
931 -- Runtime Function: int __bid_unordsd2 (_Decimal32 A, _Decimal32 B)
932 -- Runtime Function: int __dpd_unorddd2 (_Decimal64 A, _Decimal64 B)
933 -- Runtime Function: int __bid_unorddd2 (_Decimal64 A, _Decimal64 B)
934 -- Runtime Function: int __dpd_unordtd2 (_Decimal128 A, _Decimal128 B)
935 -- Runtime Function: int __bid_unordtd2 (_Decimal128 A, _Decimal128 B)
936 These functions return a nonzero value if either argument is NaN,
939 There is also a complete group of higher level functions which
940 correspond directly to comparison operators. They implement the ISO C
941 semantics for floating-point comparisons, taking NaN into account. Pay
942 careful attention to the return values defined for each set. Under the
943 hood, all of these routines are implemented as
945 if (__bid_unordXd2 (a, b))
947 return __bid_cmpXd2 (a, b);
949 where E is a constant chosen to give the proper behavior for NaN.
950 Thus, the meaning of the return value is different for each set. Do
951 not rely on this implementation; only the semantics documented below
954 -- Runtime Function: int __dpd_eqsd2 (_Decimal32 A, _Decimal32 B)
955 -- Runtime Function: int __bid_eqsd2 (_Decimal32 A, _Decimal32 B)
956 -- Runtime Function: int __dpd_eqdd2 (_Decimal64 A, _Decimal64 B)
957 -- Runtime Function: int __bid_eqdd2 (_Decimal64 A, _Decimal64 B)
958 -- Runtime Function: int __dpd_eqtd2 (_Decimal128 A, _Decimal128 B)
959 -- Runtime Function: int __bid_eqtd2 (_Decimal128 A, _Decimal128 B)
960 These functions return zero if neither argument is NaN, and A and
963 -- Runtime Function: int __dpd_nesd2 (_Decimal32 A, _Decimal32 B)
964 -- Runtime Function: int __bid_nesd2 (_Decimal32 A, _Decimal32 B)
965 -- Runtime Function: int __dpd_nedd2 (_Decimal64 A, _Decimal64 B)
966 -- Runtime Function: int __bid_nedd2 (_Decimal64 A, _Decimal64 B)
967 -- Runtime Function: int __dpd_netd2 (_Decimal128 A, _Decimal128 B)
968 -- Runtime Function: int __bid_netd2 (_Decimal128 A, _Decimal128 B)
969 These functions return a nonzero value if either argument is NaN,
970 or if A and B are unequal.
972 -- Runtime Function: int __dpd_gesd2 (_Decimal32 A, _Decimal32 B)
973 -- Runtime Function: int __bid_gesd2 (_Decimal32 A, _Decimal32 B)
974 -- Runtime Function: int __dpd_gedd2 (_Decimal64 A, _Decimal64 B)
975 -- Runtime Function: int __bid_gedd2 (_Decimal64 A, _Decimal64 B)
976 -- Runtime Function: int __dpd_getd2 (_Decimal128 A, _Decimal128 B)
977 -- Runtime Function: int __bid_getd2 (_Decimal128 A, _Decimal128 B)
978 These functions return a value greater than or equal to zero if
979 neither argument is NaN, and A is greater than or equal to B.
981 -- Runtime Function: int __dpd_ltsd2 (_Decimal32 A, _Decimal32 B)
982 -- Runtime Function: int __bid_ltsd2 (_Decimal32 A, _Decimal32 B)
983 -- Runtime Function: int __dpd_ltdd2 (_Decimal64 A, _Decimal64 B)
984 -- Runtime Function: int __bid_ltdd2 (_Decimal64 A, _Decimal64 B)
985 -- Runtime Function: int __dpd_lttd2 (_Decimal128 A, _Decimal128 B)
986 -- Runtime Function: int __bid_lttd2 (_Decimal128 A, _Decimal128 B)
987 These functions return a value less than zero if neither argument
988 is NaN, and A is strictly less than B.
990 -- Runtime Function: int __dpd_lesd2 (_Decimal32 A, _Decimal32 B)
991 -- Runtime Function: int __bid_lesd2 (_Decimal32 A, _Decimal32 B)
992 -- Runtime Function: int __dpd_ledd2 (_Decimal64 A, _Decimal64 B)
993 -- Runtime Function: int __bid_ledd2 (_Decimal64 A, _Decimal64 B)
994 -- Runtime Function: int __dpd_letd2 (_Decimal128 A, _Decimal128 B)
995 -- Runtime Function: int __bid_letd2 (_Decimal128 A, _Decimal128 B)
996 These functions return a value less than or equal to zero if
997 neither argument is NaN, and A is less than or equal to B.
999 -- Runtime Function: int __dpd_gtsd2 (_Decimal32 A, _Decimal32 B)
1000 -- Runtime Function: int __bid_gtsd2 (_Decimal32 A, _Decimal32 B)
1001 -- Runtime Function: int __dpd_gtdd2 (_Decimal64 A, _Decimal64 B)
1002 -- Runtime Function: int __bid_gtdd2 (_Decimal64 A, _Decimal64 B)
1003 -- Runtime Function: int __dpd_gttd2 (_Decimal128 A, _Decimal128 B)
1004 -- Runtime Function: int __bid_gttd2 (_Decimal128 A, _Decimal128 B)
1005 These functions return a value greater than zero if neither
1006 argument is NaN, and A is strictly greater than B.
1009 File: gccint.info, Node: Fixed-point fractional library routines, Next: Exception handling routines, Prev: Decimal float library routines, Up: Libgcc
1011 4.4 Routines for fixed-point fractional emulation
1012 =================================================
1014 The software fixed-point library implements fixed-point fractional
1015 arithmetic, and is only activated on selected targets.
1017 For ease of comprehension `fract' is an alias for the `_Fract' type,
1018 `accum' an alias for `_Accum', and `sat' an alias for `_Sat'.
1020 For illustrative purposes, in this section the fixed-point fractional
1021 type `short fract' is assumed to correspond to machine mode `QQmode';
1022 `unsigned short fract' to `UQQmode'; `fract' to `HQmode';
1023 `unsigned fract' to `UHQmode'; `long fract' to `SQmode';
1024 `unsigned long fract' to `USQmode'; `long long fract' to `DQmode'; and
1025 `unsigned long long fract' to `UDQmode'. Similarly the fixed-point
1026 accumulator type `short accum' corresponds to `HAmode';
1027 `unsigned short accum' to `UHAmode'; `accum' to `SAmode';
1028 `unsigned accum' to `USAmode'; `long accum' to `DAmode';
1029 `unsigned long accum' to `UDAmode'; `long long accum' to `TAmode'; and
1030 `unsigned long long accum' to `UTAmode'.
1032 4.4.1 Arithmetic functions
1033 --------------------------
1035 -- Runtime Function: short fract __addqq3 (short fract A, short fract
1037 -- Runtime Function: fract __addhq3 (fract A, fract B)
1038 -- Runtime Function: long fract __addsq3 (long fract A, long fract B)
1039 -- Runtime Function: long long fract __adddq3 (long long fract A, long
1041 -- Runtime Function: unsigned short fract __adduqq3 (unsigned short
1042 fract A, unsigned short fract B)
1043 -- Runtime Function: unsigned fract __adduhq3 (unsigned fract A,
1045 -- Runtime Function: unsigned long fract __addusq3 (unsigned long
1046 fract A, unsigned long fract B)
1047 -- Runtime Function: unsigned long long fract __addudq3 (unsigned long
1048 long fract A, unsigned long long fract B)
1049 -- Runtime Function: short accum __addha3 (short accum A, short accum
1051 -- Runtime Function: accum __addsa3 (accum A, accum B)
1052 -- Runtime Function: long accum __addda3 (long accum A, long accum B)
1053 -- Runtime Function: long long accum __addta3 (long long accum A, long
1055 -- Runtime Function: unsigned short accum __adduha3 (unsigned short
1056 accum A, unsigned short accum B)
1057 -- Runtime Function: unsigned accum __addusa3 (unsigned accum A,
1059 -- Runtime Function: unsigned long accum __adduda3 (unsigned long
1060 accum A, unsigned long accum B)
1061 -- Runtime Function: unsigned long long accum __adduta3 (unsigned long
1062 long accum A, unsigned long long accum B)
1063 These functions return the sum of A and B.
1065 -- Runtime Function: short fract __ssaddqq3 (short fract A, short
1067 -- Runtime Function: fract __ssaddhq3 (fract A, fract B)
1068 -- Runtime Function: long fract __ssaddsq3 (long fract A, long fract B)
1069 -- Runtime Function: long long fract __ssadddq3 (long long fract A,
1071 -- Runtime Function: short accum __ssaddha3 (short accum A, short
1073 -- Runtime Function: accum __ssaddsa3 (accum A, accum B)
1074 -- Runtime Function: long accum __ssaddda3 (long accum A, long accum B)
1075 -- Runtime Function: long long accum __ssaddta3 (long long accum A,
1077 These functions return the sum of A and B with signed saturation.
1079 -- Runtime Function: unsigned short fract __usadduqq3 (unsigned short
1080 fract A, unsigned short fract B)
1081 -- Runtime Function: unsigned fract __usadduhq3 (unsigned fract A,
1083 -- Runtime Function: unsigned long fract __usaddusq3 (unsigned long
1084 fract A, unsigned long fract B)
1085 -- Runtime Function: unsigned long long fract __usaddudq3 (unsigned
1086 long long fract A, unsigned long long fract B)
1087 -- Runtime Function: unsigned short accum __usadduha3 (unsigned short
1088 accum A, unsigned short accum B)
1089 -- Runtime Function: unsigned accum __usaddusa3 (unsigned accum A,
1091 -- Runtime Function: unsigned long accum __usadduda3 (unsigned long
1092 accum A, unsigned long accum B)
1093 -- Runtime Function: unsigned long long accum __usadduta3 (unsigned
1094 long long accum A, unsigned long long accum B)
1095 These functions return the sum of A and B with unsigned saturation.
1097 -- Runtime Function: short fract __subqq3 (short fract A, short fract
1099 -- Runtime Function: fract __subhq3 (fract A, fract B)
1100 -- Runtime Function: long fract __subsq3 (long fract A, long fract B)
1101 -- Runtime Function: long long fract __subdq3 (long long fract A, long
1103 -- Runtime Function: unsigned short fract __subuqq3 (unsigned short
1104 fract A, unsigned short fract B)
1105 -- Runtime Function: unsigned fract __subuhq3 (unsigned fract A,
1107 -- Runtime Function: unsigned long fract __subusq3 (unsigned long
1108 fract A, unsigned long fract B)
1109 -- Runtime Function: unsigned long long fract __subudq3 (unsigned long
1110 long fract A, unsigned long long fract B)
1111 -- Runtime Function: short accum __subha3 (short accum A, short accum
1113 -- Runtime Function: accum __subsa3 (accum A, accum B)
1114 -- Runtime Function: long accum __subda3 (long accum A, long accum B)
1115 -- Runtime Function: long long accum __subta3 (long long accum A, long
1117 -- Runtime Function: unsigned short accum __subuha3 (unsigned short
1118 accum A, unsigned short accum B)
1119 -- Runtime Function: unsigned accum __subusa3 (unsigned accum A,
1121 -- Runtime Function: unsigned long accum __subuda3 (unsigned long
1122 accum A, unsigned long accum B)
1123 -- Runtime Function: unsigned long long accum __subuta3 (unsigned long
1124 long accum A, unsigned long long accum B)
1125 These functions return the difference of A and B; that is, `A - B'.
1127 -- Runtime Function: short fract __sssubqq3 (short fract A, short
1129 -- Runtime Function: fract __sssubhq3 (fract A, fract B)
1130 -- Runtime Function: long fract __sssubsq3 (long fract A, long fract B)
1131 -- Runtime Function: long long fract __sssubdq3 (long long fract A,
1133 -- Runtime Function: short accum __sssubha3 (short accum A, short
1135 -- Runtime Function: accum __sssubsa3 (accum A, accum B)
1136 -- Runtime Function: long accum __sssubda3 (long accum A, long accum B)
1137 -- Runtime Function: long long accum __sssubta3 (long long accum A,
1139 These functions return the difference of A and B with signed
1140 saturation; that is, `A - B'.
1142 -- Runtime Function: unsigned short fract __ussubuqq3 (unsigned short
1143 fract A, unsigned short fract B)
1144 -- Runtime Function: unsigned fract __ussubuhq3 (unsigned fract A,
1146 -- Runtime Function: unsigned long fract __ussubusq3 (unsigned long
1147 fract A, unsigned long fract B)
1148 -- Runtime Function: unsigned long long fract __ussubudq3 (unsigned
1149 long long fract A, unsigned long long fract B)
1150 -- Runtime Function: unsigned short accum __ussubuha3 (unsigned short
1151 accum A, unsigned short accum B)
1152 -- Runtime Function: unsigned accum __ussubusa3 (unsigned accum A,
1154 -- Runtime Function: unsigned long accum __ussubuda3 (unsigned long
1155 accum A, unsigned long accum B)
1156 -- Runtime Function: unsigned long long accum __ussubuta3 (unsigned
1157 long long accum A, unsigned long long accum B)
1158 These functions return the difference of A and B with unsigned
1159 saturation; that is, `A - B'.
1161 -- Runtime Function: short fract __mulqq3 (short fract A, short fract
1163 -- Runtime Function: fract __mulhq3 (fract A, fract B)
1164 -- Runtime Function: long fract __mulsq3 (long fract A, long fract B)
1165 -- Runtime Function: long long fract __muldq3 (long long fract A, long
1167 -- Runtime Function: unsigned short fract __muluqq3 (unsigned short
1168 fract A, unsigned short fract B)
1169 -- Runtime Function: unsigned fract __muluhq3 (unsigned fract A,
1171 -- Runtime Function: unsigned long fract __mulusq3 (unsigned long
1172 fract A, unsigned long fract B)
1173 -- Runtime Function: unsigned long long fract __muludq3 (unsigned long
1174 long fract A, unsigned long long fract B)
1175 -- Runtime Function: short accum __mulha3 (short accum A, short accum
1177 -- Runtime Function: accum __mulsa3 (accum A, accum B)
1178 -- Runtime Function: long accum __mulda3 (long accum A, long accum B)
1179 -- Runtime Function: long long accum __multa3 (long long accum A, long
1181 -- Runtime Function: unsigned short accum __muluha3 (unsigned short
1182 accum A, unsigned short accum B)
1183 -- Runtime Function: unsigned accum __mulusa3 (unsigned accum A,
1185 -- Runtime Function: unsigned long accum __muluda3 (unsigned long
1186 accum A, unsigned long accum B)
1187 -- Runtime Function: unsigned long long accum __muluta3 (unsigned long
1188 long accum A, unsigned long long accum B)
1189 These functions return the product of A and B.
1191 -- Runtime Function: short fract __ssmulqq3 (short fract A, short
1193 -- Runtime Function: fract __ssmulhq3 (fract A, fract B)
1194 -- Runtime Function: long fract __ssmulsq3 (long fract A, long fract B)
1195 -- Runtime Function: long long fract __ssmuldq3 (long long fract A,
1197 -- Runtime Function: short accum __ssmulha3 (short accum A, short
1199 -- Runtime Function: accum __ssmulsa3 (accum A, accum B)
1200 -- Runtime Function: long accum __ssmulda3 (long accum A, long accum B)
1201 -- Runtime Function: long long accum __ssmulta3 (long long accum A,
1203 These functions return the product of A and B with signed
1206 -- Runtime Function: unsigned short fract __usmuluqq3 (unsigned short
1207 fract A, unsigned short fract B)
1208 -- Runtime Function: unsigned fract __usmuluhq3 (unsigned fract A,
1210 -- Runtime Function: unsigned long fract __usmulusq3 (unsigned long
1211 fract A, unsigned long fract B)
1212 -- Runtime Function: unsigned long long fract __usmuludq3 (unsigned
1213 long long fract A, unsigned long long fract B)
1214 -- Runtime Function: unsigned short accum __usmuluha3 (unsigned short
1215 accum A, unsigned short accum B)
1216 -- Runtime Function: unsigned accum __usmulusa3 (unsigned accum A,
1218 -- Runtime Function: unsigned long accum __usmuluda3 (unsigned long
1219 accum A, unsigned long accum B)
1220 -- Runtime Function: unsigned long long accum __usmuluta3 (unsigned
1221 long long accum A, unsigned long long accum B)
1222 These functions return the product of A and B with unsigned
1225 -- Runtime Function: short fract __divqq3 (short fract A, short fract
1227 -- Runtime Function: fract __divhq3 (fract A, fract B)
1228 -- Runtime Function: long fract __divsq3 (long fract A, long fract B)
1229 -- Runtime Function: long long fract __divdq3 (long long fract A, long
1231 -- Runtime Function: short accum __divha3 (short accum A, short accum
1233 -- Runtime Function: accum __divsa3 (accum A, accum B)
1234 -- Runtime Function: long accum __divda3 (long accum A, long accum B)
1235 -- Runtime Function: long long accum __divta3 (long long accum A, long
1237 These functions return the quotient of the signed division of A
1240 -- Runtime Function: unsigned short fract __udivuqq3 (unsigned short
1241 fract A, unsigned short fract B)
1242 -- Runtime Function: unsigned fract __udivuhq3 (unsigned fract A,
1244 -- Runtime Function: unsigned long fract __udivusq3 (unsigned long
1245 fract A, unsigned long fract B)
1246 -- Runtime Function: unsigned long long fract __udivudq3 (unsigned
1247 long long fract A, unsigned long long fract B)
1248 -- Runtime Function: unsigned short accum __udivuha3 (unsigned short
1249 accum A, unsigned short accum B)
1250 -- Runtime Function: unsigned accum __udivusa3 (unsigned accum A,
1252 -- Runtime Function: unsigned long accum __udivuda3 (unsigned long
1253 accum A, unsigned long accum B)
1254 -- Runtime Function: unsigned long long accum __udivuta3 (unsigned
1255 long long accum A, unsigned long long accum B)
1256 These functions return the quotient of the unsigned division of A
1259 -- Runtime Function: short fract __ssdivqq3 (short fract A, short
1261 -- Runtime Function: fract __ssdivhq3 (fract A, fract B)
1262 -- Runtime Function: long fract __ssdivsq3 (long fract A, long fract B)
1263 -- Runtime Function: long long fract __ssdivdq3 (long long fract A,
1265 -- Runtime Function: short accum __ssdivha3 (short accum A, short
1267 -- Runtime Function: accum __ssdivsa3 (accum A, accum B)
1268 -- Runtime Function: long accum __ssdivda3 (long accum A, long accum B)
1269 -- Runtime Function: long long accum __ssdivta3 (long long accum A,
1271 These functions return the quotient of the signed division of A
1272 and B with signed saturation.
1274 -- Runtime Function: unsigned short fract __usdivuqq3 (unsigned short
1275 fract A, unsigned short fract B)
1276 -- Runtime Function: unsigned fract __usdivuhq3 (unsigned fract A,
1278 -- Runtime Function: unsigned long fract __usdivusq3 (unsigned long
1279 fract A, unsigned long fract B)
1280 -- Runtime Function: unsigned long long fract __usdivudq3 (unsigned
1281 long long fract A, unsigned long long fract B)
1282 -- Runtime Function: unsigned short accum __usdivuha3 (unsigned short
1283 accum A, unsigned short accum B)
1284 -- Runtime Function: unsigned accum __usdivusa3 (unsigned accum A,
1286 -- Runtime Function: unsigned long accum __usdivuda3 (unsigned long
1287 accum A, unsigned long accum B)
1288 -- Runtime Function: unsigned long long accum __usdivuta3 (unsigned
1289 long long accum A, unsigned long long accum B)
1290 These functions return the quotient of the unsigned division of A
1291 and B with unsigned saturation.
1293 -- Runtime Function: short fract __negqq2 (short fract A)
1294 -- Runtime Function: fract __neghq2 (fract A)
1295 -- Runtime Function: long fract __negsq2 (long fract A)
1296 -- Runtime Function: long long fract __negdq2 (long long fract A)
1297 -- Runtime Function: unsigned short fract __neguqq2 (unsigned short
1299 -- Runtime Function: unsigned fract __neguhq2 (unsigned fract A)
1300 -- Runtime Function: unsigned long fract __negusq2 (unsigned long
1302 -- Runtime Function: unsigned long long fract __negudq2 (unsigned long
1304 -- Runtime Function: short accum __negha2 (short accum A)
1305 -- Runtime Function: accum __negsa2 (accum A)
1306 -- Runtime Function: long accum __negda2 (long accum A)
1307 -- Runtime Function: long long accum __negta2 (long long accum A)
1308 -- Runtime Function: unsigned short accum __neguha2 (unsigned short
1310 -- Runtime Function: unsigned accum __negusa2 (unsigned accum A)
1311 -- Runtime Function: unsigned long accum __neguda2 (unsigned long
1313 -- Runtime Function: unsigned long long accum __neguta2 (unsigned long
1315 These functions return the negation of A.
1317 -- Runtime Function: short fract __ssnegqq2 (short fract A)
1318 -- Runtime Function: fract __ssneghq2 (fract A)
1319 -- Runtime Function: long fract __ssnegsq2 (long fract A)
1320 -- Runtime Function: long long fract __ssnegdq2 (long long fract A)
1321 -- Runtime Function: short accum __ssnegha2 (short accum A)
1322 -- Runtime Function: accum __ssnegsa2 (accum A)
1323 -- Runtime Function: long accum __ssnegda2 (long accum A)
1324 -- Runtime Function: long long accum __ssnegta2 (long long accum A)
1325 These functions return the negation of A with signed saturation.
1327 -- Runtime Function: unsigned short fract __usneguqq2 (unsigned short
1329 -- Runtime Function: unsigned fract __usneguhq2 (unsigned fract A)
1330 -- Runtime Function: unsigned long fract __usnegusq2 (unsigned long
1332 -- Runtime Function: unsigned long long fract __usnegudq2 (unsigned
1334 -- Runtime Function: unsigned short accum __usneguha2 (unsigned short
1336 -- Runtime Function: unsigned accum __usnegusa2 (unsigned accum A)
1337 -- Runtime Function: unsigned long accum __usneguda2 (unsigned long
1339 -- Runtime Function: unsigned long long accum __usneguta2 (unsigned
1341 These functions return the negation of A with unsigned saturation.
1343 -- Runtime Function: short fract __ashlqq3 (short fract A, int B)
1344 -- Runtime Function: fract __ashlhq3 (fract A, int B)
1345 -- Runtime Function: long fract __ashlsq3 (long fract A, int B)
1346 -- Runtime Function: long long fract __ashldq3 (long long fract A, int
1348 -- Runtime Function: unsigned short fract __ashluqq3 (unsigned short
1350 -- Runtime Function: unsigned fract __ashluhq3 (unsigned fract A, int
1352 -- Runtime Function: unsigned long fract __ashlusq3 (unsigned long
1354 -- Runtime Function: unsigned long long fract __ashludq3 (unsigned
1355 long long fract A, int B)
1356 -- Runtime Function: short accum __ashlha3 (short accum A, int B)
1357 -- Runtime Function: accum __ashlsa3 (accum A, int B)
1358 -- Runtime Function: long accum __ashlda3 (long accum A, int B)
1359 -- Runtime Function: long long accum __ashlta3 (long long accum A, int
1361 -- Runtime Function: unsigned short accum __ashluha3 (unsigned short
1363 -- Runtime Function: unsigned accum __ashlusa3 (unsigned accum A, int
1365 -- Runtime Function: unsigned long accum __ashluda3 (unsigned long
1367 -- Runtime Function: unsigned long long accum __ashluta3 (unsigned
1368 long long accum A, int B)
1369 These functions return the result of shifting A left by B bits.
1371 -- Runtime Function: short fract __ashrqq3 (short fract A, int B)
1372 -- Runtime Function: fract __ashrhq3 (fract A, int B)
1373 -- Runtime Function: long fract __ashrsq3 (long fract A, int B)
1374 -- Runtime Function: long long fract __ashrdq3 (long long fract A, int
1376 -- Runtime Function: short accum __ashrha3 (short accum A, int B)
1377 -- Runtime Function: accum __ashrsa3 (accum A, int B)
1378 -- Runtime Function: long accum __ashrda3 (long accum A, int B)
1379 -- Runtime Function: long long accum __ashrta3 (long long accum A, int
1381 These functions return the result of arithmetically shifting A
1384 -- Runtime Function: unsigned short fract __lshruqq3 (unsigned short
1386 -- Runtime Function: unsigned fract __lshruhq3 (unsigned fract A, int
1388 -- Runtime Function: unsigned long fract __lshrusq3 (unsigned long
1390 -- Runtime Function: unsigned long long fract __lshrudq3 (unsigned
1391 long long fract A, int B)
1392 -- Runtime Function: unsigned short accum __lshruha3 (unsigned short
1394 -- Runtime Function: unsigned accum __lshrusa3 (unsigned accum A, int
1396 -- Runtime Function: unsigned long accum __lshruda3 (unsigned long
1398 -- Runtime Function: unsigned long long accum __lshruta3 (unsigned
1399 long long accum A, int B)
1400 These functions return the result of logically shifting A right by
1403 -- Runtime Function: fract __ssashlhq3 (fract A, int B)
1404 -- Runtime Function: long fract __ssashlsq3 (long fract A, int B)
1405 -- Runtime Function: long long fract __ssashldq3 (long long fract A,
1407 -- Runtime Function: short accum __ssashlha3 (short accum A, int B)
1408 -- Runtime Function: accum __ssashlsa3 (accum A, int B)
1409 -- Runtime Function: long accum __ssashlda3 (long accum A, int B)
1410 -- Runtime Function: long long accum __ssashlta3 (long long accum A,
1412 These functions return the result of shifting A left by B bits
1413 with signed saturation.
1415 -- Runtime Function: unsigned short fract __usashluqq3 (unsigned short
1417 -- Runtime Function: unsigned fract __usashluhq3 (unsigned fract A,
1419 -- Runtime Function: unsigned long fract __usashlusq3 (unsigned long
1421 -- Runtime Function: unsigned long long fract __usashludq3 (unsigned
1422 long long fract A, int B)
1423 -- Runtime Function: unsigned short accum __usashluha3 (unsigned short
1425 -- Runtime Function: unsigned accum __usashlusa3 (unsigned accum A,
1427 -- Runtime Function: unsigned long accum __usashluda3 (unsigned long
1429 -- Runtime Function: unsigned long long accum __usashluta3 (unsigned
1430 long long accum A, int B)
1431 These functions return the result of shifting A left by B bits
1432 with unsigned saturation.
1434 4.4.2 Comparison functions
1435 --------------------------
1437 The following functions implement fixed-point comparisons. These
1438 functions implement a low-level compare, upon which the higher level
1439 comparison operators (such as less than and greater than or equal to)
1440 can be constructed. The returned values lie in the range zero to two,
1441 to allow the high-level operators to be implemented by testing the
1442 returned result using either signed or unsigned comparison.
1444 -- Runtime Function: int __cmpqq2 (short fract A, short fract B)
1445 -- Runtime Function: int __cmphq2 (fract A, fract B)
1446 -- Runtime Function: int __cmpsq2 (long fract A, long fract B)
1447 -- Runtime Function: int __cmpdq2 (long long fract A, long long fract
1449 -- Runtime Function: int __cmpuqq2 (unsigned short fract A, unsigned
1451 -- Runtime Function: int __cmpuhq2 (unsigned fract A, unsigned fract B)
1452 -- Runtime Function: int __cmpusq2 (unsigned long fract A, unsigned
1454 -- Runtime Function: int __cmpudq2 (unsigned long long fract A,
1455 unsigned long long fract B)
1456 -- Runtime Function: int __cmpha2 (short accum A, short accum B)
1457 -- Runtime Function: int __cmpsa2 (accum A, accum B)
1458 -- Runtime Function: int __cmpda2 (long accum A, long accum B)
1459 -- Runtime Function: int __cmpta2 (long long accum A, long long accum
1461 -- Runtime Function: int __cmpuha2 (unsigned short accum A, unsigned
1463 -- Runtime Function: int __cmpusa2 (unsigned accum A, unsigned accum B)
1464 -- Runtime Function: int __cmpuda2 (unsigned long accum A, unsigned
1466 -- Runtime Function: int __cmputa2 (unsigned long long accum A,
1467 unsigned long long accum B)
1468 These functions perform a signed or unsigned comparison of A and B
1469 (depending on the selected machine mode). If A is less than B,
1470 they return 0; if A is greater than B, they return 2; and if A and
1471 B are equal they return 1.
1473 4.4.3 Conversion functions
1474 --------------------------
1476 -- Runtime Function: fract __fractqqhq2 (short fract A)
1477 -- Runtime Function: long fract __fractqqsq2 (short fract A)
1478 -- Runtime Function: long long fract __fractqqdq2 (short fract A)
1479 -- Runtime Function: short accum __fractqqha (short fract A)
1480 -- Runtime Function: accum __fractqqsa (short fract A)
1481 -- Runtime Function: long accum __fractqqda (short fract A)
1482 -- Runtime Function: long long accum __fractqqta (short fract A)
1483 -- Runtime Function: unsigned short fract __fractqquqq (short fract A)
1484 -- Runtime Function: unsigned fract __fractqquhq (short fract A)
1485 -- Runtime Function: unsigned long fract __fractqqusq (short fract A)
1486 -- Runtime Function: unsigned long long fract __fractqqudq (short
1488 -- Runtime Function: unsigned short accum __fractqquha (short fract A)
1489 -- Runtime Function: unsigned accum __fractqqusa (short fract A)
1490 -- Runtime Function: unsigned long accum __fractqquda (short fract A)
1491 -- Runtime Function: unsigned long long accum __fractqquta (short
1493 -- Runtime Function: signed char __fractqqqi (short fract A)
1494 -- Runtime Function: short __fractqqhi (short fract A)
1495 -- Runtime Function: int __fractqqsi (short fract A)
1496 -- Runtime Function: long __fractqqdi (short fract A)
1497 -- Runtime Function: long long __fractqqti (short fract A)
1498 -- Runtime Function: float __fractqqsf (short fract A)
1499 -- Runtime Function: double __fractqqdf (short fract A)
1500 -- Runtime Function: short fract __fracthqqq2 (fract A)
1501 -- Runtime Function: long fract __fracthqsq2 (fract A)
1502 -- Runtime Function: long long fract __fracthqdq2 (fract A)
1503 -- Runtime Function: short accum __fracthqha (fract A)
1504 -- Runtime Function: accum __fracthqsa (fract A)
1505 -- Runtime Function: long accum __fracthqda (fract A)
1506 -- Runtime Function: long long accum __fracthqta (fract A)
1507 -- Runtime Function: unsigned short fract __fracthquqq (fract A)
1508 -- Runtime Function: unsigned fract __fracthquhq (fract A)
1509 -- Runtime Function: unsigned long fract __fracthqusq (fract A)
1510 -- Runtime Function: unsigned long long fract __fracthqudq (fract A)
1511 -- Runtime Function: unsigned short accum __fracthquha (fract A)
1512 -- Runtime Function: unsigned accum __fracthqusa (fract A)
1513 -- Runtime Function: unsigned long accum __fracthquda (fract A)
1514 -- Runtime Function: unsigned long long accum __fracthquta (fract A)
1515 -- Runtime Function: signed char __fracthqqi (fract A)
1516 -- Runtime Function: short __fracthqhi (fract A)
1517 -- Runtime Function: int __fracthqsi (fract A)
1518 -- Runtime Function: long __fracthqdi (fract A)
1519 -- Runtime Function: long long __fracthqti (fract A)
1520 -- Runtime Function: float __fracthqsf (fract A)
1521 -- Runtime Function: double __fracthqdf (fract A)
1522 -- Runtime Function: short fract __fractsqqq2 (long fract A)
1523 -- Runtime Function: fract __fractsqhq2 (long fract A)
1524 -- Runtime Function: long long fract __fractsqdq2 (long fract A)
1525 -- Runtime Function: short accum __fractsqha (long fract A)
1526 -- Runtime Function: accum __fractsqsa (long fract A)
1527 -- Runtime Function: long accum __fractsqda (long fract A)
1528 -- Runtime Function: long long accum __fractsqta (long fract A)
1529 -- Runtime Function: unsigned short fract __fractsquqq (long fract A)
1530 -- Runtime Function: unsigned fract __fractsquhq (long fract A)
1531 -- Runtime Function: unsigned long fract __fractsqusq (long fract A)
1532 -- Runtime Function: unsigned long long fract __fractsqudq (long fract
1534 -- Runtime Function: unsigned short accum __fractsquha (long fract A)
1535 -- Runtime Function: unsigned accum __fractsqusa (long fract A)
1536 -- Runtime Function: unsigned long accum __fractsquda (long fract A)
1537 -- Runtime Function: unsigned long long accum __fractsquta (long fract
1539 -- Runtime Function: signed char __fractsqqi (long fract A)
1540 -- Runtime Function: short __fractsqhi (long fract A)
1541 -- Runtime Function: int __fractsqsi (long fract A)
1542 -- Runtime Function: long __fractsqdi (long fract A)
1543 -- Runtime Function: long long __fractsqti (long fract A)
1544 -- Runtime Function: float __fractsqsf (long fract A)
1545 -- Runtime Function: double __fractsqdf (long fract A)
1546 -- Runtime Function: short fract __fractdqqq2 (long long fract A)
1547 -- Runtime Function: fract __fractdqhq2 (long long fract A)
1548 -- Runtime Function: long fract __fractdqsq2 (long long fract A)
1549 -- Runtime Function: short accum __fractdqha (long long fract A)
1550 -- Runtime Function: accum __fractdqsa (long long fract A)
1551 -- Runtime Function: long accum __fractdqda (long long fract A)
1552 -- Runtime Function: long long accum __fractdqta (long long fract A)
1553 -- Runtime Function: unsigned short fract __fractdquqq (long long
1555 -- Runtime Function: unsigned fract __fractdquhq (long long fract A)
1556 -- Runtime Function: unsigned long fract __fractdqusq (long long fract
1558 -- Runtime Function: unsigned long long fract __fractdqudq (long long
1560 -- Runtime Function: unsigned short accum __fractdquha (long long
1562 -- Runtime Function: unsigned accum __fractdqusa (long long fract A)
1563 -- Runtime Function: unsigned long accum __fractdquda (long long fract
1565 -- Runtime Function: unsigned long long accum __fractdquta (long long
1567 -- Runtime Function: signed char __fractdqqi (long long fract A)
1568 -- Runtime Function: short __fractdqhi (long long fract A)
1569 -- Runtime Function: int __fractdqsi (long long fract A)
1570 -- Runtime Function: long __fractdqdi (long long fract A)
1571 -- Runtime Function: long long __fractdqti (long long fract A)
1572 -- Runtime Function: float __fractdqsf (long long fract A)
1573 -- Runtime Function: double __fractdqdf (long long fract A)
1574 -- Runtime Function: short fract __fracthaqq (short accum A)
1575 -- Runtime Function: fract __fracthahq (short accum A)
1576 -- Runtime Function: long fract __fracthasq (short accum A)
1577 -- Runtime Function: long long fract __fracthadq (short accum A)
1578 -- Runtime Function: accum __fracthasa2 (short accum A)
1579 -- Runtime Function: long accum __fracthada2 (short accum A)
1580 -- Runtime Function: long long accum __fracthata2 (short accum A)
1581 -- Runtime Function: unsigned short fract __fracthauqq (short accum A)
1582 -- Runtime Function: unsigned fract __fracthauhq (short accum A)
1583 -- Runtime Function: unsigned long fract __fracthausq (short accum A)
1584 -- Runtime Function: unsigned long long fract __fracthaudq (short
1586 -- Runtime Function: unsigned short accum __fracthauha (short accum A)
1587 -- Runtime Function: unsigned accum __fracthausa (short accum A)
1588 -- Runtime Function: unsigned long accum __fracthauda (short accum A)
1589 -- Runtime Function: unsigned long long accum __fracthauta (short
1591 -- Runtime Function: signed char __fracthaqi (short accum A)
1592 -- Runtime Function: short __fracthahi (short accum A)
1593 -- Runtime Function: int __fracthasi (short accum A)
1594 -- Runtime Function: long __fracthadi (short accum A)
1595 -- Runtime Function: long long __fracthati (short accum A)
1596 -- Runtime Function: float __fracthasf (short accum A)
1597 -- Runtime Function: double __fracthadf (short accum A)
1598 -- Runtime Function: short fract __fractsaqq (accum A)
1599 -- Runtime Function: fract __fractsahq (accum A)
1600 -- Runtime Function: long fract __fractsasq (accum A)
1601 -- Runtime Function: long long fract __fractsadq (accum A)
1602 -- Runtime Function: short accum __fractsaha2 (accum A)
1603 -- Runtime Function: long accum __fractsada2 (accum A)
1604 -- Runtime Function: long long accum __fractsata2 (accum A)
1605 -- Runtime Function: unsigned short fract __fractsauqq (accum A)
1606 -- Runtime Function: unsigned fract __fractsauhq (accum A)
1607 -- Runtime Function: unsigned long fract __fractsausq (accum A)
1608 -- Runtime Function: unsigned long long fract __fractsaudq (accum A)
1609 -- Runtime Function: unsigned short accum __fractsauha (accum A)
1610 -- Runtime Function: unsigned accum __fractsausa (accum A)
1611 -- Runtime Function: unsigned long accum __fractsauda (accum A)
1612 -- Runtime Function: unsigned long long accum __fractsauta (accum A)
1613 -- Runtime Function: signed char __fractsaqi (accum A)
1614 -- Runtime Function: short __fractsahi (accum A)
1615 -- Runtime Function: int __fractsasi (accum A)
1616 -- Runtime Function: long __fractsadi (accum A)
1617 -- Runtime Function: long long __fractsati (accum A)
1618 -- Runtime Function: float __fractsasf (accum A)
1619 -- Runtime Function: double __fractsadf (accum A)
1620 -- Runtime Function: short fract __fractdaqq (long accum A)
1621 -- Runtime Function: fract __fractdahq (long accum A)
1622 -- Runtime Function: long fract __fractdasq (long accum A)
1623 -- Runtime Function: long long fract __fractdadq (long accum A)
1624 -- Runtime Function: short accum __fractdaha2 (long accum A)
1625 -- Runtime Function: accum __fractdasa2 (long accum A)
1626 -- Runtime Function: long long accum __fractdata2 (long accum A)
1627 -- Runtime Function: unsigned short fract __fractdauqq (long accum A)
1628 -- Runtime Function: unsigned fract __fractdauhq (long accum A)
1629 -- Runtime Function: unsigned long fract __fractdausq (long accum A)
1630 -- Runtime Function: unsigned long long fract __fractdaudq (long accum
1632 -- Runtime Function: unsigned short accum __fractdauha (long accum A)
1633 -- Runtime Function: unsigned accum __fractdausa (long accum A)
1634 -- Runtime Function: unsigned long accum __fractdauda (long accum A)
1635 -- Runtime Function: unsigned long long accum __fractdauta (long accum
1637 -- Runtime Function: signed char __fractdaqi (long accum A)
1638 -- Runtime Function: short __fractdahi (long accum A)
1639 -- Runtime Function: int __fractdasi (long accum A)
1640 -- Runtime Function: long __fractdadi (long accum A)
1641 -- Runtime Function: long long __fractdati (long accum A)
1642 -- Runtime Function: float __fractdasf (long accum A)
1643 -- Runtime Function: double __fractdadf (long accum A)
1644 -- Runtime Function: short fract __fracttaqq (long long accum A)
1645 -- Runtime Function: fract __fracttahq (long long accum A)
1646 -- Runtime Function: long fract __fracttasq (long long accum A)
1647 -- Runtime Function: long long fract __fracttadq (long long accum A)
1648 -- Runtime Function: short accum __fracttaha2 (long long accum A)
1649 -- Runtime Function: accum __fracttasa2 (long long accum A)
1650 -- Runtime Function: long accum __fracttada2 (long long accum A)
1651 -- Runtime Function: unsigned short fract __fracttauqq (long long
1653 -- Runtime Function: unsigned fract __fracttauhq (long long accum A)
1654 -- Runtime Function: unsigned long fract __fracttausq (long long accum
1656 -- Runtime Function: unsigned long long fract __fracttaudq (long long
1658 -- Runtime Function: unsigned short accum __fracttauha (long long
1660 -- Runtime Function: unsigned accum __fracttausa (long long accum A)
1661 -- Runtime Function: unsigned long accum __fracttauda (long long accum
1663 -- Runtime Function: unsigned long long accum __fracttauta (long long
1665 -- Runtime Function: signed char __fracttaqi (long long accum A)
1666 -- Runtime Function: short __fracttahi (long long accum A)
1667 -- Runtime Function: int __fracttasi (long long accum A)
1668 -- Runtime Function: long __fracttadi (long long accum A)
1669 -- Runtime Function: long long __fracttati (long long accum A)
1670 -- Runtime Function: float __fracttasf (long long accum A)
1671 -- Runtime Function: double __fracttadf (long long accum A)
1672 -- Runtime Function: short fract __fractuqqqq (unsigned short fract A)
1673 -- Runtime Function: fract __fractuqqhq (unsigned short fract A)
1674 -- Runtime Function: long fract __fractuqqsq (unsigned short fract A)
1675 -- Runtime Function: long long fract __fractuqqdq (unsigned short
1677 -- Runtime Function: short accum __fractuqqha (unsigned short fract A)
1678 -- Runtime Function: accum __fractuqqsa (unsigned short fract A)
1679 -- Runtime Function: long accum __fractuqqda (unsigned short fract A)
1680 -- Runtime Function: long long accum __fractuqqta (unsigned short
1682 -- Runtime Function: unsigned fract __fractuqquhq2 (unsigned short
1684 -- Runtime Function: unsigned long fract __fractuqqusq2 (unsigned
1686 -- Runtime Function: unsigned long long fract __fractuqqudq2 (unsigned
1688 -- Runtime Function: unsigned short accum __fractuqquha (unsigned
1690 -- Runtime Function: unsigned accum __fractuqqusa (unsigned short
1692 -- Runtime Function: unsigned long accum __fractuqquda (unsigned short
1694 -- Runtime Function: unsigned long long accum __fractuqquta (unsigned
1696 -- Runtime Function: signed char __fractuqqqi (unsigned short fract A)
1697 -- Runtime Function: short __fractuqqhi (unsigned short fract A)
1698 -- Runtime Function: int __fractuqqsi (unsigned short fract A)
1699 -- Runtime Function: long __fractuqqdi (unsigned short fract A)
1700 -- Runtime Function: long long __fractuqqti (unsigned short fract A)
1701 -- Runtime Function: float __fractuqqsf (unsigned short fract A)
1702 -- Runtime Function: double __fractuqqdf (unsigned short fract A)
1703 -- Runtime Function: short fract __fractuhqqq (unsigned fract A)
1704 -- Runtime Function: fract __fractuhqhq (unsigned fract A)
1705 -- Runtime Function: long fract __fractuhqsq (unsigned fract A)
1706 -- Runtime Function: long long fract __fractuhqdq (unsigned fract A)
1707 -- Runtime Function: short accum __fractuhqha (unsigned fract A)
1708 -- Runtime Function: accum __fractuhqsa (unsigned fract A)
1709 -- Runtime Function: long accum __fractuhqda (unsigned fract A)
1710 -- Runtime Function: long long accum __fractuhqta (unsigned fract A)
1711 -- Runtime Function: unsigned short fract __fractuhquqq2 (unsigned
1713 -- Runtime Function: unsigned long fract __fractuhqusq2 (unsigned
1715 -- Runtime Function: unsigned long long fract __fractuhqudq2 (unsigned
1717 -- Runtime Function: unsigned short accum __fractuhquha (unsigned
1719 -- Runtime Function: unsigned accum __fractuhqusa (unsigned fract A)
1720 -- Runtime Function: unsigned long accum __fractuhquda (unsigned fract
1722 -- Runtime Function: unsigned long long accum __fractuhquta (unsigned
1724 -- Runtime Function: signed char __fractuhqqi (unsigned fract A)
1725 -- Runtime Function: short __fractuhqhi (unsigned fract A)
1726 -- Runtime Function: int __fractuhqsi (unsigned fract A)
1727 -- Runtime Function: long __fractuhqdi (unsigned fract A)
1728 -- Runtime Function: long long __fractuhqti (unsigned fract A)
1729 -- Runtime Function: float __fractuhqsf (unsigned fract A)
1730 -- Runtime Function: double __fractuhqdf (unsigned fract A)
1731 -- Runtime Function: short fract __fractusqqq (unsigned long fract A)
1732 -- Runtime Function: fract __fractusqhq (unsigned long fract A)
1733 -- Runtime Function: long fract __fractusqsq (unsigned long fract A)
1734 -- Runtime Function: long long fract __fractusqdq (unsigned long fract
1736 -- Runtime Function: short accum __fractusqha (unsigned long fract A)
1737 -- Runtime Function: accum __fractusqsa (unsigned long fract A)
1738 -- Runtime Function: long accum __fractusqda (unsigned long fract A)
1739 -- Runtime Function: long long accum __fractusqta (unsigned long fract
1741 -- Runtime Function: unsigned short fract __fractusquqq2 (unsigned
1743 -- Runtime Function: unsigned fract __fractusquhq2 (unsigned long
1745 -- Runtime Function: unsigned long long fract __fractusqudq2 (unsigned
1747 -- Runtime Function: unsigned short accum __fractusquha (unsigned long
1749 -- Runtime Function: unsigned accum __fractusqusa (unsigned long fract
1751 -- Runtime Function: unsigned long accum __fractusquda (unsigned long
1753 -- Runtime Function: unsigned long long accum __fractusquta (unsigned
1755 -- Runtime Function: signed char __fractusqqi (unsigned long fract A)
1756 -- Runtime Function: short __fractusqhi (unsigned long fract A)
1757 -- Runtime Function: int __fractusqsi (unsigned long fract A)
1758 -- Runtime Function: long __fractusqdi (unsigned long fract A)
1759 -- Runtime Function: long long __fractusqti (unsigned long fract A)
1760 -- Runtime Function: float __fractusqsf (unsigned long fract A)
1761 -- Runtime Function: double __fractusqdf (unsigned long fract A)
1762 -- Runtime Function: short fract __fractudqqq (unsigned long long
1764 -- Runtime Function: fract __fractudqhq (unsigned long long fract A)
1765 -- Runtime Function: long fract __fractudqsq (unsigned long long fract
1767 -- Runtime Function: long long fract __fractudqdq (unsigned long long
1769 -- Runtime Function: short accum __fractudqha (unsigned long long
1771 -- Runtime Function: accum __fractudqsa (unsigned long long fract A)
1772 -- Runtime Function: long accum __fractudqda (unsigned long long fract
1774 -- Runtime Function: long long accum __fractudqta (unsigned long long
1776 -- Runtime Function: unsigned short fract __fractudquqq2 (unsigned
1778 -- Runtime Function: unsigned fract __fractudquhq2 (unsigned long long
1780 -- Runtime Function: unsigned long fract __fractudqusq2 (unsigned long
1782 -- Runtime Function: unsigned short accum __fractudquha (unsigned long
1784 -- Runtime Function: unsigned accum __fractudqusa (unsigned long long
1786 -- Runtime Function: unsigned long accum __fractudquda (unsigned long
1788 -- Runtime Function: unsigned long long accum __fractudquta (unsigned
1790 -- Runtime Function: signed char __fractudqqi (unsigned long long
1792 -- Runtime Function: short __fractudqhi (unsigned long long fract A)
1793 -- Runtime Function: int __fractudqsi (unsigned long long fract A)
1794 -- Runtime Function: long __fractudqdi (unsigned long long fract A)
1795 -- Runtime Function: long long __fractudqti (unsigned long long fract
1797 -- Runtime Function: float __fractudqsf (unsigned long long fract A)
1798 -- Runtime Function: double __fractudqdf (unsigned long long fract A)
1799 -- Runtime Function: short fract __fractuhaqq (unsigned short accum A)
1800 -- Runtime Function: fract __fractuhahq (unsigned short accum A)
1801 -- Runtime Function: long fract __fractuhasq (unsigned short accum A)
1802 -- Runtime Function: long long fract __fractuhadq (unsigned short
1804 -- Runtime Function: short accum __fractuhaha (unsigned short accum A)
1805 -- Runtime Function: accum __fractuhasa (unsigned short accum A)
1806 -- Runtime Function: long accum __fractuhada (unsigned short accum A)
1807 -- Runtime Function: long long accum __fractuhata (unsigned short
1809 -- Runtime Function: unsigned short fract __fractuhauqq (unsigned
1811 -- Runtime Function: unsigned fract __fractuhauhq (unsigned short
1813 -- Runtime Function: unsigned long fract __fractuhausq (unsigned short
1815 -- Runtime Function: unsigned long long fract __fractuhaudq (unsigned
1817 -- Runtime Function: unsigned accum __fractuhausa2 (unsigned short
1819 -- Runtime Function: unsigned long accum __fractuhauda2 (unsigned
1821 -- Runtime Function: unsigned long long accum __fractuhauta2 (unsigned
1823 -- Runtime Function: signed char __fractuhaqi (unsigned short accum A)
1824 -- Runtime Function: short __fractuhahi (unsigned short accum A)
1825 -- Runtime Function: int __fractuhasi (unsigned short accum A)
1826 -- Runtime Function: long __fractuhadi (unsigned short accum A)
1827 -- Runtime Function: long long __fractuhati (unsigned short accum A)
1828 -- Runtime Function: float __fractuhasf (unsigned short accum A)
1829 -- Runtime Function: double __fractuhadf (unsigned short accum A)
1830 -- Runtime Function: short fract __fractusaqq (unsigned accum A)
1831 -- Runtime Function: fract __fractusahq (unsigned accum A)
1832 -- Runtime Function: long fract __fractusasq (unsigned accum A)
1833 -- Runtime Function: long long fract __fractusadq (unsigned accum A)
1834 -- Runtime Function: short accum __fractusaha (unsigned accum A)
1835 -- Runtime Function: accum __fractusasa (unsigned accum A)
1836 -- Runtime Function: long accum __fractusada (unsigned accum A)
1837 -- Runtime Function: long long accum __fractusata (unsigned accum A)
1838 -- Runtime Function: unsigned short fract __fractusauqq (unsigned
1840 -- Runtime Function: unsigned fract __fractusauhq (unsigned accum A)
1841 -- Runtime Function: unsigned long fract __fractusausq (unsigned accum
1843 -- Runtime Function: unsigned long long fract __fractusaudq (unsigned
1845 -- Runtime Function: unsigned short accum __fractusauha2 (unsigned
1847 -- Runtime Function: unsigned long accum __fractusauda2 (unsigned
1849 -- Runtime Function: unsigned long long accum __fractusauta2 (unsigned
1851 -- Runtime Function: signed char __fractusaqi (unsigned accum A)
1852 -- Runtime Function: short __fractusahi (unsigned accum A)
1853 -- Runtime Function: int __fractusasi (unsigned accum A)
1854 -- Runtime Function: long __fractusadi (unsigned accum A)
1855 -- Runtime Function: long long __fractusati (unsigned accum A)
1856 -- Runtime Function: float __fractusasf (unsigned accum A)
1857 -- Runtime Function: double __fractusadf (unsigned accum A)
1858 -- Runtime Function: short fract __fractudaqq (unsigned long accum A)
1859 -- Runtime Function: fract __fractudahq (unsigned long accum A)
1860 -- Runtime Function: long fract __fractudasq (unsigned long accum A)
1861 -- Runtime Function: long long fract __fractudadq (unsigned long accum
1863 -- Runtime Function: short accum __fractudaha (unsigned long accum A)
1864 -- Runtime Function: accum __fractudasa (unsigned long accum A)
1865 -- Runtime Function: long accum __fractudada (unsigned long accum A)
1866 -- Runtime Function: long long accum __fractudata (unsigned long accum
1868 -- Runtime Function: unsigned short fract __fractudauqq (unsigned long
1870 -- Runtime Function: unsigned fract __fractudauhq (unsigned long accum
1872 -- Runtime Function: unsigned long fract __fractudausq (unsigned long
1874 -- Runtime Function: unsigned long long fract __fractudaudq (unsigned
1876 -- Runtime Function: unsigned short accum __fractudauha2 (unsigned
1878 -- Runtime Function: unsigned accum __fractudausa2 (unsigned long
1880 -- Runtime Function: unsigned long long accum __fractudauta2 (unsigned
1882 -- Runtime Function: signed char __fractudaqi (unsigned long accum A)
1883 -- Runtime Function: short __fractudahi (unsigned long accum A)
1884 -- Runtime Function: int __fractudasi (unsigned long accum A)
1885 -- Runtime Function: long __fractudadi (unsigned long accum A)
1886 -- Runtime Function: long long __fractudati (unsigned long accum A)
1887 -- Runtime Function: float __fractudasf (unsigned long accum A)
1888 -- Runtime Function: double __fractudadf (unsigned long accum A)
1889 -- Runtime Function: short fract __fractutaqq (unsigned long long
1891 -- Runtime Function: fract __fractutahq (unsigned long long accum A)
1892 -- Runtime Function: long fract __fractutasq (unsigned long long accum
1894 -- Runtime Function: long long fract __fractutadq (unsigned long long
1896 -- Runtime Function: short accum __fractutaha (unsigned long long
1898 -- Runtime Function: accum __fractutasa (unsigned long long accum A)
1899 -- Runtime Function: long accum __fractutada (unsigned long long accum
1901 -- Runtime Function: long long accum __fractutata (unsigned long long
1903 -- Runtime Function: unsigned short fract __fractutauqq (unsigned long
1905 -- Runtime Function: unsigned fract __fractutauhq (unsigned long long
1907 -- Runtime Function: unsigned long fract __fractutausq (unsigned long
1909 -- Runtime Function: unsigned long long fract __fractutaudq (unsigned
1911 -- Runtime Function: unsigned short accum __fractutauha2 (unsigned
1913 -- Runtime Function: unsigned accum __fractutausa2 (unsigned long long
1915 -- Runtime Function: unsigned long accum __fractutauda2 (unsigned long
1917 -- Runtime Function: signed char __fractutaqi (unsigned long long
1919 -- Runtime Function: short __fractutahi (unsigned long long accum A)
1920 -- Runtime Function: int __fractutasi (unsigned long long accum A)
1921 -- Runtime Function: long __fractutadi (unsigned long long accum A)
1922 -- Runtime Function: long long __fractutati (unsigned long long accum
1924 -- Runtime Function: float __fractutasf (unsigned long long accum A)
1925 -- Runtime Function: double __fractutadf (unsigned long long accum A)
1926 -- Runtime Function: short fract __fractqiqq (signed char A)
1927 -- Runtime Function: fract __fractqihq (signed char A)
1928 -- Runtime Function: long fract __fractqisq (signed char A)
1929 -- Runtime Function: long long fract __fractqidq (signed char A)
1930 -- Runtime Function: short accum __fractqiha (signed char A)
1931 -- Runtime Function: accum __fractqisa (signed char A)
1932 -- Runtime Function: long accum __fractqida (signed char A)
1933 -- Runtime Function: long long accum __fractqita (signed char A)
1934 -- Runtime Function: unsigned short fract __fractqiuqq (signed char A)
1935 -- Runtime Function: unsigned fract __fractqiuhq (signed char A)
1936 -- Runtime Function: unsigned long fract __fractqiusq (signed char A)
1937 -- Runtime Function: unsigned long long fract __fractqiudq (signed
1939 -- Runtime Function: unsigned short accum __fractqiuha (signed char A)
1940 -- Runtime Function: unsigned accum __fractqiusa (signed char A)
1941 -- Runtime Function: unsigned long accum __fractqiuda (signed char A)
1942 -- Runtime Function: unsigned long long accum __fractqiuta (signed
1944 -- Runtime Function: short fract __fracthiqq (short A)
1945 -- Runtime Function: fract __fracthihq (short A)
1946 -- Runtime Function: long fract __fracthisq (short A)
1947 -- Runtime Function: long long fract __fracthidq (short A)
1948 -- Runtime Function: short accum __fracthiha (short A)
1949 -- Runtime Function: accum __fracthisa (short A)
1950 -- Runtime Function: long accum __fracthida (short A)
1951 -- Runtime Function: long long accum __fracthita (short A)
1952 -- Runtime Function: unsigned short fract __fracthiuqq (short A)
1953 -- Runtime Function: unsigned fract __fracthiuhq (short A)
1954 -- Runtime Function: unsigned long fract __fracthiusq (short A)
1955 -- Runtime Function: unsigned long long fract __fracthiudq (short A)
1956 -- Runtime Function: unsigned short accum __fracthiuha (short A)
1957 -- Runtime Function: unsigned accum __fracthiusa (short A)
1958 -- Runtime Function: unsigned long accum __fracthiuda (short A)
1959 -- Runtime Function: unsigned long long accum __fracthiuta (short A)
1960 -- Runtime Function: short fract __fractsiqq (int A)
1961 -- Runtime Function: fract __fractsihq (int A)
1962 -- Runtime Function: long fract __fractsisq (int A)
1963 -- Runtime Function: long long fract __fractsidq (int A)
1964 -- Runtime Function: short accum __fractsiha (int A)
1965 -- Runtime Function: accum __fractsisa (int A)
1966 -- Runtime Function: long accum __fractsida (int A)
1967 -- Runtime Function: long long accum __fractsita (int A)
1968 -- Runtime Function: unsigned short fract __fractsiuqq (int A)
1969 -- Runtime Function: unsigned fract __fractsiuhq (int A)
1970 -- Runtime Function: unsigned long fract __fractsiusq (int A)
1971 -- Runtime Function: unsigned long long fract __fractsiudq (int A)
1972 -- Runtime Function: unsigned short accum __fractsiuha (int A)
1973 -- Runtime Function: unsigned accum __fractsiusa (int A)
1974 -- Runtime Function: unsigned long accum __fractsiuda (int A)
1975 -- Runtime Function: unsigned long long accum __fractsiuta (int A)
1976 -- Runtime Function: short fract __fractdiqq (long A)
1977 -- Runtime Function: fract __fractdihq (long A)
1978 -- Runtime Function: long fract __fractdisq (long A)
1979 -- Runtime Function: long long fract __fractdidq (long A)
1980 -- Runtime Function: short accum __fractdiha (long A)
1981 -- Runtime Function: accum __fractdisa (long A)
1982 -- Runtime Function: long accum __fractdida (long A)
1983 -- Runtime Function: long long accum __fractdita (long A)
1984 -- Runtime Function: unsigned short fract __fractdiuqq (long A)
1985 -- Runtime Function: unsigned fract __fractdiuhq (long A)
1986 -- Runtime Function: unsigned long fract __fractdiusq (long A)
1987 -- Runtime Function: unsigned long long fract __fractdiudq (long A)
1988 -- Runtime Function: unsigned short accum __fractdiuha (long A)
1989 -- Runtime Function: unsigned accum __fractdiusa (long A)
1990 -- Runtime Function: unsigned long accum __fractdiuda (long A)
1991 -- Runtime Function: unsigned long long accum __fractdiuta (long A)
1992 -- Runtime Function: short fract __fracttiqq (long long A)
1993 -- Runtime Function: fract __fracttihq (long long A)
1994 -- Runtime Function: long fract __fracttisq (long long A)
1995 -- Runtime Function: long long fract __fracttidq (long long A)
1996 -- Runtime Function: short accum __fracttiha (long long A)
1997 -- Runtime Function: accum __fracttisa (long long A)
1998 -- Runtime Function: long accum __fracttida (long long A)
1999 -- Runtime Function: long long accum __fracttita (long long A)
2000 -- Runtime Function: unsigned short fract __fracttiuqq (long long A)
2001 -- Runtime Function: unsigned fract __fracttiuhq (long long A)
2002 -- Runtime Function: unsigned long fract __fracttiusq (long long A)
2003 -- Runtime Function: unsigned long long fract __fracttiudq (long long
2005 -- Runtime Function: unsigned short accum __fracttiuha (long long A)
2006 -- Runtime Function: unsigned accum __fracttiusa (long long A)
2007 -- Runtime Function: unsigned long accum __fracttiuda (long long A)
2008 -- Runtime Function: unsigned long long accum __fracttiuta (long long
2010 -- Runtime Function: short fract __fractsfqq (float A)
2011 -- Runtime Function: fract __fractsfhq (float A)
2012 -- Runtime Function: long fract __fractsfsq (float A)
2013 -- Runtime Function: long long fract __fractsfdq (float A)
2014 -- Runtime Function: short accum __fractsfha (float A)
2015 -- Runtime Function: accum __fractsfsa (float A)
2016 -- Runtime Function: long accum __fractsfda (float A)
2017 -- Runtime Function: long long accum __fractsfta (float A)
2018 -- Runtime Function: unsigned short fract __fractsfuqq (float A)
2019 -- Runtime Function: unsigned fract __fractsfuhq (float A)
2020 -- Runtime Function: unsigned long fract __fractsfusq (float A)
2021 -- Runtime Function: unsigned long long fract __fractsfudq (float A)
2022 -- Runtime Function: unsigned short accum __fractsfuha (float A)
2023 -- Runtime Function: unsigned accum __fractsfusa (float A)
2024 -- Runtime Function: unsigned long accum __fractsfuda (float A)
2025 -- Runtime Function: unsigned long long accum __fractsfuta (float A)
2026 -- Runtime Function: short fract __fractdfqq (double A)
2027 -- Runtime Function: fract __fractdfhq (double A)
2028 -- Runtime Function: long fract __fractdfsq (double A)
2029 -- Runtime Function: long long fract __fractdfdq (double A)
2030 -- Runtime Function: short accum __fractdfha (double A)
2031 -- Runtime Function: accum __fractdfsa (double A)
2032 -- Runtime Function: long accum __fractdfda (double A)
2033 -- Runtime Function: long long accum __fractdfta (double A)
2034 -- Runtime Function: unsigned short fract __fractdfuqq (double A)
2035 -- Runtime Function: unsigned fract __fractdfuhq (double A)
2036 -- Runtime Function: unsigned long fract __fractdfusq (double A)
2037 -- Runtime Function: unsigned long long fract __fractdfudq (double A)
2038 -- Runtime Function: unsigned short accum __fractdfuha (double A)
2039 -- Runtime Function: unsigned accum __fractdfusa (double A)
2040 -- Runtime Function: unsigned long accum __fractdfuda (double A)
2041 -- Runtime Function: unsigned long long accum __fractdfuta (double A)
2042 These functions convert from fractional and signed non-fractionals
2043 to fractionals and signed non-fractionals, without saturation.
2045 -- Runtime Function: fract __satfractqqhq2 (short fract A)
2046 -- Runtime Function: long fract __satfractqqsq2 (short fract A)
2047 -- Runtime Function: long long fract __satfractqqdq2 (short fract A)
2048 -- Runtime Function: short accum __satfractqqha (short fract A)
2049 -- Runtime Function: accum __satfractqqsa (short fract A)
2050 -- Runtime Function: long accum __satfractqqda (short fract A)
2051 -- Runtime Function: long long accum __satfractqqta (short fract A)
2052 -- Runtime Function: unsigned short fract __satfractqquqq (short fract
2054 -- Runtime Function: unsigned fract __satfractqquhq (short fract A)
2055 -- Runtime Function: unsigned long fract __satfractqqusq (short fract
2057 -- Runtime Function: unsigned long long fract __satfractqqudq (short
2059 -- Runtime Function: unsigned short accum __satfractqquha (short fract
2061 -- Runtime Function: unsigned accum __satfractqqusa (short fract A)
2062 -- Runtime Function: unsigned long accum __satfractqquda (short fract
2064 -- Runtime Function: unsigned long long accum __satfractqquta (short
2066 -- Runtime Function: short fract __satfracthqqq2 (fract A)
2067 -- Runtime Function: long fract __satfracthqsq2 (fract A)
2068 -- Runtime Function: long long fract __satfracthqdq2 (fract A)
2069 -- Runtime Function: short accum __satfracthqha (fract A)
2070 -- Runtime Function: accum __satfracthqsa (fract A)
2071 -- Runtime Function: long accum __satfracthqda (fract A)
2072 -- Runtime Function: long long accum __satfracthqta (fract A)
2073 -- Runtime Function: unsigned short fract __satfracthquqq (fract A)
2074 -- Runtime Function: unsigned fract __satfracthquhq (fract A)
2075 -- Runtime Function: unsigned long fract __satfracthqusq (fract A)
2076 -- Runtime Function: unsigned long long fract __satfracthqudq (fract A)
2077 -- Runtime Function: unsigned short accum __satfracthquha (fract A)
2078 -- Runtime Function: unsigned accum __satfracthqusa (fract A)
2079 -- Runtime Function: unsigned long accum __satfracthquda (fract A)
2080 -- Runtime Function: unsigned long long accum __satfracthquta (fract A)
2081 -- Runtime Function: short fract __satfractsqqq2 (long fract A)
2082 -- Runtime Function: fract __satfractsqhq2 (long fract A)
2083 -- Runtime Function: long long fract __satfractsqdq2 (long fract A)
2084 -- Runtime Function: short accum __satfractsqha (long fract A)
2085 -- Runtime Function: accum __satfractsqsa (long fract A)
2086 -- Runtime Function: long accum __satfractsqda (long fract A)
2087 -- Runtime Function: long long accum __satfractsqta (long fract A)
2088 -- Runtime Function: unsigned short fract __satfractsquqq (long fract
2090 -- Runtime Function: unsigned fract __satfractsquhq (long fract A)
2091 -- Runtime Function: unsigned long fract __satfractsqusq (long fract A)
2092 -- Runtime Function: unsigned long long fract __satfractsqudq (long
2094 -- Runtime Function: unsigned short accum __satfractsquha (long fract
2096 -- Runtime Function: unsigned accum __satfractsqusa (long fract A)
2097 -- Runtime Function: unsigned long accum __satfractsquda (long fract A)
2098 -- Runtime Function: unsigned long long accum __satfractsquta (long
2100 -- Runtime Function: short fract __satfractdqqq2 (long long fract A)
2101 -- Runtime Function: fract __satfractdqhq2 (long long fract A)
2102 -- Runtime Function: long fract __satfractdqsq2 (long long fract A)
2103 -- Runtime Function: short accum __satfractdqha (long long fract A)
2104 -- Runtime Function: accum __satfractdqsa (long long fract A)
2105 -- Runtime Function: long accum __satfractdqda (long long fract A)
2106 -- Runtime Function: long long accum __satfractdqta (long long fract A)
2107 -- Runtime Function: unsigned short fract __satfractdquqq (long long
2109 -- Runtime Function: unsigned fract __satfractdquhq (long long fract A)
2110 -- Runtime Function: unsigned long fract __satfractdqusq (long long
2112 -- Runtime Function: unsigned long long fract __satfractdqudq (long
2114 -- Runtime Function: unsigned short accum __satfractdquha (long long
2116 -- Runtime Function: unsigned accum __satfractdqusa (long long fract A)
2117 -- Runtime Function: unsigned long accum __satfractdquda (long long
2119 -- Runtime Function: unsigned long long accum __satfractdquta (long
2121 -- Runtime Function: short fract __satfracthaqq (short accum A)
2122 -- Runtime Function: fract __satfracthahq (short accum A)
2123 -- Runtime Function: long fract __satfracthasq (short accum A)
2124 -- Runtime Function: long long fract __satfracthadq (short accum A)
2125 -- Runtime Function: accum __satfracthasa2 (short accum A)
2126 -- Runtime Function: long accum __satfracthada2 (short accum A)
2127 -- Runtime Function: long long accum __satfracthata2 (short accum A)
2128 -- Runtime Function: unsigned short fract __satfracthauqq (short accum
2130 -- Runtime Function: unsigned fract __satfracthauhq (short accum A)
2131 -- Runtime Function: unsigned long fract __satfracthausq (short accum
2133 -- Runtime Function: unsigned long long fract __satfracthaudq (short
2135 -- Runtime Function: unsigned short accum __satfracthauha (short accum
2137 -- Runtime Function: unsigned accum __satfracthausa (short accum A)
2138 -- Runtime Function: unsigned long accum __satfracthauda (short accum
2140 -- Runtime Function: unsigned long long accum __satfracthauta (short
2142 -- Runtime Function: short fract __satfractsaqq (accum A)
2143 -- Runtime Function: fract __satfractsahq (accum A)
2144 -- Runtime Function: long fract __satfractsasq (accum A)
2145 -- Runtime Function: long long fract __satfractsadq (accum A)
2146 -- Runtime Function: short accum __satfractsaha2 (accum A)
2147 -- Runtime Function: long accum __satfractsada2 (accum A)
2148 -- Runtime Function: long long accum __satfractsata2 (accum A)
2149 -- Runtime Function: unsigned short fract __satfractsauqq (accum A)
2150 -- Runtime Function: unsigned fract __satfractsauhq (accum A)
2151 -- Runtime Function: unsigned long fract __satfractsausq (accum A)
2152 -- Runtime Function: unsigned long long fract __satfractsaudq (accum A)
2153 -- Runtime Function: unsigned short accum __satfractsauha (accum A)
2154 -- Runtime Function: unsigned accum __satfractsausa (accum A)
2155 -- Runtime Function: unsigned long accum __satfractsauda (accum A)
2156 -- Runtime Function: unsigned long long accum __satfractsauta (accum A)
2157 -- Runtime Function: short fract __satfractdaqq (long accum A)
2158 -- Runtime Function: fract __satfractdahq (long accum A)
2159 -- Runtime Function: long fract __satfractdasq (long accum A)
2160 -- Runtime Function: long long fract __satfractdadq (long accum A)
2161 -- Runtime Function: short accum __satfractdaha2 (long accum A)
2162 -- Runtime Function: accum __satfractdasa2 (long accum A)
2163 -- Runtime Function: long long accum __satfractdata2 (long accum A)
2164 -- Runtime Function: unsigned short fract __satfractdauqq (long accum
2166 -- Runtime Function: unsigned fract __satfractdauhq (long accum A)
2167 -- Runtime Function: unsigned long fract __satfractdausq (long accum A)
2168 -- Runtime Function: unsigned long long fract __satfractdaudq (long
2170 -- Runtime Function: unsigned short accum __satfractdauha (long accum
2172 -- Runtime Function: unsigned accum __satfractdausa (long accum A)
2173 -- Runtime Function: unsigned long accum __satfractdauda (long accum A)
2174 -- Runtime Function: unsigned long long accum __satfractdauta (long
2176 -- Runtime Function: short fract __satfracttaqq (long long accum A)
2177 -- Runtime Function: fract __satfracttahq (long long accum A)
2178 -- Runtime Function: long fract __satfracttasq (long long accum A)
2179 -- Runtime Function: long long fract __satfracttadq (long long accum A)
2180 -- Runtime Function: short accum __satfracttaha2 (long long accum A)
2181 -- Runtime Function: accum __satfracttasa2 (long long accum A)
2182 -- Runtime Function: long accum __satfracttada2 (long long accum A)
2183 -- Runtime Function: unsigned short fract __satfracttauqq (long long
2185 -- Runtime Function: unsigned fract __satfracttauhq (long long accum A)
2186 -- Runtime Function: unsigned long fract __satfracttausq (long long
2188 -- Runtime Function: unsigned long long fract __satfracttaudq (long
2190 -- Runtime Function: unsigned short accum __satfracttauha (long long
2192 -- Runtime Function: unsigned accum __satfracttausa (long long accum A)
2193 -- Runtime Function: unsigned long accum __satfracttauda (long long
2195 -- Runtime Function: unsigned long long accum __satfracttauta (long
2197 -- Runtime Function: short fract __satfractuqqqq (unsigned short fract
2199 -- Runtime Function: fract __satfractuqqhq (unsigned short fract A)
2200 -- Runtime Function: long fract __satfractuqqsq (unsigned short fract
2202 -- Runtime Function: long long fract __satfractuqqdq (unsigned short
2204 -- Runtime Function: short accum __satfractuqqha (unsigned short fract
2206 -- Runtime Function: accum __satfractuqqsa (unsigned short fract A)
2207 -- Runtime Function: long accum __satfractuqqda (unsigned short fract
2209 -- Runtime Function: long long accum __satfractuqqta (unsigned short
2211 -- Runtime Function: unsigned fract __satfractuqquhq2 (unsigned short
2213 -- Runtime Function: unsigned long fract __satfractuqqusq2 (unsigned
2215 -- Runtime Function: unsigned long long fract __satfractuqqudq2
2216 (unsigned short fract A)
2217 -- Runtime Function: unsigned short accum __satfractuqquha (unsigned
2219 -- Runtime Function: unsigned accum __satfractuqqusa (unsigned short
2221 -- Runtime Function: unsigned long accum __satfractuqquda (unsigned
2223 -- Runtime Function: unsigned long long accum __satfractuqquta
2224 (unsigned short fract A)
2225 -- Runtime Function: short fract __satfractuhqqq (unsigned fract A)
2226 -- Runtime Function: fract __satfractuhqhq (unsigned fract A)
2227 -- Runtime Function: long fract __satfractuhqsq (unsigned fract A)
2228 -- Runtime Function: long long fract __satfractuhqdq (unsigned fract A)
2229 -- Runtime Function: short accum __satfractuhqha (unsigned fract A)
2230 -- Runtime Function: accum __satfractuhqsa (unsigned fract A)
2231 -- Runtime Function: long accum __satfractuhqda (unsigned fract A)
2232 -- Runtime Function: long long accum __satfractuhqta (unsigned fract A)
2233 -- Runtime Function: unsigned short fract __satfractuhquqq2 (unsigned
2235 -- Runtime Function: unsigned long fract __satfractuhqusq2 (unsigned
2237 -- Runtime Function: unsigned long long fract __satfractuhqudq2
2239 -- Runtime Function: unsigned short accum __satfractuhquha (unsigned
2241 -- Runtime Function: unsigned accum __satfractuhqusa (unsigned fract A)
2242 -- Runtime Function: unsigned long accum __satfractuhquda (unsigned
2244 -- Runtime Function: unsigned long long accum __satfractuhquta
2246 -- Runtime Function: short fract __satfractusqqq (unsigned long fract
2248 -- Runtime Function: fract __satfractusqhq (unsigned long fract A)
2249 -- Runtime Function: long fract __satfractusqsq (unsigned long fract A)
2250 -- Runtime Function: long long fract __satfractusqdq (unsigned long
2252 -- Runtime Function: short accum __satfractusqha (unsigned long fract
2254 -- Runtime Function: accum __satfractusqsa (unsigned long fract A)
2255 -- Runtime Function: long accum __satfractusqda (unsigned long fract A)
2256 -- Runtime Function: long long accum __satfractusqta (unsigned long
2258 -- Runtime Function: unsigned short fract __satfractusquqq2 (unsigned
2260 -- Runtime Function: unsigned fract __satfractusquhq2 (unsigned long
2262 -- Runtime Function: unsigned long long fract __satfractusqudq2
2263 (unsigned long fract A)
2264 -- Runtime Function: unsigned short accum __satfractusquha (unsigned
2266 -- Runtime Function: unsigned accum __satfractusqusa (unsigned long
2268 -- Runtime Function: unsigned long accum __satfractusquda (unsigned
2270 -- Runtime Function: unsigned long long accum __satfractusquta
2271 (unsigned long fract A)
2272 -- Runtime Function: short fract __satfractudqqq (unsigned long long
2274 -- Runtime Function: fract __satfractudqhq (unsigned long long fract A)
2275 -- Runtime Function: long fract __satfractudqsq (unsigned long long
2277 -- Runtime Function: long long fract __satfractudqdq (unsigned long
2279 -- Runtime Function: short accum __satfractudqha (unsigned long long
2281 -- Runtime Function: accum __satfractudqsa (unsigned long long fract A)
2282 -- Runtime Function: long accum __satfractudqda (unsigned long long
2284 -- Runtime Function: long long accum __satfractudqta (unsigned long
2286 -- Runtime Function: unsigned short fract __satfractudquqq2 (unsigned
2288 -- Runtime Function: unsigned fract __satfractudquhq2 (unsigned long
2290 -- Runtime Function: unsigned long fract __satfractudqusq2 (unsigned
2292 -- Runtime Function: unsigned short accum __satfractudquha (unsigned
2294 -- Runtime Function: unsigned accum __satfractudqusa (unsigned long
2296 -- Runtime Function: unsigned long accum __satfractudquda (unsigned
2298 -- Runtime Function: unsigned long long accum __satfractudquta
2299 (unsigned long long fract A)
2300 -- Runtime Function: short fract __satfractuhaqq (unsigned short accum
2302 -- Runtime Function: fract __satfractuhahq (unsigned short accum A)
2303 -- Runtime Function: long fract __satfractuhasq (unsigned short accum
2305 -- Runtime Function: long long fract __satfractuhadq (unsigned short
2307 -- Runtime Function: short accum __satfractuhaha (unsigned short accum
2309 -- Runtime Function: accum __satfractuhasa (unsigned short accum A)
2310 -- Runtime Function: long accum __satfractuhada (unsigned short accum
2312 -- Runtime Function: long long accum __satfractuhata (unsigned short
2314 -- Runtime Function: unsigned short fract __satfractuhauqq (unsigned
2316 -- Runtime Function: unsigned fract __satfractuhauhq (unsigned short
2318 -- Runtime Function: unsigned long fract __satfractuhausq (unsigned
2320 -- Runtime Function: unsigned long long fract __satfractuhaudq
2321 (unsigned short accum A)
2322 -- Runtime Function: unsigned accum __satfractuhausa2 (unsigned short
2324 -- Runtime Function: unsigned long accum __satfractuhauda2 (unsigned
2326 -- Runtime Function: unsigned long long accum __satfractuhauta2
2327 (unsigned short accum A)
2328 -- Runtime Function: short fract __satfractusaqq (unsigned accum A)
2329 -- Runtime Function: fract __satfractusahq (unsigned accum A)
2330 -- Runtime Function: long fract __satfractusasq (unsigned accum A)
2331 -- Runtime Function: long long fract __satfractusadq (unsigned accum A)
2332 -- Runtime Function: short accum __satfractusaha (unsigned accum A)
2333 -- Runtime Function: accum __satfractusasa (unsigned accum A)
2334 -- Runtime Function: long accum __satfractusada (unsigned accum A)
2335 -- Runtime Function: long long accum __satfractusata (unsigned accum A)
2336 -- Runtime Function: unsigned short fract __satfractusauqq (unsigned
2338 -- Runtime Function: unsigned fract __satfractusauhq (unsigned accum A)
2339 -- Runtime Function: unsigned long fract __satfractusausq (unsigned
2341 -- Runtime Function: unsigned long long fract __satfractusaudq
2343 -- Runtime Function: unsigned short accum __satfractusauha2 (unsigned
2345 -- Runtime Function: unsigned long accum __satfractusauda2 (unsigned
2347 -- Runtime Function: unsigned long long accum __satfractusauta2
2349 -- Runtime Function: short fract __satfractudaqq (unsigned long accum
2351 -- Runtime Function: fract __satfractudahq (unsigned long accum A)
2352 -- Runtime Function: long fract __satfractudasq (unsigned long accum A)
2353 -- Runtime Function: long long fract __satfractudadq (unsigned long
2355 -- Runtime Function: short accum __satfractudaha (unsigned long accum
2357 -- Runtime Function: accum __satfractudasa (unsigned long accum A)
2358 -- Runtime Function: long accum __satfractudada (unsigned long accum A)
2359 -- Runtime Function: long long accum __satfractudata (unsigned long
2361 -- Runtime Function: unsigned short fract __satfractudauqq (unsigned
2363 -- Runtime Function: unsigned fract __satfractudauhq (unsigned long
2365 -- Runtime Function: unsigned long fract __satfractudausq (unsigned
2367 -- Runtime Function: unsigned long long fract __satfractudaudq
2368 (unsigned long accum A)
2369 -- Runtime Function: unsigned short accum __satfractudauha2 (unsigned
2371 -- Runtime Function: unsigned accum __satfractudausa2 (unsigned long
2373 -- Runtime Function: unsigned long long accum __satfractudauta2
2374 (unsigned long accum A)
2375 -- Runtime Function: short fract __satfractutaqq (unsigned long long
2377 -- Runtime Function: fract __satfractutahq (unsigned long long accum A)
2378 -- Runtime Function: long fract __satfractutasq (unsigned long long
2380 -- Runtime Function: long long fract __satfractutadq (unsigned long
2382 -- Runtime Function: short accum __satfractutaha (unsigned long long
2384 -- Runtime Function: accum __satfractutasa (unsigned long long accum A)
2385 -- Runtime Function: long accum __satfractutada (unsigned long long
2387 -- Runtime Function: long long accum __satfractutata (unsigned long
2389 -- Runtime Function: unsigned short fract __satfractutauqq (unsigned
2391 -- Runtime Function: unsigned fract __satfractutauhq (unsigned long
2393 -- Runtime Function: unsigned long fract __satfractutausq (unsigned
2395 -- Runtime Function: unsigned long long fract __satfractutaudq
2396 (unsigned long long accum A)
2397 -- Runtime Function: unsigned short accum __satfractutauha2 (unsigned
2399 -- Runtime Function: unsigned accum __satfractutausa2 (unsigned long
2401 -- Runtime Function: unsigned long accum __satfractutauda2 (unsigned
2403 -- Runtime Function: short fract __satfractqiqq (signed char A)
2404 -- Runtime Function: fract __satfractqihq (signed char A)
2405 -- Runtime Function: long fract __satfractqisq (signed char A)
2406 -- Runtime Function: long long fract __satfractqidq (signed char A)
2407 -- Runtime Function: short accum __satfractqiha (signed char A)
2408 -- Runtime Function: accum __satfractqisa (signed char A)
2409 -- Runtime Function: long accum __satfractqida (signed char A)
2410 -- Runtime Function: long long accum __satfractqita (signed char A)
2411 -- Runtime Function: unsigned short fract __satfractqiuqq (signed char
2413 -- Runtime Function: unsigned fract __satfractqiuhq (signed char A)
2414 -- Runtime Function: unsigned long fract __satfractqiusq (signed char
2416 -- Runtime Function: unsigned long long fract __satfractqiudq (signed
2418 -- Runtime Function: unsigned short accum __satfractqiuha (signed char
2420 -- Runtime Function: unsigned accum __satfractqiusa (signed char A)
2421 -- Runtime Function: unsigned long accum __satfractqiuda (signed char
2423 -- Runtime Function: unsigned long long accum __satfractqiuta (signed
2425 -- Runtime Function: short fract __satfracthiqq (short A)
2426 -- Runtime Function: fract __satfracthihq (short A)
2427 -- Runtime Function: long fract __satfracthisq (short A)
2428 -- Runtime Function: long long fract __satfracthidq (short A)
2429 -- Runtime Function: short accum __satfracthiha (short A)
2430 -- Runtime Function: accum __satfracthisa (short A)
2431 -- Runtime Function: long accum __satfracthida (short A)
2432 -- Runtime Function: long long accum __satfracthita (short A)
2433 -- Runtime Function: unsigned short fract __satfracthiuqq (short A)
2434 -- Runtime Function: unsigned fract __satfracthiuhq (short A)
2435 -- Runtime Function: unsigned long fract __satfracthiusq (short A)
2436 -- Runtime Function: unsigned long long fract __satfracthiudq (short A)
2437 -- Runtime Function: unsigned short accum __satfracthiuha (short A)
2438 -- Runtime Function: unsigned accum __satfracthiusa (short A)
2439 -- Runtime Function: unsigned long accum __satfracthiuda (short A)
2440 -- Runtime Function: unsigned long long accum __satfracthiuta (short A)
2441 -- Runtime Function: short fract __satfractsiqq (int A)
2442 -- Runtime Function: fract __satfractsihq (int A)
2443 -- Runtime Function: long fract __satfractsisq (int A)
2444 -- Runtime Function: long long fract __satfractsidq (int A)
2445 -- Runtime Function: short accum __satfractsiha (int A)
2446 -- Runtime Function: accum __satfractsisa (int A)
2447 -- Runtime Function: long accum __satfractsida (int A)
2448 -- Runtime Function: long long accum __satfractsita (int A)
2449 -- Runtime Function: unsigned short fract __satfractsiuqq (int A)
2450 -- Runtime Function: unsigned fract __satfractsiuhq (int A)
2451 -- Runtime Function: unsigned long fract __satfractsiusq (int A)
2452 -- Runtime Function: unsigned long long fract __satfractsiudq (int A)
2453 -- Runtime Function: unsigned short accum __satfractsiuha (int A)
2454 -- Runtime Function: unsigned accum __satfractsiusa (int A)
2455 -- Runtime Function: unsigned long accum __satfractsiuda (int A)
2456 -- Runtime Function: unsigned long long accum __satfractsiuta (int A)
2457 -- Runtime Function: short fract __satfractdiqq (long A)
2458 -- Runtime Function: fract __satfractdihq (long A)
2459 -- Runtime Function: long fract __satfractdisq (long A)
2460 -- Runtime Function: long long fract __satfractdidq (long A)
2461 -- Runtime Function: short accum __satfractdiha (long A)
2462 -- Runtime Function: accum __satfractdisa (long A)
2463 -- Runtime Function: long accum __satfractdida (long A)
2464 -- Runtime Function: long long accum __satfractdita (long A)
2465 -- Runtime Function: unsigned short fract __satfractdiuqq (long A)
2466 -- Runtime Function: unsigned fract __satfractdiuhq (long A)
2467 -- Runtime Function: unsigned long fract __satfractdiusq (long A)
2468 -- Runtime Function: unsigned long long fract __satfractdiudq (long A)
2469 -- Runtime Function: unsigned short accum __satfractdiuha (long A)
2470 -- Runtime Function: unsigned accum __satfractdiusa (long A)
2471 -- Runtime Function: unsigned long accum __satfractdiuda (long A)
2472 -- Runtime Function: unsigned long long accum __satfractdiuta (long A)
2473 -- Runtime Function: short fract __satfracttiqq (long long A)
2474 -- Runtime Function: fract __satfracttihq (long long A)
2475 -- Runtime Function: long fract __satfracttisq (long long A)
2476 -- Runtime Function: long long fract __satfracttidq (long long A)
2477 -- Runtime Function: short accum __satfracttiha (long long A)
2478 -- Runtime Function: accum __satfracttisa (long long A)
2479 -- Runtime Function: long accum __satfracttida (long long A)
2480 -- Runtime Function: long long accum __satfracttita (long long A)
2481 -- Runtime Function: unsigned short fract __satfracttiuqq (long long A)
2482 -- Runtime Function: unsigned fract __satfracttiuhq (long long A)
2483 -- Runtime Function: unsigned long fract __satfracttiusq (long long A)
2484 -- Runtime Function: unsigned long long fract __satfracttiudq (long
2486 -- Runtime Function: unsigned short accum __satfracttiuha (long long A)
2487 -- Runtime Function: unsigned accum __satfracttiusa (long long A)
2488 -- Runtime Function: unsigned long accum __satfracttiuda (long long A)
2489 -- Runtime Function: unsigned long long accum __satfracttiuta (long
2491 -- Runtime Function: short fract __satfractsfqq (float A)
2492 -- Runtime Function: fract __satfractsfhq (float A)
2493 -- Runtime Function: long fract __satfractsfsq (float A)
2494 -- Runtime Function: long long fract __satfractsfdq (float A)
2495 -- Runtime Function: short accum __satfractsfha (float A)
2496 -- Runtime Function: accum __satfractsfsa (float A)
2497 -- Runtime Function: long accum __satfractsfda (float A)
2498 -- Runtime Function: long long accum __satfractsfta (float A)
2499 -- Runtime Function: unsigned short fract __satfractsfuqq (float A)
2500 -- Runtime Function: unsigned fract __satfractsfuhq (float A)
2501 -- Runtime Function: unsigned long fract __satfractsfusq (float A)
2502 -- Runtime Function: unsigned long long fract __satfractsfudq (float A)
2503 -- Runtime Function: unsigned short accum __satfractsfuha (float A)
2504 -- Runtime Function: unsigned accum __satfractsfusa (float A)
2505 -- Runtime Function: unsigned long accum __satfractsfuda (float A)
2506 -- Runtime Function: unsigned long long accum __satfractsfuta (float A)
2507 -- Runtime Function: short fract __satfractdfqq (double A)
2508 -- Runtime Function: fract __satfractdfhq (double A)
2509 -- Runtime Function: long fract __satfractdfsq (double A)
2510 -- Runtime Function: long long fract __satfractdfdq (double A)
2511 -- Runtime Function: short accum __satfractdfha (double A)
2512 -- Runtime Function: accum __satfractdfsa (double A)
2513 -- Runtime Function: long accum __satfractdfda (double A)
2514 -- Runtime Function: long long accum __satfractdfta (double A)
2515 -- Runtime Function: unsigned short fract __satfractdfuqq (double A)
2516 -- Runtime Function: unsigned fract __satfractdfuhq (double A)
2517 -- Runtime Function: unsigned long fract __satfractdfusq (double A)
2518 -- Runtime Function: unsigned long long fract __satfractdfudq (double
2520 -- Runtime Function: unsigned short accum __satfractdfuha (double A)
2521 -- Runtime Function: unsigned accum __satfractdfusa (double A)
2522 -- Runtime Function: unsigned long accum __satfractdfuda (double A)
2523 -- Runtime Function: unsigned long long accum __satfractdfuta (double
2525 The functions convert from fractional and signed non-fractionals to
2526 fractionals, with saturation.
2528 -- Runtime Function: unsigned char __fractunsqqqi (short fract A)
2529 -- Runtime Function: unsigned short __fractunsqqhi (short fract A)
2530 -- Runtime Function: unsigned int __fractunsqqsi (short fract A)
2531 -- Runtime Function: unsigned long __fractunsqqdi (short fract A)
2532 -- Runtime Function: unsigned long long __fractunsqqti (short fract A)
2533 -- Runtime Function: unsigned char __fractunshqqi (fract A)
2534 -- Runtime Function: unsigned short __fractunshqhi (fract A)
2535 -- Runtime Function: unsigned int __fractunshqsi (fract A)
2536 -- Runtime Function: unsigned long __fractunshqdi (fract A)
2537 -- Runtime Function: unsigned long long __fractunshqti (fract A)
2538 -- Runtime Function: unsigned char __fractunssqqi (long fract A)
2539 -- Runtime Function: unsigned short __fractunssqhi (long fract A)
2540 -- Runtime Function: unsigned int __fractunssqsi (long fract A)
2541 -- Runtime Function: unsigned long __fractunssqdi (long fract A)
2542 -- Runtime Function: unsigned long long __fractunssqti (long fract A)
2543 -- Runtime Function: unsigned char __fractunsdqqi (long long fract A)
2544 -- Runtime Function: unsigned short __fractunsdqhi (long long fract A)
2545 -- Runtime Function: unsigned int __fractunsdqsi (long long fract A)
2546 -- Runtime Function: unsigned long __fractunsdqdi (long long fract A)
2547 -- Runtime Function: unsigned long long __fractunsdqti (long long
2549 -- Runtime Function: unsigned char __fractunshaqi (short accum A)
2550 -- Runtime Function: unsigned short __fractunshahi (short accum A)
2551 -- Runtime Function: unsigned int __fractunshasi (short accum A)
2552 -- Runtime Function: unsigned long __fractunshadi (short accum A)
2553 -- Runtime Function: unsigned long long __fractunshati (short accum A)
2554 -- Runtime Function: unsigned char __fractunssaqi (accum A)
2555 -- Runtime Function: unsigned short __fractunssahi (accum A)
2556 -- Runtime Function: unsigned int __fractunssasi (accum A)
2557 -- Runtime Function: unsigned long __fractunssadi (accum A)
2558 -- Runtime Function: unsigned long long __fractunssati (accum A)
2559 -- Runtime Function: unsigned char __fractunsdaqi (long accum A)
2560 -- Runtime Function: unsigned short __fractunsdahi (long accum A)
2561 -- Runtime Function: unsigned int __fractunsdasi (long accum A)
2562 -- Runtime Function: unsigned long __fractunsdadi (long accum A)
2563 -- Runtime Function: unsigned long long __fractunsdati (long accum A)
2564 -- Runtime Function: unsigned char __fractunstaqi (long long accum A)
2565 -- Runtime Function: unsigned short __fractunstahi (long long accum A)
2566 -- Runtime Function: unsigned int __fractunstasi (long long accum A)
2567 -- Runtime Function: unsigned long __fractunstadi (long long accum A)
2568 -- Runtime Function: unsigned long long __fractunstati (long long
2570 -- Runtime Function: unsigned char __fractunsuqqqi (unsigned short
2572 -- Runtime Function: unsigned short __fractunsuqqhi (unsigned short
2574 -- Runtime Function: unsigned int __fractunsuqqsi (unsigned short
2576 -- Runtime Function: unsigned long __fractunsuqqdi (unsigned short
2578 -- Runtime Function: unsigned long long __fractunsuqqti (unsigned
2580 -- Runtime Function: unsigned char __fractunsuhqqi (unsigned fract A)
2581 -- Runtime Function: unsigned short __fractunsuhqhi (unsigned fract A)
2582 -- Runtime Function: unsigned int __fractunsuhqsi (unsigned fract A)
2583 -- Runtime Function: unsigned long __fractunsuhqdi (unsigned fract A)
2584 -- Runtime Function: unsigned long long __fractunsuhqti (unsigned
2586 -- Runtime Function: unsigned char __fractunsusqqi (unsigned long
2588 -- Runtime Function: unsigned short __fractunsusqhi (unsigned long
2590 -- Runtime Function: unsigned int __fractunsusqsi (unsigned long fract
2592 -- Runtime Function: unsigned long __fractunsusqdi (unsigned long
2594 -- Runtime Function: unsigned long long __fractunsusqti (unsigned long
2596 -- Runtime Function: unsigned char __fractunsudqqi (unsigned long long
2598 -- Runtime Function: unsigned short __fractunsudqhi (unsigned long
2600 -- Runtime Function: unsigned int __fractunsudqsi (unsigned long long
2602 -- Runtime Function: unsigned long __fractunsudqdi (unsigned long long
2604 -- Runtime Function: unsigned long long __fractunsudqti (unsigned long
2606 -- Runtime Function: unsigned char __fractunsuhaqi (unsigned short
2608 -- Runtime Function: unsigned short __fractunsuhahi (unsigned short
2610 -- Runtime Function: unsigned int __fractunsuhasi (unsigned short
2612 -- Runtime Function: unsigned long __fractunsuhadi (unsigned short
2614 -- Runtime Function: unsigned long long __fractunsuhati (unsigned
2616 -- Runtime Function: unsigned char __fractunsusaqi (unsigned accum A)
2617 -- Runtime Function: unsigned short __fractunsusahi (unsigned accum A)
2618 -- Runtime Function: unsigned int __fractunsusasi (unsigned accum A)
2619 -- Runtime Function: unsigned long __fractunsusadi (unsigned accum A)
2620 -- Runtime Function: unsigned long long __fractunsusati (unsigned
2622 -- Runtime Function: unsigned char __fractunsudaqi (unsigned long
2624 -- Runtime Function: unsigned short __fractunsudahi (unsigned long
2626 -- Runtime Function: unsigned int __fractunsudasi (unsigned long accum
2628 -- Runtime Function: unsigned long __fractunsudadi (unsigned long
2630 -- Runtime Function: unsigned long long __fractunsudati (unsigned long
2632 -- Runtime Function: unsigned char __fractunsutaqi (unsigned long long
2634 -- Runtime Function: unsigned short __fractunsutahi (unsigned long
2636 -- Runtime Function: unsigned int __fractunsutasi (unsigned long long
2638 -- Runtime Function: unsigned long __fractunsutadi (unsigned long long
2640 -- Runtime Function: unsigned long long __fractunsutati (unsigned long
2642 -- Runtime Function: short fract __fractunsqiqq (unsigned char A)
2643 -- Runtime Function: fract __fractunsqihq (unsigned char A)
2644 -- Runtime Function: long fract __fractunsqisq (unsigned char A)
2645 -- Runtime Function: long long fract __fractunsqidq (unsigned char A)
2646 -- Runtime Function: short accum __fractunsqiha (unsigned char A)
2647 -- Runtime Function: accum __fractunsqisa (unsigned char A)
2648 -- Runtime Function: long accum __fractunsqida (unsigned char A)
2649 -- Runtime Function: long long accum __fractunsqita (unsigned char A)
2650 -- Runtime Function: unsigned short fract __fractunsqiuqq (unsigned
2652 -- Runtime Function: unsigned fract __fractunsqiuhq (unsigned char A)
2653 -- Runtime Function: unsigned long fract __fractunsqiusq (unsigned
2655 -- Runtime Function: unsigned long long fract __fractunsqiudq
2657 -- Runtime Function: unsigned short accum __fractunsqiuha (unsigned
2659 -- Runtime Function: unsigned accum __fractunsqiusa (unsigned char A)
2660 -- Runtime Function: unsigned long accum __fractunsqiuda (unsigned
2662 -- Runtime Function: unsigned long long accum __fractunsqiuta
2664 -- Runtime Function: short fract __fractunshiqq (unsigned short A)
2665 -- Runtime Function: fract __fractunshihq (unsigned short A)
2666 -- Runtime Function: long fract __fractunshisq (unsigned short A)
2667 -- Runtime Function: long long fract __fractunshidq (unsigned short A)
2668 -- Runtime Function: short accum __fractunshiha (unsigned short A)
2669 -- Runtime Function: accum __fractunshisa (unsigned short A)
2670 -- Runtime Function: long accum __fractunshida (unsigned short A)
2671 -- Runtime Function: long long accum __fractunshita (unsigned short A)
2672 -- Runtime Function: unsigned short fract __fractunshiuqq (unsigned
2674 -- Runtime Function: unsigned fract __fractunshiuhq (unsigned short A)
2675 -- Runtime Function: unsigned long fract __fractunshiusq (unsigned
2677 -- Runtime Function: unsigned long long fract __fractunshiudq
2679 -- Runtime Function: unsigned short accum __fractunshiuha (unsigned
2681 -- Runtime Function: unsigned accum __fractunshiusa (unsigned short A)
2682 -- Runtime Function: unsigned long accum __fractunshiuda (unsigned
2684 -- Runtime Function: unsigned long long accum __fractunshiuta
2686 -- Runtime Function: short fract __fractunssiqq (unsigned int A)
2687 -- Runtime Function: fract __fractunssihq (unsigned int A)
2688 -- Runtime Function: long fract __fractunssisq (unsigned int A)
2689 -- Runtime Function: long long fract __fractunssidq (unsigned int A)
2690 -- Runtime Function: short accum __fractunssiha (unsigned int A)
2691 -- Runtime Function: accum __fractunssisa (unsigned int A)
2692 -- Runtime Function: long accum __fractunssida (unsigned int A)
2693 -- Runtime Function: long long accum __fractunssita (unsigned int A)
2694 -- Runtime Function: unsigned short fract __fractunssiuqq (unsigned
2696 -- Runtime Function: unsigned fract __fractunssiuhq (unsigned int A)
2697 -- Runtime Function: unsigned long fract __fractunssiusq (unsigned int
2699 -- Runtime Function: unsigned long long fract __fractunssiudq
2701 -- Runtime Function: unsigned short accum __fractunssiuha (unsigned
2703 -- Runtime Function: unsigned accum __fractunssiusa (unsigned int A)
2704 -- Runtime Function: unsigned long accum __fractunssiuda (unsigned int
2706 -- Runtime Function: unsigned long long accum __fractunssiuta
2708 -- Runtime Function: short fract __fractunsdiqq (unsigned long A)
2709 -- Runtime Function: fract __fractunsdihq (unsigned long A)
2710 -- Runtime Function: long fract __fractunsdisq (unsigned long A)
2711 -- Runtime Function: long long fract __fractunsdidq (unsigned long A)
2712 -- Runtime Function: short accum __fractunsdiha (unsigned long A)
2713 -- Runtime Function: accum __fractunsdisa (unsigned long A)
2714 -- Runtime Function: long accum __fractunsdida (unsigned long A)
2715 -- Runtime Function: long long accum __fractunsdita (unsigned long A)
2716 -- Runtime Function: unsigned short fract __fractunsdiuqq (unsigned
2718 -- Runtime Function: unsigned fract __fractunsdiuhq (unsigned long A)
2719 -- Runtime Function: unsigned long fract __fractunsdiusq (unsigned
2721 -- Runtime Function: unsigned long long fract __fractunsdiudq
2723 -- Runtime Function: unsigned short accum __fractunsdiuha (unsigned
2725 -- Runtime Function: unsigned accum __fractunsdiusa (unsigned long A)
2726 -- Runtime Function: unsigned long accum __fractunsdiuda (unsigned
2728 -- Runtime Function: unsigned long long accum __fractunsdiuta
2730 -- Runtime Function: short fract __fractunstiqq (unsigned long long A)
2731 -- Runtime Function: fract __fractunstihq (unsigned long long A)
2732 -- Runtime Function: long fract __fractunstisq (unsigned long long A)
2733 -- Runtime Function: long long fract __fractunstidq (unsigned long
2735 -- Runtime Function: short accum __fractunstiha (unsigned long long A)
2736 -- Runtime Function: accum __fractunstisa (unsigned long long A)
2737 -- Runtime Function: long accum __fractunstida (unsigned long long A)
2738 -- Runtime Function: long long accum __fractunstita (unsigned long
2740 -- Runtime Function: unsigned short fract __fractunstiuqq (unsigned
2742 -- Runtime Function: unsigned fract __fractunstiuhq (unsigned long
2744 -- Runtime Function: unsigned long fract __fractunstiusq (unsigned
2746 -- Runtime Function: unsigned long long fract __fractunstiudq
2747 (unsigned long long A)
2748 -- Runtime Function: unsigned short accum __fractunstiuha (unsigned
2750 -- Runtime Function: unsigned accum __fractunstiusa (unsigned long
2752 -- Runtime Function: unsigned long accum __fractunstiuda (unsigned
2754 -- Runtime Function: unsigned long long accum __fractunstiuta
2755 (unsigned long long A)
2756 These functions convert from fractionals to unsigned
2757 non-fractionals; and from unsigned non-fractionals to fractionals,
2760 -- Runtime Function: short fract __satfractunsqiqq (unsigned char A)
2761 -- Runtime Function: fract __satfractunsqihq (unsigned char A)
2762 -- Runtime Function: long fract __satfractunsqisq (unsigned char A)
2763 -- Runtime Function: long long fract __satfractunsqidq (unsigned char
2765 -- Runtime Function: short accum __satfractunsqiha (unsigned char A)
2766 -- Runtime Function: accum __satfractunsqisa (unsigned char A)
2767 -- Runtime Function: long accum __satfractunsqida (unsigned char A)
2768 -- Runtime Function: long long accum __satfractunsqita (unsigned char
2770 -- Runtime Function: unsigned short fract __satfractunsqiuqq (unsigned
2772 -- Runtime Function: unsigned fract __satfractunsqiuhq (unsigned char
2774 -- Runtime Function: unsigned long fract __satfractunsqiusq (unsigned
2776 -- Runtime Function: unsigned long long fract __satfractunsqiudq
2778 -- Runtime Function: unsigned short accum __satfractunsqiuha (unsigned
2780 -- Runtime Function: unsigned accum __satfractunsqiusa (unsigned char
2782 -- Runtime Function: unsigned long accum __satfractunsqiuda (unsigned
2784 -- Runtime Function: unsigned long long accum __satfractunsqiuta
2786 -- Runtime Function: short fract __satfractunshiqq (unsigned short A)
2787 -- Runtime Function: fract __satfractunshihq (unsigned short A)
2788 -- Runtime Function: long fract __satfractunshisq (unsigned short A)
2789 -- Runtime Function: long long fract __satfractunshidq (unsigned short
2791 -- Runtime Function: short accum __satfractunshiha (unsigned short A)
2792 -- Runtime Function: accum __satfractunshisa (unsigned short A)
2793 -- Runtime Function: long accum __satfractunshida (unsigned short A)
2794 -- Runtime Function: long long accum __satfractunshita (unsigned short
2796 -- Runtime Function: unsigned short fract __satfractunshiuqq (unsigned
2798 -- Runtime Function: unsigned fract __satfractunshiuhq (unsigned short
2800 -- Runtime Function: unsigned long fract __satfractunshiusq (unsigned
2802 -- Runtime Function: unsigned long long fract __satfractunshiudq
2804 -- Runtime Function: unsigned short accum __satfractunshiuha (unsigned
2806 -- Runtime Function: unsigned accum __satfractunshiusa (unsigned short
2808 -- Runtime Function: unsigned long accum __satfractunshiuda (unsigned
2810 -- Runtime Function: unsigned long long accum __satfractunshiuta
2812 -- Runtime Function: short fract __satfractunssiqq (unsigned int A)
2813 -- Runtime Function: fract __satfractunssihq (unsigned int A)
2814 -- Runtime Function: long fract __satfractunssisq (unsigned int A)
2815 -- Runtime Function: long long fract __satfractunssidq (unsigned int A)
2816 -- Runtime Function: short accum __satfractunssiha (unsigned int A)
2817 -- Runtime Function: accum __satfractunssisa (unsigned int A)
2818 -- Runtime Function: long accum __satfractunssida (unsigned int A)
2819 -- Runtime Function: long long accum __satfractunssita (unsigned int A)
2820 -- Runtime Function: unsigned short fract __satfractunssiuqq (unsigned
2822 -- Runtime Function: unsigned fract __satfractunssiuhq (unsigned int A)
2823 -- Runtime Function: unsigned long fract __satfractunssiusq (unsigned
2825 -- Runtime Function: unsigned long long fract __satfractunssiudq
2827 -- Runtime Function: unsigned short accum __satfractunssiuha (unsigned
2829 -- Runtime Function: unsigned accum __satfractunssiusa (unsigned int A)
2830 -- Runtime Function: unsigned long accum __satfractunssiuda (unsigned
2832 -- Runtime Function: unsigned long long accum __satfractunssiuta
2834 -- Runtime Function: short fract __satfractunsdiqq (unsigned long A)
2835 -- Runtime Function: fract __satfractunsdihq (unsigned long A)
2836 -- Runtime Function: long fract __satfractunsdisq (unsigned long A)
2837 -- Runtime Function: long long fract __satfractunsdidq (unsigned long
2839 -- Runtime Function: short accum __satfractunsdiha (unsigned long A)
2840 -- Runtime Function: accum __satfractunsdisa (unsigned long A)
2841 -- Runtime Function: long accum __satfractunsdida (unsigned long A)
2842 -- Runtime Function: long long accum __satfractunsdita (unsigned long
2844 -- Runtime Function: unsigned short fract __satfractunsdiuqq (unsigned
2846 -- Runtime Function: unsigned fract __satfractunsdiuhq (unsigned long
2848 -- Runtime Function: unsigned long fract __satfractunsdiusq (unsigned
2850 -- Runtime Function: unsigned long long fract __satfractunsdiudq
2852 -- Runtime Function: unsigned short accum __satfractunsdiuha (unsigned
2854 -- Runtime Function: unsigned accum __satfractunsdiusa (unsigned long
2856 -- Runtime Function: unsigned long accum __satfractunsdiuda (unsigned
2858 -- Runtime Function: unsigned long long accum __satfractunsdiuta
2860 -- Runtime Function: short fract __satfractunstiqq (unsigned long long
2862 -- Runtime Function: fract __satfractunstihq (unsigned long long A)
2863 -- Runtime Function: long fract __satfractunstisq (unsigned long long
2865 -- Runtime Function: long long fract __satfractunstidq (unsigned long
2867 -- Runtime Function: short accum __satfractunstiha (unsigned long long
2869 -- Runtime Function: accum __satfractunstisa (unsigned long long A)
2870 -- Runtime Function: long accum __satfractunstida (unsigned long long
2872 -- Runtime Function: long long accum __satfractunstita (unsigned long
2874 -- Runtime Function: unsigned short fract __satfractunstiuqq (unsigned
2876 -- Runtime Function: unsigned fract __satfractunstiuhq (unsigned long
2878 -- Runtime Function: unsigned long fract __satfractunstiusq (unsigned
2880 -- Runtime Function: unsigned long long fract __satfractunstiudq
2881 (unsigned long long A)
2882 -- Runtime Function: unsigned short accum __satfractunstiuha (unsigned
2884 -- Runtime Function: unsigned accum __satfractunstiusa (unsigned long
2886 -- Runtime Function: unsigned long accum __satfractunstiuda (unsigned
2888 -- Runtime Function: unsigned long long accum __satfractunstiuta
2889 (unsigned long long A)
2890 These functions convert from unsigned non-fractionals to
2891 fractionals, with saturation.
2894 File: gccint.info, Node: Exception handling routines, Next: Miscellaneous routines, Prev: Fixed-point fractional library routines, Up: Libgcc
2896 4.5 Language-independent routines for exception handling
2897 ========================================================
2901 _Unwind_DeleteException
2903 _Unwind_ForcedUnwind
2906 _Unwind_GetLanguageSpecificData
2907 _Unwind_GetRegionStart
2908 _Unwind_GetTextRelBase
2909 _Unwind_GetDataRelBase
2910 _Unwind_RaiseException
2914 _Unwind_FindEnclosingFunction
2915 _Unwind_SjLj_Register
2916 _Unwind_SjLj_Unregister
2917 _Unwind_SjLj_RaiseException
2918 _Unwind_SjLj_ForcedUnwind
2921 __deregister_frame_info
2922 __deregister_frame_info_bases
2924 __register_frame_info
2925 __register_frame_info_bases
2926 __register_frame_info_table
2927 __register_frame_info_table_bases
2928 __register_frame_table
2931 File: gccint.info, Node: Miscellaneous routines, Prev: Exception handling routines, Up: Libgcc
2933 4.6 Miscellaneous runtime library routines
2934 ==========================================
2936 4.6.1 Cache control functions
2937 -----------------------------
2939 -- Runtime Function: void __clear_cache (char *BEG, char *END)
2940 This function clears the instruction cache between BEG and END.
2943 File: gccint.info, Node: Languages, Next: Source Tree, Prev: Libgcc, Up: Top
2945 5 Language Front Ends in GCC
2946 ****************************
2948 The interface to front ends for languages in GCC, and in particular the
2949 `tree' structure (*note Trees::), was initially designed for C, and
2950 many aspects of it are still somewhat biased towards C and C-like
2951 languages. It is, however, reasonably well suited to other procedural
2952 languages, and front ends for many such languages have been written for
2955 Writing a compiler as a front end for GCC, rather than compiling
2956 directly to assembler or generating C code which is then compiled by
2957 GCC, has several advantages:
2959 * GCC front ends benefit from the support for many different target
2960 machines already present in GCC.
2962 * GCC front ends benefit from all the optimizations in GCC. Some of
2963 these, such as alias analysis, may work better when GCC is
2964 compiling directly from source code then when it is compiling from
2967 * Better debugging information is generated when compiling directly
2968 from source code than when going via intermediate generated C code.
2970 Because of the advantages of writing a compiler as a GCC front end,
2971 GCC front ends have also been created for languages very different from
2972 those for which GCC was designed, such as the declarative
2973 logic/functional language Mercury. For these reasons, it may also be
2974 useful to implement compilers created for specialized purposes (for
2975 example, as part of a research project) as GCC front ends.
2978 File: gccint.info, Node: Source Tree, Next: Options, Prev: Languages, Up: Top
2980 6 Source Tree Structure and Build System
2981 ****************************************
2983 This chapter describes the structure of the GCC source tree, and how
2984 GCC is built. The user documentation for building and installing GCC
2985 is in a separate manual (`http://gcc.gnu.org/install/'), with which it
2986 is presumed that you are familiar.
2990 * Configure Terms:: Configuration terminology and history.
2991 * Top Level:: The top level source directory.
2992 * gcc Directory:: The `gcc' subdirectory.
2993 * Testsuites:: The GCC testsuites.
2996 File: gccint.info, Node: Configure Terms, Next: Top Level, Up: Source Tree
2998 6.1 Configure Terms and History
2999 ===============================
3001 The configure and build process has a long and colorful history, and can
3002 be confusing to anyone who doesn't know why things are the way they are.
3003 While there are other documents which describe the configuration process
3004 in detail, here are a few things that everyone working on GCC should
3007 There are three system names that the build knows about: the machine
3008 you are building on ("build"), the machine that you are building for
3009 ("host"), and the machine that GCC will produce code for ("target").
3010 When you configure GCC, you specify these with `--build=', `--host=',
3013 Specifying the host without specifying the build should be avoided, as
3014 `configure' may (and once did) assume that the host you specify is also
3015 the build, which may not be true.
3017 If build, host, and target are all the same, this is called a
3018 "native". If build and host are the same but target is different, this
3019 is called a "cross". If build, host, and target are all different this
3020 is called a "canadian" (for obscure reasons dealing with Canada's
3021 political party and the background of the person working on the build
3022 at that time). If host and target are the same, but build is
3023 different, you are using a cross-compiler to build a native for a
3024 different system. Some people call this a "host-x-host", "crossed
3025 native", or "cross-built native". If build and target are the same,
3026 but host is different, you are using a cross compiler to build a cross
3027 compiler that produces code for the machine you're building on. This
3028 is rare, so there is no common way of describing it. There is a
3029 proposal to call this a "crossback".
3031 If build and host are the same, the GCC you are building will also be
3032 used to build the target libraries (like `libstdc++'). If build and
3033 host are different, you must have already built and installed a cross
3034 compiler that will be used to build the target libraries (if you
3035 configured with `--target=foo-bar', this compiler will be called
3038 In the case of target libraries, the machine you're building for is the
3039 machine you specified with `--target'. So, build is the machine you're
3040 building on (no change there), host is the machine you're building for
3041 (the target libraries are built for the target, so host is the target
3042 you specified), and target doesn't apply (because you're not building a
3043 compiler, you're building libraries). The configure/make process will
3044 adjust these variables as needed. It also sets `$with_cross_host' to
3045 the original `--host' value in case you need it.
3047 The `libiberty' support library is built up to three times: once for
3048 the host, once for the target (even if they are the same), and once for
3049 the build if build and host are different. This allows it to be used
3050 by all programs which are generated in the course of the build process.
3053 File: gccint.info, Node: Top Level, Next: gcc Directory, Prev: Configure Terms, Up: Source Tree
3055 6.2 Top Level Source Directory
3056 ==============================
3058 The top level source directory in a GCC distribution contains several
3059 files and directories that are shared with other software distributions
3060 such as that of GNU Binutils. It also contains several subdirectories
3061 that contain parts of GCC and its runtime libraries:
3064 The Boehm conservative garbage collector, used as part of the Java
3068 Contributed scripts that may be found useful in conjunction with
3069 GCC. One of these, `contrib/texi2pod.pl', is used to generate man
3070 pages from Texinfo manuals as part of the GCC build process.
3073 An implementation of the `jar' command, used with the Java front
3077 The support for fixing system headers to work with GCC. See
3078 `fixincludes/README' for more information. The headers fixed by
3079 this mechanism are installed in `LIBSUBDIR/include-fixed'. Along
3080 with those headers, `README-fixinc' is also installed, as
3081 `LIBSUBDIR/include-fixed/README'.
3084 The main sources of GCC itself (except for runtime libraries),
3085 including optimizers, support for different target architectures,
3086 language front ends, and testsuites. *Note The `gcc'
3087 Subdirectory: gcc Directory, for details.
3090 Headers for the `libiberty' library.
3093 GNU `libintl', from GNU `gettext', for systems which do not
3097 The Ada runtime library.
3100 The C preprocessor library.
3103 The Fortran runtime library.
3106 The `libffi' library, used as part of the Java runtime library.
3109 The `libiberty' library, used for portability and for some
3110 generally useful data structures and algorithms. *Note
3111 Introduction: (libiberty)Top, for more information about this
3115 The Java runtime library.
3118 The `libmudflap' library, used for instrumenting pointer and array
3119 dereferencing operations.
3122 The Objective-C and Objective-C++ runtime library.
3125 The C++ runtime library.
3127 `maintainer-scripts'
3128 Scripts used by the `gccadmin' account on `gcc.gnu.org'.
3131 The `zlib' compression library, used by the Java front end and as
3132 part of the Java runtime library.
3134 The build system in the top level directory, including how recursion
3135 into subdirectories works and how building runtime libraries for
3136 multilibs is handled, is documented in a separate manual, included with
3137 GNU Binutils. *Note GNU configure and build system: (configure)Top,
3141 File: gccint.info, Node: gcc Directory, Next: Testsuites, Prev: Top Level, Up: Source Tree
3143 6.3 The `gcc' Subdirectory
3144 ==========================
3146 The `gcc' directory contains many files that are part of the C sources
3147 of GCC, other files used as part of the configuration and build
3148 process, and subdirectories including documentation and a testsuite.
3149 The files that are sources of GCC are documented in a separate chapter.
3150 *Note Passes and Files of the Compiler: Passes.
3154 * Subdirectories:: Subdirectories of `gcc'.
3155 * Configuration:: The configuration process, and the files it uses.
3156 * Build:: The build system in the `gcc' directory.
3157 * Makefile:: Targets in `gcc/Makefile'.
3158 * Library Files:: Library source files and headers under `gcc/'.
3159 * Headers:: Headers installed by GCC.
3160 * Documentation:: Building documentation in GCC.
3161 * Front End:: Anatomy of a language front end.
3162 * Back End:: Anatomy of a target back end.
3165 File: gccint.info, Node: Subdirectories, Next: Configuration, Up: gcc Directory
3167 6.3.1 Subdirectories of `gcc'
3168 -----------------------------
3170 The `gcc' directory contains the following subdirectories:
3173 Subdirectories for various languages. Directories containing a
3174 file `config-lang.in' are language subdirectories. The contents of
3175 the subdirectories `cp' (for C++), `objc' (for Objective-C) and
3176 `objcp' (for Objective-C++) are documented in this manual (*note
3177 Passes and Files of the Compiler: Passes.); those for other
3178 languages are not. *Note Anatomy of a Language Front End: Front
3179 End, for details of the files in these directories.
3182 Configuration files for supported architectures and operating
3183 systems. *Note Anatomy of a Target Back End: Back End, for
3184 details of the files in this directory.
3187 Texinfo documentation for GCC, together with automatically
3188 generated man pages and support for converting the installation
3189 manual to HTML. *Note Documentation::.
3192 System headers installed by GCC, mainly those required by the C
3193 standard of freestanding implementations. *Note Headers Installed
3194 by GCC: Headers, for details of when these and other headers are
3198 Message catalogs with translations of messages produced by GCC into
3199 various languages, `LANGUAGE.po'. This directory also contains
3200 `gcc.pot', the template for these message catalogues, `exgettext',
3201 a wrapper around `gettext' to extract the messages from the GCC
3202 sources and create `gcc.pot', which is run by `make gcc.pot', and
3203 `EXCLUDES', a list of files from which messages should not be
3207 The GCC testsuites (except for those for runtime libraries).
3211 File: gccint.info, Node: Configuration, Next: Build, Prev: Subdirectories, Up: gcc Directory
3213 6.3.2 Configuration in the `gcc' Directory
3214 ------------------------------------------
3216 The `gcc' directory is configured with an Autoconf-generated script
3217 `configure'. The `configure' script is generated from `configure.ac'
3218 and `aclocal.m4'. From the files `configure.ac' and `acconfig.h',
3219 Autoheader generates the file `config.in'. The file `cstamp-h.in' is
3220 used as a timestamp.
3224 * Config Fragments:: Scripts used by `configure'.
3225 * System Config:: The `config.build', `config.host', and
3227 * Configuration Files:: Files created by running `configure'.
3230 File: gccint.info, Node: Config Fragments, Next: System Config, Up: Configuration
3232 6.3.2.1 Scripts Used by `configure'
3233 ...................................
3235 `configure' uses some other scripts to help in its work:
3237 * The standard GNU `config.sub' and `config.guess' files, kept in
3238 the top level directory, are used.
3240 * The file `config.gcc' is used to handle configuration specific to
3241 the particular target machine. The file `config.build' is used to
3242 handle configuration specific to the particular build machine.
3243 The file `config.host' is used to handle configuration specific to
3244 the particular host machine. (In general, these should only be
3245 used for features that cannot reasonably be tested in Autoconf
3246 feature tests.) *Note The `config.build'; `config.host'; and
3247 `config.gcc' Files: System Config, for details of the contents of
3250 * Each language subdirectory has a file `LANGUAGE/config-lang.in'
3251 that is used for front-end-specific configuration. *Note The
3252 Front End `config-lang.in' File: Front End Config, for details of
3255 * A helper script `configure.frag' is used as part of creating the
3256 output of `configure'.
3259 File: gccint.info, Node: System Config, Next: Configuration Files, Prev: Config Fragments, Up: Configuration
3261 6.3.2.2 The `config.build'; `config.host'; and `config.gcc' Files
3262 .................................................................
3264 The `config.build' file contains specific rules for particular systems
3265 which GCC is built on. This should be used as rarely as possible, as
3266 the behavior of the build system can always be detected by autoconf.
3268 The `config.host' file contains specific rules for particular systems
3269 which GCC will run on. This is rarely needed.
3271 The `config.gcc' file contains specific rules for particular systems
3272 which GCC will generate code for. This is usually needed.
3274 Each file has a list of the shell variables it sets, with
3275 descriptions, at the top of the file.
3277 FIXME: document the contents of these files, and what variables should
3278 be set to control build, host and target configuration.
3281 File: gccint.info, Node: Configuration Files, Prev: System Config, Up: Configuration
3283 6.3.2.3 Files Created by `configure'
3284 ....................................
3286 Here we spell out what files will be set up by `configure' in the `gcc'
3287 directory. Some other files are created as temporary files in the
3288 configuration process, and are not used in the subsequent build; these
3291 * `Makefile' is constructed from `Makefile.in', together with the
3292 host and target fragments (*note Makefile Fragments: Fragments.)
3293 `t-TARGET' and `x-HOST' from `config', if any, and language
3294 Makefile fragments `LANGUAGE/Make-lang.in'.
3296 * `auto-host.h' contains information about the host machine
3297 determined by `configure'. If the host machine is different from
3298 the build machine, then `auto-build.h' is also created, containing
3299 such information about the build machine.
3301 * `config.status' is a script that may be run to recreate the
3302 current configuration.
3304 * `configargs.h' is a header containing details of the arguments
3305 passed to `configure' to configure GCC, and of the thread model
3308 * `cstamp-h' is used as a timestamp.
3310 * `fixinc/Makefile' is constructed from `fixinc/Makefile.in'.
3312 * `gccbug', a script for reporting bugs in GCC, is constructed from
3315 * `intl/Makefile' is constructed from `intl/Makefile.in'.
3317 * If a language `config-lang.in' file (*note The Front End
3318 `config-lang.in' File: Front End Config.) sets `outputs', then the
3319 files listed in `outputs' there are also generated.
3321 The following configuration headers are created from the Makefile,
3322 using `mkconfig.sh', rather than directly by `configure'. `config.h',
3323 `bconfig.h' and `tconfig.h' all contain the `xm-MACHINE.h' header, if
3324 any, appropriate to the host, build and target machines respectively,
3325 the configuration headers for the target, and some definitions; for the
3326 host and build machines, these include the autoconfigured headers
3327 generated by `configure'. The other configuration headers are
3328 determined by `config.gcc'. They also contain the typedefs for `rtx',
3331 * `config.h', for use in programs that run on the host machine.
3333 * `bconfig.h', for use in programs that run on the build machine.
3335 * `tconfig.h', for use in programs and libraries for the target
3338 * `tm_p.h', which includes the header `MACHINE-protos.h' that
3339 contains prototypes for functions in the target `.c' file. FIXME:
3340 why is such a separate header necessary?
3343 File: gccint.info, Node: Build, Next: Makefile, Prev: Configuration, Up: gcc Directory
3345 6.3.3 Build System in the `gcc' Directory
3346 -----------------------------------------
3348 FIXME: describe the build system, including what is built in what
3349 stages. Also list the various source files that are used in the build
3350 process but aren't source files of GCC itself and so aren't documented
3351 below (*note Passes::).
3354 File: gccint.info, Node: Makefile, Next: Library Files, Prev: Build, Up: gcc Directory
3356 6.3.4 Makefile Targets
3357 ----------------------
3359 These targets are available from the `gcc' directory:
3362 This is the default target. Depending on what your
3363 build/host/target configuration is, it coordinates all the things
3364 that need to be built.
3367 Produce info-formatted documentation and man pages. Essentially it
3368 calls `make man' and `make info'.
3371 Produce DVI-formatted documentation.
3374 Produce PDF-formatted documentation.
3377 Produce HTML-formatted documentation.
3383 Generate info-formatted pages.
3386 Delete the files made while building the compiler.
3389 That, and all the other files built by `make all'.
3392 That, and all the files created by `configure'.
3395 Distclean plus any file that can be generated from other files.
3396 Note that additional tools may be required beyond what is normally
3397 needed to build gcc.
3400 Generates files in the source directory that do not exist in CVS
3401 but should go into a release tarball. One example is
3402 `gcc/java/parse.c' which is generated from the CVS source file
3407 Copies the info-formatted and manpage documentation into the source
3408 directory usually for the purpose of generating a release tarball.
3414 Deletes installed files.
3417 Run the testsuite. This creates a `testsuite' subdirectory that
3418 has various `.sum' and `.log' files containing the results of the
3419 testing. You can run subsets with, for example, `make check-gcc'.
3420 You can specify specific tests by setting RUNTESTFLAGS to be the
3421 name of the `.exp' file, optionally followed by (for some tests)
3422 an equals and a file wildcard, like:
3424 make check-gcc RUNTESTFLAGS="execute.exp=19980413-*"
3426 Note that running the testsuite may require additional tools be
3427 installed, such as TCL or dejagnu.
3429 The toplevel tree from which you start GCC compilation is not the GCC
3430 directory, but rather a complex Makefile that coordinates the various
3431 steps of the build, including bootstrapping the compiler and using the
3432 new compiler to build target libraries.
3434 When GCC is configured for a native configuration, the default action
3435 for `make' is to do a full three-stage bootstrap. This means that GCC
3436 is built three times--once with the native compiler, once with the
3437 native-built compiler it just built, and once with the compiler it
3438 built the second time. In theory, the last two should produce the same
3439 results, which `make compare' can check. Each stage is configured
3440 separately and compiled into a separate directory, to minimize problems
3441 due to ABI incompatibilities between the native compiler and GCC.
3443 If you do a change, rebuilding will also start from the first stage
3444 and "bubble" up the change through the three stages. Each stage is
3445 taken from its build directory (if it had been built previously),
3446 rebuilt, and copied to its subdirectory. This will allow you to, for
3447 example, continue a bootstrap after fixing a bug which causes the
3448 stage2 build to crash. It does not provide as good coverage of the
3449 compiler as bootstrapping from scratch, but it ensures that the new
3450 code is syntactically correct (e.g., that you did not use GCC extensions
3451 by mistake), and avoids spurious bootstrap comparison failures(1).
3453 Other targets available from the top level include:
3456 Like `bootstrap', except that the various stages are removed once
3457 they're no longer needed. This saves disk space.
3461 Performs only the first two stages of bootstrap. Unlike a
3462 three-stage bootstrap, this does not perform a comparison to test
3463 that the compiler is running properly. Note that the disk space
3464 required by a "lean" bootstrap is approximately independent of the
3467 `stageN-bubble (N = 1...4)'
3468 Rebuild all the stages up to N, with the appropriate flags,
3469 "bubbling" the changes as described above.
3471 `all-stageN (N = 1...4)'
3472 Assuming that stage N has already been built, rebuild it with the
3473 appropriate flags. This is rarely needed.
3476 Remove everything (`make clean') and rebuilds (`make bootstrap').
3479 Compares the results of stages 2 and 3. This ensures that the
3480 compiler is running properly, since it should produce the same
3481 object files regardless of how it itself was compiled.
3484 Builds a compiler with profiling feedback information. For more
3485 information, see *Note Building with profile feedback:
3486 (gccinstall)Building.
3489 Restart a bootstrap, so that everything that was not built with
3490 the system compiler is rebuilt.
3492 `stageN-start (N = 1...4)'
3493 For each package that is bootstrapped, rename directories so that,
3494 for example, `gcc' points to the stageN GCC, compiled with the
3497 You will invoke this target if you need to test or debug the
3498 stageN GCC. If you only need to execute GCC (but you need not run
3499 `make' either to rebuild it or to run test suites), you should be
3500 able to work directly in the `stageN-gcc' directory. This makes
3501 it easier to debug multiple stages in parallel.
3504 For each package that is bootstrapped, relocate its build directory
3505 to indicate its stage. For example, if the `gcc' directory points
3506 to the stage2 GCC, after invoking this target it will be renamed
3510 If you wish to use non-default GCC flags when compiling the stage2 and
3511 stage3 compilers, set `BOOT_CFLAGS' on the command line when doing
3514 Usually, the first stage only builds the languages that the compiler
3515 is written in: typically, C and maybe Ada. If you are debugging a
3516 miscompilation of a different stage2 front-end (for example, of the
3517 Fortran front-end), you may want to have front-ends for other languages
3518 in the first stage as well. To do so, set `STAGE1_LANGUAGES' on the
3519 command line when doing `make'.
3521 For example, in the aforementioned scenario of debugging a Fortran
3522 front-end miscompilation caused by the stage1 compiler, you may need a
3525 make stage2-bubble STAGE1_LANGUAGES=c,fortran
3527 Alternatively, you can use per-language targets to build and test
3528 languages that are not enabled by default in stage1. For example,
3529 `make f951' will build a Fortran compiler even in the stage1 build
3532 ---------- Footnotes ----------
3534 (1) Except if the compiler was buggy and miscompiled some of the files
3535 that were not modified. In this case, it's best to use `make restrap'.
3537 (2) Customarily, the system compiler is also termed the `stage0' GCC.
3540 File: gccint.info, Node: Library Files, Next: Headers, Prev: Makefile, Up: gcc Directory
3542 6.3.5 Library Source Files and Headers under the `gcc' Directory
3543 ----------------------------------------------------------------
3545 FIXME: list here, with explanation, all the C source files and headers
3546 under the `gcc' directory that aren't built into the GCC executable but
3547 rather are part of runtime libraries and object files, such as
3548 `crtstuff.c' and `unwind-dw2.c'. *Note Headers Installed by GCC:
3549 Headers, for more information about the `ginclude' directory.
3552 File: gccint.info, Node: Headers, Next: Documentation, Prev: Library Files, Up: gcc Directory
3554 6.3.6 Headers Installed by GCC
3555 ------------------------------
3557 In general, GCC expects the system C library to provide most of the
3558 headers to be used with it. However, GCC will fix those headers if
3559 necessary to make them work with GCC, and will install some headers
3560 required of freestanding implementations. These headers are installed
3561 in `LIBSUBDIR/include'. Headers for non-C runtime libraries are also
3562 installed by GCC; these are not documented here. (FIXME: document them
3565 Several of the headers GCC installs are in the `ginclude' directory.
3566 These headers, `iso646.h', `stdarg.h', `stdbool.h', and `stddef.h', are
3567 installed in `LIBSUBDIR/include', unless the target Makefile fragment
3568 (*note Target Fragment::) overrides this by setting `USER_H'.
3570 In addition to these headers and those generated by fixing system
3571 headers to work with GCC, some other headers may also be installed in
3572 `LIBSUBDIR/include'. `config.gcc' may set `extra_headers'; this
3573 specifies additional headers under `config' to be installed on some
3576 GCC installs its own version of `<float.h>', from `ginclude/float.h'.
3577 This is done to cope with command-line options that change the
3578 representation of floating point numbers.
3580 GCC also installs its own version of `<limits.h>'; this is generated
3581 from `glimits.h', together with `limitx.h' and `limity.h' if the system
3582 also has its own version of `<limits.h>'. (GCC provides its own header
3583 because it is required of ISO C freestanding implementations, but needs
3584 to include the system header from its own header as well because other
3585 standards such as POSIX specify additional values to be defined in
3586 `<limits.h>'.) The system's `<limits.h>' header is used via
3587 `LIBSUBDIR/include/syslimits.h', which is copied from `gsyslimits.h' if
3588 it does not need fixing to work with GCC; if it needs fixing,
3589 `syslimits.h' is the fixed copy.
3591 GCC can also install `<tgmath.h>'. It will do this when `config.gcc'
3592 sets `use_gcc_tgmath' to `yes'.
3595 File: gccint.info, Node: Documentation, Next: Front End, Prev: Headers, Up: gcc Directory
3597 6.3.7 Building Documentation
3598 ----------------------------
3600 The main GCC documentation is in the form of manuals in Texinfo format.
3601 These are installed in Info format; DVI versions may be generated by
3602 `make dvi', PDF versions by `make pdf', and HTML versions by `make
3603 html'. In addition, some man pages are generated from the Texinfo
3604 manuals, there are some other text files with miscellaneous
3605 documentation, and runtime libraries have their own documentation
3606 outside the `gcc' directory. FIXME: document the documentation for
3607 runtime libraries somewhere.
3611 * Texinfo Manuals:: GCC manuals in Texinfo format.
3612 * Man Page Generation:: Generating man pages from Texinfo manuals.
3613 * Miscellaneous Docs:: Miscellaneous text files with documentation.
3616 File: gccint.info, Node: Texinfo Manuals, Next: Man Page Generation, Up: Documentation
3618 6.3.7.1 Texinfo Manuals
3619 .......................
3621 The manuals for GCC as a whole, and the C and C++ front ends, are in
3622 files `doc/*.texi'. Other front ends have their own manuals in files
3623 `LANGUAGE/*.texi'. Common files `doc/include/*.texi' are provided
3624 which may be included in multiple manuals; the following files are in
3628 The GNU Free Documentation License.
3631 The section "Funding Free Software".
3634 Common definitions for manuals.
3638 The GNU General Public License.
3641 A copy of `texinfo.tex' known to work with the GCC manuals.
3643 DVI-formatted manuals are generated by `make dvi', which uses
3644 `texi2dvi' (via the Makefile macro `$(TEXI2DVI)'). PDF-formatted
3645 manuals are generated by `make pdf', which uses `texi2pdf' (via the
3646 Makefile macro `$(TEXI2PDF)'). HTML formatted manuals are generated by
3647 `make html'. Info manuals are generated by `make info' (which is run
3648 as part of a bootstrap); this generates the manuals in the source
3649 directory, using `makeinfo' via the Makefile macro `$(MAKEINFO)', and
3650 they are included in release distributions.
3652 Manuals are also provided on the GCC web site, in both HTML and
3653 PostScript forms. This is done via the script
3654 `maintainer-scripts/update_web_docs'. Each manual to be provided
3655 online must be listed in the definition of `MANUALS' in that file; a
3656 file `NAME.texi' must only appear once in the source tree, and the
3657 output manual must have the same name as the source file. (However,
3658 other Texinfo files, included in manuals but not themselves the root
3659 files of manuals, may have names that appear more than once in the
3660 source tree.) The manual file `NAME.texi' should only include other
3661 files in its own directory or in `doc/include'. HTML manuals will be
3662 generated by `makeinfo --html', PostScript manuals by `texi2dvi' and
3663 `dvips', and PDF manuals by `texi2pdf'. All Texinfo files that are
3664 parts of manuals must be checked into SVN, even if they are generated
3665 files, for the generation of online manuals to work.
3667 The installation manual, `doc/install.texi', is also provided on the
3668 GCC web site. The HTML version is generated by the script
3669 `doc/install.texi2html'.
3672 File: gccint.info, Node: Man Page Generation, Next: Miscellaneous Docs, Prev: Texinfo Manuals, Up: Documentation
3674 6.3.7.2 Man Page Generation
3675 ...........................
3677 Because of user demand, in addition to full Texinfo manuals, man pages
3678 are provided which contain extracts from those manuals. These man
3679 pages are generated from the Texinfo manuals using
3680 `contrib/texi2pod.pl' and `pod2man'. (The man page for `g++',
3681 `cp/g++.1', just contains a `.so' reference to `gcc.1', but all the
3682 other man pages are generated from Texinfo manuals.)
3684 Because many systems may not have the necessary tools installed to
3685 generate the man pages, they are only generated if the `configure'
3686 script detects that recent enough tools are installed, and the
3687 Makefiles allow generating man pages to fail without aborting the
3688 build. Man pages are also included in release distributions. They are
3689 generated in the source directory.
3691 Magic comments in Texinfo files starting `@c man' control what parts
3692 of a Texinfo file go into a man page. Only a subset of Texinfo is
3693 supported by `texi2pod.pl', and it may be necessary to add support for
3694 more Texinfo features to this script when generating new man pages. To
3695 improve the man page output, some special Texinfo macros are provided
3696 in `doc/include/gcc-common.texi' which `texi2pod.pl' understands:
3699 Use in the form `@table @gcctabopt' for tables of options, where
3700 for printed output the effect of `@code' is better than that of
3701 `@option' but for man page output a different effect is wanted.
3704 Use for summary lists of options in manuals.
3707 Use at the end of each line inside `@gccoptlist'. This is
3708 necessary to avoid problems with differences in how the
3709 `@gccoptlist' macro is handled by different Texinfo formatters.
3711 FIXME: describe the `texi2pod.pl' input language and magic comments in
3715 File: gccint.info, Node: Miscellaneous Docs, Prev: Man Page Generation, Up: Documentation
3717 6.3.7.3 Miscellaneous Documentation
3718 ...................................
3720 In addition to the formal documentation that is installed by GCC, there
3721 are several other text files with miscellaneous documentation:
3724 Notes on GCC's Native Language Support. FIXME: this should be
3725 part of this manual rather than a separate file.
3728 Notes on the Free Translation Project.
3731 The GNU General Public License.
3734 The GNU Lesser General Public License.
3738 Change log files for various parts of GCC.
3741 Details of a few changes to the GCC front-end interface. FIXME:
3742 the information in this file should be part of general
3743 documentation of the front-end interface in this manual.
3746 Information about new features in old versions of GCC. (For recent
3747 versions, the information is on the GCC web site.)
3749 `README.Portability'
3750 Information about portability issues when writing code in GCC.
3751 FIXME: why isn't this part of this manual or of the GCC Coding
3754 FIXME: document such files in subdirectories, at least `config', `cp',
3755 `objc', `testsuite'.
3758 File: gccint.info, Node: Front End, Next: Back End, Prev: Documentation, Up: gcc Directory
3760 6.3.8 Anatomy of a Language Front End
3761 -------------------------------------
3763 A front end for a language in GCC has the following parts:
3765 * A directory `LANGUAGE' under `gcc' containing source files for
3766 that front end. *Note The Front End `LANGUAGE' Directory: Front
3767 End Directory, for details.
3769 * A mention of the language in the list of supported languages in
3770 `gcc/doc/install.texi'.
3772 * A mention of the name under which the language's runtime library is
3773 recognized by `--enable-shared=PACKAGE' in the documentation of
3774 that option in `gcc/doc/install.texi'.
3776 * A mention of any special prerequisites for building the front end
3777 in the documentation of prerequisites in `gcc/doc/install.texi'.
3779 * Details of contributors to that front end in
3780 `gcc/doc/contrib.texi'. If the details are in that front end's
3781 own manual then there should be a link to that manual's list in
3784 * Information about support for that language in
3785 `gcc/doc/frontends.texi'.
3787 * Information about standards for that language, and the front end's
3788 support for them, in `gcc/doc/standards.texi'. This may be a link
3789 to such information in the front end's own manual.
3791 * Details of source file suffixes for that language and `-x LANG'
3792 options supported, in `gcc/doc/invoke.texi'.
3794 * Entries in `default_compilers' in `gcc.c' for source file suffixes
3797 * Preferably testsuites, which may be under `gcc/testsuite' or
3798 runtime library directories. FIXME: document somewhere how to
3799 write testsuite harnesses.
3801 * Probably a runtime library for the language, outside the `gcc'
3802 directory. FIXME: document this further.
3804 * Details of the directories of any runtime libraries in
3805 `gcc/doc/sourcebuild.texi'.
3807 If the front end is added to the official GCC source repository, the
3808 following are also necessary:
3810 * At least one Bugzilla component for bugs in that front end and
3811 runtime libraries. This category needs to be mentioned in
3812 `gcc/gccbug.in', as well as being added to the Bugzilla database.
3814 * Normally, one or more maintainers of that front end listed in
3817 * Mentions on the GCC web site in `index.html' and `frontends.html',
3818 with any relevant links on `readings.html'. (Front ends that are
3819 not an official part of GCC may also be listed on
3820 `frontends.html', with relevant links.)
3822 * A news item on `index.html', and possibly an announcement on the
3823 <gcc-announce@gcc.gnu.org> mailing list.
3825 * The front end's manuals should be mentioned in
3826 `maintainer-scripts/update_web_docs' (*note Texinfo Manuals::) and
3827 the online manuals should be linked to from
3828 `onlinedocs/index.html'.
3830 * Any old releases or CVS repositories of the front end, before its
3831 inclusion in GCC, should be made available on the GCC FTP site
3832 `ftp://gcc.gnu.org/pub/gcc/old-releases/'.
3834 * The release and snapshot script `maintainer-scripts/gcc_release'
3835 should be updated to generate appropriate tarballs for this front
3836 end. The associated `maintainer-scripts/snapshot-README' and
3837 `maintainer-scripts/snapshot-index.html' files should be updated
3838 to list the tarballs and diffs for this front end.
3840 * If this front end includes its own version files that include the
3841 current date, `maintainer-scripts/update_version' should be
3842 updated accordingly.
3846 * Front End Directory:: The front end `LANGUAGE' directory.
3847 * Front End Config:: The front end `config-lang.in' file.
3850 File: gccint.info, Node: Front End Directory, Next: Front End Config, Up: Front End
3852 6.3.8.1 The Front End `LANGUAGE' Directory
3853 ..........................................
3855 A front end `LANGUAGE' directory contains the source files of that
3856 front end (but not of any runtime libraries, which should be outside
3857 the `gcc' directory). This includes documentation, and possibly some
3858 subsidiary programs build alongside the front end. Certain files are
3859 special and other parts of the compiler depend on their names:
3862 This file is required in all language subdirectories. *Note The
3863 Front End `config-lang.in' File: Front End Config, for details of
3867 This file is required in all language subdirectories. It contains
3868 targets `LANG.HOOK' (where `LANG' is the setting of `language' in
3869 `config-lang.in') for the following values of `HOOK', and any
3870 other Makefile rules required to build those targets (which may if
3871 necessary use other Makefiles specified in `outputs' in
3872 `config-lang.in', although this is deprecated). It also adds any
3873 testsuite targets that can use the standard rule in
3874 `gcc/Makefile.in' to the variable `lang_checks'.
3879 FIXME: exactly what goes in each of these targets?
3882 Build an `etags' `TAGS' file in the language subdirectory in
3886 Build info documentation for the front end, in the build
3887 directory. This target is only called by `make bootstrap' if
3888 a suitable version of `makeinfo' is available, so does not
3889 need to check for this, and should fail if an error occurs.
3892 Build DVI documentation for the front end, in the build
3893 directory. This should be done using `$(TEXI2DVI)', with
3894 appropriate `-I' arguments pointing to directories of
3898 Build PDF documentation for the front end, in the build
3899 directory. This should be done using `$(TEXI2PDF)', with
3900 appropriate `-I' arguments pointing to directories of
3904 Build HTML documentation for the front end, in the build
3908 Build generated man pages for the front end from Texinfo
3909 manuals (*note Man Page Generation::), in the build
3910 directory. This target is only called if the necessary tools
3911 are available, but should ignore errors so as not to stop the
3912 build if errors occur; man pages are optional and the tools
3913 involved may be installed in a broken way.
3916 Install everything that is part of the front end, apart from
3917 the compiler executables listed in `compilers' in
3921 Install info documentation for the front end, if it is
3922 present in the source directory. This target should have
3923 dependencies on info files that should be installed.
3926 Install man pages for the front end. This target should
3930 Install headers needed for plugins.
3933 Copies its dependencies into the source directory. This
3934 generally should be used for generated files such as Bison
3935 output files which are not present in CVS, but should be
3936 included in any release tarballs. This target will be
3937 executed during a bootstrap if
3938 `--enable-generated-files-in-srcdir' was specified as a
3943 Copies its dependencies into the source directory. These
3944 targets will be executed during a bootstrap if
3945 `--enable-generated-files-in-srcdir' was specified as a
3949 Uninstall files installed by installing the compiler. This is
3950 currently documented not to be supported, so the hook need
3957 The language parts of the standard GNU `*clean' targets.
3958 *Note Standard Targets for Users: (standards)Standard
3959 Targets, for details of the standard targets. For GCC,
3960 `maintainer-clean' should delete all generated files in the
3961 source directory that are not checked into CVS, but should
3962 not delete anything checked into CVS.
3964 `Make-lang.in' must also define a variable `LANG_OBJS' to a list
3965 of host object files that are used by that language.
3968 This file registers the set of switches that the front end accepts
3969 on the command line, and their `--help' text. *Note Options::.
3972 This file provides entries for `default_compilers' in `gcc.c'
3973 which override the default of giving an error that a compiler for
3974 that language is not installed.
3977 This file, which need not exist, defines any language-specific tree
3981 File: gccint.info, Node: Front End Config, Prev: Front End Directory, Up: Front End
3983 6.3.8.2 The Front End `config-lang.in' File
3984 ...........................................
3986 Each language subdirectory contains a `config-lang.in' file. In
3987 addition the main directory contains `c-config-lang.in', which contains
3988 limited information for the C language. This file is a shell script
3989 that may define some variables describing the language:
3992 This definition must be present, and gives the name of the language
3993 for some purposes such as arguments to `--enable-languages'.
3996 If defined, this variable lists (space-separated) language front
3997 ends other than C that this front end requires to be enabled (with
3998 the names given being their `language' settings). For example, the
3999 Java front end depends on the C++ front end, so sets
4000 `lang_requires=c++'.
4003 If defined, this variable lists (space-separated) front end
4004 directories other than C that this front end requires to be
4005 present. For example, the Objective-C++ front end uses source
4006 files from the C++ and Objective-C front ends, so sets
4007 `subdir_requires="cp objc"'.
4010 If defined, this variable lists (space-separated) targets in the
4011 top level `Makefile' to build the runtime libraries for this
4012 language, such as `target-libobjc'.
4015 If defined, this variable lists (space-separated) top level
4016 directories (parallel to `gcc'), apart from the runtime libraries,
4017 that should not be configured if this front end is not built.
4020 If defined to `no', this language front end is not built unless
4021 enabled in a `--enable-languages' argument. Otherwise, front ends
4022 are built by default, subject to any special logic in
4023 `configure.ac' (as is present to disable the Ada front end if the
4024 Ada compiler is not already installed).
4027 If defined to `yes', this front end is built in stage 1 of the
4028 bootstrap. This is only relevant to front ends written in their
4032 If defined, a space-separated list of compiler executables that
4033 will be run by the driver. The names here will each end with
4037 If defined, a space-separated list of files that should be
4038 generated by `configure' substituting values in them. This
4039 mechanism can be used to create a file `LANGUAGE/Makefile' from
4040 `LANGUAGE/Makefile.in', but this is deprecated, building
4041 everything from the single `gcc/Makefile' is preferred.
4044 If defined, a space-separated list of files that should be scanned
4045 by gengtype.c to generate the garbage collection tables and
4046 routines for this language. This excludes the files that are
4047 common to all front ends. *Note Type Information::.
4051 File: gccint.info, Node: Back End, Prev: Front End, Up: gcc Directory
4053 6.3.9 Anatomy of a Target Back End
4054 ----------------------------------
4056 A back end for a target architecture in GCC has the following parts:
4058 * A directory `MACHINE' under `gcc/config', containing a machine
4059 description `MACHINE.md' file (*note Machine Descriptions: Machine
4060 Desc.), header files `MACHINE.h' and `MACHINE-protos.h' and a
4061 source file `MACHINE.c' (*note Target Description Macros and
4062 Functions: Target Macros.), possibly a target Makefile fragment
4063 `t-MACHINE' (*note The Target Makefile Fragment: Target
4064 Fragment.), and maybe some other files. The names of these files
4065 may be changed from the defaults given by explicit specifications
4068 * If necessary, a file `MACHINE-modes.def' in the `MACHINE'
4069 directory, containing additional machine modes to represent
4070 condition codes. *Note Condition Code::, for further details.
4072 * An optional `MACHINE.opt' file in the `MACHINE' directory,
4073 containing a list of target-specific options. You can also add
4074 other option files using the `extra_options' variable in
4075 `config.gcc'. *Note Options::.
4077 * Entries in `config.gcc' (*note The `config.gcc' File: System
4078 Config.) for the systems with this target architecture.
4080 * Documentation in `gcc/doc/invoke.texi' for any command-line
4081 options supported by this target (*note Run-time Target
4082 Specification: Run-time Target.). This means both entries in the
4083 summary table of options and details of the individual options.
4085 * Documentation in `gcc/doc/extend.texi' for any target-specific
4086 attributes supported (*note Defining target-specific uses of
4087 `__attribute__': Target Attributes.), including where the same
4088 attribute is already supported on some targets, which are
4089 enumerated in the manual.
4091 * Documentation in `gcc/doc/extend.texi' for any target-specific
4094 * Documentation in `gcc/doc/extend.texi' of any target-specific
4095 built-in functions supported.
4097 * Documentation in `gcc/doc/extend.texi' of any target-specific
4098 format checking styles supported.
4100 * Documentation in `gcc/doc/md.texi' of any target-specific
4101 constraint letters (*note Constraints for Particular Machines:
4102 Machine Constraints.).
4104 * A note in `gcc/doc/contrib.texi' under the person or people who
4105 contributed the target support.
4107 * Entries in `gcc/doc/install.texi' for all target triplets
4108 supported with this target architecture, giving details of any
4109 special notes about installation for this target, or saying that
4110 there are no special notes if there are none.
4112 * Possibly other support outside the `gcc' directory for runtime
4113 libraries. FIXME: reference docs for this. The libstdc++ porting
4114 manual needs to be installed as info for this to work, or to be a
4115 chapter of this manual.
4117 If the back end is added to the official GCC source repository, the
4118 following are also necessary:
4120 * An entry for the target architecture in `readings.html' on the GCC
4121 web site, with any relevant links.
4123 * Details of the properties of the back end and target architecture
4124 in `backends.html' on the GCC web site.
4126 * A news item about the contribution of support for that target
4127 architecture, in `index.html' on the GCC web site.
4129 * Normally, one or more maintainers of that target listed in
4130 `MAINTAINERS'. Some existing architectures may be unmaintained,
4131 but it would be unusual to add support for a target that does not
4132 have a maintainer when support is added.
4135 File: gccint.info, Node: Testsuites, Prev: gcc Directory, Up: Source Tree
4140 GCC contains several testsuites to help maintain compiler quality.
4141 Most of the runtime libraries and language front ends in GCC have
4142 testsuites. Currently only the C language testsuites are documented
4143 here; FIXME: document the others.
4147 * Test Idioms:: Idioms used in testsuite code.
4148 * Test Directives:: Directives used within DejaGnu tests.
4149 * Ada Tests:: The Ada language testsuites.
4150 * C Tests:: The C language testsuites.
4151 * libgcj Tests:: The Java library testsuites.
4152 * gcov Testing:: Support for testing gcov.
4153 * profopt Testing:: Support for testing profile-directed optimizations.
4154 * compat Testing:: Support for testing binary compatibility.
4155 * Torture Tests:: Support for torture testing using multiple options.
4158 File: gccint.info, Node: Test Idioms, Next: Test Directives, Up: Testsuites
4160 6.4.1 Idioms Used in Testsuite Code
4161 -----------------------------------
4163 In general, C testcases have a trailing `-N.c', starting with `-1.c',
4164 in case other testcases with similar names are added later. If the
4165 test is a test of some well-defined feature, it should have a name
4166 referring to that feature such as `FEATURE-1.c'. If it does not test a
4167 well-defined feature but just happens to exercise a bug somewhere in
4168 the compiler, and a bug report has been filed for this bug in the GCC
4169 bug database, `prBUG-NUMBER-1.c' is the appropriate form of name.
4170 Otherwise (for miscellaneous bugs not filed in the GCC bug database),
4171 and previously more generally, test cases are named after the date on
4172 which they were added. This allows people to tell at a glance whether
4173 a test failure is because of a recently found bug that has not yet been
4174 fixed, or whether it may be a regression, but does not give any other
4175 information about the bug or where discussion of it may be found. Some
4176 other language testsuites follow similar conventions.
4178 In the `gcc.dg' testsuite, it is often necessary to test that an error
4179 is indeed a hard error and not just a warning--for example, where it is
4180 a constraint violation in the C standard, which must become an error
4181 with `-pedantic-errors'. The following idiom, where the first line
4182 shown is line LINE of the file and the line that generates the error,
4185 /* { dg-bogus "warning" "warning in place of error" } */
4186 /* { dg-error "REGEXP" "MESSAGE" { target *-*-* } LINE } */
4188 It may be necessary to check that an expression is an integer constant
4189 expression and has a certain value. To check that `E' has value `V',
4190 an idiom similar to the following is used:
4192 char x[((E) == (V) ? 1 : -1)];
4194 In `gcc.dg' tests, `__typeof__' is sometimes used to make assertions
4195 about the types of expressions. See, for example,
4196 `gcc.dg/c99-condexpr-1.c'. The more subtle uses depend on the exact
4197 rules for the types of conditional expressions in the C standard; see,
4198 for example, `gcc.dg/c99-intconst-1.c'.
4200 It is useful to be able to test that optimizations are being made
4201 properly. This cannot be done in all cases, but it can be done where
4202 the optimization will lead to code being optimized away (for example,
4203 where flow analysis or alias analysis should show that certain code
4204 cannot be called) or to functions not being called because they have
4205 been expanded as built-in functions. Such tests go in
4206 `gcc.c-torture/execute'. Where code should be optimized away, a call
4207 to a nonexistent function such as `link_failure ()' may be inserted; a
4210 #ifndef __OPTIMIZE__
4218 will also be needed so that linking still succeeds when the test is run
4219 without optimization. When all calls to a built-in function should
4220 have been optimized and no calls to the non-built-in version of the
4221 function should remain, that function may be defined as `static' to
4222 call `abort ()' (although redeclaring a function as static may not work
4225 All testcases must be portable. Target-specific testcases must have
4226 appropriate code to avoid causing failures on unsupported systems;
4227 unfortunately, the mechanisms for this differ by directory.
4229 FIXME: discuss non-C testsuites here.
4232 File: gccint.info, Node: Test Directives, Next: Ada Tests, Prev: Test Idioms, Up: Testsuites
4234 6.4.2 Directives used within DejaGnu tests
4235 ------------------------------------------
4237 Test directives appear within comments in a test source file and begin
4238 with `dg-'. Some of these are defined within DejaGnu and others are
4239 local to the GCC testsuite.
4241 The order in which test directives appear in a test can be important:
4242 directives local to GCC sometimes override information used by the
4243 DejaGnu directives, which know nothing about the GCC directives, so the
4244 DejaGnu directives must precede GCC directives.
4246 Several test directives include selectors which are usually preceded by
4247 the keyword `target' or `xfail'. A selector is: one or more target
4248 triplets, possibly including wildcard characters; a single
4249 effective-target keyword; or a logical expression. Depending on the
4250 context, the selector specifies whether a test is skipped and reported
4251 as unsupported or is expected to fail. Use `*-*-*' to match any target.
4252 Effective-target keywords are defined in `target-supports.exp' in the
4255 A selector expression appears within curly braces and uses a single
4256 logical operator: one of `!', `&&', or `||'. An operand is another
4257 selector expression, an effective-target keyword, a single target
4258 triplet, or a list of target triplets within quotes or curly braces.
4261 { target { ! "hppa*-*-* ia64*-*-*" } }
4262 { target { powerpc*-*-* && lp64 } }
4263 { xfail { lp64 || vect_no_align } }
4265 `{ dg-do DO-WHAT-KEYWORD [{ target/xfail SELECTOR }] }'
4266 DO-WHAT-KEYWORD specifies how the test is compiled and whether it
4267 is executed. It is one of:
4270 Compile with `-E' to run only the preprocessor.
4273 Compile with `-S' to produce an assembly code file.
4276 Compile with `-c' to produce a relocatable object file.
4279 Compile, assemble, and link to produce an executable file.
4282 Produce and run an executable file, which is expected to
4283 return an exit code of 0.
4285 The default is `compile'. That can be overridden for a set of
4286 tests by redefining `dg-do-what-default' within the `.exp' file
4289 If the directive includes the optional `{ target SELECTOR }' then
4290 the test is skipped unless the target system is included in the
4291 list of target triplets or matches the effective-target keyword.
4293 If `do-what-keyword' is `run' and the directive includes the
4294 optional `{ xfail SELECTOR }' and the selector is met then the
4295 test is expected to fail. The `xfail' clause is ignored for other
4296 values of `do-what-keyword'; those tests can use directive
4299 `{ dg-options OPTIONS [{ target SELECTOR }] }'
4300 This DejaGnu directive provides a list of compiler options, to be
4301 used if the target system matches SELECTOR, that replace the
4302 default options used for this set of tests.
4304 `{ dg-add-options FEATURE ... }'
4305 Add any compiler options that are needed to access certain
4306 features. This directive does nothing on targets that enable the
4307 features by default, or that don't provide them at all. It must
4308 come after all `dg-options' directives.
4310 The supported values of FEATURE are:
4312 The target's C99 runtime (both headers and libraries).
4315 `mips16' function attributes. Only MIPS targets support this
4316 feature, and only then in certain modes.
4318 `{ dg-timeout N [{target SELECTOR }] }'
4319 Set the time limit for the compilation and for the execution of
4320 the test to the specified number of seconds.
4322 `{ dg-timeout-factor X [{ target SELECTOR }] }'
4323 Multiply the normal time limit for compilation and execution of
4324 the test by the specified floating-point factor. The normal
4325 timeout limit, in seconds, is found by searching the following in
4328 * the value defined by an earlier `dg-timeout' directive in the
4331 * variable TOOL_TIMEOUT defined by the set of tests
4333 * GCC,TIMEOUT set in the target board
4337 `{ dg-skip-if COMMENT { SELECTOR } { INCLUDE-OPTS } { EXCLUDE-OPTS } }'
4338 Skip the test if the test system is included in SELECTOR and if
4339 each of the options in INCLUDE-OPTS is in the set of options with
4340 which the test would be compiled and if none of the options in
4341 EXCLUDE-OPTS is in the set of options with which the test would be
4344 Use `"*"' for an empty INCLUDE-OPTS list and `""' for an empty
4347 `{ dg-xfail-if COMMENT { SELECTOR } { INCLUDE-OPTS } { EXCLUDE-OPTS } }'
4348 Expect the test to fail if the conditions (which are the same as
4349 for `dg-skip-if') are met. This does not affect the execute step.
4351 `{ dg-xfail-run-if COMMENT { SELECTOR } { INCLUDE-OPTS } { EXCLUDE-OPTS } }'
4352 Expect the execute step of a test to fail if the conditions (which
4353 are the same as for `dg-skip-if') and `dg-xfail-if') are met.
4355 `{ dg-require-SUPPORT args }'
4356 Skip the test if the target does not provide the required support;
4357 see `gcc-dg.exp' in the GCC testsuite for the actual directives.
4358 These directives must appear after any `dg-do' directive in the
4359 test and before any `dg-additional-sources' directive. They
4360 require at least one argument, which can be an empty string if the
4361 specific procedure does not examine the argument.
4363 `{ dg-require-effective-target KEYWORD }'
4364 Skip the test if the test target, including current multilib flags,
4365 is not covered by the effective-target keyword. This directive
4366 must appear after any `dg-do' directive in the test and before any
4367 `dg-additional-sources' directive.
4369 `{ dg-shouldfail COMMENT { SELECTOR } { INCLUDE-OPTS } { EXCLUDE-OPTS } }'
4370 Expect the test executable to return a nonzero exit status if the
4371 conditions (which are the same as for `dg-skip-if') are met.
4373 `{ dg-error REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] }]] }'
4374 This DejaGnu directive appears on a source line that is expected
4375 to get an error message, or else specifies the source line
4376 associated with the message. If there is no message for that line
4377 or if the text of that message is not matched by REGEXP then the
4378 check fails and COMMENT is included in the `FAIL' message. The
4379 check does not look for the string `"error"' unless it is part of
4382 `{ dg-warning REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] }]] }'
4383 This DejaGnu directive appears on a source line that is expected
4384 to get a warning message, or else specifies the source line
4385 associated with the message. If there is no message for that line
4386 or if the text of that message is not matched by REGEXP then the
4387 check fails and COMMENT is included in the `FAIL' message. The
4388 check does not look for the string `"warning"' unless it is part
4391 `{ dg-message REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] }]] }'
4392 The line is expected to get a message other than an error or
4393 warning. If there is no message for that line or if the text of
4394 that message is not matched by REGEXP then the check fails and
4395 COMMENT is included in the `FAIL' message.
4397 `{ dg-bogus REGEXP [COMMENT [{ target/xfail SELECTOR } [LINE] }]] }'
4398 This DejaGnu directive appears on a source line that should not
4399 get a message matching REGEXP, or else specifies the source line
4400 associated with the bogus message. It is usually used with `xfail'
4401 to indicate that the message is a known problem for a particular
4404 `{ dg-excess-errors COMMENT [{ target/xfail SELECTOR }] }'
4405 This DejaGnu directive indicates that the test is expected to fail
4406 due to compiler messages that are not handled by `dg-error',
4407 `dg-warning' or `dg-bogus'. For this directive `xfail' has the
4408 same effect as `target'.
4410 `{ dg-output REGEXP [{ target/xfail SELECTOR }] }'
4411 This DejaGnu directive compares REGEXP to the combined output that
4412 the test executable writes to `stdout' and `stderr'.
4414 `{ dg-prune-output REGEXP }'
4415 Prune messages matching REGEXP from test output.
4417 `{ dg-additional-files "FILELIST" }'
4418 Specify additional files, other than source files, that must be
4419 copied to the system where the compiler runs.
4421 `{ dg-additional-sources "FILELIST" }'
4422 Specify additional source files to appear in the compile line
4423 following the main test file.
4425 `{ dg-final { LOCAL-DIRECTIVE } }'
4426 This DejaGnu directive is placed within a comment anywhere in the
4427 source file and is processed after the test has been compiled and
4428 run. Multiple `dg-final' commands are processed in the order in
4429 which they appear in the source file.
4431 The GCC testsuite defines the following directives to be used
4434 `cleanup-coverage-files'
4435 Removes coverage data files generated for this test.
4437 `cleanup-repo-files'
4438 Removes files generated for this test for `-frepo'.
4440 `cleanup-rtl-dump SUFFIX'
4441 Removes RTL dump files generated for this test.
4443 `cleanup-tree-dump SUFFIX'
4444 Removes tree dump files matching SUFFIX which were generated
4447 `cleanup-saved-temps'
4448 Removes files for the current test which were kept for
4451 `scan-file FILENAME REGEXP [{ target/xfail SELECTOR }]'
4452 Passes if REGEXP matches text in FILENAME.
4454 `scan-file-not FILENAME REGEXP [{ target/xfail SELECTOR }]'
4455 Passes if REGEXP does not match text in FILENAME.
4457 `scan-hidden SYMBOL [{ target/xfail SELECTOR }]'
4458 Passes if SYMBOL is defined as a hidden symbol in the test's
4461 `scan-not-hidden SYMBOL [{ target/xfail SELECTOR }]'
4462 Passes if SYMBOL is not defined as a hidden symbol in the
4463 test's assembly output.
4465 `scan-assembler-times REGEX NUM [{ target/xfail SELECTOR }]'
4466 Passes if REGEX is matched exactly NUM times in the test's
4469 `scan-assembler REGEX [{ target/xfail SELECTOR }]'
4470 Passes if REGEX matches text in the test's assembler output.
4472 `scan-assembler-not REGEX [{ target/xfail SELECTOR }]'
4473 Passes if REGEX does not match text in the test's assembler
4476 `scan-assembler-dem REGEX [{ target/xfail SELECTOR }]'
4477 Passes if REGEX matches text in the test's demangled
4480 `scan-assembler-dem-not REGEX [{ target/xfail SELECTOR }]'
4481 Passes if REGEX does not match text in the test's demangled
4484 `scan-tree-dump-times REGEX NUM SUFFIX [{ target/xfail SELECTOR }]'
4485 Passes if REGEX is found exactly NUM times in the dump file
4488 `scan-tree-dump REGEX SUFFIX [{ target/xfail SELECTOR }]'
4489 Passes if REGEX matches text in the dump file with suffix
4492 `scan-tree-dump-not REGEX SUFFIX [{ target/xfail SELECTOR }]'
4493 Passes if REGEX does not match text in the dump file with
4496 `scan-tree-dump-dem REGEX SUFFIX [{ target/xfail SELECTOR }]'
4497 Passes if REGEX matches demangled text in the dump file with
4500 `scan-tree-dump-dem-not REGEX SUFFIX [{ target/xfail SELECTOR }]'
4501 Passes if REGEX does not match demangled text in the dump
4502 file with suffix SUFFIX.
4504 `output-exists [{ target/xfail SELECTOR }]'
4505 Passes if compiler output file exists.
4507 `output-exists-not [{ target/xfail SELECTOR }]'
4508 Passes if compiler output file does not exist.
4510 `run-gcov SOURCEFILE'
4511 Check line counts in `gcov' tests.
4513 `run-gcov [branches] [calls] { OPTS SOURCEFILE }'
4514 Check branch and/or call counts, in addition to line counts,
4518 File: gccint.info, Node: Ada Tests, Next: C Tests, Prev: Test Directives, Up: Testsuites
4520 6.4.3 Ada Language Testsuites
4521 -----------------------------
4523 The Ada testsuite includes executable tests from the ACATS 2.5
4524 testsuite, publicly available at
4525 `http://www.adaic.org/compilers/acats/2.5'
4527 These tests are integrated in the GCC testsuite in the
4528 `gcc/testsuite/ada/acats' directory, and enabled automatically when
4529 running `make check', assuming the Ada language has been enabled when
4532 You can also run the Ada testsuite independently, using `make
4533 check-ada', or run a subset of the tests by specifying which chapter to
4536 $ make check-ada CHAPTERS="c3 c9"
4538 The tests are organized by directory, each directory corresponding to
4539 a chapter of the Ada Reference Manual. So for example, c9 corresponds
4540 to chapter 9, which deals with tasking features of the language.
4542 There is also an extra chapter called `gcc' containing a template for
4543 creating new executable tests.
4545 The tests are run using two `sh' scripts: `run_acats' and
4546 `run_all.sh'. To run the tests using a simulator or a cross target,
4547 see the small customization section at the top of `run_all.sh'.
4549 These tests are run using the build tree: they can be run without doing
4553 File: gccint.info, Node: C Tests, Next: libgcj Tests, Prev: Ada Tests, Up: Testsuites
4555 6.4.4 C Language Testsuites
4556 ---------------------------
4558 GCC contains the following C language testsuites, in the
4559 `gcc/testsuite' directory:
4562 This contains tests of particular features of the C compiler,
4563 using the more modern `dg' harness. Correctness tests for various
4564 compiler features should go here if possible.
4566 Magic comments determine whether the file is preprocessed,
4567 compiled, linked or run. In these tests, error and warning
4568 message texts are compared against expected texts or regular
4569 expressions given in comments. These tests are run with the
4570 options `-ansi -pedantic' unless other options are given in the
4571 test. Except as noted below they are not run with multiple
4572 optimization options.
4575 This subdirectory contains tests for binary compatibility using
4576 `compat.exp', which in turn uses the language-independent support
4577 (*note Support for testing binary compatibility: compat Testing.).
4580 This subdirectory contains tests of the preprocessor.
4583 This subdirectory contains tests for debug formats. Tests in this
4584 subdirectory are run for each debug format that the compiler
4588 This subdirectory contains tests of the `-Wformat' format
4589 checking. Tests in this directory are run with and without
4593 This subdirectory contains tests of code that should not compile
4594 and does not need any special compilation options. They are run
4595 with multiple optimization options, since sometimes invalid code
4596 crashes the compiler with optimization.
4599 FIXME: describe this.
4602 This contains particular code fragments which have historically
4603 broken easily. These tests are run with multiple optimization
4604 options, so tests for features which only break at some
4605 optimization levels belong here. This also contains tests to
4606 check that certain optimizations occur. It might be worthwhile to
4607 separate the correctness tests cleanly from the code quality
4608 tests, but it hasn't been done yet.
4610 `gcc.c-torture/compat'
4611 FIXME: describe this.
4613 This directory should probably not be used for new tests.
4615 `gcc.c-torture/compile'
4616 This testsuite contains test cases that should compile, but do not
4617 need to link or run. These test cases are compiled with several
4618 different combinations of optimization options. All warnings are
4619 disabled for these test cases, so this directory is not suitable if
4620 you wish to test for the presence or absence of compiler warnings.
4621 While special options can be set, and tests disabled on specific
4622 platforms, by the use of `.x' files, mostly these test cases
4623 should not contain platform dependencies. FIXME: discuss how
4624 defines such as `NO_LABEL_VALUES' and `STACK_SIZE' are used.
4626 `gcc.c-torture/execute'
4627 This testsuite contains test cases that should compile, link and
4628 run; otherwise the same comments as for `gcc.c-torture/compile'
4631 `gcc.c-torture/execute/ieee'
4632 This contains tests which are specific to IEEE floating point.
4634 `gcc.c-torture/unsorted'
4635 FIXME: describe this.
4637 This directory should probably not be used for new tests.
4639 `gcc.c-torture/misc-tests'
4640 This directory contains C tests that require special handling.
4641 Some of these tests have individual expect files, and others share
4642 special-purpose expect files:
4645 Test `-fbranch-probabilities' using `bprob.exp', which in
4646 turn uses the generic, language-independent framework (*note
4647 Support for testing profile-directed optimizations: profopt
4651 Test the testsuite itself using `dg-test.exp'.
4654 Test `gcov' output using `gcov.exp', which in turn uses the
4655 language-independent support (*note Support for testing gcov:
4659 Test i386-specific support for data prefetch using
4660 `i386-prefetch.exp'.
4663 FIXME: merge in `testsuite/README.gcc' and discuss the format of test
4664 cases and magic comments more.
4667 File: gccint.info, Node: libgcj Tests, Next: gcov Testing, Prev: C Tests, Up: Testsuites
4669 6.4.5 The Java library testsuites.
4670 ----------------------------------
4672 Runtime tests are executed via `make check' in the
4673 `TARGET/libjava/testsuite' directory in the build tree. Additional
4674 runtime tests can be checked into this testsuite.
4676 Regression testing of the core packages in libgcj is also covered by
4677 the Mauve testsuite. The Mauve Project develops tests for the Java
4678 Class Libraries. These tests are run as part of libgcj testing by
4679 placing the Mauve tree within the libjava testsuite sources at
4680 `libjava/testsuite/libjava.mauve/mauve', or by specifying the location
4681 of that tree when invoking `make', as in `make MAUVEDIR=~/mauve check'.
4683 To detect regressions, a mechanism in `mauve.exp' compares the
4684 failures for a test run against the list of expected failures in
4685 `libjava/testsuite/libjava.mauve/xfails' from the source hierarchy.
4686 Update this file when adding new failing tests to Mauve, or when fixing
4687 bugs in libgcj that had caused Mauve test failures.
4689 We encourage developers to contribute test cases to Mauve.
4692 File: gccint.info, Node: gcov Testing, Next: profopt Testing, Prev: libgcj Tests, Up: Testsuites
4694 6.4.6 Support for testing `gcov'
4695 --------------------------------
4697 Language-independent support for testing `gcov', and for checking that
4698 branch profiling produces expected values, is provided by the expect
4699 file `gcov.exp'. `gcov' tests also rely on procedures in `gcc.dg.exp'
4700 to compile and run the test program. A typical `gcov' test contains
4701 the following DejaGnu commands within comments:
4703 { dg-options "-fprofile-arcs -ftest-coverage" }
4704 { dg-do run { target native } }
4705 { dg-final { run-gcov sourcefile } }
4707 Checks of `gcov' output can include line counts, branch percentages,
4708 and call return percentages. All of these checks are requested via
4709 commands that appear in comments in the test's source file. Commands
4710 to check line counts are processed by default. Commands to check
4711 branch percentages and call return percentages are processed if the
4712 `run-gcov' command has arguments `branches' or `calls', respectively.
4713 For example, the following specifies checking both, as well as passing
4716 { dg-final { run-gcov branches calls { -b sourcefile } } }
4718 A line count command appears within a comment on the source line that
4719 is expected to get the specified count and has the form `count(CNT)'.
4720 A test should only check line counts for lines that will get the same
4721 count for any architecture.
4723 Commands to check branch percentages (`branch') and call return
4724 percentages (`returns') are very similar to each other. A beginning
4725 command appears on or before the first of a range of lines that will
4726 report the percentage, and the ending command follows that range of
4727 lines. The beginning command can include a list of percentages, all of
4728 which are expected to be found within the range. A range is terminated
4729 by the next command of the same kind. A command `branch(end)' or
4730 `returns(end)' marks the end of a range without starting a new one.
4733 if (i > 10 && j > i && j < 20) /* branch(27 50 75) */
4737 For a call return percentage, the value specified is the percentage of
4738 calls reported to return. For a branch percentage, the value is either
4739 the expected percentage or 100 minus that value, since the direction of
4740 a branch can differ depending on the target or the optimization level.
4742 Not all branches and calls need to be checked. A test should not
4743 check for branches that might be optimized away or replaced with
4744 predicated instructions. Don't check for calls inserted by the
4745 compiler or ones that might be inlined or optimized away.
4747 A single test can check for combinations of line counts, branch
4748 percentages, and call return percentages. The command to check a line
4749 count must appear on the line that will report that count, but commands
4750 to check branch percentages and call return percentages can bracket the
4751 lines that report them.
4754 File: gccint.info, Node: profopt Testing, Next: compat Testing, Prev: gcov Testing, Up: Testsuites
4756 6.4.7 Support for testing profile-directed optimizations
4757 --------------------------------------------------------
4759 The file `profopt.exp' provides language-independent support for
4760 checking correct execution of a test built with profile-directed
4761 optimization. This testing requires that a test program be built and
4762 executed twice. The first time it is compiled to generate profile
4763 data, and the second time it is compiled to use the data that was
4764 generated during the first execution. The second execution is to
4765 verify that the test produces the expected results.
4767 To check that the optimization actually generated better code, a test
4768 can be built and run a third time with normal optimizations to verify
4769 that the performance is better with the profile-directed optimizations.
4770 `profopt.exp' has the beginnings of this kind of support.
4772 `profopt.exp' provides generic support for profile-directed
4773 optimizations. Each set of tests that uses it provides information
4774 about a specific optimization:
4777 tool being tested, e.g., `gcc'
4780 options used to generate profile data
4783 options used to optimize using that profile data
4786 suffix of profile data files
4789 list of options with which to run each test, similar to the lists
4793 File: gccint.info, Node: compat Testing, Next: Torture Tests, Prev: profopt Testing, Up: Testsuites
4795 6.4.8 Support for testing binary compatibility
4796 ----------------------------------------------
4798 The file `compat.exp' provides language-independent support for binary
4799 compatibility testing. It supports testing interoperability of two
4800 compilers that follow the same ABI, or of multiple sets of compiler
4801 options that should not affect binary compatibility. It is intended to
4802 be used for testsuites that complement ABI testsuites.
4804 A test supported by this framework has three parts, each in a separate
4805 source file: a main program and two pieces that interact with each
4806 other to split up the functionality being tested.
4808 `TESTNAME_main.SUFFIX'
4809 Contains the main program, which calls a function in file
4810 `TESTNAME_x.SUFFIX'.
4813 Contains at least one call to a function in `TESTNAME_y.SUFFIX'.
4816 Shares data with, or gets arguments from, `TESTNAME_x.SUFFIX'.
4818 Within each test, the main program and one functional piece are
4819 compiled by the GCC under test. The other piece can be compiled by an
4820 alternate compiler. If no alternate compiler is specified, then all
4821 three source files are all compiled by the GCC under test. You can
4822 specify pairs of sets of compiler options. The first element of such a
4823 pair specifies options used with the GCC under test, and the second
4824 element of the pair specifies options used with the alternate compiler.
4825 Each test is compiled with each pair of options.
4827 `compat.exp' defines default pairs of compiler options. These can be
4828 overridden by defining the environment variable `COMPAT_OPTIONS' as:
4830 COMPAT_OPTIONS="[list [list {TST1} {ALT1}]
4831 ...[list {TSTN} {ALTN}]]"
4833 where TSTI and ALTI are lists of options, with TSTI used by the
4834 compiler under test and ALTI used by the alternate compiler. For
4835 example, with `[list [list {-g -O0} {-O3}] [list {-fpic} {-fPIC -O2}]]',
4836 the test is first built with `-g -O0' by the compiler under test and
4837 with `-O3' by the alternate compiler. The test is built a second time
4838 using `-fpic' by the compiler under test and `-fPIC -O2' by the
4841 An alternate compiler is specified by defining an environment variable
4842 to be the full pathname of an installed compiler; for C define
4843 `ALT_CC_UNDER_TEST', and for C++ define `ALT_CXX_UNDER_TEST'. These
4844 will be written to the `site.exp' file used by DejaGnu. The default is
4845 to build each test with the compiler under test using the first of each
4846 pair of compiler options from `COMPAT_OPTIONS'. When
4847 `ALT_CC_UNDER_TEST' or `ALT_CXX_UNDER_TEST' is `same', each test is
4848 built using the compiler under test but with combinations of the
4849 options from `COMPAT_OPTIONS'.
4851 To run only the C++ compatibility suite using the compiler under test
4852 and another version of GCC using specific compiler options, do the
4853 following from `OBJDIR/gcc':
4857 ALT_CXX_UNDER_TEST=${alt_prefix}/bin/g++ \
4858 COMPAT_OPTIONS="lists as shown above" \
4860 RUNTESTFLAGS="compat.exp"
4862 A test that fails when the source files are compiled with different
4863 compilers, but passes when the files are compiled with the same
4864 compiler, demonstrates incompatibility of the generated code or runtime
4865 support. A test that fails for the alternate compiler but passes for
4866 the compiler under test probably tests for a bug that was fixed in the
4867 compiler under test but is present in the alternate compiler.
4869 The binary compatibility tests support a small number of test framework
4870 commands that appear within comments in a test file.
4873 These commands can be used in `TESTNAME_main.SUFFIX' to skip the
4874 test if specific support is not available on the target.
4877 The specified options are used for compiling this particular source
4878 file, appended to the options from `COMPAT_OPTIONS'. When this
4879 command appears in `TESTNAME_main.SUFFIX' the options are also
4880 used to link the test program.
4883 This command can be used in a secondary source file to specify that
4884 compilation is expected to fail for particular options on
4888 File: gccint.info, Node: Torture Tests, Prev: compat Testing, Up: Testsuites
4890 6.4.9 Support for torture testing using multiple options
4891 --------------------------------------------------------
4893 Throughout the compiler testsuite there are several directories whose
4894 tests are run multiple times, each with a different set of options.
4895 These are known as torture tests.
4896 `gcc/testsuite/lib/torture-options.exp' defines procedures to set up
4900 Initialize use of torture lists.
4902 `set-torture-options'
4903 Set lists of torture options to use for tests with and without
4904 loops. Optionally combine a set of torture options with a set of
4905 other options, as is done with Objective-C runtime options.
4908 Finalize use of torture lists.
4910 The `.exp' file for a set of tests that use torture options must
4911 include calls to these three procedures if:
4913 * It calls `gcc-dg-runtest' and overrides DG_TORTURE_OPTIONS.
4915 * It calls ${TOOL}`-torture' or ${TOOL}`-torture-execute', where
4916 TOOL is `c', `fortran', or `objc'.
4918 * It calls `dg-pch'.
4920 It is not necessary for a `.exp' file that calls `gcc-dg-runtest' to
4921 call the torture procedures if the tests should use the list in
4922 DG_TORTURE_OPTIONS defined in `gcc-dg.exp'.
4924 Most uses of torture options can override the default lists by defining
4925 TORTURE_OPTIONS or add to the default list by defining
4926 ADDITIONAL_TORTURE_OPTIONS. Define these in a `.dejagnurc' file or add
4927 them to the `site.exp' file; for example
4929 set ADDITIONAL_TORTURE_OPTIONS [list \
4930 { -O2 -ftree-loop-linear } \
4931 { -O2 -fpeel-loops } ]
4934 File: gccint.info, Node: Options, Next: Passes, Prev: Source Tree, Up: Top
4936 7 Option specification files
4937 ****************************
4939 Most GCC command-line options are described by special option
4940 definition files, the names of which conventionally end in `.opt'.
4941 This chapter describes the format of these files.
4945 * Option file format:: The general layout of the files
4946 * Option properties:: Supported option properties
4949 File: gccint.info, Node: Option file format, Next: Option properties, Up: Options
4951 7.1 Option file format
4952 ======================
4954 Option files are a simple list of records in which each field occupies
4955 its own line and in which the records themselves are separated by blank
4956 lines. Comments may appear on their own line anywhere within the file
4957 and are preceded by semicolons. Whitespace is allowed before the
4960 The files can contain the following types of record:
4962 * A language definition record. These records have two fields: the
4963 string `Language' and the name of the language. Once a language
4964 has been declared in this way, it can be used as an option
4965 property. *Note Option properties::.
4967 * A target specific save record to save additional information. These
4968 records have two fields: the string `TargetSave', and a
4969 declaration type to go in the `cl_target_option' structure.
4971 * An option definition record. These records have the following
4973 1. the name of the option, with the leading "-" removed
4975 2. a space-separated list of option properties (*note Option
4978 3. the help text to use for `--help' (omitted if the second field
4979 contains the `Undocumented' property).
4981 By default, all options beginning with "f", "W" or "m" are
4982 implicitly assumed to take a "no-" form. This form should not be
4983 listed separately. If an option beginning with one of these
4984 letters does not have a "no-" form, you can use the
4985 `RejectNegative' property to reject it.
4987 The help text is automatically line-wrapped before being displayed.
4988 Normally the name of the option is printed on the left-hand side of
4989 the output and the help text is printed on the right. However, if
4990 the help text contains a tab character, the text to the left of
4991 the tab is used instead of the option's name and the text to the
4992 right of the tab forms the help text. This allows you to
4993 elaborate on what type of argument the option takes.
4995 * A target mask record. These records have one field of the form
4996 `Mask(X)'. The options-processing script will automatically
4997 allocate a bit in `target_flags' (*note Run-time Target::) for
4998 each mask name X and set the macro `MASK_X' to the appropriate
4999 bitmask. It will also declare a `TARGET_X' macro that has the
5000 value 1 when bit `MASK_X' is set and 0 otherwise.
5002 They are primarily intended to declare target masks that are not
5003 associated with user options, either because these masks represent
5004 internal switches or because the options are not available on all
5005 configurations and yet the masks always need to be defined.
5008 File: gccint.info, Node: Option properties, Prev: Option file format, Up: Options
5010 7.2 Option properties
5011 =====================
5013 The second field of an option record can specify the following
5017 The option is available for all languages and targets.
5020 The option is available for all languages but is target-specific.
5023 The option is available when compiling for the given language.
5025 It is possible to specify several different languages for the same
5026 option. Each LANGUAGE must have been declared by an earlier
5027 `Language' record. *Note Option file format::.
5030 The option does not have a "no-" form. All options beginning with
5031 "f", "W" or "m" are assumed to have a "no-" form unless this
5034 `Negative(OTHERNAME)'
5035 The option will turn off another option OTHERNAME, which is the
5036 the option name with the leading "-" removed. This chain action
5037 will propagate through the `Negative' property of the option to be
5042 The option takes a mandatory argument. `Joined' indicates that
5043 the option and argument can be included in the same `argv' entry
5044 (as with `-mflush-func=NAME', for example). `Separate' indicates
5045 that the option and argument can be separate `argv' entries (as
5046 with `-o'). An option is allowed to have both of these properties.
5049 The option takes an optional argument. If the argument is given,
5050 it will be part of the same `argv' entry as the option itself.
5052 This property cannot be used alongside `Joined' or `Separate'.
5055 The option's argument is a non-negative integer. The option parser
5056 will check and convert the argument before passing it to the
5057 relevant option handler. `UInteger' should also be used on
5058 options like `-falign-loops' where both `-falign-loops' and
5059 `-falign-loops'=N are supported to make sure the saved options are
5060 given a full integer.
5063 The state of this option should be stored in variable VAR. The
5064 way that the state is stored depends on the type of option:
5066 * If the option uses the `Mask' or `InverseMask' properties,
5067 VAR is the integer variable that contains the mask.
5069 * If the option is a normal on/off switch, VAR is an integer
5070 variable that is nonzero when the option is enabled. The
5071 options parser will set the variable to 1 when the positive
5072 form of the option is used and 0 when the "no-" form is used.
5074 * If the option takes an argument and has the `UInteger'
5075 property, VAR is an integer variable that stores the value of
5078 * Otherwise, if the option takes an argument, VAR is a pointer
5079 to the argument string. The pointer will be null if the
5080 argument is optional and wasn't given.
5082 The option-processing script will usually declare VAR in
5083 `options.c' and leave it to be zero-initialized at start-up time.
5084 You can modify this behavior using `VarExists' and `Init'.
5087 The option controls an integer variable VAR and is active when VAR
5088 equals SET. The option parser will set VAR to SET when the
5089 positive form of the option is used and `!SET' when the "no-" form
5092 VAR is declared in the same way as for the single-argument form
5096 The variable specified by the `Var' property already exists. No
5097 definition should be added to `options.c' in response to this
5100 You should use this property only if the variable is declared
5101 outside `options.c'.
5104 The variable specified by the `Var' property should be statically
5105 initialized to VALUE.
5108 The option is associated with a bit in the `target_flags' variable
5109 (*note Run-time Target::) and is active when that bit is set. You
5110 may also specify `Var' to select a variable other than
5113 The options-processing script will automatically allocate a unique
5114 bit for the option. If the option is attached to `target_flags',
5115 the script will set the macro `MASK_NAME' to the appropriate
5116 bitmask. It will also declare a `TARGET_NAME' macro that has the
5117 value 1 when the option is active and 0 otherwise. If you use
5118 `Var' to attach the option to a different variable, the associated
5119 macros are called `OPTION_MASK_NAME' and `OPTION_NAME'
5122 You can disable automatic bit allocation using `MaskExists'.
5124 `InverseMask(OTHERNAME)'
5125 `InverseMask(OTHERNAME, THISNAME)'
5126 The option is the inverse of another option that has the
5127 `Mask(OTHERNAME)' property. If THISNAME is given, the
5128 options-processing script will declare a `TARGET_THISNAME' macro
5129 that is 1 when the option is active and 0 otherwise.
5132 The mask specified by the `Mask' property already exists. No
5133 `MASK' or `TARGET' definitions should be added to `options.h' in
5134 response to this option record.
5136 The main purpose of this property is to support synonymous options.
5137 The first option should use `Mask(NAME)' and the others should use
5138 `Mask(NAME) MaskExists'.
5141 The state of the option should be printed by `-fverbose-asm'.
5144 The option is deliberately missing documentation and should not be
5145 included in the `--help' output.
5148 The option should only be accepted if preprocessor condition COND
5149 is true. Note that any C declarations associated with the option
5150 will be present even if COND is false; COND simply controls
5151 whether the option is accepted and whether it is printed in the
5155 Build the `cl_target_option' structure to hold a copy of the
5156 option, add the functions `cl_target_option_save' and
5157 `cl_target_option_restore' to save and restore the options.
5160 File: gccint.info, Node: Passes, Next: Trees, Prev: Options, Up: Top
5162 8 Passes and Files of the Compiler
5163 **********************************
5165 This chapter is dedicated to giving an overview of the optimization and
5166 code generation passes of the compiler. In the process, it describes
5167 some of the language front end interface, though this description is no
5168 where near complete.
5172 * Parsing pass:: The language front end turns text into bits.
5173 * Gimplification pass:: The bits are turned into something we can optimize.
5174 * Pass manager:: Sequencing the optimization passes.
5175 * Tree-SSA passes:: Optimizations on a high-level representation.
5176 * RTL passes:: Optimizations on a low-level representation.
5179 File: gccint.info, Node: Parsing pass, Next: Gimplification pass, Up: Passes
5184 The language front end is invoked only once, via
5185 `lang_hooks.parse_file', to parse the entire input. The language front
5186 end may use any intermediate language representation deemed
5187 appropriate. The C front end uses GENERIC trees (CROSSREF), plus a
5188 double handful of language specific tree codes defined in
5189 `c-common.def'. The Fortran front end uses a completely different
5190 private representation.
5192 At some point the front end must translate the representation used in
5193 the front end to a representation understood by the language-independent
5194 portions of the compiler. Current practice takes one of two forms.
5195 The C front end manually invokes the gimplifier (CROSSREF) on each
5196 function, and uses the gimplifier callbacks to convert the
5197 language-specific tree nodes directly to GIMPLE (CROSSREF) before
5198 passing the function off to be compiled. The Fortran front end
5199 converts from a private representation to GENERIC, which is later
5200 lowered to GIMPLE when the function is compiled. Which route to choose
5201 probably depends on how well GENERIC (plus extensions) can be made to
5202 match up with the source language and necessary parsing data structures.
5204 BUG: Gimplification must occur before nested function lowering, and
5205 nested function lowering must be done by the front end before passing
5206 the data off to cgraph.
5208 TODO: Cgraph should control nested function lowering. It would only
5209 be invoked when it is certain that the outer-most function is used.
5211 TODO: Cgraph needs a gimplify_function callback. It should be invoked
5212 when (1) it is certain that the function is used, (2) warning flags
5213 specified by the user require some amount of compilation in order to
5214 honor, (3) the language indicates that semantic analysis is not
5215 complete until gimplification occurs. Hum... this sounds overly
5216 complicated. Perhaps we should just have the front end gimplify
5217 always; in most cases it's only one function call.
5219 The front end needs to pass all function definitions and top level
5220 declarations off to the middle-end so that they can be compiled and
5221 emitted to the object file. For a simple procedural language, it is
5222 usually most convenient to do this as each top level declaration or
5223 definition is seen. There is also a distinction to be made between
5224 generating functional code and generating complete debug information.
5225 The only thing that is absolutely required for functional code is that
5226 function and data _definitions_ be passed to the middle-end. For
5227 complete debug information, function, data and type declarations should
5228 all be passed as well.
5230 In any case, the front end needs each complete top-level function or
5231 data declaration, and each data definition should be passed to
5232 `rest_of_decl_compilation'. Each complete type definition should be
5233 passed to `rest_of_type_compilation'. Each function definition should
5234 be passed to `cgraph_finalize_function'.
5236 TODO: I know rest_of_compilation currently has all sorts of
5237 rtl-generation semantics. I plan to move all code generation bits
5238 (both tree and rtl) to compile_function. Should we hide cgraph from
5239 the front ends and move back to rest_of_compilation as the official
5240 interface? Possibly we should rename all three interfaces such that
5241 the names match in some meaningful way and that is more descriptive
5244 The middle-end will, at its option, emit the function and data
5245 definitions immediately or queue them for later processing.
5248 File: gccint.info, Node: Gimplification pass, Next: Pass manager, Prev: Parsing pass, Up: Passes
5250 8.2 Gimplification pass
5251 =======================
5253 "Gimplification" is a whimsical term for the process of converting the
5254 intermediate representation of a function into the GIMPLE language
5255 (CROSSREF). The term stuck, and so words like "gimplification",
5256 "gimplify", "gimplifier" and the like are sprinkled throughout this
5259 While a front end may certainly choose to generate GIMPLE directly if
5260 it chooses, this can be a moderately complex process unless the
5261 intermediate language used by the front end is already fairly simple.
5262 Usually it is easier to generate GENERIC trees plus extensions and let
5263 the language-independent gimplifier do most of the work.
5265 The main entry point to this pass is `gimplify_function_tree' located
5266 in `gimplify.c'. From here we process the entire function gimplifying
5267 each statement in turn. The main workhorse for this pass is
5268 `gimplify_expr'. Approximately everything passes through here at least
5269 once, and it is from here that we invoke the `lang_hooks.gimplify_expr'
5272 The callback should examine the expression in question and return
5273 `GS_UNHANDLED' if the expression is not a language specific construct
5274 that requires attention. Otherwise it should alter the expression in
5275 some way to such that forward progress is made toward producing valid
5276 GIMPLE. If the callback is certain that the transformation is complete
5277 and the expression is valid GIMPLE, it should return `GS_ALL_DONE'.
5278 Otherwise it should return `GS_OK', which will cause the expression to
5279 be processed again. If the callback encounters an error during the
5280 transformation (because the front end is relying on the gimplification
5281 process to finish semantic checks), it should return `GS_ERROR'.
5284 File: gccint.info, Node: Pass manager, Next: Tree-SSA passes, Prev: Gimplification pass, Up: Passes
5289 The pass manager is located in `passes.c', `tree-optimize.c' and
5290 `tree-pass.h'. Its job is to run all of the individual passes in the
5291 correct order, and take care of standard bookkeeping that applies to
5294 The theory of operation is that each pass defines a structure that
5295 represents everything we need to know about that pass--when it should
5296 be run, how it should be run, what intermediate language form or
5297 on-the-side data structures it needs. We register the pass to be run
5298 in some particular order, and the pass manager arranges for everything
5299 to happen in the correct order.
5301 The actuality doesn't completely live up to the theory at present.
5302 Command-line switches and `timevar_id_t' enumerations must still be
5303 defined elsewhere. The pass manager validates constraints but does not
5304 attempt to (re-)generate data structures or lower intermediate language
5305 form based on the requirements of the next pass. Nevertheless, what is
5306 present is useful, and a far sight better than nothing at all.
5308 Each pass may have its own dump file (for GCC debugging purposes).
5309 Passes without any names, or with a name starting with a star, do not
5312 TODO: describe the global variables set up by the pass manager, and a
5313 brief description of how a new pass should use it. I need to look at
5314 what info rtl passes use first....
5317 File: gccint.info, Node: Tree-SSA passes, Next: RTL passes, Prev: Pass manager, Up: Passes
5322 The following briefly describes the tree optimization passes that are
5323 run after gimplification and what source files they are located in.
5325 * Remove useless statements
5327 This pass is an extremely simple sweep across the gimple code in
5328 which we identify obviously dead code and remove it. Here we do
5329 things like simplify `if' statements with constant conditions,
5330 remove exception handling constructs surrounding code that
5331 obviously cannot throw, remove lexical bindings that contain no
5332 variables, and other assorted simplistic cleanups. The idea is to
5333 get rid of the obvious stuff quickly rather than wait until later
5334 when it's more work to get rid of it. This pass is located in
5335 `tree-cfg.c' and described by `pass_remove_useless_stmts'.
5337 * Mudflap declaration registration
5339 If mudflap (*note -fmudflap -fmudflapth -fmudflapir: (gcc)Optimize
5340 Options.) is enabled, we generate code to register some variable
5341 declarations with the mudflap runtime. Specifically, the runtime
5342 tracks the lifetimes of those variable declarations that have
5343 their addresses taken, or whose bounds are unknown at compile time
5344 (`extern'). This pass generates new exception handling constructs
5345 (`try'/`finally'), and so must run before those are lowered. In
5346 addition, the pass enqueues declarations of static variables whose
5347 lifetimes extend to the entire program. The pass is located in
5348 `tree-mudflap.c' and is described by `pass_mudflap_1'.
5352 If OpenMP generation (`-fopenmp') is enabled, this pass lowers
5353 OpenMP constructs into GIMPLE.
5355 Lowering of OpenMP constructs involves creating replacement
5356 expressions for local variables that have been mapped using data
5357 sharing clauses, exposing the control flow of most synchronization
5358 directives and adding region markers to facilitate the creation of
5359 the control flow graph. The pass is located in `omp-low.c' and is
5360 described by `pass_lower_omp'.
5364 If OpenMP generation (`-fopenmp') is enabled, this pass expands
5365 parallel regions into their own functions to be invoked by the
5366 thread library. The pass is located in `omp-low.c' and is
5367 described by `pass_expand_omp'.
5369 * Lower control flow
5371 This pass flattens `if' statements (`COND_EXPR') and moves lexical
5372 bindings (`BIND_EXPR') out of line. After this pass, all `if'
5373 statements will have exactly two `goto' statements in its `then'
5374 and `else' arms. Lexical binding information for each statement
5375 will be found in `TREE_BLOCK' rather than being inferred from its
5376 position under a `BIND_EXPR'. This pass is found in
5377 `gimple-low.c' and is described by `pass_lower_cf'.
5379 * Lower exception handling control flow
5381 This pass decomposes high-level exception handling constructs
5382 (`TRY_FINALLY_EXPR' and `TRY_CATCH_EXPR') into a form that
5383 explicitly represents the control flow involved. After this pass,
5384 `lookup_stmt_eh_region' will return a non-negative number for any
5385 statement that may have EH control flow semantics; examine
5386 `tree_can_throw_internal' or `tree_can_throw_external' for exact
5387 semantics. Exact control flow may be extracted from
5388 `foreach_reachable_handler'. The EH region nesting tree is defined
5389 in `except.h' and built in `except.c'. The lowering pass itself
5390 is in `tree-eh.c' and is described by `pass_lower_eh'.
5392 * Build the control flow graph
5394 This pass decomposes a function into basic blocks and creates all
5395 of the edges that connect them. It is located in `tree-cfg.c' and
5396 is described by `pass_build_cfg'.
5398 * Find all referenced variables
5400 This pass walks the entire function and collects an array of all
5401 variables referenced in the function, `referenced_vars'. The
5402 index at which a variable is found in the array is used as a UID
5403 for the variable within this function. This data is needed by the
5404 SSA rewriting routines. The pass is located in `tree-dfa.c' and
5405 is described by `pass_referenced_vars'.
5407 * Enter static single assignment form
5409 This pass rewrites the function such that it is in SSA form. After
5410 this pass, all `is_gimple_reg' variables will be referenced by
5411 `SSA_NAME', and all occurrences of other variables will be
5412 annotated with `VDEFS' and `VUSES'; PHI nodes will have been
5413 inserted as necessary for each basic block. This pass is located
5414 in `tree-ssa.c' and is described by `pass_build_ssa'.
5416 * Warn for uninitialized variables
5418 This pass scans the function for uses of `SSA_NAME's that are fed
5419 by default definition. For non-parameter variables, such uses are
5420 uninitialized. The pass is run twice, before and after
5421 optimization (if turned on). In the first pass we only warn for
5422 uses that are positively uninitialized; in the second pass we warn
5423 for uses that are possibly uninitialized. The pass is located in
5424 `tree-ssa.c' and is defined by `pass_early_warn_uninitialized' and
5425 `pass_late_warn_uninitialized'.
5427 * Dead code elimination
5429 This pass scans the function for statements without side effects
5430 whose result is unused. It does not do memory life analysis, so
5431 any value that is stored in memory is considered used. The pass
5432 is run multiple times throughout the optimization process. It is
5433 located in `tree-ssa-dce.c' and is described by `pass_dce'.
5435 * Dominator optimizations
5437 This pass performs trivial dominator-based copy and constant
5438 propagation, expression simplification, and jump threading. It is
5439 run multiple times throughout the optimization process. It it
5440 located in `tree-ssa-dom.c' and is described by `pass_dominator'.
5442 * Forward propagation of single-use variables
5444 This pass attempts to remove redundant computation by substituting
5445 variables that are used once into the expression that uses them and
5446 seeing if the result can be simplified. It is located in
5447 `tree-ssa-forwprop.c' and is described by `pass_forwprop'.
5451 This pass attempts to change the name of compiler temporaries
5452 involved in copy operations such that SSA->normal can coalesce the
5453 copy away. When compiler temporaries are copies of user
5454 variables, it also renames the compiler temporary to the user
5455 variable resulting in better use of user symbols. It is located
5456 in `tree-ssa-copyrename.c' and is described by `pass_copyrename'.
5458 * PHI node optimizations
5460 This pass recognizes forms of PHI inputs that can be represented as
5461 conditional expressions and rewrites them into straight line code.
5462 It is located in `tree-ssa-phiopt.c' and is described by
5465 * May-alias optimization
5467 This pass performs a flow sensitive SSA-based points-to analysis.
5468 The resulting may-alias, must-alias, and escape analysis
5469 information is used to promote variables from in-memory
5470 addressable objects to non-aliased variables that can be renamed
5471 into SSA form. We also update the `VDEF'/`VUSE' memory tags for
5472 non-renameable aggregates so that we get fewer false kills. The
5473 pass is located in `tree-ssa-alias.c' and is described by
5476 Interprocedural points-to information is located in
5477 `tree-ssa-structalias.c' and described by `pass_ipa_pta'.
5481 This pass rewrites the function in order to collect runtime block
5482 and value profiling data. Such data may be fed back into the
5483 compiler on a subsequent run so as to allow optimization based on
5484 expected execution frequencies. The pass is located in
5485 `predict.c' and is described by `pass_profile'.
5487 * Lower complex arithmetic
5489 This pass rewrites complex arithmetic operations into their
5490 component scalar arithmetic operations. The pass is located in
5491 `tree-complex.c' and is described by `pass_lower_complex'.
5493 * Scalar replacement of aggregates
5495 This pass rewrites suitable non-aliased local aggregate variables
5496 into a set of scalar variables. The resulting scalar variables are
5497 rewritten into SSA form, which allows subsequent optimization
5498 passes to do a significantly better job with them. The pass is
5499 located in `tree-sra.c' and is described by `pass_sra'.
5501 * Dead store elimination
5503 This pass eliminates stores to memory that are subsequently
5504 overwritten by another store, without any intervening loads. The
5505 pass is located in `tree-ssa-dse.c' and is described by `pass_dse'.
5507 * Tail recursion elimination
5509 This pass transforms tail recursion into a loop. It is located in
5510 `tree-tailcall.c' and is described by `pass_tail_recursion'.
5512 * Forward store motion
5514 This pass sinks stores and assignments down the flowgraph closer
5515 to their use point. The pass is located in `tree-ssa-sink.c' and
5516 is described by `pass_sink_code'.
5518 * Partial redundancy elimination
5520 This pass eliminates partially redundant computations, as well as
5521 performing load motion. The pass is located in `tree-ssa-pre.c'
5522 and is described by `pass_pre'.
5524 Just before partial redundancy elimination, if
5525 `-funsafe-math-optimizations' is on, GCC tries to convert
5526 divisions to multiplications by the reciprocal. The pass is
5527 located in `tree-ssa-math-opts.c' and is described by
5528 `pass_cse_reciprocal'.
5530 * Full redundancy elimination
5532 This is a simpler form of PRE that only eliminates redundancies
5533 that occur an all paths. It is located in `tree-ssa-pre.c' and
5534 described by `pass_fre'.
5538 The main driver of the pass is placed in `tree-ssa-loop.c' and
5539 described by `pass_loop'.
5541 The optimizations performed by this pass are:
5543 Loop invariant motion. This pass moves only invariants that would
5544 be hard to handle on rtl level (function calls, operations that
5545 expand to nontrivial sequences of insns). With `-funswitch-loops'
5546 it also moves operands of conditions that are invariant out of the
5547 loop, so that we can use just trivial invariantness analysis in
5548 loop unswitching. The pass also includes store motion. The pass
5549 is implemented in `tree-ssa-loop-im.c'.
5551 Canonical induction variable creation. This pass creates a simple
5552 counter for number of iterations of the loop and replaces the exit
5553 condition of the loop using it, in case when a complicated
5554 analysis is necessary to determine the number of iterations.
5555 Later optimizations then may determine the number easily. The
5556 pass is implemented in `tree-ssa-loop-ivcanon.c'.
5558 Induction variable optimizations. This pass performs standard
5559 induction variable optimizations, including strength reduction,
5560 induction variable merging and induction variable elimination.
5561 The pass is implemented in `tree-ssa-loop-ivopts.c'.
5563 Loop unswitching. This pass moves the conditional jumps that are
5564 invariant out of the loops. To achieve this, a duplicate of the
5565 loop is created for each possible outcome of conditional jump(s).
5566 The pass is implemented in `tree-ssa-loop-unswitch.c'. This pass
5567 should eventually replace the rtl-level loop unswitching in
5568 `loop-unswitch.c', but currently the rtl-level pass is not
5569 completely redundant yet due to deficiencies in tree level alias
5572 The optimizations also use various utility functions contained in
5573 `tree-ssa-loop-manip.c', `cfgloop.c', `cfgloopanal.c' and
5576 Vectorization. This pass transforms loops to operate on vector
5577 types instead of scalar types. Data parallelism across loop
5578 iterations is exploited to group data elements from consecutive
5579 iterations into a vector and operate on them in parallel.
5580 Depending on available target support the loop is conceptually
5581 unrolled by a factor `VF' (vectorization factor), which is the
5582 number of elements operated upon in parallel in each iteration,
5583 and the `VF' copies of each scalar operation are fused to form a
5584 vector operation. Additional loop transformations such as peeling
5585 and versioning may take place to align the number of iterations,
5586 and to align the memory accesses in the loop. The pass is
5587 implemented in `tree-vectorizer.c' (the main driver and general
5588 utilities), `tree-vect-analyze.c' and `tree-vect-transform.c'.
5589 Analysis of data references is in `tree-data-ref.c'.
5591 Autoparallelization. This pass splits the loop iteration space to
5592 run into several threads. The pass is implemented in
5595 * Tree level if-conversion for vectorizer
5597 This pass applies if-conversion to simple loops to help vectorizer.
5598 We identify if convertible loops, if-convert statements and merge
5599 basic blocks in one big block. The idea is to present loop in such
5600 form so that vectorizer can have one to one mapping between
5601 statements and available vector operations. This patch
5602 re-introduces COND_EXPR at GIMPLE level. This pass is located in
5603 `tree-if-conv.c' and is described by `pass_if_conversion'.
5605 * Conditional constant propagation
5607 This pass relaxes a lattice of values in order to identify those
5608 that must be constant even in the presence of conditional branches.
5609 The pass is located in `tree-ssa-ccp.c' and is described by
5612 A related pass that works on memory loads and stores, and not just
5613 register values, is located in `tree-ssa-ccp.c' and described by
5616 * Conditional copy propagation
5618 This is similar to constant propagation but the lattice of values
5619 is the "copy-of" relation. It eliminates redundant copies from the
5620 code. The pass is located in `tree-ssa-copy.c' and described by
5623 A related pass that works on memory copies, and not just register
5624 copies, is located in `tree-ssa-copy.c' and described by
5625 `pass_store_copy_prop'.
5627 * Value range propagation
5629 This transformation is similar to constant propagation but instead
5630 of propagating single constant values, it propagates known value
5631 ranges. The implementation is based on Patterson's range
5632 propagation algorithm (Accurate Static Branch Prediction by Value
5633 Range Propagation, J. R. C. Patterson, PLDI '95). In contrast to
5634 Patterson's algorithm, this implementation does not propagate
5635 branch probabilities nor it uses more than a single range per SSA
5636 name. This means that the current implementation cannot be used
5637 for branch prediction (though adapting it would not be difficult).
5638 The pass is located in `tree-vrp.c' and is described by
5641 * Folding built-in functions
5643 This pass simplifies built-in functions, as applicable, with
5644 constant arguments or with inferable string lengths. It is
5645 located in `tree-ssa-ccp.c' and is described by
5646 `pass_fold_builtins'.
5648 * Split critical edges
5650 This pass identifies critical edges and inserts empty basic blocks
5651 such that the edge is no longer critical. The pass is located in
5652 `tree-cfg.c' and is described by `pass_split_crit_edges'.
5654 * Control dependence dead code elimination
5656 This pass is a stronger form of dead code elimination that can
5657 eliminate unnecessary control flow statements. It is located in
5658 `tree-ssa-dce.c' and is described by `pass_cd_dce'.
5660 * Tail call elimination
5662 This pass identifies function calls that may be rewritten into
5663 jumps. No code transformation is actually applied here, but the
5664 data and control flow problem is solved. The code transformation
5665 requires target support, and so is delayed until RTL. In the
5666 meantime `CALL_EXPR_TAILCALL' is set indicating the possibility.
5667 The pass is located in `tree-tailcall.c' and is described by
5668 `pass_tail_calls'. The RTL transformation is handled by
5669 `fixup_tail_calls' in `calls.c'.
5671 * Warn for function return without value
5673 For non-void functions, this pass locates return statements that do
5674 not specify a value and issues a warning. Such a statement may
5675 have been injected by falling off the end of the function. This
5676 pass is run last so that we have as much time as possible to prove
5677 that the statement is not reachable. It is located in
5678 `tree-cfg.c' and is described by `pass_warn_function_return'.
5680 * Mudflap statement annotation
5682 If mudflap is enabled, we rewrite some memory accesses with code to
5683 validate that the memory access is correct. In particular,
5684 expressions involving pointer dereferences (`INDIRECT_REF',
5685 `ARRAY_REF', etc.) are replaced by code that checks the selected
5686 address range against the mudflap runtime's database of valid
5687 regions. This check includes an inline lookup into a
5688 direct-mapped cache, based on shift/mask operations of the pointer
5689 value, with a fallback function call into the runtime. The pass
5690 is located in `tree-mudflap.c' and is described by
5693 * Leave static single assignment form
5695 This pass rewrites the function such that it is in normal form. At
5696 the same time, we eliminate as many single-use temporaries as
5697 possible, so the intermediate language is no longer GIMPLE, but
5698 GENERIC. The pass is located in `tree-outof-ssa.c' and is
5699 described by `pass_del_ssa'.
5701 * Merge PHI nodes that feed into one another
5703 This is part of the CFG cleanup passes. It attempts to join PHI
5704 nodes from a forwarder CFG block into another block with PHI
5705 nodes. The pass is located in `tree-cfgcleanup.c' and is
5706 described by `pass_merge_phi'.
5708 * Return value optimization
5710 If a function always returns the same local variable, and that
5711 local variable is an aggregate type, then the variable is replaced
5712 with the return value for the function (i.e., the function's
5713 DECL_RESULT). This is equivalent to the C++ named return value
5714 optimization applied to GIMPLE. The pass is located in
5715 `tree-nrv.c' and is described by `pass_nrv'.
5717 * Return slot optimization
5719 If a function returns a memory object and is called as `var =
5720 foo()', this pass tries to change the call so that the address of
5721 `var' is sent to the caller to avoid an extra memory copy. This
5722 pass is located in `tree-nrv.c' and is described by
5725 * Optimize calls to `__builtin_object_size'
5727 This is a propagation pass similar to CCP that tries to remove
5728 calls to `__builtin_object_size' when the size of the object can be
5729 computed at compile-time. This pass is located in
5730 `tree-object-size.c' and is described by `pass_object_sizes'.
5732 * Loop invariant motion
5734 This pass removes expensive loop-invariant computations out of
5735 loops. The pass is located in `tree-ssa-loop.c' and described by
5738 * Loop nest optimizations
5740 This is a family of loop transformations that works on loop nests.
5741 It includes loop interchange, scaling, skewing and reversal and
5742 they are all geared to the optimization of data locality in array
5743 traversals and the removal of dependencies that hamper
5744 optimizations such as loop parallelization and vectorization. The
5745 pass is located in `tree-loop-linear.c' and described by
5746 `pass_linear_transform'.
5748 * Removal of empty loops
5750 This pass removes loops with no code in them. The pass is located
5751 in `tree-ssa-loop-ivcanon.c' and described by `pass_empty_loop'.
5753 * Unrolling of small loops
5755 This pass completely unrolls loops with few iterations. The pass
5756 is located in `tree-ssa-loop-ivcanon.c' and described by
5757 `pass_complete_unroll'.
5759 * Predictive commoning
5761 This pass makes the code reuse the computations from the previous
5762 iterations of the loops, especially loads and stores to memory.
5763 It does so by storing the values of these computations to a bank
5764 of temporary variables that are rotated at the end of loop. To
5765 avoid the need for this rotation, the loop is then unrolled and
5766 the copies of the loop body are rewritten to use the appropriate
5767 version of the temporary variable. This pass is located in
5768 `tree-predcom.c' and described by `pass_predcom'.
5772 This pass issues prefetch instructions for array references inside
5773 loops. The pass is located in `tree-ssa-loop-prefetch.c' and
5774 described by `pass_loop_prefetch'.
5778 This pass rewrites arithmetic expressions to enable optimizations
5779 that operate on them, like redundancy elimination and
5780 vectorization. The pass is located in `tree-ssa-reassoc.c' and
5781 described by `pass_reassoc'.
5783 * Optimization of `stdarg' functions
5785 This pass tries to avoid the saving of register arguments into the
5786 stack on entry to `stdarg' functions. If the function doesn't use
5787 any `va_start' macros, no registers need to be saved. If
5788 `va_start' macros are used, the `va_list' variables don't escape
5789 the function, it is only necessary to save registers that will be
5790 used in `va_arg' macros. For instance, if `va_arg' is only used
5791 with integral types in the function, floating point registers
5792 don't need to be saved. This pass is located in `tree-stdarg.c'
5793 and described by `pass_stdarg'.
5797 File: gccint.info, Node: RTL passes, Prev: Tree-SSA passes, Up: Passes
5802 The following briefly describes the rtl generation and optimization
5803 passes that are run after tree optimization.
5807 The source files for RTL generation include `stmt.c', `calls.c',
5808 `expr.c', `explow.c', `expmed.c', `function.c', `optabs.c' and
5809 `emit-rtl.c'. Also, the file `insn-emit.c', generated from the
5810 machine description by the program `genemit', is used in this
5811 pass. The header file `expr.h' is used for communication within
5814 The header files `insn-flags.h' and `insn-codes.h', generated from
5815 the machine description by the programs `genflags' and `gencodes',
5816 tell this pass which standard names are available for use and
5817 which patterns correspond to them.
5819 * Generate exception handling landing pads
5821 This pass generates the glue that handles communication between the
5822 exception handling library routines and the exception handlers
5823 within the function. Entry points in the function that are
5824 invoked by the exception handling library are called "landing
5825 pads". The code for this pass is located within `except.c'.
5827 * Cleanup control flow graph
5829 This pass removes unreachable code, simplifies jumps to next,
5830 jumps to jump, jumps across jumps, etc. The pass is run multiple
5831 times. For historical reasons, it is occasionally referred to as
5832 the "jump optimization pass". The bulk of the code for this pass
5833 is in `cfgcleanup.c', and there are support routines in `cfgrtl.c'
5836 * Forward propagation of single-def values
5838 This pass attempts to remove redundant computation by substituting
5839 variables that come from a single definition, and seeing if the
5840 result can be simplified. It performs copy propagation and
5841 addressing mode selection. The pass is run twice, with values
5842 being propagated into loops only on the second run. It is located
5845 * Common subexpression elimination
5847 This pass removes redundant computation within basic blocks, and
5848 optimizes addressing modes based on cost. The pass is run twice.
5849 The source is located in `cse.c'.
5851 * Global common subexpression elimination.
5853 This pass performs two different types of GCSE depending on
5854 whether you are optimizing for size or not (LCM based GCSE tends
5855 to increase code size for a gain in speed, while Morel-Renvoise
5856 based GCSE does not). When optimizing for size, GCSE is done
5857 using Morel-Renvoise Partial Redundancy Elimination, with the
5858 exception that it does not try to move invariants out of
5859 loops--that is left to the loop optimization pass. If MR PRE
5860 GCSE is done, code hoisting (aka unification) is also done, as
5861 well as load motion. If you are optimizing for speed, LCM (lazy
5862 code motion) based GCSE is done. LCM is based on the work of
5863 Knoop, Ruthing, and Steffen. LCM based GCSE also does loop
5864 invariant code motion. We also perform load and store motion when
5865 optimizing for speed. Regardless of which type of GCSE is used,
5866 the GCSE pass also performs global constant and copy propagation.
5867 The source file for this pass is `gcse.c', and the LCM routines
5872 This pass performs several loop related optimizations. The source
5873 files `cfgloopanal.c' and `cfgloopmanip.c' contain generic loop
5874 analysis and manipulation code. Initialization and finalization
5875 of loop structures is handled by `loop-init.c'. A loop invariant
5876 motion pass is implemented in `loop-invariant.c'. Basic block
5877 level optimizations--unrolling, peeling and unswitching loops--
5878 are implemented in `loop-unswitch.c' and `loop-unroll.c'.
5879 Replacing of the exit condition of loops by special
5880 machine-dependent instructions is handled by `loop-doloop.c'.
5884 This pass is an aggressive form of GCSE that transforms the control
5885 flow graph of a function by propagating constants into conditional
5886 branch instructions. The source file for this pass is `gcse.c'.
5890 This pass attempts to replace conditional branches and surrounding
5891 assignments with arithmetic, boolean value producing comparison
5892 instructions, and conditional move instructions. In the very last
5893 invocation after reload, it will generate predicated instructions
5894 when supported by the target. The pass is located in `ifcvt.c'.
5898 This pass splits independent uses of each pseudo-register. This
5899 can improve effect of the other transformation, such as CSE or
5900 register allocation. Its source files are `web.c'.
5904 This pass computes which pseudo-registers are live at each point in
5905 the program, and makes the first instruction that uses a value
5906 point at the instruction that computed the value. It then deletes
5907 computations whose results are never used, and combines memory
5908 references with add or subtract instructions to make autoincrement
5909 or autodecrement addressing. The pass is located in `flow.c'.
5911 * Instruction combination
5913 This pass attempts to combine groups of two or three instructions
5914 that are related by data flow into single instructions. It
5915 combines the RTL expressions for the instructions by substitution,
5916 simplifies the result using algebra, and then attempts to match
5917 the result against the machine description. The pass is located
5922 This pass looks for cases where matching constraints would force an
5923 instruction to need a reload, and this reload would be a
5924 register-to-register move. It then attempts to change the
5925 registers used by the instruction to avoid the move instruction.
5926 The pass is located in `regmove.c'.
5928 * Optimize mode switching
5930 This pass looks for instructions that require the processor to be
5931 in a specific "mode" and minimizes the number of mode changes
5932 required to satisfy all users. What these modes are, and what
5933 they apply to are completely target-specific. The source is
5934 located in `mode-switching.c'.
5938 This pass looks at innermost loops and reorders their instructions
5939 by overlapping different iterations. Modulo scheduling is
5940 performed immediately before instruction scheduling. The pass is
5941 located in (`modulo-sched.c').
5943 * Instruction scheduling
5945 This pass looks for instructions whose output will not be
5946 available by the time that it is used in subsequent instructions.
5947 Memory loads and floating point instructions often have this
5948 behavior on RISC machines. It re-orders instructions within a
5949 basic block to try to separate the definition and use of items
5950 that otherwise would cause pipeline stalls. This pass is
5951 performed twice, before and after register allocation. The pass
5952 is located in `haifa-sched.c', `sched-deps.c', `sched-ebb.c',
5953 `sched-rgn.c' and `sched-vis.c'.
5955 * Register allocation
5957 These passes make sure that all occurrences of pseudo registers are
5958 eliminated, either by allocating them to a hard register, replacing
5959 them by an equivalent expression (e.g. a constant) or by placing
5960 them on the stack. This is done in several subpasses:
5962 * Register move optimizations. This pass makes some simple RTL
5963 code transformations which improve the subsequent register
5964 allocation. The source file is `regmove.c'.
5966 * The integrated register allocator (IRA). It is called
5967 integrated because coalescing, register live range splitting,
5968 and hard register preferencing are done on-the-fly during
5969 coloring. It also has better integration with the reload
5970 pass. Pseudo-registers spilled by the allocator or the
5971 reload have still a chance to get hard-registers if the
5972 reload evicts some pseudo-registers from hard-registers. The
5973 allocator helps to choose better pseudos for spilling based
5974 on their live ranges and to coalesce stack slots allocated
5975 for the spilled pseudo-registers. IRA is a regional register
5976 allocator which is transformed into Chaitin-Briggs allocator
5977 if there is one region. By default, IRA chooses regions using
5978 register pressure but the user can force it to use one region
5979 or regions corresponding to all loops.
5981 Source files of the allocator are `ira.c', `ira-build.c',
5982 `ira-costs.c', `ira-conflicts.c', `ira-color.c',
5983 `ira-emit.c', `ira-lives', plus header files `ira.h' and
5984 `ira-int.h' used for the communication between the allocator
5985 and the rest of the compiler and between the IRA files.
5987 * Reloading. This pass renumbers pseudo registers with the
5988 hardware registers numbers they were allocated. Pseudo
5989 registers that did not get hard registers are replaced with
5990 stack slots. Then it finds instructions that are invalid
5991 because a value has failed to end up in a register, or has
5992 ended up in a register of the wrong kind. It fixes up these
5993 instructions by reloading the problematical values
5994 temporarily into registers. Additional instructions are
5995 generated to do the copying.
5997 The reload pass also optionally eliminates the frame pointer
5998 and inserts instructions to save and restore call-clobbered
5999 registers around calls.
6001 Source files are `reload.c' and `reload1.c', plus the header
6002 `reload.h' used for communication between them.
6004 * Basic block reordering
6006 This pass implements profile guided code positioning. If profile
6007 information is not available, various types of static analysis are
6008 performed to make the predictions normally coming from the profile
6009 feedback (IE execution frequency, branch probability, etc). It is
6010 implemented in the file `bb-reorder.c', and the various prediction
6011 routines are in `predict.c'.
6015 This pass computes where the variables are stored at each position
6016 in code and generates notes describing the variable locations to
6017 RTL code. The location lists are then generated according to these
6018 notes to debug information if the debugging information format
6019 supports location lists.
6021 * Delayed branch scheduling
6023 This optional pass attempts to find instructions that can go into
6024 the delay slots of other instructions, usually jumps and calls.
6025 The source file name is `reorg.c'.
6029 On many RISC machines, branch instructions have a limited range.
6030 Thus, longer sequences of instructions must be used for long
6031 branches. In this pass, the compiler figures out what how far
6032 each instruction will be from each other instruction, and
6033 therefore whether the usual instructions, or the longer sequences,
6034 must be used for each branch.
6036 * Register-to-stack conversion
6038 Conversion from usage of some hard registers to usage of a register
6039 stack may be done at this point. Currently, this is supported only
6040 for the floating-point registers of the Intel 80387 coprocessor.
6041 The source file name is `reg-stack.c'.
6045 This pass outputs the assembler code for the function. The source
6046 files are `final.c' plus `insn-output.c'; the latter is generated
6047 automatically from the machine description by the tool `genoutput'.
6048 The header file `conditions.h' is used for communication between
6049 these files. If mudflap is enabled, the queue of deferred
6050 declarations and any addressed constants (e.g., string literals)
6051 is processed by `mudflap_finish_file' into a synthetic constructor
6052 function containing calls into the mudflap runtime.
6054 * Debugging information output
6056 This is run after final because it must output the stack slot
6057 offsets for pseudo registers that did not get hard registers.
6058 Source files are `dbxout.c' for DBX symbol table format,
6059 `sdbout.c' for SDB symbol table format, `dwarfout.c' for DWARF
6060 symbol table format, files `dwarf2out.c' and `dwarf2asm.c' for
6061 DWARF2 symbol table format, and `vmsdbgout.c' for VMS debug symbol
6066 File: gccint.info, Node: Trees, Next: RTL, Prev: Passes, Up: Top
6068 9 Trees: The intermediate representation used by the C and C++ front ends
6069 *************************************************************************
6071 This chapter documents the internal representation used by GCC to
6072 represent C and C++ source programs. When presented with a C or C++
6073 source program, GCC parses the program, performs semantic analysis
6074 (including the generation of error messages), and then produces the
6075 internal representation described here. This representation contains a
6076 complete representation for the entire translation unit provided as
6077 input to the front end. This representation is then typically processed
6078 by a code-generator in order to produce machine code, but could also be
6079 used in the creation of source browsers, intelligent editors, automatic
6080 documentation generators, interpreters, and any other programs needing
6081 the ability to process C or C++ code.
6083 This chapter explains the internal representation. In particular, it
6084 documents the internal representation for C and C++ source constructs,
6085 and the macros, functions, and variables that can be used to access
6086 these constructs. The C++ representation is largely a superset of the
6087 representation used in the C front end. There is only one construct
6088 used in C that does not appear in the C++ front end and that is the GNU
6089 "nested function" extension. Many of the macros documented here do not
6090 apply in C because the corresponding language constructs do not appear
6093 If you are developing a "back end", be it is a code-generator or some
6094 other tool, that uses this representation, you may occasionally find
6095 that you need to ask questions not easily answered by the functions and
6096 macros available here. If that situation occurs, it is quite likely
6097 that GCC already supports the functionality you desire, but that the
6098 interface is simply not documented here. In that case, you should ask
6099 the GCC maintainers (via mail to <gcc@gcc.gnu.org>) about documenting
6100 the functionality you require. Similarly, if you find yourself writing
6101 functions that do not deal directly with your back end, but instead
6102 might be useful to other people using the GCC front end, you should
6103 submit your patches for inclusion in GCC.
6107 * Deficiencies:: Topics net yet covered in this document.
6108 * Tree overview:: All about `tree's.
6109 * Types:: Fundamental and aggregate types.
6110 * Scopes:: Namespaces and classes.
6111 * Functions:: Overloading, function bodies, and linkage.
6112 * Declarations:: Type declarations and variables.
6113 * Attributes:: Declaration and type attributes.
6114 * Expression trees:: From `typeid' to `throw'.
6117 File: gccint.info, Node: Deficiencies, Next: Tree overview, Up: Trees
6122 There are many places in which this document is incomplet and incorrekt.
6123 It is, as of yet, only _preliminary_ documentation.
6126 File: gccint.info, Node: Tree overview, Next: Types, Prev: Deficiencies, Up: Trees
6131 The central data structure used by the internal representation is the
6132 `tree'. These nodes, while all of the C type `tree', are of many
6133 varieties. A `tree' is a pointer type, but the object to which it
6134 points may be of a variety of types. From this point forward, we will
6135 refer to trees in ordinary type, rather than in `this font', except
6136 when talking about the actual C type `tree'.
6138 You can tell what kind of node a particular tree is by using the
6139 `TREE_CODE' macro. Many, many macros take trees as input and return
6140 trees as output. However, most macros require a certain kind of tree
6141 node as input. In other words, there is a type-system for trees, but
6142 it is not reflected in the C type-system.
6144 For safety, it is useful to configure GCC with `--enable-checking'.
6145 Although this results in a significant performance penalty (since all
6146 tree types are checked at run-time), and is therefore inappropriate in a
6147 release version, it is extremely helpful during the development process.
6149 Many macros behave as predicates. Many, although not all, of these
6150 predicates end in `_P'. Do not rely on the result type of these macros
6151 being of any particular type. You may, however, rely on the fact that
6152 the type can be compared to `0', so that statements like
6153 if (TEST_P (t) && !TEST_P (y))
6156 int i = (TEST_P (t) != 0);
6157 are legal. Macros that return `int' values now may be changed to
6158 return `tree' values, or other pointers in the future. Even those that
6159 continue to return `int' may return multiple nonzero codes where
6160 previously they returned only zero and one. Therefore, you should not
6162 if (TEST_P (t) == 1)
6163 as this code is not guaranteed to work correctly in the future.
6165 You should not take the address of values returned by the macros or
6166 functions described here. In particular, no guarantee is given that the
6169 In general, the names of macros are all in uppercase, while the names
6170 of functions are entirely in lowercase. There are rare exceptions to
6171 this rule. You should assume that any macro or function whose name is
6172 made up entirely of uppercase letters may evaluate its arguments more
6173 than once. You may assume that a macro or function whose name is made
6174 up entirely of lowercase letters will evaluate its arguments only once.
6176 The `error_mark_node' is a special tree. Its tree code is
6177 `ERROR_MARK', but since there is only ever one node with that code, the
6178 usual practice is to compare the tree against `error_mark_node'. (This
6179 test is just a test for pointer equality.) If an error has occurred
6180 during front-end processing the flag `errorcount' will be set. If the
6181 front end has encountered code it cannot handle, it will issue a
6182 message to the user and set `sorrycount'. When these flags are set,
6183 any macro or function which normally returns a tree of a particular
6184 kind may instead return the `error_mark_node'. Thus, if you intend to
6185 do any processing of erroneous code, you must be prepared to deal with
6186 the `error_mark_node'.
6188 Occasionally, a particular tree slot (like an operand to an expression,
6189 or a particular field in a declaration) will be referred to as
6190 "reserved for the back end". These slots are used to store RTL when
6191 the tree is converted to RTL for use by the GCC back end. However, if
6192 that process is not taking place (e.g., if the front end is being hooked
6193 up to an intelligent editor), then those slots may be used by the back
6194 end presently in use.
6196 If you encounter situations that do not match this documentation, such
6197 as tree nodes of types not mentioned here, or macros documented to
6198 return entities of a particular kind that instead return entities of
6199 some different kind, you have found a bug, either in the front end or in
6200 the documentation. Please report these bugs as you would any other bug.
6204 * Macros and Functions::Macros and functions that can be used with all trees.
6205 * Identifiers:: The names of things.
6206 * Containers:: Lists and vectors.
6209 File: gccint.info, Node: Macros and Functions, Next: Identifiers, Up: Tree overview
6214 This section is not here yet.
6217 File: gccint.info, Node: Identifiers, Next: Containers, Prev: Macros and Functions, Up: Tree overview
6222 An `IDENTIFIER_NODE' represents a slightly more general concept that
6223 the standard C or C++ concept of identifier. In particular, an
6224 `IDENTIFIER_NODE' may contain a `$', or other extraordinary characters.
6226 There are never two distinct `IDENTIFIER_NODE's representing the same
6227 identifier. Therefore, you may use pointer equality to compare
6228 `IDENTIFIER_NODE's, rather than using a routine like `strcmp'.
6230 You can use the following macros to access identifiers:
6231 `IDENTIFIER_POINTER'
6232 The string represented by the identifier, represented as a
6233 `char*'. This string is always `NUL'-terminated, and contains no
6234 embedded `NUL' characters.
6237 The length of the string returned by `IDENTIFIER_POINTER', not
6238 including the trailing `NUL'. This value of `IDENTIFIER_LENGTH
6239 (x)' is always the same as `strlen (IDENTIFIER_POINTER (x))'.
6241 `IDENTIFIER_OPNAME_P'
6242 This predicate holds if the identifier represents the name of an
6243 overloaded operator. In this case, you should not depend on the
6244 contents of either the `IDENTIFIER_POINTER' or the
6245 `IDENTIFIER_LENGTH'.
6247 `IDENTIFIER_TYPENAME_P'
6248 This predicate holds if the identifier represents the name of a
6249 user-defined conversion operator. In this case, the `TREE_TYPE' of
6250 the `IDENTIFIER_NODE' holds the type to which the conversion
6255 File: gccint.info, Node: Containers, Prev: Identifiers, Up: Tree overview
6260 Two common container data structures can be represented directly with
6261 tree nodes. A `TREE_LIST' is a singly linked list containing two trees
6262 per node. These are the `TREE_PURPOSE' and `TREE_VALUE' of each node.
6263 (Often, the `TREE_PURPOSE' contains some kind of tag, or additional
6264 information, while the `TREE_VALUE' contains the majority of the
6265 payload. In other cases, the `TREE_PURPOSE' is simply `NULL_TREE',
6266 while in still others both the `TREE_PURPOSE' and `TREE_VALUE' are of
6267 equal stature.) Given one `TREE_LIST' node, the next node is found by
6268 following the `TREE_CHAIN'. If the `TREE_CHAIN' is `NULL_TREE', then
6269 you have reached the end of the list.
6271 A `TREE_VEC' is a simple vector. The `TREE_VEC_LENGTH' is an integer
6272 (not a tree) giving the number of nodes in the vector. The nodes
6273 themselves are accessed using the `TREE_VEC_ELT' macro, which takes two
6274 arguments. The first is the `TREE_VEC' in question; the second is an
6275 integer indicating which element in the vector is desired. The
6276 elements are indexed from zero.
6279 File: gccint.info, Node: Types, Next: Scopes, Prev: Tree overview, Up: Trees
6284 All types have corresponding tree nodes. However, you should not assume
6285 that there is exactly one tree node corresponding to each type. There
6286 are often multiple nodes corresponding to the same type.
6288 For the most part, different kinds of types have different tree codes.
6289 (For example, pointer types use a `POINTER_TYPE' code while arrays use
6290 an `ARRAY_TYPE' code.) However, pointers to member functions use the
6291 `RECORD_TYPE' code. Therefore, when writing a `switch' statement that
6292 depends on the code associated with a particular type, you should take
6293 care to handle pointers to member functions under the `RECORD_TYPE'
6296 In C++, an array type is not qualified; rather the type of the array
6297 elements is qualified. This situation is reflected in the intermediate
6298 representation. The macros described here will always examine the
6299 qualification of the underlying element type when applied to an array
6300 type. (If the element type is itself an array, then the recursion
6301 continues until a non-array type is found, and the qualification of this
6302 type is examined.) So, for example, `CP_TYPE_CONST_P' will hold of the
6303 type `const int ()[7]', denoting an array of seven `int's.
6305 The following functions and macros deal with cv-qualification of types:
6307 This macro returns the set of type qualifiers applied to this type.
6308 This value is `TYPE_UNQUALIFIED' if no qualifiers have been
6309 applied. The `TYPE_QUAL_CONST' bit is set if the type is
6310 `const'-qualified. The `TYPE_QUAL_VOLATILE' bit is set if the
6311 type is `volatile'-qualified. The `TYPE_QUAL_RESTRICT' bit is set
6312 if the type is `restrict'-qualified.
6315 This macro holds if the type is `const'-qualified.
6317 `CP_TYPE_VOLATILE_P'
6318 This macro holds if the type is `volatile'-qualified.
6320 `CP_TYPE_RESTRICT_P'
6321 This macro holds if the type is `restrict'-qualified.
6323 `CP_TYPE_CONST_NON_VOLATILE_P'
6324 This predicate holds for a type that is `const'-qualified, but
6325 _not_ `volatile'-qualified; other cv-qualifiers are ignored as
6326 well: only the `const'-ness is tested.
6329 This macro returns the unqualified version of a type. It may be
6330 applied to an unqualified type, but it is not always the identity
6331 function in that case.
6333 A few other macros and functions are usable with all types:
6335 The number of bits required to represent the type, represented as
6336 an `INTEGER_CST'. For an incomplete type, `TYPE_SIZE' will be
6340 The alignment of the type, in bits, represented as an `int'.
6343 This macro returns a declaration (in the form of a `TYPE_DECL') for
6344 the type. (Note this macro does _not_ return a `IDENTIFIER_NODE',
6345 as you might expect, given its name!) You can look at the
6346 `DECL_NAME' of the `TYPE_DECL' to obtain the actual name of the
6347 type. The `TYPE_NAME' will be `NULL_TREE' for a type that is not
6348 a built-in type, the result of a typedef, or a named class type.
6351 This predicate holds if the type is an integral type. Notice that
6352 in C++, enumerations are _not_ integral types.
6355 This predicate holds if the type is an integral type (in the C++
6356 sense) or a floating point type.
6359 This predicate holds for a class-type.
6362 This predicate holds for a built-in type.
6365 This predicate holds if the type is a pointer to data member.
6368 This predicate holds if the type is a pointer type, and the
6369 pointee is not a data member.
6372 This predicate holds for a pointer to function type.
6375 This predicate holds for a pointer to object type. Note however
6376 that it does not hold for the generic pointer to object type `void
6377 *'. You may use `TYPE_PTROBV_P' to test for a pointer to object
6378 type as well as `void *'.
6381 This macro returns the "canonical" type for the given type node.
6382 Canonical types are used to improve performance in the C++ and
6383 Objective-C++ front ends by allowing efficient comparison between
6384 two type nodes in `same_type_p': if the `TYPE_CANONICAL' values of
6385 the types are equal, the types are equivalent; otherwise, the types
6386 are not equivalent. The notion of equivalence for canonical types
6387 is the same as the notion of type equivalence in the language
6388 itself. For instance,
6390 When `TYPE_CANONICAL' is `NULL_TREE', there is no canonical type
6391 for the given type node. In this case, comparison between this
6392 type and any other type requires the compiler to perform a deep,
6393 "structural" comparison to see if the two type nodes have the same
6394 form and properties.
6396 The canonical type for a node is always the most fundamental type
6397 in the equivalence class of types. For instance, `int' is its own
6398 canonical type. A typedef `I' of `int' will have `int' as its
6399 canonical type. Similarly, `I*' and a typedef `IP' (defined to
6400 `I*') will has `int*' as their canonical type. When building a new
6401 type node, be sure to set `TYPE_CANONICAL' to the appropriate
6402 canonical type. If the new type is a compound type (built from
6403 other types), and any of those other types require structural
6404 equality, use `SET_TYPE_STRUCTURAL_EQUALITY' to ensure that the
6405 new type also requires structural equality. Finally, if for some
6406 reason you cannot guarantee that `TYPE_CANONICAL' will point to
6407 the canonical type, use `SET_TYPE_STRUCTURAL_EQUALITY' to make
6408 sure that the new type-and any type constructed based on
6409 it-requires structural equality. If you suspect that the canonical
6410 type system is miscomparing types, pass `--param
6411 verify-canonical-types=1' to the compiler or configure with
6412 `--enable-checking' to force the compiler to verify its
6413 canonical-type comparisons against the structural comparisons; the
6414 compiler will then print any warnings if the canonical types
6417 `TYPE_STRUCTURAL_EQUALITY_P'
6418 This predicate holds when the node requires structural equality
6419 checks, e.g., when `TYPE_CANONICAL' is `NULL_TREE'.
6421 `SET_TYPE_STRUCTURAL_EQUALITY'
6422 This macro states that the type node it is given requires
6423 structural equality checks, e.g., it sets `TYPE_CANONICAL' to
6427 This predicate takes two types as input, and holds if they are the
6428 same type. For example, if one type is a `typedef' for the other,
6429 or both are `typedef's for the same type. This predicate also
6430 holds if the two trees given as input are simply copies of one
6431 another; i.e., there is no difference between them at the source
6432 level, but, for whatever reason, a duplicate has been made in the
6433 representation. You should never use `==' (pointer equality) to
6434 compare types; always use `same_type_p' instead.
6436 Detailed below are the various kinds of types, and the macros that can
6437 be used to access them. Although other kinds of types are used
6438 elsewhere in G++, the types described here are the only ones that you
6439 will encounter while examining the intermediate representation.
6442 Used to represent the `void' type.
6445 Used to represent the various integral types, including `char',
6446 `short', `int', `long', and `long long'. This code is not used
6447 for enumeration types, nor for the `bool' type. The
6448 `TYPE_PRECISION' is the number of bits used in the representation,
6449 represented as an `unsigned int'. (Note that in the general case
6450 this is not the same value as `TYPE_SIZE'; suppose that there were
6451 a 24-bit integer type, but that alignment requirements for the ABI
6452 required 32-bit alignment. Then, `TYPE_SIZE' would be an
6453 `INTEGER_CST' for 32, while `TYPE_PRECISION' would be 24.) The
6454 integer type is unsigned if `TYPE_UNSIGNED' holds; otherwise, it
6457 The `TYPE_MIN_VALUE' is an `INTEGER_CST' for the smallest integer
6458 that may be represented by this type. Similarly, the
6459 `TYPE_MAX_VALUE' is an `INTEGER_CST' for the largest integer that
6460 may be represented by this type.
6463 Used to represent the `float', `double', and `long double' types.
6464 The number of bits in the floating-point representation is given
6465 by `TYPE_PRECISION', as in the `INTEGER_TYPE' case.
6468 Used to represent the `short _Fract', `_Fract', `long _Fract',
6469 `long long _Fract', `short _Accum', `_Accum', `long _Accum', and
6470 `long long _Accum' types. The number of bits in the fixed-point
6471 representation is given by `TYPE_PRECISION', as in the
6472 `INTEGER_TYPE' case. There may be padding bits, fractional bits
6473 and integral bits. The number of fractional bits is given by
6474 `TYPE_FBIT', and the number of integral bits is given by
6475 `TYPE_IBIT'. The fixed-point type is unsigned if `TYPE_UNSIGNED'
6476 holds; otherwise, it is signed. The fixed-point type is
6477 saturating if `TYPE_SATURATING' holds; otherwise, it is not
6481 Used to represent GCC built-in `__complex__' data types. The
6482 `TREE_TYPE' is the type of the real and imaginary parts.
6485 Used to represent an enumeration type. The `TYPE_PRECISION' gives
6486 (as an `int'), the number of bits used to represent the type. If
6487 there are no negative enumeration constants, `TYPE_UNSIGNED' will
6488 hold. The minimum and maximum enumeration constants may be
6489 obtained with `TYPE_MIN_VALUE' and `TYPE_MAX_VALUE', respectively;
6490 each of these macros returns an `INTEGER_CST'.
6492 The actual enumeration constants themselves may be obtained by
6493 looking at the `TYPE_VALUES'. This macro will return a
6494 `TREE_LIST', containing the constants. The `TREE_PURPOSE' of each
6495 node will be an `IDENTIFIER_NODE' giving the name of the constant;
6496 the `TREE_VALUE' will be an `INTEGER_CST' giving the value
6497 assigned to that constant. These constants will appear in the
6498 order in which they were declared. The `TREE_TYPE' of each of
6499 these constants will be the type of enumeration type itself.
6502 Used to represent the `bool' type.
6505 Used to represent pointer types, and pointer to data member types.
6506 The `TREE_TYPE' gives the type to which this type points. If the
6507 type is a pointer to data member type, then `TYPE_PTRMEM_P' will
6508 hold. For a pointer to data member type of the form `T X::*',
6509 `TYPE_PTRMEM_CLASS_TYPE' will be the type `X', while
6510 `TYPE_PTRMEM_POINTED_TO_TYPE' will be the type `T'.
6513 Used to represent reference types. The `TREE_TYPE' gives the type
6514 to which this type refers.
6517 Used to represent the type of non-member functions and of static
6518 member functions. The `TREE_TYPE' gives the return type of the
6519 function. The `TYPE_ARG_TYPES' are a `TREE_LIST' of the argument
6520 types. The `TREE_VALUE' of each node in this list is the type of
6521 the corresponding argument; the `TREE_PURPOSE' is an expression
6522 for the default argument value, if any. If the last node in the
6523 list is `void_list_node' (a `TREE_LIST' node whose `TREE_VALUE' is
6524 the `void_type_node'), then functions of this type do not take
6525 variable arguments. Otherwise, they do take a variable number of
6528 Note that in C (but not in C++) a function declared like `void f()'
6529 is an unprototyped function taking a variable number of arguments;
6530 the `TYPE_ARG_TYPES' of such a function will be `NULL'.
6533 Used to represent the type of a non-static member function. Like a
6534 `FUNCTION_TYPE', the return type is given by the `TREE_TYPE'. The
6535 type of `*this', i.e., the class of which functions of this type
6536 are a member, is given by the `TYPE_METHOD_BASETYPE'. The
6537 `TYPE_ARG_TYPES' is the parameter list, as for a `FUNCTION_TYPE',
6538 and includes the `this' argument.
6541 Used to represent array types. The `TREE_TYPE' gives the type of
6542 the elements in the array. If the array-bound is present in the
6543 type, the `TYPE_DOMAIN' is an `INTEGER_TYPE' whose
6544 `TYPE_MIN_VALUE' and `TYPE_MAX_VALUE' will be the lower and upper
6545 bounds of the array, respectively. The `TYPE_MIN_VALUE' will
6546 always be an `INTEGER_CST' for zero, while the `TYPE_MAX_VALUE'
6547 will be one less than the number of elements in the array, i.e.,
6548 the highest value which may be used to index an element in the
6552 Used to represent `struct' and `class' types, as well as pointers
6553 to member functions and similar constructs in other languages.
6554 `TYPE_FIELDS' contains the items contained in this type, each of
6555 which can be a `FIELD_DECL', `VAR_DECL', `CONST_DECL', or
6556 `TYPE_DECL'. You may not make any assumptions about the ordering
6557 of the fields in the type or whether one or more of them overlap.
6558 If `TYPE_PTRMEMFUNC_P' holds, then this type is a pointer-to-member
6559 type. In that case, the `TYPE_PTRMEMFUNC_FN_TYPE' is a
6560 `POINTER_TYPE' pointing to a `METHOD_TYPE'. The `METHOD_TYPE' is
6561 the type of a function pointed to by the pointer-to-member
6562 function. If `TYPE_PTRMEMFUNC_P' does not hold, this type is a
6563 class type. For more information, see *note Classes::.
6566 Used to represent `union' types. Similar to `RECORD_TYPE' except
6567 that all `FIELD_DECL' nodes in `TYPE_FIELD' start at bit position
6571 Used to represent part of a variant record in Ada. Similar to
6572 `UNION_TYPE' except that each `FIELD_DECL' has a `DECL_QUALIFIER'
6573 field, which contains a boolean expression that indicates whether
6574 the field is present in the object. The type will only have one
6575 field, so each field's `DECL_QUALIFIER' is only evaluated if none
6576 of the expressions in the previous fields in `TYPE_FIELDS' are
6577 nonzero. Normally these expressions will reference a field in the
6578 outer object using a `PLACEHOLDER_EXPR'.
6581 This node is used to represent a type the knowledge of which is
6582 insufficient for a sound processing.
6585 This node is used to represent a pointer-to-data member. For a
6586 data member `X::m' the `TYPE_OFFSET_BASETYPE' is `X' and the
6587 `TREE_TYPE' is the type of `m'.
6590 Used to represent a construct of the form `typename T::A'. The
6591 `TYPE_CONTEXT' is `T'; the `TYPE_NAME' is an `IDENTIFIER_NODE' for
6592 `A'. If the type is specified via a template-id, then
6593 `TYPENAME_TYPE_FULLNAME' yields a `TEMPLATE_ID_EXPR'. The
6594 `TREE_TYPE' is non-`NULL' if the node is implicitly generated in
6595 support for the implicit typename extension; in which case the
6596 `TREE_TYPE' is a type node for the base-class.
6599 Used to represent the `__typeof__' extension. The `TYPE_FIELDS'
6600 is the expression the type of which is being represented.
6602 There are variables whose values represent some of the basic types.
6610 `unsigned_type_node.'
6611 A node for `unsigned int'.
6615 It may sometimes be useful to compare one of these variables with a
6616 type in hand, using `same_type_p'.
6619 File: gccint.info, Node: Scopes, Next: Functions, Prev: Types, Up: Trees
6624 The root of the entire intermediate representation is the variable
6625 `global_namespace'. This is the namespace specified with `::' in C++
6626 source code. All other namespaces, types, variables, functions, and so
6627 forth can be found starting with this namespace.
6629 Besides namespaces, the other high-level scoping construct in C++ is
6630 the class. (Throughout this manual the term "class" is used to mean the
6631 types referred to in the ANSI/ISO C++ Standard as classes; these include
6632 types defined with the `class', `struct', and `union' keywords.)
6636 * Namespaces:: Member functions, types, etc.
6637 * Classes:: Members, bases, friends, etc.
6640 File: gccint.info, Node: Namespaces, Next: Classes, Up: Scopes
6645 A namespace is represented by a `NAMESPACE_DECL' node.
6647 However, except for the fact that it is distinguished as the root of
6648 the representation, the global namespace is no different from any other
6649 namespace. Thus, in what follows, we describe namespaces generally,
6650 rather than the global namespace in particular.
6652 The following macros and functions can be used on a `NAMESPACE_DECL':
6655 This macro is used to obtain the `IDENTIFIER_NODE' corresponding to
6656 the unqualified name of the name of the namespace (*note
6657 Identifiers::). The name of the global namespace is `::', even
6658 though in C++ the global namespace is unnamed. However, you
6659 should use comparison with `global_namespace', rather than
6660 `DECL_NAME' to determine whether or not a namespace is the global
6661 one. An unnamed namespace will have a `DECL_NAME' equal to
6662 `anonymous_namespace_name'. Within a single translation unit, all
6663 unnamed namespaces will have the same name.
6666 This macro returns the enclosing namespace. The `DECL_CONTEXT' for
6667 the `global_namespace' is `NULL_TREE'.
6669 `DECL_NAMESPACE_ALIAS'
6670 If this declaration is for a namespace alias, then
6671 `DECL_NAMESPACE_ALIAS' is the namespace for which this one is an
6674 Do not attempt to use `cp_namespace_decls' for a namespace which is
6675 an alias. Instead, follow `DECL_NAMESPACE_ALIAS' links until you
6676 reach an ordinary, non-alias, namespace, and call
6677 `cp_namespace_decls' there.
6679 `DECL_NAMESPACE_STD_P'
6680 This predicate holds if the namespace is the special `::std'
6683 `cp_namespace_decls'
6684 This function will return the declarations contained in the
6685 namespace, including types, overloaded functions, other
6686 namespaces, and so forth. If there are no declarations, this
6687 function will return `NULL_TREE'. The declarations are connected
6688 through their `TREE_CHAIN' fields.
6690 Although most entries on this list will be declarations,
6691 `TREE_LIST' nodes may also appear. In this case, the `TREE_VALUE'
6692 will be an `OVERLOAD'. The value of the `TREE_PURPOSE' is
6693 unspecified; back ends should ignore this value. As with the
6694 other kinds of declarations returned by `cp_namespace_decls', the
6695 `TREE_CHAIN' will point to the next declaration in this list.
6697 For more information on the kinds of declarations that can occur
6698 on this list, *Note Declarations::. Some declarations will not
6699 appear on this list. In particular, no `FIELD_DECL',
6700 `LABEL_DECL', or `PARM_DECL' nodes will appear here.
6702 This function cannot be used with namespaces that have
6703 `DECL_NAMESPACE_ALIAS' set.
6707 File: gccint.info, Node: Classes, Prev: Namespaces, Up: Scopes
6712 A class type is represented by either a `RECORD_TYPE' or a
6713 `UNION_TYPE'. A class declared with the `union' tag is represented by
6714 a `UNION_TYPE', while classes declared with either the `struct' or the
6715 `class' tag are represented by `RECORD_TYPE's. You can use the
6716 `CLASSTYPE_DECLARED_CLASS' macro to discern whether or not a particular
6717 type is a `class' as opposed to a `struct'. This macro will be true
6718 only for classes declared with the `class' tag.
6720 Almost all non-function members are available on the `TYPE_FIELDS'
6721 list. Given one member, the next can be found by following the
6722 `TREE_CHAIN'. You should not depend in any way on the order in which
6723 fields appear on this list. All nodes on this list will be `DECL'
6724 nodes. A `FIELD_DECL' is used to represent a non-static data member, a
6725 `VAR_DECL' is used to represent a static data member, and a `TYPE_DECL'
6726 is used to represent a type. Note that the `CONST_DECL' for an
6727 enumeration constant will appear on this list, if the enumeration type
6728 was declared in the class. (Of course, the `TYPE_DECL' for the
6729 enumeration type will appear here as well.) There are no entries for
6730 base classes on this list. In particular, there is no `FIELD_DECL' for
6731 the "base-class portion" of an object.
6733 The `TYPE_VFIELD' is a compiler-generated field used to point to
6734 virtual function tables. It may or may not appear on the `TYPE_FIELDS'
6735 list. However, back ends should handle the `TYPE_VFIELD' just like all
6736 the entries on the `TYPE_FIELDS' list.
6738 The function members are available on the `TYPE_METHODS' list. Again,
6739 subsequent members are found by following the `TREE_CHAIN' field. If a
6740 function is overloaded, each of the overloaded functions appears; no
6741 `OVERLOAD' nodes appear on the `TYPE_METHODS' list. Implicitly
6742 declared functions (including default constructors, copy constructors,
6743 assignment operators, and destructors) will appear on this list as well.
6745 Every class has an associated "binfo", which can be obtained with
6746 `TYPE_BINFO'. Binfos are used to represent base-classes. The binfo
6747 given by `TYPE_BINFO' is the degenerate case, whereby every class is
6748 considered to be its own base-class. The base binfos for a particular
6749 binfo are held in a vector, whose length is obtained with
6750 `BINFO_N_BASE_BINFOS'. The base binfos themselves are obtained with
6751 `BINFO_BASE_BINFO' and `BINFO_BASE_ITERATE'. To add a new binfo, use
6752 `BINFO_BASE_APPEND'. The vector of base binfos can be obtained with
6753 `BINFO_BASE_BINFOS', but normally you do not need to use that. The
6754 class type associated with a binfo is given by `BINFO_TYPE'. It is not
6755 always the case that `BINFO_TYPE (TYPE_BINFO (x))', because of typedefs
6756 and qualified types. Neither is it the case that `TYPE_BINFO
6757 (BINFO_TYPE (y))' is the same binfo as `y'. The reason is that if `y'
6758 is a binfo representing a base-class `B' of a derived class `D', then
6759 `BINFO_TYPE (y)' will be `B', and `TYPE_BINFO (BINFO_TYPE (y))' will be
6760 `B' as its own base-class, rather than as a base-class of `D'.
6762 The access to a base type can be found with `BINFO_BASE_ACCESS'. This
6763 will produce `access_public_node', `access_private_node' or
6764 `access_protected_node'. If bases are always public,
6765 `BINFO_BASE_ACCESSES' may be `NULL'.
6767 `BINFO_VIRTUAL_P' is used to specify whether the binfo is inherited
6768 virtually or not. The other flags, `BINFO_MARKED_P' and `BINFO_FLAG_1'
6769 to `BINFO_FLAG_6' can be used for language specific use.
6771 The following macros can be used on a tree node representing a
6775 This predicate holds if the class is local class _i.e._ declared
6776 inside a function body.
6778 `TYPE_POLYMORPHIC_P'
6779 This predicate holds if the class has at least one virtual function
6780 (declared or inherited).
6782 `TYPE_HAS_DEFAULT_CONSTRUCTOR'
6783 This predicate holds whenever its argument represents a class-type
6784 with default constructor.
6786 `CLASSTYPE_HAS_MUTABLE'
6787 `TYPE_HAS_MUTABLE_P'
6788 These predicates hold for a class-type having a mutable data
6791 `CLASSTYPE_NON_POD_P'
6792 This predicate holds only for class-types that are not PODs.
6794 `TYPE_HAS_NEW_OPERATOR'
6795 This predicate holds for a class-type that defines `operator new'.
6797 `TYPE_HAS_ARRAY_NEW_OPERATOR'
6798 This predicate holds for a class-type for which `operator new[]'
6801 `TYPE_OVERLOADS_CALL_EXPR'
6802 This predicate holds for class-type for which the function call
6803 `operator()' is overloaded.
6805 `TYPE_OVERLOADS_ARRAY_REF'
6806 This predicate holds for a class-type that overloads `operator[]'
6808 `TYPE_OVERLOADS_ARROW'
6809 This predicate holds for a class-type for which `operator->' is
6814 File: gccint.info, Node: Declarations, Next: Attributes, Prev: Functions, Up: Trees
6819 This section covers the various kinds of declarations that appear in the
6820 internal representation, except for declarations of functions
6821 (represented by `FUNCTION_DECL' nodes), which are described in *Note
6826 * Working with declarations:: Macros and functions that work on
6828 * Internal structure:: How declaration nodes are represented.
6831 File: gccint.info, Node: Working with declarations, Next: Internal structure, Up: Declarations
6833 9.5.1 Working with declarations
6834 -------------------------------
6836 Some macros can be used with any kind of declaration. These include:
6838 This macro returns an `IDENTIFIER_NODE' giving the name of the
6842 This macro returns the type of the entity declared.
6845 This macro returns the name of the file in which the entity was
6846 declared, as a `char*'. For an entity declared implicitly by the
6847 compiler (like `__builtin_memcpy'), this will be the string
6851 This macro returns the line number at which the entity was
6852 declared, as an `int'.
6855 This predicate holds if the declaration was implicitly generated
6856 by the compiler. For example, this predicate will hold of an
6857 implicitly declared member function, or of the `TYPE_DECL'
6858 implicitly generated for a class type. Recall that in C++ code
6861 is roughly equivalent to C code like:
6864 The implicitly generated `typedef' declaration is represented by a
6865 `TYPE_DECL' for which `DECL_ARTIFICIAL' holds.
6867 `DECL_NAMESPACE_SCOPE_P'
6868 This predicate holds if the entity was declared at a namespace
6871 `DECL_CLASS_SCOPE_P'
6872 This predicate holds if the entity was declared at a class scope.
6874 `DECL_FUNCTION_SCOPE_P'
6875 This predicate holds if the entity was declared inside a function
6879 The various kinds of declarations include:
6881 These nodes are used to represent labels in function bodies. For
6882 more information, see *Note Functions::. These nodes only appear
6886 These nodes are used to represent enumeration constants. The
6887 value of the constant is given by `DECL_INITIAL' which will be an
6888 `INTEGER_CST' with the same type as the `TREE_TYPE' of the
6889 `CONST_DECL', i.e., an `ENUMERAL_TYPE'.
6892 These nodes represent the value returned by a function. When a
6893 value is assigned to a `RESULT_DECL', that indicates that the
6894 value should be returned, via bitwise copy, by the function. You
6895 can use `DECL_SIZE' and `DECL_ALIGN' on a `RESULT_DECL', just as
6899 These nodes represent `typedef' declarations. The `TREE_TYPE' is
6900 the type declared to have the name given by `DECL_NAME'. In some
6901 cases, there is no associated name.
6904 These nodes represent variables with namespace or block scope, as
6905 well as static data members. The `DECL_SIZE' and `DECL_ALIGN' are
6906 analogous to `TYPE_SIZE' and `TYPE_ALIGN'. For a declaration, you
6907 should always use the `DECL_SIZE' and `DECL_ALIGN' rather than the
6908 `TYPE_SIZE' and `TYPE_ALIGN' given by the `TREE_TYPE', since
6909 special attributes may have been applied to the variable to give
6910 it a particular size and alignment. You may use the predicates
6911 `DECL_THIS_STATIC' or `DECL_THIS_EXTERN' to test whether the
6912 storage class specifiers `static' or `extern' were used to declare
6915 If this variable is initialized (but does not require a
6916 constructor), the `DECL_INITIAL' will be an expression for the
6917 initializer. The initializer should be evaluated, and a bitwise
6918 copy into the variable performed. If the `DECL_INITIAL' is the
6919 `error_mark_node', there is an initializer, but it is given by an
6920 explicit statement later in the code; no bitwise copy is required.
6922 GCC provides an extension that allows either automatic variables,
6923 or global variables, to be placed in particular registers. This
6924 extension is being used for a particular `VAR_DECL' if
6925 `DECL_REGISTER' holds for the `VAR_DECL', and if
6926 `DECL_ASSEMBLER_NAME' is not equal to `DECL_NAME'. In that case,
6927 `DECL_ASSEMBLER_NAME' is the name of the register into which the
6928 variable will be placed.
6931 Used to represent a parameter to a function. Treat these nodes
6932 similarly to `VAR_DECL' nodes. These nodes only appear in the
6933 `DECL_ARGUMENTS' for a `FUNCTION_DECL'.
6935 The `DECL_ARG_TYPE' for a `PARM_DECL' is the type that will
6936 actually be used when a value is passed to this function. It may
6937 be a wider type than the `TREE_TYPE' of the parameter; for
6938 example, the ordinary type might be `short' while the
6939 `DECL_ARG_TYPE' is `int'.
6942 These nodes represent non-static data members. The `DECL_SIZE' and
6943 `DECL_ALIGN' behave as for `VAR_DECL' nodes. The position of the
6944 field within the parent record is specified by a combination of
6945 three attributes. `DECL_FIELD_OFFSET' is the position, counting
6946 in bytes, of the `DECL_OFFSET_ALIGN'-bit sized word containing the
6947 bit of the field closest to the beginning of the structure.
6948 `DECL_FIELD_BIT_OFFSET' is the bit offset of the first bit of the
6949 field within this word; this may be nonzero even for fields that
6950 are not bit-fields, since `DECL_OFFSET_ALIGN' may be greater than
6951 the natural alignment of the field's type.
6953 If `DECL_C_BIT_FIELD' holds, this field is a bit-field. In a
6954 bit-field, `DECL_BIT_FIELD_TYPE' also contains the type that was
6955 originally specified for it, while DECL_TYPE may be a modified
6956 type with lesser precision, according to the size of the bit field.
6962 These nodes are used to represent class, function, and variable
6963 (static data member) templates. The
6964 `DECL_TEMPLATE_SPECIALIZATIONS' are a `TREE_LIST'. The
6965 `TREE_VALUE' of each node in the list is a `TEMPLATE_DECL's or
6966 `FUNCTION_DECL's representing specializations (including
6967 instantiations) of this template. Back ends can safely ignore
6968 `TEMPLATE_DECL's, but should examine `FUNCTION_DECL' nodes on the
6969 specializations list just as they would ordinary `FUNCTION_DECL'
6972 For a class template, the `DECL_TEMPLATE_INSTANTIATIONS' list
6973 contains the instantiations. The `TREE_VALUE' of each node is an
6974 instantiation of the class. The `DECL_TEMPLATE_SPECIALIZATIONS'
6975 contains partial specializations of the class.
6978 Back ends can safely ignore these nodes.
6982 File: gccint.info, Node: Internal structure, Prev: Working with declarations, Up: Declarations
6984 9.5.2 Internal structure
6985 ------------------------
6987 `DECL' nodes are represented internally as a hierarchy of structures.
6991 * Current structure hierarchy:: The current DECL node structure
6993 * Adding new DECL node types:: How to add a new DECL node to a
6997 File: gccint.info, Node: Current structure hierarchy, Next: Adding new DECL node types, Up: Internal structure
6999 9.5.2.1 Current structure hierarchy
7000 ...................................
7002 `struct tree_decl_minimal'
7003 This is the minimal structure to inherit from in order for common
7004 `DECL' macros to work. The fields it contains are a unique ID,
7005 source location, context, and name.
7007 `struct tree_decl_common'
7008 This structure inherits from `struct tree_decl_minimal'. It
7009 contains fields that most `DECL' nodes need, such as a field to
7010 store alignment, machine mode, size, and attributes.
7012 `struct tree_field_decl'
7013 This structure inherits from `struct tree_decl_common'. It is
7014 used to represent `FIELD_DECL'.
7016 `struct tree_label_decl'
7017 This structure inherits from `struct tree_decl_common'. It is
7018 used to represent `LABEL_DECL'.
7020 `struct tree_translation_unit_decl'
7021 This structure inherits from `struct tree_decl_common'. It is
7022 used to represent `TRANSLATION_UNIT_DECL'.
7024 `struct tree_decl_with_rtl'
7025 This structure inherits from `struct tree_decl_common'. It
7026 contains a field to store the low-level RTL associated with a
7029 `struct tree_result_decl'
7030 This structure inherits from `struct tree_decl_with_rtl'. It is
7031 used to represent `RESULT_DECL'.
7033 `struct tree_const_decl'
7034 This structure inherits from `struct tree_decl_with_rtl'. It is
7035 used to represent `CONST_DECL'.
7037 `struct tree_parm_decl'
7038 This structure inherits from `struct tree_decl_with_rtl'. It is
7039 used to represent `PARM_DECL'.
7041 `struct tree_decl_with_vis'
7042 This structure inherits from `struct tree_decl_with_rtl'. It
7043 contains fields necessary to store visibility information, as well
7044 as a section name and assembler name.
7046 `struct tree_var_decl'
7047 This structure inherits from `struct tree_decl_with_vis'. It is
7048 used to represent `VAR_DECL'.
7050 `struct tree_function_decl'
7051 This structure inherits from `struct tree_decl_with_vis'. It is
7052 used to represent `FUNCTION_DECL'.
7056 File: gccint.info, Node: Adding new DECL node types, Prev: Current structure hierarchy, Up: Internal structure
7058 9.5.2.2 Adding new DECL node types
7059 ..................................
7061 Adding a new `DECL' tree consists of the following steps
7063 Add a new tree code for the `DECL' node
7064 For language specific `DECL' nodes, there is a `.def' file in each
7065 frontend directory where the tree code should be added. For
7066 `DECL' nodes that are part of the middle-end, the code should be
7067 added to `tree.def'.
7069 Create a new structure type for the `DECL' node
7070 These structures should inherit from one of the existing
7071 structures in the language hierarchy by using that structure as
7074 struct tree_foo_decl
7076 struct tree_decl_with_vis common;
7079 Would create a structure name `tree_foo_decl' that inherits from
7080 `struct tree_decl_with_vis'.
7082 For language specific `DECL' nodes, this new structure type should
7083 go in the appropriate `.h' file. For `DECL' nodes that are part
7084 of the middle-end, the structure type should go in `tree.h'.
7086 Add a member to the tree structure enumerator for the node
7087 For garbage collection and dynamic checking purposes, each `DECL'
7088 node structure type is required to have a unique enumerator value
7089 specified with it. For language specific `DECL' nodes, this new
7090 enumerator value should go in the appropriate `.def' file. For
7091 `DECL' nodes that are part of the middle-end, the enumerator
7092 values are specified in `treestruct.def'.
7094 Update `union tree_node'
7095 In order to make your new structure type usable, it must be added
7096 to `union tree_node'. For language specific `DECL' nodes, a new
7097 entry should be added to the appropriate `.h' file of the form
7098 struct tree_foo_decl GTY ((tag ("TS_VAR_DECL"))) foo_decl;
7099 For `DECL' nodes that are part of the middle-end, the additional
7100 member goes directly into `union tree_node' in `tree.h'.
7102 Update dynamic checking info
7103 In order to be able to check whether accessing a named portion of
7104 `union tree_node' is legal, and whether a certain `DECL' node
7105 contains one of the enumerated `DECL' node structures in the
7106 hierarchy, a simple lookup table is used. This lookup table needs
7107 to be kept up to date with the tree structure hierarchy, or else
7108 checking and containment macros will fail inappropriately.
7110 For language specific `DECL' nodes, their is an `init_ts' function
7111 in an appropriate `.c' file, which initializes the lookup table.
7112 Code setting up the table for new `DECL' nodes should be added
7113 there. For each `DECL' tree code and enumerator value
7114 representing a member of the inheritance hierarchy, the table
7115 should contain 1 if that tree code inherits (directly or
7116 indirectly) from that member. Thus, a `FOO_DECL' node derived
7117 from `struct decl_with_rtl', and enumerator value `TS_FOO_DECL',
7118 would be set up as follows
7119 tree_contains_struct[FOO_DECL][TS_FOO_DECL] = 1;
7120 tree_contains_struct[FOO_DECL][TS_DECL_WRTL] = 1;
7121 tree_contains_struct[FOO_DECL][TS_DECL_COMMON] = 1;
7122 tree_contains_struct[FOO_DECL][TS_DECL_MINIMAL] = 1;
7124 For `DECL' nodes that are part of the middle-end, the setup code
7127 Add macros to access any new fields and flags
7128 Each added field or flag should have a macro that is used to access
7129 it, that performs appropriate checking to ensure only the right
7130 type of `DECL' nodes access the field.
7132 These macros generally take the following form
7133 #define FOO_DECL_FIELDNAME(NODE) FOO_DECL_CHECK(NODE)->foo_decl.fieldname
7134 However, if the structure is simply a base class for further
7135 structures, something like the following should be used
7136 #define BASE_STRUCT_CHECK(T) CONTAINS_STRUCT_CHECK(T, TS_BASE_STRUCT)
7137 #define BASE_STRUCT_FIELDNAME(NODE) \
7138 (BASE_STRUCT_CHECK(NODE)->base_struct.fieldname
7142 File: gccint.info, Node: Functions, Next: Declarations, Prev: Scopes, Up: Trees
7147 A function is represented by a `FUNCTION_DECL' node. A set of
7148 overloaded functions is sometimes represented by a `OVERLOAD' node.
7150 An `OVERLOAD' node is not a declaration, so none of the `DECL_' macros
7151 should be used on an `OVERLOAD'. An `OVERLOAD' node is similar to a
7152 `TREE_LIST'. Use `OVL_CURRENT' to get the function associated with an
7153 `OVERLOAD' node; use `OVL_NEXT' to get the next `OVERLOAD' node in the
7154 list of overloaded functions. The macros `OVL_CURRENT' and `OVL_NEXT'
7155 are actually polymorphic; you can use them to work with `FUNCTION_DECL'
7156 nodes as well as with overloads. In the case of a `FUNCTION_DECL',
7157 `OVL_CURRENT' will always return the function itself, and `OVL_NEXT'
7158 will always be `NULL_TREE'.
7160 To determine the scope of a function, you can use the `DECL_CONTEXT'
7161 macro. This macro will return the class (either a `RECORD_TYPE' or a
7162 `UNION_TYPE') or namespace (a `NAMESPACE_DECL') of which the function
7163 is a member. For a virtual function, this macro returns the class in
7164 which the function was actually defined, not the base class in which
7165 the virtual declaration occurred.
7167 If a friend function is defined in a class scope, the
7168 `DECL_FRIEND_CONTEXT' macro can be used to determine the class in which
7169 it was defined. For example, in
7170 class C { friend void f() {} };
7171 the `DECL_CONTEXT' for `f' will be the `global_namespace', but the
7172 `DECL_FRIEND_CONTEXT' will be the `RECORD_TYPE' for `C'.
7174 In C, the `DECL_CONTEXT' for a function maybe another function. This
7175 representation indicates that the GNU nested function extension is in
7176 use. For details on the semantics of nested functions, see the GCC
7177 Manual. The nested function can refer to local variables in its
7178 containing function. Such references are not explicitly marked in the
7179 tree structure; back ends must look at the `DECL_CONTEXT' for the
7180 referenced `VAR_DECL'. If the `DECL_CONTEXT' for the referenced
7181 `VAR_DECL' is not the same as the function currently being processed,
7182 and neither `DECL_EXTERNAL' nor `TREE_STATIC' hold, then the reference
7183 is to a local variable in a containing function, and the back end must
7184 take appropriate action.
7188 * Function Basics:: Function names, linkage, and so forth.
7189 * Function Bodies:: The statements that make up a function body.
7192 File: gccint.info, Node: Function Basics, Next: Function Bodies, Up: Functions
7194 9.6.1 Function Basics
7195 ---------------------
7197 The following macros and functions can be used on a `FUNCTION_DECL':
7199 This predicate holds for a function that is the program entry point
7203 This macro returns the unqualified name of the function, as an
7204 `IDENTIFIER_NODE'. For an instantiation of a function template,
7205 the `DECL_NAME' is the unqualified name of the template, not
7206 something like `f<int>'. The value of `DECL_NAME' is undefined
7207 when used on a constructor, destructor, overloaded operator, or
7208 type-conversion operator, or any function that is implicitly
7209 generated by the compiler. See below for macros that can be used
7210 to distinguish these cases.
7212 `DECL_ASSEMBLER_NAME'
7213 This macro returns the mangled name of the function, also an
7214 `IDENTIFIER_NODE'. This name does not contain leading underscores
7215 on systems that prefix all identifiers with underscores. The
7216 mangled name is computed in the same way on all platforms; if
7217 special processing is required to deal with the object file format
7218 used on a particular platform, it is the responsibility of the
7219 back end to perform those modifications. (Of course, the back end
7220 should not modify `DECL_ASSEMBLER_NAME' itself.)
7222 Using `DECL_ASSEMBLER_NAME' will cause additional memory to be
7223 allocated (for the mangled name of the entity) so it should be used
7224 only when emitting assembly code. It should not be used within the
7225 optimizers to determine whether or not two declarations are the
7226 same, even though some of the existing optimizers do use it in
7227 that way. These uses will be removed over time.
7230 This predicate holds if the function is undefined.
7233 This predicate holds if the function has external linkage.
7235 `DECL_LOCAL_FUNCTION_P'
7236 This predicate holds if the function was declared at block scope,
7237 even though it has a global scope.
7240 This predicate holds if the function is a built-in function but its
7241 prototype is not yet explicitly declared.
7243 `DECL_EXTERN_C_FUNCTION_P'
7244 This predicate holds if the function is declared as an ``extern
7248 This macro holds if multiple copies of this function may be
7249 emitted in various translation units. It is the responsibility of
7250 the linker to merge the various copies. Template instantiations
7251 are the most common example of functions for which
7252 `DECL_LINKONCE_P' holds; G++ instantiates needed templates in all
7253 translation units which require them, and then relies on the
7254 linker to remove duplicate instantiations.
7256 FIXME: This macro is not yet implemented.
7258 `DECL_FUNCTION_MEMBER_P'
7259 This macro holds if the function is a member of a class, rather
7260 than a member of a namespace.
7262 `DECL_STATIC_FUNCTION_P'
7263 This predicate holds if the function a static member function.
7265 `DECL_NONSTATIC_MEMBER_FUNCTION_P'
7266 This macro holds for a non-static member function.
7268 `DECL_CONST_MEMFUNC_P'
7269 This predicate holds for a `const'-member function.
7271 `DECL_VOLATILE_MEMFUNC_P'
7272 This predicate holds for a `volatile'-member function.
7274 `DECL_CONSTRUCTOR_P'
7275 This macro holds if the function is a constructor.
7277 `DECL_NONCONVERTING_P'
7278 This predicate holds if the constructor is a non-converting
7281 `DECL_COMPLETE_CONSTRUCTOR_P'
7282 This predicate holds for a function which is a constructor for an
7283 object of a complete type.
7285 `DECL_BASE_CONSTRUCTOR_P'
7286 This predicate holds for a function which is a constructor for a
7287 base class sub-object.
7289 `DECL_COPY_CONSTRUCTOR_P'
7290 This predicate holds for a function which is a copy-constructor.
7293 This macro holds if the function is a destructor.
7295 `DECL_COMPLETE_DESTRUCTOR_P'
7296 This predicate holds if the function is the destructor for an
7297 object a complete type.
7299 `DECL_OVERLOADED_OPERATOR_P'
7300 This macro holds if the function is an overloaded operator.
7303 This macro holds if the function is a type-conversion operator.
7305 `DECL_GLOBAL_CTOR_P'
7306 This predicate holds if the function is a file-scope initialization
7309 `DECL_GLOBAL_DTOR_P'
7310 This predicate holds if the function is a file-scope finalization
7314 This predicate holds if the function is a thunk.
7316 These functions represent stub code that adjusts the `this' pointer
7317 and then jumps to another function. When the jumped-to function
7318 returns, control is transferred directly to the caller, without
7319 returning to the thunk. The first parameter to the thunk is
7320 always the `this' pointer; the thunk should add `THUNK_DELTA' to
7321 this value. (The `THUNK_DELTA' is an `int', not an `INTEGER_CST'.)
7323 Then, if `THUNK_VCALL_OFFSET' (an `INTEGER_CST') is nonzero the
7324 adjusted `this' pointer must be adjusted again. The complete
7325 calculation is given by the following pseudo-code:
7328 if (THUNK_VCALL_OFFSET)
7329 this += (*((ptrdiff_t **) this))[THUNK_VCALL_OFFSET]
7331 Finally, the thunk should jump to the location given by
7332 `DECL_INITIAL'; this will always be an expression for the address
7335 `DECL_NON_THUNK_FUNCTION_P'
7336 This predicate holds if the function is _not_ a thunk function.
7338 `GLOBAL_INIT_PRIORITY'
7339 If either `DECL_GLOBAL_CTOR_P' or `DECL_GLOBAL_DTOR_P' holds, then
7340 this gives the initialization priority for the function. The
7341 linker will arrange that all functions for which
7342 `DECL_GLOBAL_CTOR_P' holds are run in increasing order of priority
7343 before `main' is called. When the program exits, all functions for
7344 which `DECL_GLOBAL_DTOR_P' holds are run in the reverse order.
7347 This macro holds if the function was implicitly generated by the
7348 compiler, rather than explicitly declared. In addition to
7349 implicitly generated class member functions, this macro holds for
7350 the special functions created to implement static initialization
7351 and destruction, to compute run-time type information, and so
7355 This macro returns the `PARM_DECL' for the first argument to the
7356 function. Subsequent `PARM_DECL' nodes can be obtained by
7357 following the `TREE_CHAIN' links.
7360 This macro returns the `RESULT_DECL' for the function.
7363 This macro returns the `FUNCTION_TYPE' or `METHOD_TYPE' for the
7366 `TYPE_RAISES_EXCEPTIONS'
7367 This macro returns the list of exceptions that a (member-)function
7368 can raise. The returned list, if non `NULL', is comprised of nodes
7369 whose `TREE_VALUE' represents a type.
7372 This predicate holds when the exception-specification of its
7373 arguments is of the form ``()''.
7375 `DECL_ARRAY_DELETE_OPERATOR_P'
7376 This predicate holds if the function an overloaded `operator
7379 `DECL_FUNCTION_SPECIFIC_TARGET'
7380 This macro returns a tree node that holds the target options that
7381 are to be used to compile this particular function or `NULL_TREE'
7382 if the function is to be compiled with the target options
7383 specified on the command line.
7385 `DECL_FUNCTION_SPECIFIC_OPTIMIZATION'
7386 This macro returns a tree node that holds the optimization options
7387 that are to be used to compile this particular function or
7388 `NULL_TREE' if the function is to be compiled with the
7389 optimization options specified on the command line.
7392 File: gccint.info, Node: Function Bodies, Prev: Function Basics, Up: Functions
7394 9.6.2 Function Bodies
7395 ---------------------
7397 A function that has a definition in the current translation unit will
7398 have a non-`NULL' `DECL_INITIAL'. However, back ends should not make
7399 use of the particular value given by `DECL_INITIAL'.
7401 The `DECL_SAVED_TREE' macro will give the complete body of the
7407 There are tree nodes corresponding to all of the source-level statement
7408 constructs, used within the C and C++ frontends. These are enumerated
7409 here, together with a list of the various macros that can be used to
7410 obtain information about them. There are a few macros that can be used
7411 with all statements:
7413 `STMT_IS_FULL_EXPR_P'
7414 In C++, statements normally constitute "full expressions";
7415 temporaries created during a statement are destroyed when the
7416 statement is complete. However, G++ sometimes represents
7417 expressions by statements; these statements will not have
7418 `STMT_IS_FULL_EXPR_P' set. Temporaries created during such
7419 statements should be destroyed when the innermost enclosing
7420 statement with `STMT_IS_FULL_EXPR_P' set is exited.
7423 Here is the list of the various statement nodes, and the macros used to
7424 access them. This documentation describes the use of these nodes in
7425 non-template functions (including instantiations of template functions).
7426 In template functions, the same nodes are used, but sometimes in
7427 slightly different ways.
7429 Many of the statements have substatements. For example, a `while'
7430 loop will have a body, which is itself a statement. If the substatement
7431 is `NULL_TREE', it is considered equivalent to a statement consisting
7432 of a single `;', i.e., an expression statement in which the expression
7433 has been omitted. A substatement may in fact be a list of statements,
7434 connected via their `TREE_CHAIN's. So, you should always process the
7435 statement tree by looping over substatements, like this:
7436 void process_stmt (stmt)
7441 switch (TREE_CODE (stmt))
7444 process_stmt (THEN_CLAUSE (stmt));
7445 /* More processing here. */
7451 stmt = TREE_CHAIN (stmt);
7454 In other words, while the `then' clause of an `if' statement in C++
7455 can be only one statement (although that one statement may be a
7456 compound statement), the intermediate representation will sometimes use
7457 several statements chained together.
7460 Used to represent an inline assembly statement. For an inline
7461 assembly statement like:
7463 The `ASM_STRING' macro will return a `STRING_CST' node for `"mov
7464 x, y"'. If the original statement made use of the
7465 extended-assembly syntax, then `ASM_OUTPUTS', `ASM_INPUTS', and
7466 `ASM_CLOBBERS' will be the outputs, inputs, and clobbers for the
7467 statement, represented as `STRING_CST' nodes. The
7468 extended-assembly syntax looks like:
7469 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
7470 The first string is the `ASM_STRING', containing the instruction
7471 template. The next two strings are the output and inputs,
7472 respectively; this statement has no clobbers. As this example
7473 indicates, "plain" assembly statements are merely a special case
7474 of extended assembly statements; they have no cv-qualifiers,
7475 outputs, inputs, or clobbers. All of the strings will be
7476 `NUL'-terminated, and will contain no embedded `NUL'-characters.
7478 If the assembly statement is declared `volatile', or if the
7479 statement was not an extended assembly statement, and is therefore
7480 implicitly volatile, then the predicate `ASM_VOLATILE_P' will hold
7484 Used to represent a `break' statement. There are no additional
7488 Use to represent a `case' label, range of `case' labels, or a
7489 `default' label. If `CASE_LOW' is `NULL_TREE', then this is a
7490 `default' label. Otherwise, if `CASE_HIGH' is `NULL_TREE', then
7491 this is an ordinary `case' label. In this case, `CASE_LOW' is an
7492 expression giving the value of the label. Both `CASE_LOW' and
7493 `CASE_HIGH' are `INTEGER_CST' nodes. These values will have the
7494 same type as the condition expression in the switch statement.
7496 Otherwise, if both `CASE_LOW' and `CASE_HIGH' are defined, the
7497 statement is a range of case labels. Such statements originate
7498 with the extension that allows users to write things of the form:
7500 The first value will be `CASE_LOW', while the second will be
7504 Used to represent an action that should take place upon exit from
7505 the enclosing scope. Typically, these actions are calls to
7506 destructors for local objects, but back ends cannot rely on this
7507 fact. If these nodes are in fact representing such destructors,
7508 `CLEANUP_DECL' will be the `VAR_DECL' destroyed. Otherwise,
7509 `CLEANUP_DECL' will be `NULL_TREE'. In any case, the
7510 `CLEANUP_EXPR' is the expression to execute. The cleanups
7511 executed on exit from a scope should be run in the reverse order
7512 of the order in which the associated `CLEANUP_STMT's were
7516 Used to represent a `continue' statement. There are no additional
7520 Used to mark the beginning (if `CTOR_BEGIN_P' holds) or end (if
7521 `CTOR_END_P' holds of the main body of a constructor. See also
7522 `SUBOBJECT' for more information on how to use these nodes.
7525 Used to represent a local declaration. The `DECL_STMT_DECL' macro
7526 can be used to obtain the entity declared. This declaration may
7527 be a `LABEL_DECL', indicating that the label declared is a local
7528 label. (As an extension, GCC allows the declaration of labels
7529 with scope.) In C, this declaration may be a `FUNCTION_DECL',
7530 indicating the use of the GCC nested function extension. For more
7531 information, *note Functions::.
7534 Used to represent a `do' loop. The body of the loop is given by
7535 `DO_BODY' while the termination condition for the loop is given by
7536 `DO_COND'. The condition for a `do'-statement is always an
7540 Used to represent a temporary object of a class with no data whose
7541 address is never taken. (All such objects are interchangeable.)
7542 The `TREE_TYPE' represents the type of the object.
7545 Used to represent an expression statement. Use `EXPR_STMT_EXPR' to
7546 obtain the expression.
7549 Used to represent a `for' statement. The `FOR_INIT_STMT' is the
7550 initialization statement for the loop. The `FOR_COND' is the
7551 termination condition. The `FOR_EXPR' is the expression executed
7552 right before the `FOR_COND' on each loop iteration; often, this
7553 expression increments a counter. The body of the loop is given by
7554 `FOR_BODY'. Note that `FOR_INIT_STMT' and `FOR_BODY' return
7555 statements, while `FOR_COND' and `FOR_EXPR' return expressions.
7558 Used to represent a `goto' statement. The `GOTO_DESTINATION' will
7559 usually be a `LABEL_DECL'. However, if the "computed goto"
7560 extension has been used, the `GOTO_DESTINATION' will be an
7561 arbitrary expression indicating the destination. This expression
7562 will always have pointer type.
7565 Used to represent a C++ `catch' block. The `HANDLER_TYPE' is the
7566 type of exception that will be caught by this handler; it is equal
7567 (by pointer equality) to `NULL' if this handler is for all types.
7568 `HANDLER_PARMS' is the `DECL_STMT' for the catch parameter, and
7569 `HANDLER_BODY' is the code for the block itself.
7572 Used to represent an `if' statement. The `IF_COND' is the
7575 If the condition is a `TREE_LIST', then the `TREE_PURPOSE' is a
7576 statement (usually a `DECL_STMT'). Each time the condition is
7577 evaluated, the statement should be executed. Then, the
7578 `TREE_VALUE' should be used as the conditional expression itself.
7579 This representation is used to handle C++ code like this:
7583 where there is a new local variable (or variables) declared within
7586 The `THEN_CLAUSE' represents the statement given by the `then'
7587 condition, while the `ELSE_CLAUSE' represents the statement given
7588 by the `else' condition.
7591 Used to represent a label. The `LABEL_DECL' declared by this
7592 statement can be obtained with the `LABEL_EXPR_LABEL' macro. The
7593 `IDENTIFIER_NODE' giving the name of the label can be obtained from
7594 the `LABEL_DECL' with `DECL_NAME'.
7597 Used to represent a `return' statement. The `RETURN_EXPR' is the
7598 expression returned; it will be `NULL_TREE' if the statement was
7603 In a constructor, these nodes are used to mark the point at which a
7604 subobject of `this' is fully constructed. If, after this point, an
7605 exception is thrown before a `CTOR_STMT' with `CTOR_END_P' set is
7606 encountered, the `SUBOBJECT_CLEANUP' must be executed. The
7607 cleanups must be executed in the reverse order in which they
7611 Used to represent a `switch' statement. The `SWITCH_STMT_COND' is
7612 the expression on which the switch is occurring. See the
7613 documentation for an `IF_STMT' for more information on the
7614 representation used for the condition. The `SWITCH_STMT_BODY' is
7615 the body of the switch statement. The `SWITCH_STMT_TYPE' is the
7616 original type of switch expression as given in the source, before
7617 any compiler conversions.
7620 Used to represent a `try' block. The body of the try block is
7621 given by `TRY_STMTS'. Each of the catch blocks is a `HANDLER'
7622 node. The first handler is given by `TRY_HANDLERS'. Subsequent
7623 handlers are obtained by following the `TREE_CHAIN' link from one
7624 handler to the next. The body of the handler is given by
7627 If `CLEANUP_P' holds of the `TRY_BLOCK', then the `TRY_HANDLERS'
7628 will not be a `HANDLER' node. Instead, it will be an expression
7629 that should be executed if an exception is thrown in the try
7630 block. It must rethrow the exception after executing that code.
7631 And, if an exception is thrown while the expression is executing,
7632 `terminate' must be called.
7635 Used to represent a `using' directive. The namespace is given by
7636 `USING_STMT_NAMESPACE', which will be a NAMESPACE_DECL. This node
7637 is needed inside template functions, to implement using directives
7638 during instantiation.
7641 Used to represent a `while' loop. The `WHILE_COND' is the
7642 termination condition for the loop. See the documentation for an
7643 `IF_STMT' for more information on the representation used for the
7646 The `WHILE_BODY' is the body of the loop.
7650 File: gccint.info, Node: Attributes, Next: Expression trees, Prev: Declarations, Up: Trees
7652 9.7 Attributes in trees
7653 =======================
7655 Attributes, as specified using the `__attribute__' keyword, are
7656 represented internally as a `TREE_LIST'. The `TREE_PURPOSE' is the
7657 name of the attribute, as an `IDENTIFIER_NODE'. The `TREE_VALUE' is a
7658 `TREE_LIST' of the arguments of the attribute, if any, or `NULL_TREE'
7659 if there are no arguments; the arguments are stored as the `TREE_VALUE'
7660 of successive entries in the list, and may be identifiers or
7661 expressions. The `TREE_CHAIN' of the attribute is the next attribute
7662 in a list of attributes applying to the same declaration or type, or
7663 `NULL_TREE' if there are no further attributes in the list.
7665 Attributes may be attached to declarations and to types; these
7666 attributes may be accessed with the following macros. All attributes
7667 are stored in this way, and many also cause other changes to the
7668 declaration or type or to other internal compiler data structures.
7670 -- Tree Macro: tree DECL_ATTRIBUTES (tree DECL)
7671 This macro returns the attributes on the declaration DECL.
7673 -- Tree Macro: tree TYPE_ATTRIBUTES (tree TYPE)
7674 This macro returns the attributes on the type TYPE.
7677 File: gccint.info, Node: Expression trees, Prev: Attributes, Up: Trees
7682 The internal representation for expressions is for the most part quite
7683 straightforward. However, there are a few facts that one must bear in
7684 mind. In particular, the expression "tree" is actually a directed
7685 acyclic graph. (For example there may be many references to the integer
7686 constant zero throughout the source program; many of these will be
7687 represented by the same expression node.) You should not rely on
7688 certain kinds of node being shared, nor should you rely on certain
7689 kinds of nodes being unshared.
7691 The following macros can be used with all expression nodes:
7694 Returns the type of the expression. This value may not be
7695 precisely the same type that would be given the expression in the
7698 In what follows, some nodes that one might expect to always have type
7699 `bool' are documented to have either integral or boolean type. At some
7700 point in the future, the C front end may also make use of this same
7701 intermediate representation, and at this point these nodes will
7702 certainly have integral type. The previous sentence is not meant to
7703 imply that the C++ front end does not or will not give these nodes
7706 Below, we list the various kinds of expression nodes. Except where
7707 noted otherwise, the operands to an expression are accessed using the
7708 `TREE_OPERAND' macro. For example, to access the first operand to a
7709 binary plus expression `expr', use:
7711 TREE_OPERAND (expr, 0)
7712 As this example indicates, the operands are zero-indexed.
7714 All the expressions starting with `OMP_' represent directives and
7715 clauses used by the OpenMP API `http://www.openmp.org/'.
7717 The table below begins with constants, moves on to unary expressions,
7718 then proceeds to binary expressions, and concludes with various other
7719 kinds of expressions:
7722 These nodes represent integer constants. Note that the type of
7723 these constants is obtained with `TREE_TYPE'; they are not always
7724 of type `int'. In particular, `char' constants are represented
7725 with `INTEGER_CST' nodes. The value of the integer constant `e' is
7727 ((TREE_INT_CST_HIGH (e) << HOST_BITS_PER_WIDE_INT)
7728 + TREE_INST_CST_LOW (e))
7729 HOST_BITS_PER_WIDE_INT is at least thirty-two on all platforms.
7730 Both `TREE_INT_CST_HIGH' and `TREE_INT_CST_LOW' return a
7731 `HOST_WIDE_INT'. The value of an `INTEGER_CST' is interpreted as
7732 a signed or unsigned quantity depending on the type of the
7733 constant. In general, the expression given above will overflow,
7734 so it should not be used to calculate the value of the constant.
7736 The variable `integer_zero_node' is an integer constant with value
7737 zero. Similarly, `integer_one_node' is an integer constant with
7738 value one. The `size_zero_node' and `size_one_node' variables are
7739 analogous, but have type `size_t' rather than `int'.
7741 The function `tree_int_cst_lt' is a predicate which holds if its
7742 first argument is less than its second. Both constants are
7743 assumed to have the same signedness (i.e., either both should be
7744 signed or both should be unsigned.) The full width of the
7745 constant is used when doing the comparison; the usual rules about
7746 promotions and conversions are ignored. Similarly,
7747 `tree_int_cst_equal' holds if the two constants are equal. The
7748 `tree_int_cst_sgn' function returns the sign of a constant. The
7749 value is `1', `0', or `-1' according on whether the constant is
7750 greater than, equal to, or less than zero. Again, the signedness
7751 of the constant's type is taken into account; an unsigned constant
7752 is never less than zero, no matter what its bit-pattern.
7755 FIXME: Talk about how to obtain representations of this constant,
7756 do comparisons, and so forth.
7759 These nodes represent fixed-point constants. The type of these
7760 constants is obtained with `TREE_TYPE'. `TREE_FIXED_CST_PTR'
7761 points to to struct fixed_value; `TREE_FIXED_CST' returns the
7762 structure itself. Struct fixed_value contains `data' with the
7763 size of two HOST_BITS_PER_WIDE_INT and `mode' as the associated
7764 fixed-point machine mode for `data'.
7767 These nodes are used to represent complex number constants, that
7768 is a `__complex__' whose parts are constant nodes. The
7769 `TREE_REALPART' and `TREE_IMAGPART' return the real and the
7770 imaginary parts respectively.
7773 These nodes are used to represent vector constants, whose parts are
7774 constant nodes. Each individual constant node is either an
7775 integer or a double constant node. The first operand is a
7776 `TREE_LIST' of the constant nodes and is accessed through
7777 `TREE_VECTOR_CST_ELTS'.
7780 These nodes represent string-constants. The `TREE_STRING_LENGTH'
7781 returns the length of the string, as an `int'. The
7782 `TREE_STRING_POINTER' is a `char*' containing the string itself.
7783 The string may not be `NUL'-terminated, and it may contain
7784 embedded `NUL' characters. Therefore, the `TREE_STRING_LENGTH'
7785 includes the trailing `NUL' if it is present.
7787 For wide string constants, the `TREE_STRING_LENGTH' is the number
7788 of bytes in the string, and the `TREE_STRING_POINTER' points to an
7789 array of the bytes of the string, as represented on the target
7790 system (that is, as integers in the target endianness). Wide and
7791 non-wide string constants are distinguished only by the `TREE_TYPE'
7792 of the `STRING_CST'.
7794 FIXME: The formats of string constants are not well-defined when
7795 the target system bytes are not the same width as host system
7799 These nodes are used to represent pointer-to-member constants. The
7800 `PTRMEM_CST_CLASS' is the class type (either a `RECORD_TYPE' or
7801 `UNION_TYPE' within which the pointer points), and the
7802 `PTRMEM_CST_MEMBER' is the declaration for the pointed to object.
7803 Note that the `DECL_CONTEXT' for the `PTRMEM_CST_MEMBER' is in
7804 general different from the `PTRMEM_CST_CLASS'. For example, given:
7805 struct B { int i; };
7806 struct D : public B {};
7808 The `PTRMEM_CST_CLASS' for `&D::i' is `D', even though the
7809 `DECL_CONTEXT' for the `PTRMEM_CST_MEMBER' is `B', since `B::i' is
7810 a member of `B', not `D'.
7813 These nodes represent variables, including static data members.
7814 For more information, *note Declarations::.
7817 These nodes represent unary negation of the single operand, for
7818 both integer and floating-point types. The type of negation can be
7819 determined by looking at the type of the expression.
7821 The behavior of this operation on signed arithmetic overflow is
7822 controlled by the `flag_wrapv' and `flag_trapv' variables.
7825 These nodes represent the absolute value of the single operand, for
7826 both integer and floating-point types. This is typically used to
7827 implement the `abs', `labs' and `llabs' builtins for integer
7828 types, and the `fabs', `fabsf' and `fabsl' builtins for floating
7829 point types. The type of abs operation can be determined by
7830 looking at the type of the expression.
7832 This node is not used for complex types. To represent the modulus
7833 or complex abs of a complex value, use the `BUILT_IN_CABS',
7834 `BUILT_IN_CABSF' or `BUILT_IN_CABSL' builtins, as used to
7835 implement the C99 `cabs', `cabsf' and `cabsl' built-in functions.
7838 These nodes represent bitwise complement, and will always have
7839 integral type. The only operand is the value to be complemented.
7842 These nodes represent logical negation, and will always have
7843 integral (or boolean) type. The operand is the value being
7844 negated. The type of the operand and that of the result are
7845 always of `BOOLEAN_TYPE' or `INTEGER_TYPE'.
7849 `POSTDECREMENT_EXPR'
7850 `POSTINCREMENT_EXPR'
7851 These nodes represent increment and decrement expressions. The
7852 value of the single operand is computed, and the operand
7853 incremented or decremented. In the case of `PREDECREMENT_EXPR' and
7854 `PREINCREMENT_EXPR', the value of the expression is the value
7855 resulting after the increment or decrement; in the case of
7856 `POSTDECREMENT_EXPR' and `POSTINCREMENT_EXPR' is the value before
7857 the increment or decrement occurs. The type of the operand, like
7858 that of the result, will be either integral, boolean, or
7862 These nodes are used to represent the address of an object. (These
7863 expressions will always have pointer or reference type.) The
7864 operand may be another expression, or it may be a declaration.
7866 As an extension, GCC allows users to take the address of a label.
7867 In this case, the operand of the `ADDR_EXPR' will be a
7868 `LABEL_DECL'. The type of such an expression is `void*'.
7870 If the object addressed is not an lvalue, a temporary is created,
7871 and the address of the temporary is used.
7874 These nodes are used to represent the object pointed to by a
7875 pointer. The operand is the pointer being dereferenced; it will
7876 always have pointer or reference type.
7879 These nodes represent conversion of a floating-point value to an
7880 integer. The single operand will have a floating-point type, while
7881 the complete expression will have an integral (or boolean) type.
7882 The operand is rounded towards zero.
7885 These nodes represent conversion of an integral (or boolean) value
7886 to a floating-point value. The single operand will have integral
7887 type, while the complete expression will have a floating-point
7890 FIXME: How is the operand supposed to be rounded? Is this
7891 dependent on `-mieee'?
7894 These nodes are used to represent complex numbers constructed from
7895 two expressions of the same (integer or real) type. The first
7896 operand is the real part and the second operand is the imaginary
7900 These nodes represent the conjugate of their operand.
7904 These nodes represent respectively the real and the imaginary parts
7905 of complex numbers (their sole argument).
7908 These nodes indicate that their one and only operand is not an
7909 lvalue. A back end can treat these identically to the single
7913 These nodes are used to represent conversions that do not require
7914 any code-generation. For example, conversion of a `char*' to an
7915 `int*' does not require any code be generated; such a conversion is
7916 represented by a `NOP_EXPR'. The single operand is the expression
7917 to be converted. The conversion from a pointer to a reference is
7918 also represented with a `NOP_EXPR'.
7921 These nodes are similar to `NOP_EXPR's, but are used in those
7922 situations where code may need to be generated. For example, if an
7923 `int*' is converted to an `int' code may need to be generated on
7924 some platforms. These nodes are never used for C++-specific
7925 conversions, like conversions between pointers to different
7926 classes in an inheritance hierarchy. Any adjustments that need to
7927 be made in such cases are always indicated explicitly. Similarly,
7928 a user-defined conversion is never represented by a
7929 `CONVERT_EXPR'; instead, the function calls are made explicit.
7931 `FIXED_CONVERT_EXPR'
7932 These nodes are used to represent conversions that involve
7933 fixed-point values. For example, from a fixed-point value to
7934 another fixed-point value, from an integer to a fixed-point value,
7935 from a fixed-point value to an integer, from a floating-point
7936 value to a fixed-point value, or from a fixed-point value to a
7937 floating-point value.
7940 These nodes represent `throw' expressions. The single operand is
7941 an expression for the code that should be executed to throw the
7942 exception. However, there is one implicit action not represented
7943 in that expression; namely the call to `__throw'. This function
7944 takes no arguments. If `setjmp'/`longjmp' exceptions are used, the
7945 function `__sjthrow' is called instead. The normal GCC back end
7946 uses the function `emit_throw' to generate this code; you can
7947 examine this function to see what needs to be done.
7951 These nodes represent left and right shifts, respectively. The
7952 first operand is the value to shift; it will always be of integral
7953 type. The second operand is an expression for the number of bits
7954 by which to shift. Right shift should be treated as arithmetic,
7955 i.e., the high-order bits should be zero-filled when the
7956 expression has unsigned type and filled with the sign bit when the
7957 expression has signed type. Note that the result is undefined if
7958 the second operand is larger than or equal to the first operand's
7964 These nodes represent bitwise inclusive or, bitwise exclusive or,
7965 and bitwise and, respectively. Both operands will always have
7970 These nodes represent logical "and" and logical "or", respectively.
7971 These operators are not strict; i.e., the second operand is
7972 evaluated only if the value of the expression is not determined by
7973 evaluation of the first operand. The type of the operands and
7974 that of the result are always of `BOOLEAN_TYPE' or `INTEGER_TYPE'.
7979 These nodes represent logical and, logical or, and logical
7980 exclusive or. They are strict; both arguments are always
7981 evaluated. There are no corresponding operators in C or C++, but
7982 the front end will sometimes generate these expressions anyhow, if
7983 it can tell that strictness does not matter. The type of the
7984 operands and that of the result are always of `BOOLEAN_TYPE' or
7988 This node represents pointer arithmetic. The first operand is
7989 always a pointer/reference type. The second operand is always an
7990 unsigned integer type compatible with sizetype. This is the only
7991 binary arithmetic operand that can operate on pointer types.
7996 These nodes represent various binary arithmetic operations.
7997 Respectively, these operations are addition, subtraction (of the
7998 second operand from the first) and multiplication. Their operands
7999 may have either integral or floating type, but there will never be
8000 case in which one operand is of floating type and the other is of
8003 The behavior of these operations on signed arithmetic overflow is
8004 controlled by the `flag_wrapv' and `flag_trapv' variables.
8007 This node represents a floating point division operation.
8013 These nodes represent integer division operations that return an
8014 integer result. `TRUNC_DIV_EXPR' rounds towards zero,
8015 `FLOOR_DIV_EXPR' rounds towards negative infinity, `CEIL_DIV_EXPR'
8016 rounds towards positive infinity and `ROUND_DIV_EXPR' rounds to
8017 the closest integer. Integer division in C and C++ is truncating,
8018 i.e. `TRUNC_DIV_EXPR'.
8020 The behavior of these operations on signed arithmetic overflow,
8021 when dividing the minimum signed integer by minus one, is
8022 controlled by the `flag_wrapv' and `flag_trapv' variables.
8028 These nodes represent the integer remainder or modulus operation.
8029 The integer modulus of two operands `a' and `b' is defined as `a -
8030 (a/b)*b' where the division calculated using the corresponding
8031 division operator. Hence for `TRUNC_MOD_EXPR' this definition
8032 assumes division using truncation towards zero, i.e.
8033 `TRUNC_DIV_EXPR'. Integer remainder in C and C++ uses truncating
8034 division, i.e. `TRUNC_MOD_EXPR'.
8037 The `EXACT_DIV_EXPR' code is used to represent integer divisions
8038 where the numerator is known to be an exact multiple of the
8039 denominator. This allows the backend to choose between the faster
8040 of `TRUNC_DIV_EXPR', `CEIL_DIV_EXPR' and `FLOOR_DIV_EXPR' for the
8044 These nodes represent array accesses. The first operand is the
8045 array; the second is the index. To calculate the address of the
8046 memory accessed, you must scale the index by the size of the type
8047 of the array elements. The type of these expressions must be the
8048 type of a component of the array. The third and fourth operands
8049 are used after gimplification to represent the lower bound and
8050 component size but should not be used directly; call
8051 `array_ref_low_bound' and `array_ref_element_size' instead.
8054 These nodes represent access to a range (or "slice") of an array.
8055 The operands are the same as that for `ARRAY_REF' and have the same
8056 meanings. The type of these expressions must be an array whose
8057 component type is the same as that of the first operand. The
8058 range of that array type determines the amount of data these
8062 These nodes represent memory accesses whose address directly map to
8063 an addressing mode of the target architecture. The first argument
8064 is `TMR_SYMBOL' and must be a `VAR_DECL' of an object with a fixed
8065 address. The second argument is `TMR_BASE' and the third one is
8066 `TMR_INDEX'. The fourth argument is `TMR_STEP' and must be an
8067 `INTEGER_CST'. The fifth argument is `TMR_OFFSET' and must be an
8068 `INTEGER_CST'. Any of the arguments may be NULL if the
8069 appropriate component does not appear in the address. Address of
8070 the `TARGET_MEM_REF' is determined in the following way.
8072 &TMR_SYMBOL + TMR_BASE + TMR_INDEX * TMR_STEP + TMR_OFFSET
8074 The sixth argument is the reference to the original memory access,
8075 which is preserved for the purposes of the RTL alias analysis.
8076 The seventh argument is a tag representing the results of tree
8077 level alias analysis.
8085 These nodes represent the less than, less than or equal to, greater
8086 than, greater than or equal to, equal, and not equal comparison
8087 operators. The first and second operand with either be both of
8088 integral type or both of floating type. The result type of these
8089 expressions will always be of integral or boolean type. These
8090 operations return the result type's zero value for false, and the
8091 result type's one value for true.
8093 For floating point comparisons, if we honor IEEE NaNs and either
8094 operand is NaN, then `NE_EXPR' always returns true and the
8095 remaining operators always return false. On some targets,
8096 comparisons against an IEEE NaN, other than equality and
8097 inequality, may generate a floating point exception.
8101 These nodes represent non-trapping ordered and unordered comparison
8102 operators. These operations take two floating point operands and
8103 determine whether they are ordered or unordered relative to each
8104 other. If either operand is an IEEE NaN, their comparison is
8105 defined to be unordered, otherwise the comparison is defined to be
8106 ordered. The result type of these expressions will always be of
8107 integral or boolean type. These operations return the result
8108 type's zero value for false, and the result type's one value for
8117 These nodes represent the unordered comparison operators. These
8118 operations take two floating point operands and determine whether
8119 the operands are unordered or are less than, less than or equal to,
8120 greater than, greater than or equal to, or equal respectively. For
8121 example, `UNLT_EXPR' returns true if either operand is an IEEE NaN
8122 or the first operand is less than the second. With the possible
8123 exception of `LTGT_EXPR', all of these operations are guaranteed
8124 not to generate a floating point exception. The result type of
8125 these expressions will always be of integral or boolean type.
8126 These operations return the result type's zero value for false,
8127 and the result type's one value for true.
8130 These nodes represent assignment. The left-hand side is the first
8131 operand; the right-hand side is the second operand. The left-hand
8132 side will be a `VAR_DECL', `INDIRECT_REF', `COMPONENT_REF', or
8135 These nodes are used to represent not only assignment with `=' but
8136 also compound assignments (like `+='), by reduction to `='
8137 assignment. In other words, the representation for `i += 3' looks
8138 just like that for `i = i + 3'.
8141 These nodes are just like `MODIFY_EXPR', but are used only when a
8142 variable is initialized, rather than assigned to subsequently.
8143 This means that we can assume that the target of the
8144 initialization is not used in computing its own value; any
8145 reference to the lhs in computing the rhs is undefined.
8148 These nodes represent non-static data member accesses. The first
8149 operand is the object (rather than a pointer to it); the second
8150 operand is the `FIELD_DECL' for the data member. The third
8151 operand represents the byte offset of the field, but should not be
8152 used directly; call `component_ref_field_offset' instead.
8155 These nodes represent comma-expressions. The first operand is an
8156 expression whose value is computed and thrown away prior to the
8157 evaluation of the second operand. The value of the entire
8158 expression is the value of the second operand.
8161 These nodes represent `?:' expressions. The first operand is of
8162 boolean or integral type. If it evaluates to a nonzero value, the
8163 second operand should be evaluated, and returned as the value of
8164 the expression. Otherwise, the third operand is evaluated, and
8165 returned as the value of the expression.
8167 The second operand must have the same type as the entire
8168 expression, unless it unconditionally throws an exception or calls
8169 a noreturn function, in which case it should have void type. The
8170 same constraints apply to the third operand. This allows array
8171 bounds checks to be represented conveniently as `(i >= 0 && i <
8174 As a GNU extension, the C language front-ends allow the second
8175 operand of the `?:' operator may be omitted in the source. For
8176 example, `x ? : 3' is equivalent to `x ? x : 3', assuming that `x'
8177 is an expression without side-effects. In the tree
8178 representation, however, the second operand is always present,
8179 possibly protected by `SAVE_EXPR' if the first argument does cause
8183 These nodes are used to represent calls to functions, including
8184 non-static member functions. `CALL_EXPR's are implemented as
8185 expression nodes with a variable number of operands. Rather than
8186 using `TREE_OPERAND' to extract them, it is preferable to use the
8187 specialized accessor macros and functions that operate
8188 specifically on `CALL_EXPR' nodes.
8190 `CALL_EXPR_FN' returns a pointer to the function to call; it is
8191 always an expression whose type is a `POINTER_TYPE'.
8193 The number of arguments to the call is returned by
8194 `call_expr_nargs', while the arguments themselves can be accessed
8195 with the `CALL_EXPR_ARG' macro. The arguments are zero-indexed
8196 and numbered left-to-right. You can iterate over the arguments
8197 using `FOR_EACH_CALL_EXPR_ARG', as in:
8200 call_expr_arg_iterator iter;
8201 FOR_EACH_CALL_EXPR_ARG (arg, iter, call)
8202 /* arg is bound to successive arguments of call. */
8205 For non-static member functions, there will be an operand
8206 corresponding to the `this' pointer. There will always be
8207 expressions corresponding to all of the arguments, even if the
8208 function is declared with default arguments and some arguments are
8209 not explicitly provided at the call sites.
8211 `CALL_EXPR's also have a `CALL_EXPR_STATIC_CHAIN' operand that is
8212 used to implement nested functions. This operand is otherwise
8216 These nodes are used to represent GCC's statement-expression
8217 extension. The statement-expression extension allows code like
8219 int f() { return ({ int j; j = 3; j + 7; }); }
8220 In other words, an sequence of statements may occur where a single
8221 expression would normally appear. The `STMT_EXPR' node represents
8222 such an expression. The `STMT_EXPR_STMT' gives the statement
8223 contained in the expression. The value of the expression is the
8224 value of the last sub-statement in the body. More precisely, the
8225 value is the value computed by the last statement nested inside
8226 `BIND_EXPR', `TRY_FINALLY_EXPR', or `TRY_CATCH_EXPR'. For
8229 the value is `3' while in:
8231 there is no value. If the `STMT_EXPR' does not yield a value,
8232 it's type will be `void'.
8235 These nodes represent local blocks. The first operand is a list of
8236 variables, connected via their `TREE_CHAIN' field. These will
8237 never require cleanups. The scope of these variables is just the
8238 body of the `BIND_EXPR'. The body of the `BIND_EXPR' is the
8242 These nodes represent "infinite" loops. The `LOOP_EXPR_BODY'
8243 represents the body of the loop. It should be executed forever,
8244 unless an `EXIT_EXPR' is encountered.
8247 These nodes represent conditional exits from the nearest enclosing
8248 `LOOP_EXPR'. The single operand is the condition; if it is
8249 nonzero, then the loop should be exited. An `EXIT_EXPR' will only
8250 appear within a `LOOP_EXPR'.
8252 `CLEANUP_POINT_EXPR'
8253 These nodes represent full-expressions. The single operand is an
8254 expression to evaluate. Any destructor calls engendered by the
8255 creation of temporaries during the evaluation of that expression
8256 should be performed immediately after the expression is evaluated.
8259 These nodes represent the brace-enclosed initializers for a
8260 structure or array. The first operand is reserved for use by the
8261 back end. The second operand is a `TREE_LIST'. If the
8262 `TREE_TYPE' of the `CONSTRUCTOR' is a `RECORD_TYPE' or
8263 `UNION_TYPE', then the `TREE_PURPOSE' of each node in the
8264 `TREE_LIST' will be a `FIELD_DECL' and the `TREE_VALUE' of each
8265 node will be the expression used to initialize that field.
8267 If the `TREE_TYPE' of the `CONSTRUCTOR' is an `ARRAY_TYPE', then
8268 the `TREE_PURPOSE' of each element in the `TREE_LIST' will be an
8269 `INTEGER_CST' or a `RANGE_EXPR' of two `INTEGER_CST's. A single
8270 `INTEGER_CST' indicates which element of the array (indexed from
8271 zero) is being assigned to. A `RANGE_EXPR' indicates an inclusive
8272 range of elements to initialize. In both cases the `TREE_VALUE'
8273 is the corresponding initializer. It is re-evaluated for each
8274 element of a `RANGE_EXPR'. If the `TREE_PURPOSE' is `NULL_TREE',
8275 then the initializer is for the next available array element.
8277 In the front end, you should not depend on the fields appearing in
8278 any particular order. However, in the middle end, fields must
8279 appear in declaration order. You should not assume that all
8280 fields will be represented. Unrepresented fields will be set to
8283 `COMPOUND_LITERAL_EXPR'
8284 These nodes represent ISO C99 compound literals. The
8285 `COMPOUND_LITERAL_EXPR_DECL_STMT' is a `DECL_STMT' containing an
8286 anonymous `VAR_DECL' for the unnamed object represented by the
8287 compound literal; the `DECL_INITIAL' of that `VAR_DECL' is a
8288 `CONSTRUCTOR' representing the brace-enclosed list of initializers
8289 in the compound literal. That anonymous `VAR_DECL' can also be
8290 accessed directly by the `COMPOUND_LITERAL_EXPR_DECL' macro.
8293 A `SAVE_EXPR' represents an expression (possibly involving
8294 side-effects) that is used more than once. The side-effects should
8295 occur only the first time the expression is evaluated. Subsequent
8296 uses should just reuse the computed value. The first operand to
8297 the `SAVE_EXPR' is the expression to evaluate. The side-effects
8298 should be executed where the `SAVE_EXPR' is first encountered in a
8299 depth-first preorder traversal of the expression tree.
8302 A `TARGET_EXPR' represents a temporary object. The first operand
8303 is a `VAR_DECL' for the temporary variable. The second operand is
8304 the initializer for the temporary. The initializer is evaluated
8305 and, if non-void, copied (bitwise) into the temporary. If the
8306 initializer is void, that means that it will perform the
8307 initialization itself.
8309 Often, a `TARGET_EXPR' occurs on the right-hand side of an
8310 assignment, or as the second operand to a comma-expression which is
8311 itself the right-hand side of an assignment, etc. In this case,
8312 we say that the `TARGET_EXPR' is "normal"; otherwise, we say it is
8313 "orphaned". For a normal `TARGET_EXPR' the temporary variable
8314 should be treated as an alias for the left-hand side of the
8315 assignment, rather than as a new temporary variable.
8317 The third operand to the `TARGET_EXPR', if present, is a
8318 cleanup-expression (i.e., destructor call) for the temporary. If
8319 this expression is orphaned, then this expression must be executed
8320 when the statement containing this expression is complete. These
8321 cleanups must always be executed in the order opposite to that in
8322 which they were encountered. Note that if a temporary is created
8323 on one branch of a conditional operator (i.e., in the second or
8324 third operand to a `COND_EXPR'), the cleanup must be run only if
8325 that branch is actually executed.
8327 See `STMT_IS_FULL_EXPR_P' for more information about running these
8331 An `AGGR_INIT_EXPR' represents the initialization as the return
8332 value of a function call, or as the result of a constructor. An
8333 `AGGR_INIT_EXPR' will only appear as a full-expression, or as the
8334 second operand of a `TARGET_EXPR'. `AGGR_INIT_EXPR's have a
8335 representation similar to that of `CALL_EXPR's. You can use the
8336 `AGGR_INIT_EXPR_FN' and `AGGR_INIT_EXPR_ARG' macros to access the
8337 function to call and the arguments to pass.
8339 If `AGGR_INIT_VIA_CTOR_P' holds of the `AGGR_INIT_EXPR', then the
8340 initialization is via a constructor call. The address of the
8341 `AGGR_INIT_EXPR_SLOT' operand, which is always a `VAR_DECL', is
8342 taken, and this value replaces the first argument in the argument
8345 In either case, the expression is void.
8348 This node is used to implement support for the C/C++ variable
8349 argument-list mechanism. It represents expressions like `va_arg
8350 (ap, type)'. Its `TREE_TYPE' yields the tree representation for
8351 `type' and its sole argument yields the representation for `ap'.
8353 `CHANGE_DYNAMIC_TYPE_EXPR'
8354 Indicates the special aliasing required by C++ placement new. It
8355 has two operands: a type and a location. It means that the
8356 dynamic type of the location is changing to be the specified type.
8357 The alias analysis code takes this into account when doing type
8358 based alias analysis.
8361 Represents `#pragma omp parallel [clause1 ... clauseN]'. It has
8364 Operand `OMP_PARALLEL_BODY' is valid while in GENERIC and High
8365 GIMPLE forms. It contains the body of code to be executed by all
8366 the threads. During GIMPLE lowering, this operand becomes `NULL'
8367 and the body is emitted linearly after `OMP_PARALLEL'.
8369 Operand `OMP_PARALLEL_CLAUSES' is the list of clauses associated
8372 Operand `OMP_PARALLEL_FN' is created by `pass_lower_omp', it
8373 contains the `FUNCTION_DECL' for the function that will contain
8374 the body of the parallel region.
8376 Operand `OMP_PARALLEL_DATA_ARG' is also created by
8377 `pass_lower_omp'. If there are shared variables to be communicated
8378 to the children threads, this operand will contain the `VAR_DECL'
8379 that contains all the shared values and variables.
8382 Represents `#pragma omp for [clause1 ... clauseN]'. It has 5
8385 Operand `OMP_FOR_BODY' contains the loop body.
8387 Operand `OMP_FOR_CLAUSES' is the list of clauses associated with
8390 Operand `OMP_FOR_INIT' is the loop initialization code of the form
8393 Operand `OMP_FOR_COND' is the loop conditional expression of the
8394 form `VAR {<,>,<=,>=} N2'.
8396 Operand `OMP_FOR_INCR' is the loop index increment of the form
8399 Operand `OMP_FOR_PRE_BODY' contains side-effect code from operands
8400 `OMP_FOR_INIT', `OMP_FOR_COND' and `OMP_FOR_INC'. These
8401 side-effects are part of the `OMP_FOR' block but must be evaluated
8402 before the start of loop body.
8404 The loop index variable `VAR' must be a signed integer variable,
8405 which is implicitly private to each thread. Bounds `N1' and `N2'
8406 and the increment expression `INCR' are required to be loop
8407 invariant integer expressions that are evaluated without any
8408 synchronization. The evaluation order, frequency of evaluation and
8409 side-effects are unspecified by the standard.
8412 Represents `#pragma omp sections [clause1 ... clauseN]'.
8414 Operand `OMP_SECTIONS_BODY' contains the sections body, which in
8415 turn contains a set of `OMP_SECTION' nodes for each of the
8416 concurrent sections delimited by `#pragma omp section'.
8418 Operand `OMP_SECTIONS_CLAUSES' is the list of clauses associated
8422 Section delimiter for `OMP_SECTIONS'.
8425 Represents `#pragma omp single'.
8427 Operand `OMP_SINGLE_BODY' contains the body of code to be executed
8430 Operand `OMP_SINGLE_CLAUSES' is the list of clauses associated
8434 Represents `#pragma omp master'.
8436 Operand `OMP_MASTER_BODY' contains the body of code to be executed
8437 by the master thread.
8440 Represents `#pragma omp ordered'.
8442 Operand `OMP_ORDERED_BODY' contains the body of code to be
8443 executed in the sequential order dictated by the loop index
8447 Represents `#pragma omp critical [name]'.
8449 Operand `OMP_CRITICAL_BODY' is the critical section.
8451 Operand `OMP_CRITICAL_NAME' is an optional identifier to label the
8455 This does not represent any OpenMP directive, it is an artificial
8456 marker to indicate the end of the body of an OpenMP. It is used by
8457 the flow graph (`tree-cfg.c') and OpenMP region building code
8461 Similarly, this instruction does not represent an OpenMP
8462 directive, it is used by `OMP_FOR' and `OMP_SECTIONS' to mark the
8463 place where the code needs to loop to the next iteration (in the
8464 case of `OMP_FOR') or the next section (in the case of
8467 In some cases, `OMP_CONTINUE' is placed right before `OMP_RETURN'.
8468 But if there are cleanups that need to occur right after the
8469 looping body, it will be emitted between `OMP_CONTINUE' and
8473 Represents `#pragma omp atomic'.
8475 Operand 0 is the address at which the atomic operation is to be
8478 Operand 1 is the expression to evaluate. The gimplifier tries
8479 three alternative code generation strategies. Whenever possible,
8480 an atomic update built-in is used. If that fails, a
8481 compare-and-swap loop is attempted. If that also fails, a regular
8482 critical section around the expression is used.
8485 Represents clauses associated with one of the `OMP_' directives.
8486 Clauses are represented by separate sub-codes defined in `tree.h'.
8487 Clauses codes can be one of: `OMP_CLAUSE_PRIVATE',
8488 `OMP_CLAUSE_SHARED', `OMP_CLAUSE_FIRSTPRIVATE',
8489 `OMP_CLAUSE_LASTPRIVATE', `OMP_CLAUSE_COPYIN',
8490 `OMP_CLAUSE_COPYPRIVATE', `OMP_CLAUSE_IF',
8491 `OMP_CLAUSE_NUM_THREADS', `OMP_CLAUSE_SCHEDULE',
8492 `OMP_CLAUSE_NOWAIT', `OMP_CLAUSE_ORDERED', `OMP_CLAUSE_DEFAULT',
8493 and `OMP_CLAUSE_REDUCTION'. Each code represents the
8494 corresponding OpenMP clause.
8496 Clauses associated with the same directive are chained together
8497 via `OMP_CLAUSE_CHAIN'. Those clauses that accept a list of
8498 variables are restricted to exactly one, accessed with
8499 `OMP_CLAUSE_VAR'. Therefore, multiple variables under the same
8500 clause `C' need to be represented as multiple `C' clauses chained
8501 together. This facilitates adding new clauses during compilation.
8506 These nodes represent whole vector left and right shifts,
8507 respectively. The first operand is the vector to shift; it will
8508 always be of vector type. The second operand is an expression for
8509 the number of bits by which to shift. Note that the result is
8510 undefined if the second operand is larger than or equal to the
8511 first operand's type size.
8513 `VEC_WIDEN_MULT_HI_EXPR'
8515 `VEC_WIDEN_MULT_LO_EXPR'
8516 These nodes represent widening vector multiplication of the high
8517 and low parts of the two input vectors, respectively. Their
8518 operands are vectors that contain the same number of elements
8519 (`N') of the same integral type. The result is a vector that
8520 contains half as many elements, of an integral type whose size is
8521 twice as wide. In the case of `VEC_WIDEN_MULT_HI_EXPR' the high
8522 `N/2' elements of the two vector are multiplied to produce the
8523 vector of `N/2' products. In the case of `VEC_WIDEN_MULT_LO_EXPR'
8524 the low `N/2' elements of the two vector are multiplied to produce
8525 the vector of `N/2' products.
8527 `VEC_UNPACK_HI_EXPR'
8529 `VEC_UNPACK_LO_EXPR'
8530 These nodes represent unpacking of the high and low parts of the
8531 input vector, respectively. The single operand is a vector that
8532 contains `N' elements of the same integral or floating point type.
8533 The result is a vector that contains half as many elements, of an
8534 integral or floating point type whose size is twice as wide. In
8535 the case of `VEC_UNPACK_HI_EXPR' the high `N/2' elements of the
8536 vector are extracted and widened (promoted). In the case of
8537 `VEC_UNPACK_LO_EXPR' the low `N/2' elements of the vector are
8538 extracted and widened (promoted).
8540 `VEC_UNPACK_FLOAT_HI_EXPR'
8542 `VEC_UNPACK_FLOAT_LO_EXPR'
8543 These nodes represent unpacking of the high and low parts of the
8544 input vector, where the values are converted from fixed point to
8545 floating point. The single operand is a vector that contains `N'
8546 elements of the same integral type. The result is a vector that
8547 contains half as many elements of a floating point type whose size
8548 is twice as wide. In the case of `VEC_UNPACK_HI_EXPR' the high
8549 `N/2' elements of the vector are extracted, converted and widened.
8550 In the case of `VEC_UNPACK_LO_EXPR' the low `N/2' elements of the
8551 vector are extracted, converted and widened.
8553 `VEC_PACK_TRUNC_EXPR'
8554 This node represents packing of truncated elements of the two
8555 input vectors into the output vector. Input operands are vectors
8556 that contain the same number of elements of the same integral or
8557 floating point type. The result is a vector that contains twice
8558 as many elements of an integral or floating point type whose size
8559 is half as wide. The elements of the two vectors are demoted and
8560 merged (concatenated) to form the output vector.
8563 This node represents packing of elements of the two input vectors
8564 into the output vector using saturation. Input operands are
8565 vectors that contain the same number of elements of the same
8566 integral type. The result is a vector that contains twice as many
8567 elements of an integral type whose size is half as wide. The
8568 elements of the two vectors are demoted and merged (concatenated)
8569 to form the output vector.
8571 `VEC_PACK_FIX_TRUNC_EXPR'
8572 This node represents packing of elements of the two input vectors
8573 into the output vector, where the values are converted from
8574 floating point to fixed point. Input operands are vectors that
8575 contain the same number of elements of a floating point type. The
8576 result is a vector that contains twice as many elements of an
8577 integral type whose size is half as wide. The elements of the two
8578 vectors are merged (concatenated) to form the output vector.
8580 `VEC_EXTRACT_EVEN_EXPR'
8582 `VEC_EXTRACT_ODD_EXPR'
8583 These nodes represent extracting of the even/odd elements of the
8584 two input vectors, respectively. Their operands and result are
8585 vectors that contain the same number of elements of the same type.
8587 `VEC_INTERLEAVE_HIGH_EXPR'
8589 `VEC_INTERLEAVE_LOW_EXPR'
8590 These nodes represent merging and interleaving of the high/low
8591 elements of the two input vectors, respectively. The operands and
8592 the result are vectors that contain the same number of elements
8593 (`N') of the same type. In the case of
8594 `VEC_INTERLEAVE_HIGH_EXPR', the high `N/2' elements of the first
8595 input vector are interleaved with the high `N/2' elements of the
8596 second input vector. In the case of `VEC_INTERLEAVE_LOW_EXPR', the
8597 low `N/2' elements of the first input vector are interleaved with
8598 the low `N/2' elements of the second input vector.
8602 File: gccint.info, Node: RTL, Next: GENERIC, Prev: Trees, Up: Top
8604 10 RTL Representation
8605 *********************
8607 Most of the work of the compiler is done on an intermediate
8608 representation called register transfer language. In this language,
8609 the instructions to be output are described, pretty much one by one, in
8610 an algebraic form that describes what the instruction does.
8612 RTL is inspired by Lisp lists. It has both an internal form, made up
8613 of structures that point at other structures, and a textual form that
8614 is used in the machine description and in printed debugging dumps. The
8615 textual form uses nested parentheses to indicate the pointers in the
8620 * RTL Objects:: Expressions vs vectors vs strings vs integers.
8621 * RTL Classes:: Categories of RTL expression objects, and their structure.
8622 * Accessors:: Macros to access expression operands or vector elts.
8623 * Special Accessors:: Macros to access specific annotations on RTL.
8624 * Flags:: Other flags in an RTL expression.
8625 * Machine Modes:: Describing the size and format of a datum.
8626 * Constants:: Expressions with constant values.
8627 * Regs and Memory:: Expressions representing register contents or memory.
8628 * Arithmetic:: Expressions representing arithmetic on other expressions.
8629 * Comparisons:: Expressions representing comparison of expressions.
8630 * Bit-Fields:: Expressions representing bit-fields in memory or reg.
8631 * Vector Operations:: Expressions involving vector datatypes.
8632 * Conversions:: Extending, truncating, floating or fixing.
8633 * RTL Declarations:: Declaring volatility, constancy, etc.
8634 * Side Effects:: Expressions for storing in registers, etc.
8635 * Incdec:: Embedded side-effects for autoincrement addressing.
8636 * Assembler:: Representing `asm' with operands.
8637 * Insns:: Expression types for entire insns.
8638 * Calls:: RTL representation of function call insns.
8639 * Sharing:: Some expressions are unique; others *must* be copied.
8640 * Reading RTL:: Reading textual RTL from a file.
8643 File: gccint.info, Node: RTL Objects, Next: RTL Classes, Up: RTL
8645 10.1 RTL Object Types
8646 =====================
8648 RTL uses five kinds of objects: expressions, integers, wide integers,
8649 strings and vectors. Expressions are the most important ones. An RTL
8650 expression ("RTX", for short) is a C structure, but it is usually
8651 referred to with a pointer; a type that is given the typedef name `rtx'.
8653 An integer is simply an `int'; their written form uses decimal digits.
8654 A wide integer is an integral object whose type is `HOST_WIDE_INT';
8655 their written form uses decimal digits.
8657 A string is a sequence of characters. In core it is represented as a
8658 `char *' in usual C fashion, and it is written in C syntax as well.
8659 However, strings in RTL may never be null. If you write an empty
8660 string in a machine description, it is represented in core as a null
8661 pointer rather than as a pointer to a null character. In certain
8662 contexts, these null pointers instead of strings are valid. Within RTL
8663 code, strings are most commonly found inside `symbol_ref' expressions,
8664 but they appear in other contexts in the RTL expressions that make up
8665 machine descriptions.
8667 In a machine description, strings are normally written with double
8668 quotes, as you would in C. However, strings in machine descriptions may
8669 extend over many lines, which is invalid C, and adjacent string
8670 constants are not concatenated as they are in C. Any string constant
8671 may be surrounded with a single set of parentheses. Sometimes this
8672 makes the machine description easier to read.
8674 There is also a special syntax for strings, which can be useful when C
8675 code is embedded in a machine description. Wherever a string can
8676 appear, it is also valid to write a C-style brace block. The entire
8677 brace block, including the outermost pair of braces, is considered to be
8678 the string constant. Double quote characters inside the braces are not
8679 special. Therefore, if you write string constants in the C code, you
8680 need not escape each quote character with a backslash.
8682 A vector contains an arbitrary number of pointers to expressions. The
8683 number of elements in the vector is explicitly present in the vector.
8684 The written form of a vector consists of square brackets (`[...]')
8685 surrounding the elements, in sequence and with whitespace separating
8686 them. Vectors of length zero are not created; null pointers are used
8689 Expressions are classified by "expression codes" (also called RTX
8690 codes). The expression code is a name defined in `rtl.def', which is
8691 also (in uppercase) a C enumeration constant. The possible expression
8692 codes and their meanings are machine-independent. The code of an RTX
8693 can be extracted with the macro `GET_CODE (X)' and altered with
8694 `PUT_CODE (X, NEWCODE)'.
8696 The expression code determines how many operands the expression
8697 contains, and what kinds of objects they are. In RTL, unlike Lisp, you
8698 cannot tell by looking at an operand what kind of object it is.
8699 Instead, you must know from its context--from the expression code of
8700 the containing expression. For example, in an expression of code
8701 `subreg', the first operand is to be regarded as an expression and the
8702 second operand as an integer. In an expression of code `plus', there
8703 are two operands, both of which are to be regarded as expressions. In
8704 a `symbol_ref' expression, there is one operand, which is to be
8705 regarded as a string.
8707 Expressions are written as parentheses containing the name of the
8708 expression type, its flags and machine mode if any, and then the
8709 operands of the expression (separated by spaces).
8711 Expression code names in the `md' file are written in lowercase, but
8712 when they appear in C code they are written in uppercase. In this
8713 manual, they are shown as follows: `const_int'.
8715 In a few contexts a null pointer is valid where an expression is
8716 normally wanted. The written form of this is `(nil)'.
8719 File: gccint.info, Node: RTL Classes, Next: Accessors, Prev: RTL Objects, Up: RTL
8721 10.2 RTL Classes and Formats
8722 ============================
8724 The various expression codes are divided into several "classes", which
8725 are represented by single characters. You can determine the class of
8726 an RTX code with the macro `GET_RTX_CLASS (CODE)'. Currently,
8727 `rtl.def' defines these classes:
8730 An RTX code that represents an actual object, such as a register
8731 (`REG') or a memory location (`MEM', `SYMBOL_REF'). `LO_SUM') is
8732 also included; instead, `SUBREG' and `STRICT_LOW_PART' are not in
8733 this class, but in class `x'.
8736 An RTX code that represents a constant object. `HIGH' is also
8737 included in this class.
8740 An RTX code for a non-symmetric comparison, such as `GEU' or `LT'.
8743 An RTX code for a symmetric (commutative) comparison, such as `EQ'
8747 An RTX code for a unary arithmetic operation, such as `NEG',
8748 `NOT', or `ABS'. This category also includes value extension
8749 (sign or zero) and conversions between integer and floating point.
8752 An RTX code for a commutative binary operation, such as `PLUS' or
8753 `AND'. `NE' and `EQ' are comparisons, so they have class `<'.
8756 An RTX code for a non-commutative binary operation, such as
8757 `MINUS', `DIV', or `ASHIFTRT'.
8760 An RTX code for a bit-field operation. Currently only
8761 `ZERO_EXTRACT' and `SIGN_EXTRACT'. These have three inputs and
8762 are lvalues (so they can be used for insertion as well). *Note
8766 An RTX code for other three input operations. Currently only
8767 `IF_THEN_ELSE' and `VEC_MERGE'.
8770 An RTX code for an entire instruction: `INSN', `JUMP_INSN', and
8771 `CALL_INSN'. *Note Insns::.
8774 An RTX code for something that matches in insns, such as
8775 `MATCH_DUP'. These only occur in machine descriptions.
8778 An RTX code for an auto-increment addressing mode, such as
8782 All other RTX codes. This category includes the remaining codes
8783 used only in machine descriptions (`DEFINE_*', etc.). It also
8784 includes all the codes describing side effects (`SET', `USE',
8785 `CLOBBER', etc.) and the non-insns that may appear on an insn
8786 chain, such as `NOTE', `BARRIER', and `CODE_LABEL'. `SUBREG' is
8787 also part of this class.
8789 For each expression code, `rtl.def' specifies the number of contained
8790 objects and their kinds using a sequence of characters called the
8791 "format" of the expression code. For example, the format of `subreg'
8794 These are the most commonly used format characters:
8797 An expression (actually a pointer to an expression).
8809 A vector of expressions.
8811 A few other format characters are used occasionally:
8814 `u' is equivalent to `e' except that it is printed differently in
8815 debugging dumps. It is used for pointers to insns.
8818 `n' is equivalent to `i' except that it is printed differently in
8819 debugging dumps. It is used for the line number or code number of
8823 `S' indicates a string which is optional. In the RTL objects in
8824 core, `S' is equivalent to `s', but when the object is read, from
8825 an `md' file, the string value of this operand may be omitted. An
8826 omitted string is taken to be the null string.
8829 `V' indicates a vector which is optional. In the RTL objects in
8830 core, `V' is equivalent to `E', but when the object is read from
8831 an `md' file, the vector value of this operand may be omitted. An
8832 omitted vector is effectively the same as a vector of no elements.
8835 `B' indicates a pointer to basic block structure.
8838 `0' means a slot whose contents do not fit any normal category.
8839 `0' slots are not printed at all in dumps, and are often used in
8840 special ways by small parts of the compiler.
8842 There are macros to get the number of operands and the format of an
8845 `GET_RTX_LENGTH (CODE)'
8846 Number of operands of an RTX of code CODE.
8848 `GET_RTX_FORMAT (CODE)'
8849 The format of an RTX of code CODE, as a C string.
8851 Some classes of RTX codes always have the same format. For example, it
8852 is safe to assume that all comparison operations have format `ee'.
8855 All codes of this class have format `e'.
8860 All codes of these classes have format `ee'.
8864 All codes of these classes have format `eee'.
8867 All codes of this class have formats that begin with `iuueiee'.
8868 *Note Insns::. Note that not all RTL objects linked onto an insn
8869 chain are of class `i'.
8874 You can make no assumptions about the format of these codes.
8877 File: gccint.info, Node: Accessors, Next: Special Accessors, Prev: RTL Classes, Up: RTL
8879 10.3 Access to Operands
8880 =======================
8882 Operands of expressions are accessed using the macros `XEXP', `XINT',
8883 `XWINT' and `XSTR'. Each of these macros takes two arguments: an
8884 expression-pointer (RTX) and an operand number (counting from zero).
8889 accesses operand 2 of expression X, as an expression.
8893 accesses the same operand as an integer. `XSTR', used in the same
8894 fashion, would access it as a string.
8896 Any operand can be accessed as an integer, as an expression or as a
8897 string. You must choose the correct method of access for the kind of
8898 value actually stored in the operand. You would do this based on the
8899 expression code of the containing expression. That is also how you
8900 would know how many operands there are.
8902 For example, if X is a `subreg' expression, you know that it has two
8903 operands which can be correctly accessed as `XEXP (X, 0)' and `XINT (X,
8904 1)'. If you did `XINT (X, 0)', you would get the address of the
8905 expression operand but cast as an integer; that might occasionally be
8906 useful, but it would be cleaner to write `(int) XEXP (X, 0)'. `XEXP
8907 (X, 1)' would also compile without error, and would return the second,
8908 integer operand cast as an expression pointer, which would probably
8909 result in a crash when accessed. Nothing stops you from writing `XEXP
8910 (X, 28)' either, but this will access memory past the end of the
8911 expression with unpredictable results.
8913 Access to operands which are vectors is more complicated. You can use
8914 the macro `XVEC' to get the vector-pointer itself, or the macros
8915 `XVECEXP' and `XVECLEN' to access the elements and length of a vector.
8918 Access the vector-pointer which is operand number IDX in EXP.
8920 `XVECLEN (EXP, IDX)'
8921 Access the length (number of elements) in the vector which is in
8922 operand number IDX in EXP. This value is an `int'.
8924 `XVECEXP (EXP, IDX, ELTNUM)'
8925 Access element number ELTNUM in the vector which is in operand
8926 number IDX in EXP. This value is an RTX.
8928 It is up to you to make sure that ELTNUM is not negative and is
8929 less than `XVECLEN (EXP, IDX)'.
8931 All the macros defined in this section expand into lvalues and
8932 therefore can be used to assign the operands, lengths and vector
8933 elements as well as to access them.
8936 File: gccint.info, Node: Special Accessors, Next: Flags, Prev: Accessors, Up: RTL
8938 10.4 Access to Special Operands
8939 ===============================
8941 Some RTL nodes have special annotations associated with them.
8946 If 0, X is not in any alias set, and may alias anything.
8947 Otherwise, X can only alias `MEM's in a conflicting alias
8948 set. This value is set in a language-dependent manner in the
8949 front-end, and should not be altered in the back-end. In
8950 some front-ends, these numbers may correspond in some way to
8951 types, or other language-level entities, but they need not,
8952 and the back-end makes no such assumptions. These set
8953 numbers are tested with `alias_sets_conflict_p'.
8956 If this register is known to hold the value of some user-level
8957 declaration, this is that tree node. It may also be a
8958 `COMPONENT_REF', in which case this is some field reference,
8959 and `TREE_OPERAND (X, 0)' contains the declaration, or
8960 another `COMPONENT_REF', or null if there is no compile-time
8961 object associated with the reference.
8964 The offset from the start of `MEM_EXPR' as a `CONST_INT' rtx.
8967 The size in bytes of the memory reference as a `CONST_INT'
8968 rtx. This is mostly relevant for `BLKmode' references as
8969 otherwise the size is implied by the mode.
8972 The known alignment in bits of the memory reference.
8976 `ORIGINAL_REGNO (X)'
8977 This field holds the number the register "originally" had;
8978 for a pseudo register turned into a hard reg this will hold
8979 the old pseudo register number.
8982 If this register is known to hold the value of some user-level
8983 declaration, this is that tree node.
8986 If this register is known to hold the value of some user-level
8987 declaration, this is the offset into that logical storage.
8991 `SYMBOL_REF_DECL (X)'
8992 If the `symbol_ref' X was created for a `VAR_DECL' or a
8993 `FUNCTION_DECL', that tree is recorded here. If this value is
8994 null, then X was created by back end code generation routines,
8995 and there is no associated front end symbol table entry.
8997 `SYMBOL_REF_DECL' may also point to a tree of class `'c'',
8998 that is, some sort of constant. In this case, the
8999 `symbol_ref' is an entry in the per-file constant pool;
9000 again, there is no associated front end symbol table entry.
9002 `SYMBOL_REF_CONSTANT (X)'
9003 If `CONSTANT_POOL_ADDRESS_P (X)' is true, this is the constant
9004 pool entry for X. It is null otherwise.
9006 `SYMBOL_REF_DATA (X)'
9007 A field of opaque type used to store `SYMBOL_REF_DECL' or
9008 `SYMBOL_REF_CONSTANT'.
9010 `SYMBOL_REF_FLAGS (X)'
9011 In a `symbol_ref', this is used to communicate various
9012 predicates about the symbol. Some of these are common enough
9013 to be computed by common code, some are specific to the
9014 target. The common bits are:
9016 `SYMBOL_FLAG_FUNCTION'
9017 Set if the symbol refers to a function.
9020 Set if the symbol is local to this "module". See
9021 `TARGET_BINDS_LOCAL_P'.
9023 `SYMBOL_FLAG_EXTERNAL'
9024 Set if this symbol is not defined in this translation
9025 unit. Note that this is not the inverse of
9026 `SYMBOL_FLAG_LOCAL'.
9029 Set if the symbol is located in the small data section.
9030 See `TARGET_IN_SMALL_DATA_P'.
9032 `SYMBOL_REF_TLS_MODEL (X)'
9033 This is a multi-bit field accessor that returns the
9034 `tls_model' to be used for a thread-local storage
9035 symbol. It returns zero for non-thread-local symbols.
9037 `SYMBOL_FLAG_HAS_BLOCK_INFO'
9038 Set if the symbol has `SYMBOL_REF_BLOCK' and
9039 `SYMBOL_REF_BLOCK_OFFSET' fields.
9041 `SYMBOL_FLAG_ANCHOR'
9042 Set if the symbol is used as a section anchor. "Section
9043 anchors" are symbols that have a known position within
9044 an `object_block' and that can be used to access nearby
9045 members of that block. They are used to implement
9046 `-fsection-anchors'.
9048 If this flag is set, then `SYMBOL_FLAG_HAS_BLOCK_INFO'
9051 Bits beginning with `SYMBOL_FLAG_MACH_DEP' are available for
9054 `SYMBOL_REF_BLOCK (X)'
9055 If `SYMBOL_REF_HAS_BLOCK_INFO_P (X)', this is the `object_block'
9056 structure to which the symbol belongs, or `NULL' if it has not
9057 been assigned a block.
9059 `SYMBOL_REF_BLOCK_OFFSET (X)'
9060 If `SYMBOL_REF_HAS_BLOCK_INFO_P (X)', this is the offset of X from
9061 the first object in `SYMBOL_REF_BLOCK (X)'. The value is negative
9062 if X has not yet been assigned to a block, or it has not been
9063 given an offset within that block.
9066 File: gccint.info, Node: Flags, Next: Machine Modes, Prev: Special Accessors, Up: RTL
9068 10.5 Flags in an RTL Expression
9069 ===============================
9071 RTL expressions contain several flags (one-bit bit-fields) that are
9072 used in certain types of expression. Most often they are accessed with
9073 the following macros, which expand into lvalues.
9075 `CONSTANT_POOL_ADDRESS_P (X)'
9076 Nonzero in a `symbol_ref' if it refers to part of the current
9077 function's constant pool. For most targets these addresses are in
9078 a `.rodata' section entirely separate from the function, but for
9079 some targets the addresses are close to the beginning of the
9080 function. In either case GCC assumes these addresses can be
9081 addressed directly, perhaps with the help of base registers.
9082 Stored in the `unchanging' field and printed as `/u'.
9084 `RTL_CONST_CALL_P (X)'
9085 In a `call_insn' indicates that the insn represents a call to a
9086 const function. Stored in the `unchanging' field and printed as
9089 `RTL_PURE_CALL_P (X)'
9090 In a `call_insn' indicates that the insn represents a call to a
9091 pure function. Stored in the `return_val' field and printed as
9094 `RTL_CONST_OR_PURE_CALL_P (X)'
9095 In a `call_insn', true if `RTL_CONST_CALL_P' or `RTL_PURE_CALL_P'
9098 `RTL_LOOPING_CONST_OR_PURE_CALL_P (X)'
9099 In a `call_insn' indicates that the insn represents a possibly
9100 infinite looping call to a const or pure function. Stored in the
9101 `call' field and printed as `/c'. Only true if one of
9102 `RTL_CONST_CALL_P' or `RTL_PURE_CALL_P' is true.
9104 `INSN_ANNULLED_BRANCH_P (X)'
9105 In a `jump_insn', `call_insn', or `insn' indicates that the branch
9106 is an annulling one. See the discussion under `sequence' below.
9107 Stored in the `unchanging' field and printed as `/u'.
9109 `INSN_DELETED_P (X)'
9110 In an `insn', `call_insn', `jump_insn', `code_label', `barrier',
9111 or `note', nonzero if the insn has been deleted. Stored in the
9112 `volatil' field and printed as `/v'.
9114 `INSN_FROM_TARGET_P (X)'
9115 In an `insn' or `jump_insn' or `call_insn' in a delay slot of a
9116 branch, indicates that the insn is from the target of the branch.
9117 If the branch insn has `INSN_ANNULLED_BRANCH_P' set, this insn
9118 will only be executed if the branch is taken. For annulled
9119 branches with `INSN_FROM_TARGET_P' clear, the insn will be
9120 executed only if the branch is not taken. When
9121 `INSN_ANNULLED_BRANCH_P' is not set, this insn will always be
9122 executed. Stored in the `in_struct' field and printed as `/s'.
9124 `LABEL_PRESERVE_P (X)'
9125 In a `code_label' or `note', indicates that the label is
9126 referenced by code or data not visible to the RTL of a given
9127 function. Labels referenced by a non-local goto will have this
9128 bit set. Stored in the `in_struct' field and printed as `/s'.
9130 `LABEL_REF_NONLOCAL_P (X)'
9131 In `label_ref' and `reg_label' expressions, nonzero if this is a
9132 reference to a non-local label. Stored in the `volatil' field and
9135 `MEM_IN_STRUCT_P (X)'
9136 In `mem' expressions, nonzero for reference to an entire structure,
9137 union or array, or to a component of one. Zero for references to a
9138 scalar variable or through a pointer to a scalar. If both this
9139 flag and `MEM_SCALAR_P' are clear, then we don't know whether this
9140 `mem' is in a structure or not. Both flags should never be
9141 simultaneously set. Stored in the `in_struct' field and printed
9144 `MEM_KEEP_ALIAS_SET_P (X)'
9145 In `mem' expressions, 1 if we should keep the alias set for this
9146 mem unchanged when we access a component. Set to 1, for example,
9147 when we are already in a non-addressable component of an aggregate.
9148 Stored in the `jump' field and printed as `/j'.
9151 In `mem' expressions, nonzero for reference to a scalar known not
9152 to be a member of a structure, union, or array. Zero for such
9153 references and for indirections through pointers, even pointers
9154 pointing to scalar types. If both this flag and `MEM_IN_STRUCT_P'
9155 are clear, then we don't know whether this `mem' is in a structure
9156 or not. Both flags should never be simultaneously set. Stored in
9157 the `return_val' field and printed as `/i'.
9159 `MEM_VOLATILE_P (X)'
9160 In `mem', `asm_operands', and `asm_input' expressions, nonzero for
9161 volatile memory references. Stored in the `volatil' field and
9165 In `mem', nonzero for memory references that will not trap.
9166 Stored in the `call' field and printed as `/c'.
9169 Nonzero in a `mem' if the memory reference holds a pointer.
9170 Stored in the `frame_related' field and printed as `/f'.
9172 `REG_FUNCTION_VALUE_P (X)'
9173 Nonzero in a `reg' if it is the place in which this function's
9174 value is going to be returned. (This happens only in a hard
9175 register.) Stored in the `return_val' field and printed as `/i'.
9178 Nonzero in a `reg' if the register holds a pointer. Stored in the
9179 `frame_related' field and printed as `/f'.
9182 In a `reg', nonzero if it corresponds to a variable present in the
9183 user's source code. Zero for temporaries generated internally by
9184 the compiler. Stored in the `volatil' field and printed as `/v'.
9186 The same hard register may be used also for collecting the values
9187 of functions called by this one, but `REG_FUNCTION_VALUE_P' is zero
9188 in this kind of use.
9190 `RTX_FRAME_RELATED_P (X)'
9191 Nonzero in an `insn', `call_insn', `jump_insn', `barrier', or
9192 `set' which is part of a function prologue and sets the stack
9193 pointer, sets the frame pointer, or saves a register. This flag
9194 should also be set on an instruction that sets up a temporary
9195 register to use in place of the frame pointer. Stored in the
9196 `frame_related' field and printed as `/f'.
9198 In particular, on RISC targets where there are limits on the sizes
9199 of immediate constants, it is sometimes impossible to reach the
9200 register save area directly from the stack pointer. In that case,
9201 a temporary register is used that is near enough to the register
9202 save area, and the Canonical Frame Address, i.e., DWARF2's logical
9203 frame pointer, register must (temporarily) be changed to be this
9204 temporary register. So, the instruction that sets this temporary
9205 register must be marked as `RTX_FRAME_RELATED_P'.
9207 If the marked instruction is overly complex (defined in terms of
9208 what `dwarf2out_frame_debug_expr' can handle), you will also have
9209 to create a `REG_FRAME_RELATED_EXPR' note and attach it to the
9210 instruction. This note should contain a simple expression of the
9211 computation performed by this instruction, i.e., one that
9212 `dwarf2out_frame_debug_expr' can handle.
9214 This flag is required for exception handling support on targets
9217 `MEM_READONLY_P (X)'
9218 Nonzero in a `mem', if the memory is statically allocated and
9221 Read-only in this context means never modified during the lifetime
9222 of the program, not necessarily in ROM or in write-disabled pages.
9223 A common example of the later is a shared library's global offset
9224 table. This table is initialized by the runtime loader, so the
9225 memory is technically writable, but after control is transfered
9226 from the runtime loader to the application, this memory will never
9227 be subsequently modified.
9229 Stored in the `unchanging' field and printed as `/u'.
9232 During instruction scheduling, in an `insn', `call_insn' or
9233 `jump_insn', indicates that the previous insn must be scheduled
9234 together with this insn. This is used to ensure that certain
9235 groups of instructions will not be split up by the instruction
9236 scheduling pass, for example, `use' insns before a `call_insn' may
9237 not be separated from the `call_insn'. Stored in the `in_struct'
9238 field and printed as `/s'.
9240 `SET_IS_RETURN_P (X)'
9241 For a `set', nonzero if it is for a return. Stored in the `jump'
9242 field and printed as `/j'.
9244 `SIBLING_CALL_P (X)'
9245 For a `call_insn', nonzero if the insn is a sibling call. Stored
9246 in the `jump' field and printed as `/j'.
9248 `STRING_POOL_ADDRESS_P (X)'
9249 For a `symbol_ref' expression, nonzero if it addresses this
9250 function's string constant pool. Stored in the `frame_related'
9251 field and printed as `/f'.
9253 `SUBREG_PROMOTED_UNSIGNED_P (X)'
9254 Returns a value greater then zero for a `subreg' that has
9255 `SUBREG_PROMOTED_VAR_P' nonzero if the object being referenced is
9256 kept zero-extended, zero if it is kept sign-extended, and less
9257 then zero if it is extended some other way via the `ptr_extend'
9258 instruction. Stored in the `unchanging' field and `volatil'
9259 field, printed as `/u' and `/v'. This macro may only be used to
9260 get the value it may not be used to change the value. Use
9261 `SUBREG_PROMOTED_UNSIGNED_SET' to change the value.
9263 `SUBREG_PROMOTED_UNSIGNED_SET (X)'
9264 Set the `unchanging' and `volatil' fields in a `subreg' to reflect
9265 zero, sign, or other extension. If `volatil' is zero, then
9266 `unchanging' as nonzero means zero extension and as zero means
9267 sign extension. If `volatil' is nonzero then some other type of
9268 extension was done via the `ptr_extend' instruction.
9270 `SUBREG_PROMOTED_VAR_P (X)'
9271 Nonzero in a `subreg' if it was made when accessing an object that
9272 was promoted to a wider mode in accord with the `PROMOTED_MODE'
9273 machine description macro (*note Storage Layout::). In this case,
9274 the mode of the `subreg' is the declared mode of the object and
9275 the mode of `SUBREG_REG' is the mode of the register that holds
9276 the object. Promoted variables are always either sign- or
9277 zero-extended to the wider mode on every assignment. Stored in
9278 the `in_struct' field and printed as `/s'.
9280 `SYMBOL_REF_USED (X)'
9281 In a `symbol_ref', indicates that X has been used. This is
9282 normally only used to ensure that X is only declared external
9283 once. Stored in the `used' field.
9285 `SYMBOL_REF_WEAK (X)'
9286 In a `symbol_ref', indicates that X has been declared weak.
9287 Stored in the `return_val' field and printed as `/i'.
9289 `SYMBOL_REF_FLAG (X)'
9290 In a `symbol_ref', this is used as a flag for machine-specific
9291 purposes. Stored in the `volatil' field and printed as `/v'.
9293 Most uses of `SYMBOL_REF_FLAG' are historic and may be subsumed by
9294 `SYMBOL_REF_FLAGS'. Certainly use of `SYMBOL_REF_FLAGS' is
9295 mandatory if the target requires more than one bit of storage.
9297 These are the fields to which the above macros refer:
9300 In a `mem', 1 means that the memory reference will not trap.
9302 In a `call', 1 means that this pure or const call may possibly
9305 In an RTL dump, this flag is represented as `/c'.
9308 In an `insn' or `set' expression, 1 means that it is part of a
9309 function prologue and sets the stack pointer, sets the frame
9310 pointer, saves a register, or sets up a temporary register to use
9311 in place of the frame pointer.
9313 In `reg' expressions, 1 means that the register holds a pointer.
9315 In `mem' expressions, 1 means that the memory reference holds a
9318 In `symbol_ref' expressions, 1 means that the reference addresses
9319 this function's string constant pool.
9321 In an RTL dump, this flag is represented as `/f'.
9324 In `mem' expressions, it is 1 if the memory datum referred to is
9325 all or part of a structure or array; 0 if it is (or might be) a
9326 scalar variable. A reference through a C pointer has 0 because
9327 the pointer might point to a scalar variable. This information
9328 allows the compiler to determine something about possible cases of
9331 In `reg' expressions, it is 1 if the register has its entire life
9332 contained within the test expression of some loop.
9334 In `subreg' expressions, 1 means that the `subreg' is accessing an
9335 object that has had its mode promoted from a wider mode.
9337 In `label_ref' expressions, 1 means that the referenced label is
9338 outside the innermost loop containing the insn in which the
9339 `label_ref' was found.
9341 In `code_label' expressions, it is 1 if the label may never be
9342 deleted. This is used for labels which are the target of
9343 non-local gotos. Such a label that would have been deleted is
9344 replaced with a `note' of type `NOTE_INSN_DELETED_LABEL'.
9346 In an `insn' during dead-code elimination, 1 means that the insn is
9349 In an `insn' or `jump_insn' during reorg for an insn in the delay
9350 slot of a branch, 1 means that this insn is from the target of the
9353 In an `insn' during instruction scheduling, 1 means that this insn
9354 must be scheduled as part of a group together with the previous
9357 In an RTL dump, this flag is represented as `/s'.
9360 In `reg' expressions, 1 means the register contains the value to
9361 be returned by the current function. On machines that pass
9362 parameters in registers, the same register number may be used for
9363 parameters as well, but this flag is not set on such uses.
9365 In `mem' expressions, 1 means the memory reference is to a scalar
9366 known not to be a member of a structure, union, or array.
9368 In `symbol_ref' expressions, 1 means the referenced symbol is weak.
9370 In `call' expressions, 1 means the call is pure.
9372 In an RTL dump, this flag is represented as `/i'.
9375 In a `mem' expression, 1 means we should keep the alias set for
9376 this mem unchanged when we access a component.
9378 In a `set', 1 means it is for a return.
9380 In a `call_insn', 1 means it is a sibling call.
9382 In an RTL dump, this flag is represented as `/j'.
9385 In `reg' and `mem' expressions, 1 means that the value of the
9386 expression never changes.
9388 In `subreg' expressions, it is 1 if the `subreg' references an
9389 unsigned object whose mode has been promoted to a wider mode.
9391 In an `insn' or `jump_insn' in the delay slot of a branch
9392 instruction, 1 means an annulling branch should be used.
9394 In a `symbol_ref' expression, 1 means that this symbol addresses
9395 something in the per-function constant pool.
9397 In a `call_insn' 1 means that this instruction is a call to a const
9400 In an RTL dump, this flag is represented as `/u'.
9403 This flag is used directly (without an access macro) at the end of
9404 RTL generation for a function, to count the number of times an
9405 expression appears in insns. Expressions that appear more than
9406 once are copied, according to the rules for shared structure
9409 For a `reg', it is used directly (without an access macro) by the
9410 leaf register renumbering code to ensure that each register is only
9413 In a `symbol_ref', it indicates that an external declaration for
9414 the symbol has already been written.
9417 In a `mem', `asm_operands', or `asm_input' expression, it is 1 if
9418 the memory reference is volatile. Volatile memory references may
9419 not be deleted, reordered or combined.
9421 In a `symbol_ref' expression, it is used for machine-specific
9424 In a `reg' expression, it is 1 if the value is a user-level
9425 variable. 0 indicates an internal compiler temporary.
9427 In an `insn', 1 means the insn has been deleted.
9429 In `label_ref' and `reg_label' expressions, 1 means a reference to
9432 In an RTL dump, this flag is represented as `/v'.
9435 File: gccint.info, Node: Machine Modes, Next: Constants, Prev: Flags, Up: RTL
9440 A machine mode describes a size of data object and the representation
9441 used for it. In the C code, machine modes are represented by an
9442 enumeration type, `enum machine_mode', defined in `machmode.def'. Each
9443 RTL expression has room for a machine mode and so do certain kinds of
9444 tree expressions (declarations and types, to be precise).
9446 In debugging dumps and machine descriptions, the machine mode of an RTL
9447 expression is written after the expression code with a colon to separate
9448 them. The letters `mode' which appear at the end of each machine mode
9449 name are omitted. For example, `(reg:SI 38)' is a `reg' expression
9450 with machine mode `SImode'. If the mode is `VOIDmode', it is not
9453 Here is a table of machine modes. The term "byte" below refers to an
9454 object of `BITS_PER_UNIT' bits (*note Storage Layout::).
9457 "Bit" mode represents a single bit, for predicate registers.
9460 "Quarter-Integer" mode represents a single byte treated as an
9464 "Half-Integer" mode represents a two-byte integer.
9467 "Partial Single Integer" mode represents an integer which occupies
9468 four bytes but which doesn't really use all four. On some
9469 machines, this is the right mode to use for pointers.
9472 "Single Integer" mode represents a four-byte integer.
9475 "Partial Double Integer" mode represents an integer which occupies
9476 eight bytes but which doesn't really use all eight. On some
9477 machines, this is the right mode to use for certain pointers.
9480 "Double Integer" mode represents an eight-byte integer.
9483 "Tetra Integer" (?) mode represents a sixteen-byte integer.
9486 "Octa Integer" (?) mode represents a thirty-two-byte integer.
9489 "Quarter-Floating" mode represents a quarter-precision (single
9490 byte) floating point number.
9493 "Half-Floating" mode represents a half-precision (two byte)
9494 floating point number.
9497 "Three-Quarter-Floating" (?) mode represents a
9498 three-quarter-precision (three byte) floating point number.
9501 "Single Floating" mode represents a four byte floating point
9502 number. In the common case, of a processor with IEEE arithmetic
9503 and 8-bit bytes, this is a single-precision IEEE floating point
9504 number; it can also be used for double-precision (on processors
9505 with 16-bit bytes) and single-precision VAX and IBM types.
9508 "Double Floating" mode represents an eight byte floating point
9509 number. In the common case, of a processor with IEEE arithmetic
9510 and 8-bit bytes, this is a double-precision IEEE floating point
9514 "Extended Floating" mode represents an IEEE extended floating point
9515 number. This mode only has 80 meaningful bits (ten bytes). Some
9516 processors require such numbers to be padded to twelve bytes,
9517 others to sixteen; this mode is used for either.
9520 "Single Decimal Floating" mode represents a four byte decimal
9521 floating point number (as distinct from conventional binary
9525 "Double Decimal Floating" mode represents an eight byte decimal
9526 floating point number.
9529 "Tetra Decimal Floating" mode represents a sixteen byte decimal
9530 floating point number all 128 of whose bits are meaningful.
9533 "Tetra Floating" mode represents a sixteen byte floating point
9534 number all 128 of whose bits are meaningful. One common use is the
9535 IEEE quad-precision format.
9538 "Quarter-Fractional" mode represents a single byte treated as a
9539 signed fractional number. The default format is "s.7".
9542 "Half-Fractional" mode represents a two-byte signed fractional
9543 number. The default format is "s.15".
9546 "Single Fractional" mode represents a four-byte signed fractional
9547 number. The default format is "s.31".
9550 "Double Fractional" mode represents an eight-byte signed
9551 fractional number. The default format is "s.63".
9554 "Tetra Fractional" mode represents a sixteen-byte signed
9555 fractional number. The default format is "s.127".
9558 "Unsigned Quarter-Fractional" mode represents a single byte
9559 treated as an unsigned fractional number. The default format is
9563 "Unsigned Half-Fractional" mode represents a two-byte unsigned
9564 fractional number. The default format is ".16".
9567 "Unsigned Single Fractional" mode represents a four-byte unsigned
9568 fractional number. The default format is ".32".
9571 "Unsigned Double Fractional" mode represents an eight-byte unsigned
9572 fractional number. The default format is ".64".
9575 "Unsigned Tetra Fractional" mode represents a sixteen-byte unsigned
9576 fractional number. The default format is ".128".
9579 "Half-Accumulator" mode represents a two-byte signed accumulator.
9580 The default format is "s8.7".
9583 "Single Accumulator" mode represents a four-byte signed
9584 accumulator. The default format is "s16.15".
9587 "Double Accumulator" mode represents an eight-byte signed
9588 accumulator. The default format is "s32.31".
9591 "Tetra Accumulator" mode represents a sixteen-byte signed
9592 accumulator. The default format is "s64.63".
9595 "Unsigned Half-Accumulator" mode represents a two-byte unsigned
9596 accumulator. The default format is "8.8".
9599 "Unsigned Single Accumulator" mode represents a four-byte unsigned
9600 accumulator. The default format is "16.16".
9603 "Unsigned Double Accumulator" mode represents an eight-byte
9604 unsigned accumulator. The default format is "32.32".
9607 "Unsigned Tetra Accumulator" mode represents a sixteen-byte
9608 unsigned accumulator. The default format is "64.64".
9611 "Condition Code" mode represents the value of a condition code,
9612 which is a machine-specific set of bits used to represent the
9613 result of a comparison operation. Other machine-specific modes
9614 may also be used for the condition code. These modes are not used
9615 on machines that use `cc0' (see *note Condition Code::).
9618 "Block" mode represents values that are aggregates to which none of
9619 the other modes apply. In RTL, only memory references can have
9620 this mode, and only if they appear in string-move or vector
9621 instructions. On machines which have no such instructions,
9622 `BLKmode' will not appear in RTL.
9625 Void mode means the absence of a mode or an unspecified mode. For
9626 example, RTL expressions of code `const_int' have mode `VOIDmode'
9627 because they can be taken to have whatever mode the context
9628 requires. In debugging dumps of RTL, `VOIDmode' is expressed by
9629 the absence of any mode.
9631 `QCmode, HCmode, SCmode, DCmode, XCmode, TCmode'
9632 These modes stand for a complex number represented as a pair of
9633 floating point values. The floating point values are in `QFmode',
9634 `HFmode', `SFmode', `DFmode', `XFmode', and `TFmode', respectively.
9636 `CQImode, CHImode, CSImode, CDImode, CTImode, COImode'
9637 These modes stand for a complex number represented as a pair of
9638 integer values. The integer values are in `QImode', `HImode',
9639 `SImode', `DImode', `TImode', and `OImode', respectively.
9641 The machine description defines `Pmode' as a C macro which expands
9642 into the machine mode used for addresses. Normally this is the mode
9643 whose size is `BITS_PER_WORD', `SImode' on 32-bit machines.
9645 The only modes which a machine description must support are `QImode',
9646 and the modes corresponding to `BITS_PER_WORD', `FLOAT_TYPE_SIZE' and
9647 `DOUBLE_TYPE_SIZE'. The compiler will attempt to use `DImode' for
9648 8-byte structures and unions, but this can be prevented by overriding
9649 the definition of `MAX_FIXED_MODE_SIZE'. Alternatively, you can have
9650 the compiler use `TImode' for 16-byte structures and unions. Likewise,
9651 you can arrange for the C type `short int' to avoid using `HImode'.
9653 Very few explicit references to machine modes remain in the compiler
9654 and these few references will soon be removed. Instead, the machine
9655 modes are divided into mode classes. These are represented by the
9656 enumeration type `enum mode_class' defined in `machmode.h'. The
9657 possible mode classes are:
9660 Integer modes. By default these are `BImode', `QImode', `HImode',
9661 `SImode', `DImode', `TImode', and `OImode'.
9664 The "partial integer" modes, `PQImode', `PHImode', `PSImode' and
9668 Floating point modes. By default these are `QFmode', `HFmode',
9669 `TQFmode', `SFmode', `DFmode', `XFmode' and `TFmode'.
9671 `MODE_DECIMAL_FLOAT'
9672 Decimal floating point modes. By default these are `SDmode',
9673 `DDmode' and `TDmode'.
9676 Signed fractional modes. By default these are `QQmode', `HQmode',
9677 `SQmode', `DQmode' and `TQmode'.
9680 Unsigned fractional modes. By default these are `UQQmode',
9681 `UHQmode', `USQmode', `UDQmode' and `UTQmode'.
9684 Signed accumulator modes. By default these are `HAmode',
9685 `SAmode', `DAmode' and `TAmode'.
9688 Unsigned accumulator modes. By default these are `UHAmode',
9689 `USAmode', `UDAmode' and `UTAmode'.
9692 Complex integer modes. (These are not currently implemented).
9694 `MODE_COMPLEX_FLOAT'
9695 Complex floating point modes. By default these are `QCmode',
9696 `HCmode', `SCmode', `DCmode', `XCmode', and `TCmode'.
9699 Algol or Pascal function variables including a static chain.
9700 (These are not currently implemented).
9703 Modes representing condition code values. These are `CCmode' plus
9704 any `CC_MODE' modes listed in the `MACHINE-modes.def'. *Note Jump
9705 Patterns::, also see *Note Condition Code::.
9708 This is a catchall mode class for modes which don't fit into the
9709 above classes. Currently `VOIDmode' and `BLKmode' are in
9712 Here are some C macros that relate to machine modes:
9715 Returns the machine mode of the RTX X.
9717 `PUT_MODE (X, NEWMODE)'
9718 Alters the machine mode of the RTX X to be NEWMODE.
9721 Stands for the number of machine modes available on the target
9722 machine. This is one greater than the largest numeric value of any
9726 Returns the name of mode M as a string.
9728 `GET_MODE_CLASS (M)'
9729 Returns the mode class of mode M.
9731 `GET_MODE_WIDER_MODE (M)'
9732 Returns the next wider natural mode. For example, the expression
9733 `GET_MODE_WIDER_MODE (QImode)' returns `HImode'.
9736 Returns the size in bytes of a datum of mode M.
9738 `GET_MODE_BITSIZE (M)'
9739 Returns the size in bits of a datum of mode M.
9742 Returns the number of integral bits of a datum of fixed-point mode
9746 Returns the number of fractional bits of a datum of fixed-point
9750 Returns a bitmask containing 1 for all bits in a word that fit
9751 within mode M. This macro can only be used for modes whose
9752 bitsize is less than or equal to `HOST_BITS_PER_INT'.
9754 `GET_MODE_ALIGNMENT (M)'
9755 Return the required alignment, in bits, for an object of mode M.
9757 `GET_MODE_UNIT_SIZE (M)'
9758 Returns the size in bytes of the subunits of a datum of mode M.
9759 This is the same as `GET_MODE_SIZE' except in the case of complex
9760 modes. For them, the unit size is the size of the real or
9763 `GET_MODE_NUNITS (M)'
9764 Returns the number of units contained in a mode, i.e.,
9765 `GET_MODE_SIZE' divided by `GET_MODE_UNIT_SIZE'.
9767 `GET_CLASS_NARROWEST_MODE (C)'
9768 Returns the narrowest mode in mode class C.
9770 The global variables `byte_mode' and `word_mode' contain modes whose
9771 classes are `MODE_INT' and whose bitsizes are either `BITS_PER_UNIT' or
9772 `BITS_PER_WORD', respectively. On 32-bit machines, these are `QImode'
9773 and `SImode', respectively.
9776 File: gccint.info, Node: Constants, Next: Regs and Memory, Prev: Machine Modes, Up: RTL
9778 10.7 Constant Expression Types
9779 ==============================
9781 The simplest RTL expressions are those that represent constant values.
9784 This type of expression represents the integer value I. I is
9785 customarily accessed with the macro `INTVAL' as in `INTVAL (EXP)',
9786 which is equivalent to `XWINT (EXP, 0)'.
9788 Constants generated for modes with fewer bits than `HOST_WIDE_INT'
9789 must be sign extended to full width (e.g., with `gen_int_mode').
9791 There is only one expression object for the integer value zero; it
9792 is the value of the variable `const0_rtx'. Likewise, the only
9793 expression for integer value one is found in `const1_rtx', the only
9794 expression for integer value two is found in `const2_rtx', and the
9795 only expression for integer value negative one is found in
9796 `constm1_rtx'. Any attempt to create an expression of code
9797 `const_int' and value zero, one, two or negative one will return
9798 `const0_rtx', `const1_rtx', `const2_rtx' or `constm1_rtx' as
9801 Similarly, there is only one object for the integer whose value is
9802 `STORE_FLAG_VALUE'. It is found in `const_true_rtx'. If
9803 `STORE_FLAG_VALUE' is one, `const_true_rtx' and `const1_rtx' will
9804 point to the same object. If `STORE_FLAG_VALUE' is -1,
9805 `const_true_rtx' and `constm1_rtx' will point to the same object.
9807 `(const_double:M I0 I1 ...)'
9808 Represents either a floating-point constant of mode M or an
9809 integer constant too large to fit into `HOST_BITS_PER_WIDE_INT'
9810 bits but small enough to fit within twice that number of bits (GCC
9811 does not provide a mechanism to represent even larger constants).
9812 In the latter case, M will be `VOIDmode'.
9814 If M is `VOIDmode', the bits of the value are stored in I0 and I1.
9815 I0 is customarily accessed with the macro `CONST_DOUBLE_LOW' and
9816 I1 with `CONST_DOUBLE_HIGH'.
9818 If the constant is floating point (regardless of its precision),
9819 then the number of integers used to store the value depends on the
9820 size of `REAL_VALUE_TYPE' (*note Floating Point::). The integers
9821 represent a floating point number, but not precisely in the target
9822 machine's or host machine's floating point format. To convert
9823 them to the precise bit pattern used by the target machine, use
9824 the macro `REAL_VALUE_TO_TARGET_DOUBLE' and friends (*note Data
9827 `(const_fixed:M ...)'
9828 Represents a fixed-point constant of mode M. The operand is a
9829 data structure of type `struct fixed_value' and is accessed with
9830 the macro `CONST_FIXED_VALUE'. The high part of data is accessed
9831 with `CONST_FIXED_VALUE_HIGH'; the low part is accessed with
9832 `CONST_FIXED_VALUE_LOW'.
9834 `(const_vector:M [X0 X1 ...])'
9835 Represents a vector constant. The square brackets stand for the
9836 vector containing the constant elements. X0, X1 and so on are the
9837 `const_int', `const_double' or `const_fixed' elements.
9839 The number of units in a `const_vector' is obtained with the macro
9840 `CONST_VECTOR_NUNITS' as in `CONST_VECTOR_NUNITS (V)'.
9842 Individual elements in a vector constant are accessed with the
9843 macro `CONST_VECTOR_ELT' as in `CONST_VECTOR_ELT (V, N)' where V
9844 is the vector constant and N is the element desired.
9846 `(const_string STR)'
9847 Represents a constant string with value STR. Currently this is
9848 used only for insn attributes (*note Insn Attributes::) since
9849 constant strings in C are placed in memory.
9851 `(symbol_ref:MODE SYMBOL)'
9852 Represents the value of an assembler label for data. SYMBOL is a
9853 string that describes the name of the assembler label. If it
9854 starts with a `*', the label is the rest of SYMBOL not including
9855 the `*'. Otherwise, the label is SYMBOL, usually prefixed with
9858 The `symbol_ref' contains a mode, which is usually `Pmode'.
9859 Usually that is the only mode for which a symbol is directly valid.
9861 `(label_ref:MODE LABEL)'
9862 Represents the value of an assembler label for code. It contains
9863 one operand, an expression, which must be a `code_label' or a
9864 `note' of type `NOTE_INSN_DELETED_LABEL' that appears in the
9865 instruction sequence to identify the place where the label should
9868 The reason for using a distinct expression type for code label
9869 references is so that jump optimization can distinguish them.
9871 The `label_ref' contains a mode, which is usually `Pmode'.
9872 Usually that is the only mode for which a label is directly valid.
9875 Represents a constant that is the result of an assembly-time
9876 arithmetic computation. The operand, EXP, is an expression that
9877 contains only constants (`const_int', `symbol_ref' and `label_ref'
9878 expressions) combined with `plus' and `minus'. However, not all
9879 combinations are valid, since the assembler cannot do arbitrary
9880 arithmetic on relocatable symbols.
9882 M should be `Pmode'.
9885 Represents the high-order bits of EXP, usually a `symbol_ref'.
9886 The number of bits is machine-dependent and is normally the number
9887 of bits specified in an instruction that initializes the high
9888 order bits of a register. It is used with `lo_sum' to represent
9889 the typical two-instruction sequence used in RISC machines to
9890 reference a global memory location.
9892 M should be `Pmode'.
9894 The macro `CONST0_RTX (MODE)' refers to an expression with value 0 in
9895 mode MODE. If mode MODE is of mode class `MODE_INT', it returns
9896 `const0_rtx'. If mode MODE is of mode class `MODE_FLOAT', it returns a
9897 `CONST_DOUBLE' expression in mode MODE. Otherwise, it returns a
9898 `CONST_VECTOR' expression in mode MODE. Similarly, the macro
9899 `CONST1_RTX (MODE)' refers to an expression with value 1 in mode MODE
9900 and similarly for `CONST2_RTX'. The `CONST1_RTX' and `CONST2_RTX'
9901 macros are undefined for vector modes.
9904 File: gccint.info, Node: Regs and Memory, Next: Arithmetic, Prev: Constants, Up: RTL
9906 10.8 Registers and Memory
9907 =========================
9909 Here are the RTL expression types for describing access to machine
9910 registers and to main memory.
9913 For small values of the integer N (those that are less than
9914 `FIRST_PSEUDO_REGISTER'), this stands for a reference to machine
9915 register number N: a "hard register". For larger values of N, it
9916 stands for a temporary value or "pseudo register". The compiler's
9917 strategy is to generate code assuming an unlimited number of such
9918 pseudo registers, and later convert them into hard registers or
9919 into memory references.
9921 M is the machine mode of the reference. It is necessary because
9922 machines can generally refer to each register in more than one
9923 mode. For example, a register may contain a full word but there
9924 may be instructions to refer to it as a half word or as a single
9925 byte, as well as instructions to refer to it as a floating point
9926 number of various precisions.
9928 Even for a register that the machine can access in only one mode,
9929 the mode must always be specified.
9931 The symbol `FIRST_PSEUDO_REGISTER' is defined by the machine
9932 description, since the number of hard registers on the machine is
9933 an invariant characteristic of the machine. Note, however, that
9934 not all of the machine registers must be general registers. All
9935 the machine registers that can be used for storage of data are
9936 given hard register numbers, even those that can be used only in
9937 certain instructions or can hold only certain types of data.
9939 A hard register may be accessed in various modes throughout one
9940 function, but each pseudo register is given a natural mode and is
9941 accessed only in that mode. When it is necessary to describe an
9942 access to a pseudo register using a nonnatural mode, a `subreg'
9945 A `reg' expression with a machine mode that specifies more than
9946 one word of data may actually stand for several consecutive
9947 registers. If in addition the register number specifies a
9948 hardware register, then it actually represents several consecutive
9949 hardware registers starting with the specified one.
9951 Each pseudo register number used in a function's RTL code is
9952 represented by a unique `reg' expression.
9954 Some pseudo register numbers, those within the range of
9955 `FIRST_VIRTUAL_REGISTER' to `LAST_VIRTUAL_REGISTER' only appear
9956 during the RTL generation phase and are eliminated before the
9957 optimization phases. These represent locations in the stack frame
9958 that cannot be determined until RTL generation for the function
9959 has been completed. The following virtual register numbers are
9962 `VIRTUAL_INCOMING_ARGS_REGNUM'
9963 This points to the first word of the incoming arguments
9964 passed on the stack. Normally these arguments are placed
9965 there by the caller, but the callee may have pushed some
9966 arguments that were previously passed in registers.
9968 When RTL generation is complete, this virtual register is
9969 replaced by the sum of the register given by
9970 `ARG_POINTER_REGNUM' and the value of `FIRST_PARM_OFFSET'.
9972 `VIRTUAL_STACK_VARS_REGNUM'
9973 If `FRAME_GROWS_DOWNWARD' is defined to a nonzero value, this
9974 points to immediately above the first variable on the stack.
9975 Otherwise, it points to the first variable on the stack.
9977 `VIRTUAL_STACK_VARS_REGNUM' is replaced with the sum of the
9978 register given by `FRAME_POINTER_REGNUM' and the value
9979 `STARTING_FRAME_OFFSET'.
9981 `VIRTUAL_STACK_DYNAMIC_REGNUM'
9982 This points to the location of dynamically allocated memory
9983 on the stack immediately after the stack pointer has been
9984 adjusted by the amount of memory desired.
9986 This virtual register is replaced by the sum of the register
9987 given by `STACK_POINTER_REGNUM' and the value
9988 `STACK_DYNAMIC_OFFSET'.
9990 `VIRTUAL_OUTGOING_ARGS_REGNUM'
9991 This points to the location in the stack at which outgoing
9992 arguments should be written when the stack is pre-pushed
9993 (arguments pushed using push insns should always use
9994 `STACK_POINTER_REGNUM').
9996 This virtual register is replaced by the sum of the register
9997 given by `STACK_POINTER_REGNUM' and the value
9998 `STACK_POINTER_OFFSET'.
10000 `(subreg:M1 REG:M2 BYTENUM)'
10001 `subreg' expressions are used to refer to a register in a machine
10002 mode other than its natural one, or to refer to one register of a
10003 multi-part `reg' that actually refers to several registers.
10005 Each pseudo register has a natural mode. If it is necessary to
10006 operate on it in a different mode, the register must be enclosed
10009 There are currently three supported types for the first operand of
10011 * pseudo registers This is the most common case. Most
10012 `subreg's have pseudo `reg's as their first operand.
10014 * mem `subreg's of `mem' were common in earlier versions of GCC
10015 and are still supported. During the reload pass these are
10016 replaced by plain `mem's. On machines that do not do
10017 instruction scheduling, use of `subreg's of `mem' are still
10018 used, but this is no longer recommended. Such `subreg's are
10019 considered to be `register_operand's rather than
10020 `memory_operand's before and during reload. Because of this,
10021 the scheduling passes cannot properly schedule instructions
10022 with `subreg's of `mem', so for machines that do scheduling,
10023 `subreg's of `mem' should never be used. To support this,
10024 the combine and recog passes have explicit code to inhibit
10025 the creation of `subreg's of `mem' when `INSN_SCHEDULING' is
10028 The use of `subreg's of `mem' after the reload pass is an area
10029 that is not well understood and should be avoided. There is
10030 still some code in the compiler to support this, but this
10031 code has possibly rotted. This use of `subreg's is
10032 discouraged and will most likely not be supported in the
10035 * hard registers It is seldom necessary to wrap hard registers
10036 in `subreg's; such registers would normally reduce to a
10037 single `reg' rtx. This use of `subreg's is discouraged and
10038 may not be supported in the future.
10041 `subreg's of `subreg's are not supported. Using
10042 `simplify_gen_subreg' is the recommended way to avoid this problem.
10044 `subreg's come in two distinct flavors, each having its own usage
10047 Paradoxical subregs
10048 When M1 is strictly wider than M2, the `subreg' expression is
10049 called "paradoxical". The canonical test for this class of
10052 GET_MODE_SIZE (M1) > GET_MODE_SIZE (M2)
10054 Paradoxical `subreg's can be used as both lvalues and rvalues.
10055 When used as an lvalue, the low-order bits of the source value
10056 are stored in REG and the high-order bits are discarded.
10057 When used as an rvalue, the low-order bits of the `subreg' are
10058 taken from REG while the high-order bits may or may not be
10061 The high-order bits of rvalues are in the following
10064 * `subreg's of `mem' When M2 is smaller than a word, the
10065 macro `LOAD_EXTEND_OP', can control how the high-order
10068 * `subreg' of `reg's The upper bits are defined when
10069 `SUBREG_PROMOTED_VAR_P' is true.
10070 `SUBREG_PROMOTED_UNSIGNED_P' describes what the upper
10071 bits hold. Such subregs usually represent local
10072 variables, register variables and parameter pseudo
10073 variables that have been promoted to a wider mode.
10076 BYTENUM is always zero for a paradoxical `subreg', even on
10077 big-endian targets.
10079 For example, the paradoxical `subreg':
10081 (set (subreg:SI (reg:HI X) 0) Y)
10083 stores the lower 2 bytes of Y in X and discards the upper 2
10084 bytes. A subsequent:
10086 (set Z (subreg:SI (reg:HI X) 0))
10088 would set the lower two bytes of Z to Y and set the upper two
10089 bytes to an unknown value assuming `SUBREG_PROMOTED_VAR_P' is
10093 When M1 is at least as narrow as M2 the `subreg' expression
10094 is called "normal".
10096 Normal `subreg's restrict consideration to certain bits of
10097 REG. There are two cases. If M1 is smaller than a word, the
10098 `subreg' refers to the least-significant part (or "lowpart")
10099 of one word of REG. If M1 is word-sized or greater, the
10100 `subreg' refers to one or more complete words.
10102 When used as an lvalue, `subreg' is a word-based accessor.
10103 Storing to a `subreg' modifies all the words of REG that
10104 overlap the `subreg', but it leaves the other words of REG
10107 When storing to a normal `subreg' that is smaller than a word,
10108 the other bits of the referenced word are usually left in an
10109 undefined state. This laxity makes it easier to generate
10110 efficient code for such instructions. To represent an
10111 instruction that preserves all the bits outside of those in
10112 the `subreg', use `strict_low_part' or `zero_extract' around
10115 BYTENUM must identify the offset of the first byte of the
10116 `subreg' from the start of REG, assuming that REG is laid out
10117 in memory order. The memory order of bytes is defined by two
10118 target macros, `WORDS_BIG_ENDIAN' and `BYTES_BIG_ENDIAN':
10120 * `WORDS_BIG_ENDIAN', if set to 1, says that byte number
10121 zero is part of the most significant word; otherwise, it
10122 is part of the least significant word.
10124 * `BYTES_BIG_ENDIAN', if set to 1, says that byte number
10125 zero is the most significant byte within a word;
10126 otherwise, it is the least significant byte within a
10129 On a few targets, `FLOAT_WORDS_BIG_ENDIAN' disagrees with
10130 `WORDS_BIG_ENDIAN'. However, most parts of the compiler treat
10131 floating point values as if they had the same endianness as
10132 integer values. This works because they handle them solely
10133 as a collection of integer values, with no particular
10134 numerical value. Only real.c and the runtime libraries care
10135 about `FLOAT_WORDS_BIG_ENDIAN'.
10139 (subreg:HI (reg:SI X) 2)
10141 on a `BYTES_BIG_ENDIAN', `UNITS_PER_WORD == 4' target is the
10144 (subreg:HI (reg:SI X) 0)
10146 on a little-endian, `UNITS_PER_WORD == 4' target. Both
10147 `subreg's access the lower two bytes of register X.
10150 A `MODE_PARTIAL_INT' mode behaves as if it were as wide as the
10151 corresponding `MODE_INT' mode, except that it has an unknown
10152 number of undefined bits. For example:
10154 (subreg:PSI (reg:SI 0) 0)
10156 accesses the whole of `(reg:SI 0)', but the exact relationship
10157 between the `PSImode' value and the `SImode' value is not defined.
10158 If we assume `UNITS_PER_WORD <= 4', then the following two
10161 (subreg:PSI (reg:DI 0) 0)
10162 (subreg:PSI (reg:DI 0) 4)
10164 represent independent 4-byte accesses to the two halves of
10165 `(reg:DI 0)'. Both `subreg's have an unknown number of undefined
10168 If `UNITS_PER_WORD <= 2' then these two `subreg's:
10170 (subreg:HI (reg:PSI 0) 0)
10171 (subreg:HI (reg:PSI 0) 2)
10173 represent independent 2-byte accesses that together span the whole
10174 of `(reg:PSI 0)'. Storing to the first `subreg' does not affect
10175 the value of the second, and vice versa. `(reg:PSI 0)' has an
10176 unknown number of undefined bits, so the assignment:
10178 (set (subreg:HI (reg:PSI 0) 0) (reg:HI 4))
10180 does not guarantee that `(subreg:HI (reg:PSI 0) 0)' has the value
10183 The rules above apply to both pseudo REGs and hard REGs. If the
10184 semantics are not correct for particular combinations of M1, M2
10185 and hard REG, the target-specific code must ensure that those
10186 combinations are never used. For example:
10188 CANNOT_CHANGE_MODE_CLASS (M2, M1, CLASS)
10190 must be true for every class CLASS that includes REG.
10192 The first operand of a `subreg' expression is customarily accessed
10193 with the `SUBREG_REG' macro and the second operand is customarily
10194 accessed with the `SUBREG_BYTE' macro.
10196 It has been several years since a platform in which
10197 `BYTES_BIG_ENDIAN' not equal to `WORDS_BIG_ENDIAN' has been
10198 tested. Anyone wishing to support such a platform in the future
10199 may be confronted with code rot.
10202 This represents a scratch register that will be required for the
10203 execution of a single instruction and not used subsequently. It is
10204 converted into a `reg' by either the local register allocator or
10207 `scratch' is usually present inside a `clobber' operation (*note
10211 This refers to the machine's condition code register. It has no
10212 operands and may not have a machine mode. There are two ways to
10215 * To stand for a complete set of condition code flags. This is
10216 best on most machines, where each comparison sets the entire
10219 With this technique, `(cc0)' may be validly used in only two
10220 contexts: as the destination of an assignment (in test and
10221 compare instructions) and in comparison operators comparing
10222 against zero (`const_int' with value zero; that is to say,
10225 * To stand for a single flag that is the result of a single
10226 condition. This is useful on machines that have only a
10227 single flag bit, and in which comparison instructions must
10228 specify the condition to test.
10230 With this technique, `(cc0)' may be validly used in only two
10231 contexts: as the destination of an assignment (in test and
10232 compare instructions) where the source is a comparison
10233 operator, and as the first operand of `if_then_else' (in a
10234 conditional branch).
10236 There is only one expression object of code `cc0'; it is the value
10237 of the variable `cc0_rtx'. Any attempt to create an expression of
10238 code `cc0' will return `cc0_rtx'.
10240 Instructions can set the condition code implicitly. On many
10241 machines, nearly all instructions set the condition code based on
10242 the value that they compute or store. It is not necessary to
10243 record these actions explicitly in the RTL because the machine
10244 description includes a prescription for recognizing the
10245 instructions that do so (by means of the macro
10246 `NOTICE_UPDATE_CC'). *Note Condition Code::. Only instructions
10247 whose sole purpose is to set the condition code, and instructions
10248 that use the condition code, need mention `(cc0)'.
10250 On some machines, the condition code register is given a register
10251 number and a `reg' is used instead of `(cc0)'. This is usually the
10252 preferable approach if only a small subset of instructions modify
10253 the condition code. Other machines store condition codes in
10254 general registers; in such cases a pseudo register should be used.
10256 Some machines, such as the SPARC and RS/6000, have two sets of
10257 arithmetic instructions, one that sets and one that does not set
10258 the condition code. This is best handled by normally generating
10259 the instruction that does not set the condition code, and making a
10260 pattern that both performs the arithmetic and sets the condition
10261 code register (which would not be `(cc0)' in this case). For
10262 examples, search for `addcc' and `andcc' in `sparc.md'.
10265 This represents the machine's program counter. It has no operands
10266 and may not have a machine mode. `(pc)' may be validly used only
10267 in certain specific contexts in jump instructions.
10269 There is only one expression object of code `pc'; it is the value
10270 of the variable `pc_rtx'. Any attempt to create an expression of
10271 code `pc' will return `pc_rtx'.
10273 All instructions that do not jump alter the program counter
10274 implicitly by incrementing it, but there is no need to mention
10277 `(mem:M ADDR ALIAS)'
10278 This RTX represents a reference to main memory at an address
10279 represented by the expression ADDR. M specifies how large a unit
10280 of memory is accessed. ALIAS specifies an alias set for the
10281 reference. In general two items are in different alias sets if
10282 they cannot reference the same memory address.
10284 The construct `(mem:BLK (scratch))' is considered to alias all
10285 other memories. Thus it may be used as a memory barrier in
10286 epilogue stack deallocation patterns.
10288 `(concatM RTX RTX)'
10289 This RTX represents the concatenation of two other RTXs. This is
10290 used for complex values. It should only appear in the RTL
10291 attached to declarations and during RTL generation. It should not
10292 appear in the ordinary insn chain.
10294 `(concatnM [RTX ...])'
10295 This RTX represents the concatenation of all the RTX to make a
10296 single value. Like `concat', this should only appear in
10297 declarations, and not in the insn chain.
10300 File: gccint.info, Node: Arithmetic, Next: Comparisons, Prev: Regs and Memory, Up: RTL
10302 10.9 RTL Expressions for Arithmetic
10303 ===================================
10305 Unless otherwise specified, all the operands of arithmetic expressions
10306 must be valid for mode M. An operand is valid for mode M if it has
10307 mode M, or if it is a `const_int' or `const_double' and M is a mode of
10310 For commutative binary operations, constants should be placed in the
10316 These three expressions all represent the sum of the values
10317 represented by X and Y carried out in machine mode M. They differ
10318 in their behavior on overflow of integer modes. `plus' wraps
10319 round modulo the width of M; `ss_plus' saturates at the maximum
10320 signed value representable in M; `us_plus' saturates at the
10321 maximum unsigned value.
10324 This expression represents the sum of X and the low-order bits of
10325 Y. It is used with `high' (*note Constants::) to represent the
10326 typical two-instruction sequence used in RISC machines to
10327 reference a global memory location.
10329 The number of low order bits is machine-dependent but is normally
10330 the number of bits in a `Pmode' item minus the number of bits set
10333 M should be `Pmode'.
10338 These three expressions represent the result of subtracting Y from
10339 X, carried out in mode M. Behavior on overflow is the same as for
10340 the three variants of `plus' (see above).
10343 Represents the result of subtracting Y from X for purposes of
10344 comparison. The result is computed without overflow, as if with
10345 infinite precision.
10347 Of course, machines can't really subtract with infinite precision.
10348 However, they can pretend to do so when only the sign of the
10349 result will be used, which is the case when the result is stored
10350 in the condition code. And that is the _only_ way this kind of
10351 expression may validly be used: as a value to be stored in the
10352 condition codes, either `(cc0)' or a register. *Note
10355 The mode M is not related to the modes of X and Y, but instead is
10356 the mode of the condition code value. If `(cc0)' is used, it is
10357 `VOIDmode'. Otherwise it is some mode in class `MODE_CC', often
10358 `CCmode'. *Note Condition Code::. If M is `VOIDmode' or
10359 `CCmode', the operation returns sufficient information (in an
10360 unspecified format) so that any comparison operator can be applied
10361 to the result of the `COMPARE' operation. For other modes in
10362 class `MODE_CC', the operation only returns a subset of this
10365 Normally, X and Y must have the same mode. Otherwise, `compare'
10366 is valid only if the mode of X is in class `MODE_INT' and Y is a
10367 `const_int' or `const_double' with mode `VOIDmode'. The mode of X
10368 determines what mode the comparison is to be done in; thus it must
10371 If one of the operands is a constant, it should be placed in the
10372 second operand and the comparison code adjusted as appropriate.
10374 A `compare' specifying two `VOIDmode' constants is not valid since
10375 there is no way to know in what mode the comparison is to be
10376 performed; the comparison must either be folded during the
10377 compilation or the first operand must be loaded into a register
10378 while its mode is still known.
10383 These two expressions represent the negation (subtraction from
10384 zero) of the value represented by X, carried out in mode M. They
10385 differ in the behavior on overflow of integer modes. In the case
10386 of `neg', the negation of the operand may be a number not
10387 representable in mode M, in which case it is truncated to M.
10388 `ss_neg' and `us_neg' ensure that an out-of-bounds result
10389 saturates to the maximum or minimum signed or unsigned value.
10394 Represents the signed product of the values represented by X and Y
10395 carried out in machine mode M. `ss_mult' and `us_mult' ensure
10396 that an out-of-bounds result saturates to the maximum or minimum
10397 signed or unsigned value.
10399 Some machines support a multiplication that generates a product
10400 wider than the operands. Write the pattern for this as
10402 (mult:M (sign_extend:M X) (sign_extend:M Y))
10404 where M is wider than the modes of X and Y, which need not be the
10407 For unsigned widening multiplication, use the same idiom, but with
10408 `zero_extend' instead of `sign_extend'.
10412 Represents the quotient in signed division of X by Y, carried out
10413 in machine mode M. If M is a floating point mode, it represents
10414 the exact quotient; otherwise, the integerized quotient. `ss_div'
10415 ensures that an out-of-bounds result saturates to the maximum or
10416 minimum signed value.
10418 Some machines have division instructions in which the operands and
10419 quotient widths are not all the same; you should represent such
10420 instructions using `truncate' and `sign_extend' as in,
10422 (truncate:M1 (div:M2 X (sign_extend:M2 Y)))
10426 Like `div' but represents unsigned division. `us_div' ensures
10427 that an out-of-bounds result saturates to the maximum or minimum
10432 Like `div' and `udiv' but represent the remainder instead of the
10437 Represents the smaller (for `smin') or larger (for `smax') of X
10438 and Y, interpreted as signed values in mode M. When used with
10439 floating point, if both operands are zeros, or if either operand
10440 is `NaN', then it is unspecified which of the two operands is
10441 returned as the result.
10445 Like `smin' and `smax', but the values are interpreted as unsigned
10449 Represents the bitwise complement of the value represented by X,
10450 carried out in mode M, which must be a fixed-point machine mode.
10453 Represents the bitwise logical-and of the values represented by X
10454 and Y, carried out in machine mode M, which must be a fixed-point
10458 Represents the bitwise inclusive-or of the values represented by X
10459 and Y, carried out in machine mode M, which must be a fixed-point
10463 Represents the bitwise exclusive-or of the values represented by X
10464 and Y, carried out in machine mode M, which must be a fixed-point
10468 `(ss_ashift:M X C)'
10469 `(us_ashift:M X C)'
10470 These three expressions represent the result of arithmetically
10471 shifting X left by C places. They differ in their behavior on
10472 overflow of integer modes. An `ashift' operation is a plain shift
10473 with no special behavior in case of a change in the sign bit;
10474 `ss_ashift' and `us_ashift' saturates to the minimum or maximum
10475 representable value if any of the bits shifted out differs from
10476 the final sign bit.
10478 X have mode M, a fixed-point machine mode. C be a fixed-point
10479 mode or be a constant with mode `VOIDmode'; which mode is
10480 determined by the mode called for in the machine description entry
10481 for the left-shift instruction. For example, on the VAX, the mode
10482 of C is `QImode' regardless of M.
10486 Like `ashift' but for right shift. Unlike the case for left shift,
10487 these two operations are distinct.
10491 Similar but represent left and right rotate. If C is a constant,
10495 Represents the absolute value of X, computed in mode M.
10498 Represents the square root of X, computed in mode M. Most often M
10499 will be a floating point mode.
10502 Represents one plus the index of the least significant 1-bit in X,
10503 represented as an integer of mode M. (The value is zero if X is
10504 zero.) The mode of X need not be M; depending on the target
10505 machine, various mode combinations may be valid.
10508 Represents the number of leading 0-bits in X, represented as an
10509 integer of mode M, starting at the most significant bit position.
10510 If X is zero, the value is determined by
10511 `CLZ_DEFINED_VALUE_AT_ZERO' (*note Misc::). Note that this is one
10512 of the few expressions that is not invariant under widening. The
10513 mode of X will usually be an integer mode.
10516 Represents the number of trailing 0-bits in X, represented as an
10517 integer of mode M, starting at the least significant bit position.
10518 If X is zero, the value is determined by
10519 `CTZ_DEFINED_VALUE_AT_ZERO' (*note Misc::). Except for this case,
10520 `ctz(x)' is equivalent to `ffs(X) - 1'. The mode of X will
10521 usually be an integer mode.
10524 Represents the number of 1-bits in X, represented as an integer of
10525 mode M. The mode of X will usually be an integer mode.
10528 Represents the number of 1-bits modulo 2 in X, represented as an
10529 integer of mode M. The mode of X will usually be an integer mode.
10532 Represents the value X with the order of bytes reversed, carried
10533 out in mode M, which must be a fixed-point machine mode.
10536 File: gccint.info, Node: Comparisons, Next: Bit-Fields, Prev: Arithmetic, Up: RTL
10538 10.10 Comparison Operations
10539 ===========================
10541 Comparison operators test a relation on two operands and are considered
10542 to represent a machine-dependent nonzero value described by, but not
10543 necessarily equal to, `STORE_FLAG_VALUE' (*note Misc::) if the relation
10544 holds, or zero if it does not, for comparison operators whose results
10545 have a `MODE_INT' mode, `FLOAT_STORE_FLAG_VALUE' (*note Misc::) if the
10546 relation holds, or zero if it does not, for comparison operators that
10547 return floating-point values, and a vector of either
10548 `VECTOR_STORE_FLAG_VALUE' (*note Misc::) if the relation holds, or of
10549 zeros if it does not, for comparison operators that return vector
10550 results. The mode of the comparison operation is independent of the
10551 mode of the data being compared. If the comparison operation is being
10552 tested (e.g., the first operand of an `if_then_else'), the mode must be
10555 There are two ways that comparison operations may be used. The
10556 comparison operators may be used to compare the condition codes `(cc0)'
10557 against zero, as in `(eq (cc0) (const_int 0))'. Such a construct
10558 actually refers to the result of the preceding instruction in which the
10559 condition codes were set. The instruction setting the condition code
10560 must be adjacent to the instruction using the condition code; only
10561 `note' insns may separate them.
10563 Alternatively, a comparison operation may directly compare two data
10564 objects. The mode of the comparison is determined by the operands; they
10565 must both be valid for a common machine mode. A comparison with both
10566 operands constant would be invalid as the machine mode could not be
10567 deduced from it, but such a comparison should never exist in RTL due to
10570 In the example above, if `(cc0)' were last set to `(compare X Y)', the
10571 comparison operation is identical to `(eq X Y)'. Usually only one style
10572 of comparisons is supported on a particular machine, but the combine
10573 pass will try to merge the operations to produce the `eq' shown in case
10574 it exists in the context of the particular insn involved.
10576 Inequality comparisons come in two flavors, signed and unsigned. Thus,
10577 there are distinct expression codes `gt' and `gtu' for signed and
10578 unsigned greater-than. These can produce different results for the same
10579 pair of integer values: for example, 1 is signed greater-than -1 but not
10580 unsigned greater-than, because -1 when regarded as unsigned is actually
10581 `0xffffffff' which is greater than 1.
10583 The signed comparisons are also used for floating point values.
10584 Floating point comparisons are distinguished by the machine modes of
10588 `STORE_FLAG_VALUE' if the values represented by X and Y are equal,
10592 `STORE_FLAG_VALUE' if the values represented by X and Y are not
10593 equal, otherwise 0.
10596 `STORE_FLAG_VALUE' if the X is greater than Y. If they are
10597 fixed-point, the comparison is done in a signed sense.
10600 Like `gt' but does unsigned comparison, on fixed-point numbers
10605 Like `gt' and `gtu' but test for "less than".
10609 Like `gt' and `gtu' but test for "greater than or equal".
10613 Like `gt' and `gtu' but test for "less than or equal".
10615 `(if_then_else COND THEN ELSE)'
10616 This is not a comparison operation but is listed here because it is
10617 always used in conjunction with a comparison operation. To be
10618 precise, COND is a comparison expression. This expression
10619 represents a choice, according to COND, between the value
10620 represented by THEN and the one represented by ELSE.
10622 On most machines, `if_then_else' expressions are valid only to
10623 express conditional jumps.
10625 `(cond [TEST1 VALUE1 TEST2 VALUE2 ...] DEFAULT)'
10626 Similar to `if_then_else', but more general. Each of TEST1,
10627 TEST2, ... is performed in turn. The result of this expression is
10628 the VALUE corresponding to the first nonzero test, or DEFAULT if
10629 none of the tests are nonzero expressions.
10631 This is currently not valid for instruction patterns and is
10632 supported only for insn attributes. *Note Insn Attributes::.
10635 File: gccint.info, Node: Bit-Fields, Next: Vector Operations, Prev: Comparisons, Up: RTL
10640 Special expression codes exist to represent bit-field instructions.
10642 `(sign_extract:M LOC SIZE POS)'
10643 This represents a reference to a sign-extended bit-field contained
10644 or starting in LOC (a memory or register reference). The bit-field
10645 is SIZE bits wide and starts at bit POS. The compilation option
10646 `BITS_BIG_ENDIAN' says which end of the memory unit POS counts
10649 If LOC is in memory, its mode must be a single-byte integer mode.
10650 If LOC is in a register, the mode to use is specified by the
10651 operand of the `insv' or `extv' pattern (*note Standard Names::)
10652 and is usually a full-word integer mode, which is the default if
10655 The mode of POS is machine-specific and is also specified in the
10656 `insv' or `extv' pattern.
10658 The mode M is the same as the mode that would be used for LOC if
10659 it were a register.
10661 A `sign_extract' can not appear as an lvalue, or part thereof, in
10664 `(zero_extract:M LOC SIZE POS)'
10665 Like `sign_extract' but refers to an unsigned or zero-extended
10666 bit-field. The same sequence of bits are extracted, but they are
10667 filled to an entire word with zeros instead of by sign-extension.
10669 Unlike `sign_extract', this type of expressions can be lvalues in
10670 RTL; they may appear on the left side of an assignment, indicating
10671 insertion of a value into the specified bit-field.
10674 File: gccint.info, Node: Vector Operations, Next: Conversions, Prev: Bit-Fields, Up: RTL
10676 10.12 Vector Operations
10677 =======================
10679 All normal RTL expressions can be used with vector modes; they are
10680 interpreted as operating on each part of the vector independently.
10681 Additionally, there are a few new expressions to describe specific
10684 `(vec_merge:M VEC1 VEC2 ITEMS)'
10685 This describes a merge operation between two vectors. The result
10686 is a vector of mode M; its elements are selected from either VEC1
10687 or VEC2. Which elements are selected is described by ITEMS, which
10688 is a bit mask represented by a `const_int'; a zero bit indicates
10689 the corresponding element in the result vector is taken from VEC2
10690 while a set bit indicates it is taken from VEC1.
10692 `(vec_select:M VEC1 SELECTION)'
10693 This describes an operation that selects parts of a vector. VEC1
10694 is the source vector, SELECTION is a `parallel' that contains a
10695 `const_int' for each of the subparts of the result vector, giving
10696 the number of the source subpart that should be stored into it.
10698 `(vec_concat:M VEC1 VEC2)'
10699 Describes a vector concat operation. The result is a
10700 concatenation of the vectors VEC1 and VEC2; its length is the sum
10701 of the lengths of the two inputs.
10703 `(vec_duplicate:M VEC)'
10704 This operation converts a small vector into a larger one by
10705 duplicating the input values. The output vector mode must have
10706 the same submodes as the input vector mode, and the number of
10707 output parts must be an integer multiple of the number of input
10712 File: gccint.info, Node: Conversions, Next: RTL Declarations, Prev: Vector Operations, Up: RTL
10717 All conversions between machine modes must be represented by explicit
10718 conversion operations. For example, an expression which is the sum of
10719 a byte and a full word cannot be written as `(plus:SI (reg:QI 34)
10720 (reg:SI 80))' because the `plus' operation requires two operands of the
10721 same machine mode. Therefore, the byte-sized operand is enclosed in a
10722 conversion operation, as in
10724 (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
10726 The conversion operation is not a mere placeholder, because there may
10727 be more than one way of converting from a given starting mode to the
10728 desired final mode. The conversion operation code says how to do it.
10730 For all conversion operations, X must not be `VOIDmode' because the
10731 mode in which to do the conversion would not be known. The conversion
10732 must either be done at compile-time or X must be placed into a register.
10734 `(sign_extend:M X)'
10735 Represents the result of sign-extending the value X to machine
10736 mode M. M must be a fixed-point mode and X a fixed-point value of
10737 a mode narrower than M.
10739 `(zero_extend:M X)'
10740 Represents the result of zero-extending the value X to machine
10741 mode M. M must be a fixed-point mode and X a fixed-point value of
10742 a mode narrower than M.
10744 `(float_extend:M X)'
10745 Represents the result of extending the value X to machine mode M.
10746 M must be a floating point mode and X a floating point value of a
10747 mode narrower than M.
10750 Represents the result of truncating the value X to machine mode M.
10751 M must be a fixed-point mode and X a fixed-point value of a mode
10754 `(ss_truncate:M X)'
10755 Represents the result of truncating the value X to machine mode M,
10756 using signed saturation in the case of overflow. Both M and the
10757 mode of X must be fixed-point modes.
10759 `(us_truncate:M X)'
10760 Represents the result of truncating the value X to machine mode M,
10761 using unsigned saturation in the case of overflow. Both M and the
10762 mode of X must be fixed-point modes.
10764 `(float_truncate:M X)'
10765 Represents the result of truncating the value X to machine mode M.
10766 M must be a floating point mode and X a floating point value of a
10770 Represents the result of converting fixed point value X, regarded
10771 as signed, to floating point mode M.
10773 `(unsigned_float:M X)'
10774 Represents the result of converting fixed point value X, regarded
10775 as unsigned, to floating point mode M.
10778 When M is a floating-point mode, represents the result of
10779 converting floating point value X (valid for mode M) to an
10780 integer, still represented in floating point mode M, by rounding
10783 When M is a fixed-point mode, represents the result of converting
10784 floating point value X to mode M, regarded as signed. How
10785 rounding is done is not specified, so this operation may be used
10786 validly in compiling C code only for integer-valued operands.
10788 `(unsigned_fix:M X)'
10789 Represents the result of converting floating point value X to
10790 fixed point mode M, regarded as unsigned. How rounding is done is
10793 `(fract_convert:M X)'
10794 Represents the result of converting fixed-point value X to
10795 fixed-point mode M, signed integer value X to fixed-point mode M,
10796 floating-point value X to fixed-point mode M, fixed-point value X
10797 to integer mode M regarded as signed, or fixed-point value X to
10798 floating-point mode M. When overflows or underflows happen, the
10799 results are undefined.
10802 Represents the result of converting fixed-point value X to
10803 fixed-point mode M, signed integer value X to fixed-point mode M,
10804 or floating-point value X to fixed-point mode M. When overflows
10805 or underflows happen, the results are saturated to the maximum or
10808 `(unsigned_fract_convert:M X)'
10809 Represents the result of converting fixed-point value X to integer
10810 mode M regarded as unsigned, or unsigned integer value X to
10811 fixed-point mode M. When overflows or underflows happen, the
10812 results are undefined.
10814 `(unsigned_sat_fract:M X)'
10815 Represents the result of converting unsigned integer value X to
10816 fixed-point mode M. When overflows or underflows happen, the
10817 results are saturated to the maximum or the minimum.
10820 File: gccint.info, Node: RTL Declarations, Next: Side Effects, Prev: Conversions, Up: RTL
10825 Declaration expression codes do not represent arithmetic operations but
10826 rather state assertions about their operands.
10828 `(strict_low_part (subreg:M (reg:N R) 0))'
10829 This expression code is used in only one context: as the
10830 destination operand of a `set' expression. In addition, the
10831 operand of this expression must be a non-paradoxical `subreg'
10834 The presence of `strict_low_part' says that the part of the
10835 register which is meaningful in mode N, but is not part of mode M,
10836 is not to be altered. Normally, an assignment to such a subreg is
10837 allowed to have undefined effects on the rest of the register when
10838 M is less than a word.
10841 File: gccint.info, Node: Side Effects, Next: Incdec, Prev: RTL Declarations, Up: RTL
10843 10.15 Side Effect Expressions
10844 =============================
10846 The expression codes described so far represent values, not actions.
10847 But machine instructions never produce values; they are meaningful only
10848 for their side effects on the state of the machine. Special expression
10849 codes are used to represent side effects.
10851 The body of an instruction is always one of these side effect codes;
10852 the codes described above, which represent values, appear only as the
10856 Represents the action of storing the value of X into the place
10857 represented by LVAL. LVAL must be an expression representing a
10858 place that can be stored in: `reg' (or `subreg', `strict_low_part'
10859 or `zero_extract'), `mem', `pc', `parallel', or `cc0'.
10861 If LVAL is a `reg', `subreg' or `mem', it has a machine mode; then
10862 X must be valid for that mode.
10864 If LVAL is a `reg' whose machine mode is less than the full width
10865 of the register, then it means that the part of the register
10866 specified by the machine mode is given the specified value and the
10867 rest of the register receives an undefined value. Likewise, if
10868 LVAL is a `subreg' whose machine mode is narrower than the mode of
10869 the register, the rest of the register can be changed in an
10872 If LVAL is a `strict_low_part' of a subreg, then the part of the
10873 register specified by the machine mode of the `subreg' is given
10874 the value X and the rest of the register is not changed.
10876 If LVAL is a `zero_extract', then the referenced part of the
10877 bit-field (a memory or register reference) specified by the
10878 `zero_extract' is given the value X and the rest of the bit-field
10879 is not changed. Note that `sign_extract' can not appear in LVAL.
10881 If LVAL is `(cc0)', it has no machine mode, and X may be either a
10882 `compare' expression or a value that may have any mode. The
10883 latter case represents a "test" instruction. The expression `(set
10884 (cc0) (reg:M N))' is equivalent to `(set (cc0) (compare (reg:M N)
10885 (const_int 0)))'. Use the former expression to save space during
10888 If LVAL is a `parallel', it is used to represent the case of a
10889 function returning a structure in multiple registers. Each element
10890 of the `parallel' is an `expr_list' whose first operand is a `reg'
10891 and whose second operand is a `const_int' representing the offset
10892 (in bytes) into the structure at which the data in that register
10893 corresponds. The first element may be null to indicate that the
10894 structure is also passed partly in memory.
10896 If LVAL is `(pc)', we have a jump instruction, and the
10897 possibilities for X are very limited. It may be a `label_ref'
10898 expression (unconditional jump). It may be an `if_then_else'
10899 (conditional jump), in which case either the second or the third
10900 operand must be `(pc)' (for the case which does not jump) and the
10901 other of the two must be a `label_ref' (for the case which does
10902 jump). X may also be a `mem' or `(plus:SI (pc) Y)', where Y may
10903 be a `reg' or a `mem'; these unusual patterns are used to
10904 represent jumps through branch tables.
10906 If LVAL is neither `(cc0)' nor `(pc)', the mode of LVAL must not
10907 be `VOIDmode' and the mode of X must be valid for the mode of LVAL.
10909 LVAL is customarily accessed with the `SET_DEST' macro and X with
10910 the `SET_SRC' macro.
10913 As the sole expression in a pattern, represents a return from the
10914 current function, on machines where this can be done with one
10915 instruction, such as VAXen. On machines where a multi-instruction
10916 "epilogue" must be executed in order to return from the function,
10917 returning is done by jumping to a label which precedes the
10918 epilogue, and the `return' expression code is never used.
10920 Inside an `if_then_else' expression, represents the value to be
10921 placed in `pc' to return to the caller.
10923 Note that an insn pattern of `(return)' is logically equivalent to
10924 `(set (pc) (return))', but the latter form is never used.
10926 `(call FUNCTION NARGS)'
10927 Represents a function call. FUNCTION is a `mem' expression whose
10928 address is the address of the function to be called. NARGS is an
10929 expression which can be used for two purposes: on some machines it
10930 represents the number of bytes of stack argument; on others, it
10931 represents the number of argument registers.
10933 Each machine has a standard machine mode which FUNCTION must have.
10934 The machine description defines macro `FUNCTION_MODE' to expand
10935 into the requisite mode name. The purpose of this mode is to
10936 specify what kind of addressing is allowed, on machines where the
10937 allowed kinds of addressing depend on the machine mode being
10941 Represents the storing or possible storing of an unpredictable,
10942 undescribed value into X, which must be a `reg', `scratch',
10943 `parallel' or `mem' expression.
10945 One place this is used is in string instructions that store
10946 standard values into particular hard registers. It may not be
10947 worth the trouble to describe the values that are stored, but it
10948 is essential to inform the compiler that the registers will be
10949 altered, lest it attempt to keep data in them across the string
10952 If X is `(mem:BLK (const_int 0))' or `(mem:BLK (scratch))', it
10953 means that all memory locations must be presumed clobbered. If X
10954 is a `parallel', it has the same meaning as a `parallel' in a
10957 Note that the machine description classifies certain hard
10958 registers as "call-clobbered". All function call instructions are
10959 assumed by default to clobber these registers, so there is no need
10960 to use `clobber' expressions to indicate this fact. Also, each
10961 function call is assumed to have the potential to alter any memory
10962 location, unless the function is declared `const'.
10964 If the last group of expressions in a `parallel' are each a
10965 `clobber' expression whose arguments are `reg' or `match_scratch'
10966 (*note RTL Template::) expressions, the combiner phase can add the
10967 appropriate `clobber' expressions to an insn it has constructed
10968 when doing so will cause a pattern to be matched.
10970 This feature can be used, for example, on a machine that whose
10971 multiply and add instructions don't use an MQ register but which
10972 has an add-accumulate instruction that does clobber the MQ
10973 register. Similarly, a combined instruction might require a
10974 temporary register while the constituent instructions might not.
10976 When a `clobber' expression for a register appears inside a
10977 `parallel' with other side effects, the register allocator
10978 guarantees that the register is unoccupied both before and after
10979 that insn if it is a hard register clobber. For pseudo-register
10980 clobber, the register allocator and the reload pass do not assign
10981 the same hard register to the clobber and the input operands if
10982 there is an insn alternative containing the `&' constraint (*note
10983 Modifiers::) for the clobber and the hard register is in register
10984 classes of the clobber in the alternative. You can clobber either
10985 a specific hard register, a pseudo register, or a `scratch'
10986 expression; in the latter two cases, GCC will allocate a hard
10987 register that is available there for use as a temporary.
10989 For instructions that require a temporary register, you should use
10990 `scratch' instead of a pseudo-register because this will allow the
10991 combiner phase to add the `clobber' when required. You do this by
10992 coding (`clobber' (`match_scratch' ...)). If you do clobber a
10993 pseudo register, use one which appears nowhere else--generate a
10994 new one each time. Otherwise, you may confuse CSE.
10996 There is one other known use for clobbering a pseudo register in a
10997 `parallel': when one of the input operands of the insn is also
10998 clobbered by the insn. In this case, using the same pseudo
10999 register in the clobber and elsewhere in the insn produces the
11003 Represents the use of the value of X. It indicates that the value
11004 in X at this point in the program is needed, even though it may
11005 not be apparent why this is so. Therefore, the compiler will not
11006 attempt to delete previous instructions whose only effect is to
11007 store a value in X. X must be a `reg' expression.
11009 In some situations, it may be tempting to add a `use' of a
11010 register in a `parallel' to describe a situation where the value
11011 of a special register will modify the behavior of the instruction.
11012 An hypothetical example might be a pattern for an addition that can
11013 either wrap around or use saturating addition depending on the
11014 value of a special control register:
11016 (parallel [(set (reg:SI 2) (unspec:SI [(reg:SI 3)
11020 This will not work, several of the optimizers only look at
11021 expressions locally; it is very likely that if you have multiple
11022 insns with identical inputs to the `unspec', they will be
11023 optimized away even if register 1 changes in between.
11025 This means that `use' can _only_ be used to describe that the
11026 register is live. You should think twice before adding `use'
11027 statements, more often you will want to use `unspec' instead. The
11028 `use' RTX is most commonly useful to describe that a fixed
11029 register is implicitly used in an insn. It is also safe to use in
11030 patterns where the compiler knows for other reasons that the result
11031 of the whole pattern is variable, such as `movmemM' or `call'
11034 During the reload phase, an insn that has a `use' as pattern can
11035 carry a reg_equal note. These `use' insns will be deleted before
11036 the reload phase exits.
11038 During the delayed branch scheduling phase, X may be an insn.
11039 This indicates that X previously was located at this place in the
11040 code and its data dependencies need to be taken into account.
11041 These `use' insns will be deleted before the delayed branch
11042 scheduling phase exits.
11044 `(parallel [X0 X1 ...])'
11045 Represents several side effects performed in parallel. The square
11046 brackets stand for a vector; the operand of `parallel' is a vector
11047 of expressions. X0, X1 and so on are individual side effect
11048 expressions--expressions of code `set', `call', `return',
11049 `clobber' or `use'.
11051 "In parallel" means that first all the values used in the
11052 individual side-effects are computed, and second all the actual
11053 side-effects are performed. For example,
11055 (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
11056 (set (mem:SI (reg:SI 1)) (reg:SI 1))])
11058 says unambiguously that the values of hard register 1 and the
11059 memory location addressed by it are interchanged. In both places
11060 where `(reg:SI 1)' appears as a memory address it refers to the
11061 value in register 1 _before_ the execution of the insn.
11063 It follows that it is _incorrect_ to use `parallel' and expect the
11064 result of one `set' to be available for the next one. For
11065 example, people sometimes attempt to represent a jump-if-zero
11066 instruction this way:
11068 (parallel [(set (cc0) (reg:SI 34))
11069 (set (pc) (if_then_else
11070 (eq (cc0) (const_int 0))
11074 But this is incorrect, because it says that the jump condition
11075 depends on the condition code value _before_ this instruction, not
11076 on the new value that is set by this instruction.
11078 Peephole optimization, which takes place together with final
11079 assembly code output, can produce insns whose patterns consist of
11080 a `parallel' whose elements are the operands needed to output the
11081 resulting assembler code--often `reg', `mem' or constant
11082 expressions. This would not be well-formed RTL at any other stage
11083 in compilation, but it is ok then because no further optimization
11084 remains to be done. However, the definition of the macro
11085 `NOTICE_UPDATE_CC', if any, must deal with such insns if you
11086 define any peephole optimizations.
11088 `(cond_exec [COND EXPR])'
11089 Represents a conditionally executed expression. The EXPR is
11090 executed only if the COND is nonzero. The COND expression must
11091 not have side-effects, but the EXPR may very well have
11094 `(sequence [INSNS ...])'
11095 Represents a sequence of insns. Each of the INSNS that appears in
11096 the vector is suitable for appearing in the chain of insns, so it
11097 must be an `insn', `jump_insn', `call_insn', `code_label',
11098 `barrier' or `note'.
11100 A `sequence' RTX is never placed in an actual insn during RTL
11101 generation. It represents the sequence of insns that result from a
11102 `define_expand' _before_ those insns are passed to `emit_insn' to
11103 insert them in the chain of insns. When actually inserted, the
11104 individual sub-insns are separated out and the `sequence' is
11107 After delay-slot scheduling is completed, an insn and all the
11108 insns that reside in its delay slots are grouped together into a
11109 `sequence'. The insn requiring the delay slot is the first insn
11110 in the vector; subsequent insns are to be placed in the delay slot.
11112 `INSN_ANNULLED_BRANCH_P' is set on an insn in a delay slot to
11113 indicate that a branch insn should be used that will conditionally
11114 annul the effect of the insns in the delay slots. In such a case,
11115 `INSN_FROM_TARGET_P' indicates that the insn is from the target of
11116 the branch and should be executed only if the branch is taken;
11117 otherwise the insn should be executed only if the branch is not
11118 taken. *Note Delay Slots::.
11120 These expression codes appear in place of a side effect, as the body of
11121 an insn, though strictly speaking they do not always describe side
11125 Represents literal assembler code as described by the string S.
11127 `(unspec [OPERANDS ...] INDEX)'
11128 `(unspec_volatile [OPERANDS ...] INDEX)'
11129 Represents a machine-specific operation on OPERANDS. INDEX
11130 selects between multiple machine-specific operations.
11131 `unspec_volatile' is used for volatile operations and operations
11132 that may trap; `unspec' is used for other operations.
11134 These codes may appear inside a `pattern' of an insn, inside a
11135 `parallel', or inside an expression.
11137 `(addr_vec:M [LR0 LR1 ...])'
11138 Represents a table of jump addresses. The vector elements LR0,
11139 etc., are `label_ref' expressions. The mode M specifies how much
11140 space is given to each address; normally M would be `Pmode'.
11142 `(addr_diff_vec:M BASE [LR0 LR1 ...] MIN MAX FLAGS)'
11143 Represents a table of jump addresses expressed as offsets from
11144 BASE. The vector elements LR0, etc., are `label_ref' expressions
11145 and so is BASE. The mode M specifies how much space is given to
11146 each address-difference. MIN and MAX are set up by branch
11147 shortening and hold a label with a minimum and a maximum address,
11148 respectively. FLAGS indicates the relative position of BASE, MIN
11149 and MAX to the containing insn and of MIN and MAX to BASE. See
11150 rtl.def for details.
11152 `(prefetch:M ADDR RW LOCALITY)'
11153 Represents prefetch of memory at address ADDR. Operand RW is 1 if
11154 the prefetch is for data to be written, 0 otherwise; targets that
11155 do not support write prefetches should treat this as a normal
11156 prefetch. Operand LOCALITY specifies the amount of temporal
11157 locality; 0 if there is none or 1, 2, or 3 for increasing levels
11158 of temporal locality; targets that do not support locality hints
11159 should ignore this.
11161 This insn is used to minimize cache-miss latency by moving data
11162 into a cache before it is accessed. It should use only
11163 non-faulting data prefetch instructions.
11166 File: gccint.info, Node: Incdec, Next: Assembler, Prev: Side Effects, Up: RTL
11168 10.16 Embedded Side-Effects on Addresses
11169 ========================================
11171 Six special side-effect expression codes appear as memory addresses.
11174 Represents the side effect of decrementing X by a standard amount
11175 and represents also the value that X has after being decremented.
11176 X must be a `reg' or `mem', but most machines allow only a `reg'.
11177 M must be the machine mode for pointers on the machine in use.
11178 The amount X is decremented by is the length in bytes of the
11179 machine mode of the containing memory reference of which this
11180 expression serves as the address. Here is an example of its use:
11182 (mem:DF (pre_dec:SI (reg:SI 39)))
11184 This says to decrement pseudo register 39 by the length of a
11185 `DFmode' value and use the result to address a `DFmode' value.
11188 Similar, but specifies incrementing X instead of decrementing it.
11191 Represents the same side effect as `pre_dec' but a different
11192 value. The value represented here is the value X has before being
11196 Similar, but specifies incrementing X instead of decrementing it.
11198 `(post_modify:M X Y)'
11199 Represents the side effect of setting X to Y and represents X
11200 before X is modified. X must be a `reg' or `mem', but most
11201 machines allow only a `reg'. M must be the machine mode for
11202 pointers on the machine in use.
11204 The expression Y must be one of three forms: `(plus:M X Z)',
11205 `(minus:M X Z)', or `(plus:M X I)', where Z is an index register
11206 and I is a constant.
11208 Here is an example of its use:
11210 (mem:SF (post_modify:SI (reg:SI 42) (plus (reg:SI 42)
11213 This says to modify pseudo register 42 by adding the contents of
11214 pseudo register 48 to it, after the use of what ever 42 points to.
11216 `(pre_modify:M X EXPR)'
11217 Similar except side effects happen before the use.
11219 These embedded side effect expressions must be used with care.
11220 Instruction patterns may not use them. Until the `flow' pass of the
11221 compiler, they may occur only to represent pushes onto the stack. The
11222 `flow' pass finds cases where registers are incremented or decremented
11223 in one instruction and used as an address shortly before or after;
11224 these cases are then transformed to use pre- or post-increment or
11227 If a register used as the operand of these expressions is used in
11228 another address in an insn, the original value of the register is used.
11229 Uses of the register outside of an address are not permitted within the
11230 same insn as a use in an embedded side effect expression because such
11231 insns behave differently on different machines and hence must be treated
11232 as ambiguous and disallowed.
11234 An instruction that can be represented with an embedded side effect
11235 could also be represented using `parallel' containing an additional
11236 `set' to describe how the address register is altered. This is not
11237 done because machines that allow these operations at all typically
11238 allow them wherever a memory address is called for. Describing them as
11239 additional parallel stores would require doubling the number of entries
11240 in the machine description.
11243 File: gccint.info, Node: Assembler, Next: Insns, Prev: Incdec, Up: RTL
11245 10.17 Assembler Instructions as Expressions
11246 ===========================================
11248 The RTX code `asm_operands' represents a value produced by a
11249 user-specified assembler instruction. It is used to represent an `asm'
11250 statement with arguments. An `asm' statement with a single output
11251 operand, like this:
11253 asm ("foo %1,%2,%0" : "=a" (outputvar) : "g" (x + y), "di" (*z));
11255 is represented using a single `asm_operands' RTX which represents the
11256 value that is stored in `outputvar':
11258 (set RTX-FOR-OUTPUTVAR
11259 (asm_operands "foo %1,%2,%0" "a" 0
11260 [RTX-FOR-ADDITION-RESULT RTX-FOR-*Z]
11261 [(asm_input:M1 "g")
11262 (asm_input:M2 "di")]))
11264 Here the operands of the `asm_operands' RTX are the assembler template
11265 string, the output-operand's constraint, the index-number of the output
11266 operand among the output operands specified, a vector of input operand
11267 RTX's, and a vector of input-operand modes and constraints. The mode
11268 M1 is the mode of the sum `x+y'; M2 is that of `*z'.
11270 When an `asm' statement has multiple output values, its insn has
11271 several such `set' RTX's inside of a `parallel'. Each `set' contains a
11272 `asm_operands'; all of these share the same assembler template and
11273 vectors, but each contains the constraint for the respective output
11274 operand. They are also distinguished by the output-operand index
11275 number, which is 0, 1, ... for successive output operands.
11278 File: gccint.info, Node: Insns, Next: Calls, Prev: Assembler, Up: RTL
11283 The RTL representation of the code for a function is a doubly-linked
11284 chain of objects called "insns". Insns are expressions with special
11285 codes that are used for no other purpose. Some insns are actual
11286 instructions; others represent dispatch tables for `switch' statements;
11287 others represent labels to jump to or various sorts of declarative
11290 In addition to its own specific data, each insn must have a unique
11291 id-number that distinguishes it from all other insns in the current
11292 function (after delayed branch scheduling, copies of an insn with the
11293 same id-number may be present in multiple places in a function, but
11294 these copies will always be identical and will only appear inside a
11295 `sequence'), and chain pointers to the preceding and following insns.
11296 These three fields occupy the same position in every insn, independent
11297 of the expression code of the insn. They could be accessed with `XEXP'
11298 and `XINT', but instead three special macros are always used:
11301 Accesses the unique id of insn I.
11304 Accesses the chain pointer to the insn preceding I. If I is the
11305 first insn, this is a null pointer.
11308 Accesses the chain pointer to the insn following I. If I is the
11309 last insn, this is a null pointer.
11311 The first insn in the chain is obtained by calling `get_insns'; the
11312 last insn is the result of calling `get_last_insn'. Within the chain
11313 delimited by these insns, the `NEXT_INSN' and `PREV_INSN' pointers must
11314 always correspond: if INSN is not the first insn,
11316 NEXT_INSN (PREV_INSN (INSN)) == INSN
11318 is always true and if INSN is not the last insn,
11320 PREV_INSN (NEXT_INSN (INSN)) == INSN
11324 After delay slot scheduling, some of the insns in the chain might be
11325 `sequence' expressions, which contain a vector of insns. The value of
11326 `NEXT_INSN' in all but the last of these insns is the next insn in the
11327 vector; the value of `NEXT_INSN' of the last insn in the vector is the
11328 same as the value of `NEXT_INSN' for the `sequence' in which it is
11329 contained. Similar rules apply for `PREV_INSN'.
11331 This means that the above invariants are not necessarily true for insns
11332 inside `sequence' expressions. Specifically, if INSN is the first insn
11333 in a `sequence', `NEXT_INSN (PREV_INSN (INSN))' is the insn containing
11334 the `sequence' expression, as is the value of `PREV_INSN (NEXT_INSN
11335 (INSN))' if INSN is the last insn in the `sequence' expression. You
11336 can use these expressions to find the containing `sequence' expression.
11338 Every insn has one of the following six expression codes:
11341 The expression code `insn' is used for instructions that do not
11342 jump and do not do function calls. `sequence' expressions are
11343 always contained in insns with code `insn' even if one of those
11344 insns should jump or do function calls.
11346 Insns with code `insn' have four additional fields beyond the three
11347 mandatory ones listed above. These four are described in a table
11351 The expression code `jump_insn' is used for instructions that may
11352 jump (or, more generally, may contain `label_ref' expressions to
11353 which `pc' can be set in that instruction). If there is an
11354 instruction to return from the current function, it is recorded as
11357 `jump_insn' insns have the same extra fields as `insn' insns,
11358 accessed in the same way and in addition contain a field
11359 `JUMP_LABEL' which is defined once jump optimization has completed.
11361 For simple conditional and unconditional jumps, this field contains
11362 the `code_label' to which this insn will (possibly conditionally)
11363 branch. In a more complex jump, `JUMP_LABEL' records one of the
11364 labels that the insn refers to; other jump target labels are
11365 recorded as `REG_LABEL_TARGET' notes. The exception is `addr_vec'
11366 and `addr_diff_vec', where `JUMP_LABEL' is `NULL_RTX' and the only
11367 way to find the labels is to scan the entire body of the insn.
11369 Return insns count as jumps, but since they do not refer to any
11370 labels, their `JUMP_LABEL' is `NULL_RTX'.
11373 The expression code `call_insn' is used for instructions that may
11374 do function calls. It is important to distinguish these
11375 instructions because they imply that certain registers and memory
11376 locations may be altered unpredictably.
11378 `call_insn' insns have the same extra fields as `insn' insns,
11379 accessed in the same way and in addition contain a field
11380 `CALL_INSN_FUNCTION_USAGE', which contains a list (chain of
11381 `expr_list' expressions) containing `use' and `clobber'
11382 expressions that denote hard registers and `MEM's used or
11383 clobbered by the called function.
11385 A `MEM' generally points to a stack slots in which arguments passed
11386 to the libcall by reference (*note TARGET_PASS_BY_REFERENCE:
11387 Register Arguments.) are stored. If the argument is caller-copied
11388 (*note TARGET_CALLEE_COPIES: Register Arguments.), the stack slot
11389 will be mentioned in `CLOBBER' and `USE' entries; if it's
11390 callee-copied, only a `USE' will appear, and the `MEM' may point
11391 to addresses that are not stack slots.
11393 `CLOBBER'ed registers in this list augment registers specified in
11394 `CALL_USED_REGISTERS' (*note Register Basics::).
11397 A `code_label' insn represents a label that a jump insn can jump
11398 to. It contains two special fields of data in addition to the
11399 three standard ones. `CODE_LABEL_NUMBER' is used to hold the
11400 "label number", a number that identifies this label uniquely among
11401 all the labels in the compilation (not just in the current
11402 function). Ultimately, the label is represented in the assembler
11403 output as an assembler label, usually of the form `LN' where N is
11406 When a `code_label' appears in an RTL expression, it normally
11407 appears within a `label_ref' which represents the address of the
11408 label, as a number.
11410 Besides as a `code_label', a label can also be represented as a
11411 `note' of type `NOTE_INSN_DELETED_LABEL'.
11413 The field `LABEL_NUSES' is only defined once the jump optimization
11414 phase is completed. It contains the number of times this label is
11415 referenced in the current function.
11417 The field `LABEL_KIND' differentiates four different types of
11418 labels: `LABEL_NORMAL', `LABEL_STATIC_ENTRY',
11419 `LABEL_GLOBAL_ENTRY', and `LABEL_WEAK_ENTRY'. The only labels
11420 that do not have type `LABEL_NORMAL' are "alternate entry points"
11421 to the current function. These may be static (visible only in the
11422 containing translation unit), global (exposed to all translation
11423 units), or weak (global, but can be overridden by another symbol
11424 with the same name).
11426 Much of the compiler treats all four kinds of label identically.
11427 Some of it needs to know whether or not a label is an alternate
11428 entry point; for this purpose, the macro `LABEL_ALT_ENTRY_P' is
11429 provided. It is equivalent to testing whether `LABEL_KIND (label)
11430 == LABEL_NORMAL'. The only place that cares about the distinction
11431 between static, global, and weak alternate entry points, besides
11432 the front-end code that creates them, is the function
11433 `output_alternate_entry_point', in `final.c'.
11435 To set the kind of a label, use the `SET_LABEL_KIND' macro.
11438 Barriers are placed in the instruction stream when control cannot
11439 flow past them. They are placed after unconditional jump
11440 instructions to indicate that the jumps are unconditional and
11441 after calls to `volatile' functions, which do not return (e.g.,
11442 `exit'). They contain no information beyond the three standard
11446 `note' insns are used to represent additional debugging and
11447 declarative information. They contain two nonstandard fields, an
11448 integer which is accessed with the macro `NOTE_LINE_NUMBER' and a
11449 string accessed with `NOTE_SOURCE_FILE'.
11451 If `NOTE_LINE_NUMBER' is positive, the note represents the
11452 position of a source line and `NOTE_SOURCE_FILE' is the source
11453 file name that the line came from. These notes control generation
11454 of line number data in the assembler output.
11456 Otherwise, `NOTE_LINE_NUMBER' is not really a line number but a
11457 code with one of the following values (and `NOTE_SOURCE_FILE' must
11458 contain a null pointer):
11460 `NOTE_INSN_DELETED'
11461 Such a note is completely ignorable. Some passes of the
11462 compiler delete insns by altering them into notes of this
11465 `NOTE_INSN_DELETED_LABEL'
11466 This marks what used to be a `code_label', but was not used
11467 for other purposes than taking its address and was
11468 transformed to mark that no code jumps to it.
11470 `NOTE_INSN_BLOCK_BEG'
11471 `NOTE_INSN_BLOCK_END'
11472 These types of notes indicate the position of the beginning
11473 and end of a level of scoping of variable names. They
11474 control the output of debugging information.
11476 `NOTE_INSN_EH_REGION_BEG'
11477 `NOTE_INSN_EH_REGION_END'
11478 These types of notes indicate the position of the beginning
11479 and end of a level of scoping for exception handling.
11480 `NOTE_BLOCK_NUMBER' identifies which `CODE_LABEL' or `note'
11481 of type `NOTE_INSN_DELETED_LABEL' is associated with the
11484 `NOTE_INSN_LOOP_BEG'
11485 `NOTE_INSN_LOOP_END'
11486 These types of notes indicate the position of the beginning
11487 and end of a `while' or `for' loop. They enable the loop
11488 optimizer to find loops quickly.
11490 `NOTE_INSN_LOOP_CONT'
11491 Appears at the place in a loop that `continue' statements
11494 `NOTE_INSN_LOOP_VTOP'
11495 This note indicates the place in a loop where the exit test
11496 begins for those loops in which the exit test has been
11497 duplicated. This position becomes another virtual start of
11498 the loop when considering loop invariants.
11500 `NOTE_INSN_FUNCTION_BEG'
11501 Appears at the start of the function body, after the function
11505 These codes are printed symbolically when they appear in debugging
11508 The machine mode of an insn is normally `VOIDmode', but some phases
11509 use the mode for various purposes.
11511 The common subexpression elimination pass sets the mode of an insn to
11512 `QImode' when it is the first insn in a block that has already been
11515 The second Haifa scheduling pass, for targets that can multiple issue,
11516 sets the mode of an insn to `TImode' when it is believed that the
11517 instruction begins an issue group. That is, when the instruction
11518 cannot issue simultaneously with the previous. This may be relied on
11519 by later passes, in particular machine-dependent reorg.
11521 Here is a table of the extra fields of `insn', `jump_insn' and
11525 An expression for the side effect performed by this insn. This
11526 must be one of the following codes: `set', `call', `use',
11527 `clobber', `return', `asm_input', `asm_output', `addr_vec',
11528 `addr_diff_vec', `trap_if', `unspec', `unspec_volatile',
11529 `parallel', `cond_exec', or `sequence'. If it is a `parallel',
11530 each element of the `parallel' must be one these codes, except that
11531 `parallel' expressions cannot be nested and `addr_vec' and
11532 `addr_diff_vec' are not permitted inside a `parallel' expression.
11535 An integer that says which pattern in the machine description
11536 matches this insn, or -1 if the matching has not yet been
11539 Such matching is never attempted and this field remains -1 on an
11540 insn whose pattern consists of a single `use', `clobber',
11541 `asm_input', `addr_vec' or `addr_diff_vec' expression.
11543 Matching is also never attempted on insns that result from an `asm'
11544 statement. These contain at least one `asm_operands' expression.
11545 The function `asm_noperands' returns a non-negative value for such
11548 In the debugging output, this field is printed as a number
11549 followed by a symbolic representation that locates the pattern in
11550 the `md' file as some small positive or negative offset from a
11554 A list (chain of `insn_list' expressions) giving information about
11555 dependencies between instructions within a basic block. Neither a
11556 jump nor a label may come between the related insns. These are
11557 only used by the schedulers and by combine. This is a deprecated
11558 data structure. Def-use and use-def chains are now preferred.
11561 A list (chain of `expr_list' and `insn_list' expressions) giving
11562 miscellaneous information about the insn. It is often information
11563 pertaining to the registers used in this insn.
11565 The `LOG_LINKS' field of an insn is a chain of `insn_list'
11566 expressions. Each of these has two operands: the first is an insn, and
11567 the second is another `insn_list' expression (the next one in the
11568 chain). The last `insn_list' in the chain has a null pointer as second
11569 operand. The significant thing about the chain is which insns appear
11570 in it (as first operands of `insn_list' expressions). Their order is
11573 This list is originally set up by the flow analysis pass; it is a null
11574 pointer until then. Flow only adds links for those data dependencies
11575 which can be used for instruction combination. For each insn, the flow
11576 analysis pass adds a link to insns which store into registers values
11577 that are used for the first time in this insn.
11579 The `REG_NOTES' field of an insn is a chain similar to the `LOG_LINKS'
11580 field but it includes `expr_list' expressions in addition to
11581 `insn_list' expressions. There are several kinds of register notes,
11582 which are distinguished by the machine mode, which in a register note
11583 is really understood as being an `enum reg_note'. The first operand OP
11584 of the note is data whose meaning depends on the kind of note.
11586 The macro `REG_NOTE_KIND (X)' returns the kind of register note. Its
11587 counterpart, the macro `PUT_REG_NOTE_KIND (X, NEWKIND)' sets the
11588 register note type of X to be NEWKIND.
11590 Register notes are of three classes: They may say something about an
11591 input to an insn, they may say something about an output of an insn, or
11592 they may create a linkage between two insns. There are also a set of
11593 values that are only used in `LOG_LINKS'.
11595 These register notes annotate inputs to an insn:
11598 The value in OP dies in this insn; that is to say, altering the
11599 value immediately after this insn would not affect the future
11600 behavior of the program.
11602 It does not follow that the register OP has no useful value after
11603 this insn since OP is not necessarily modified by this insn.
11604 Rather, no subsequent instruction uses the contents of OP.
11607 The register OP being set by this insn will not be used in a
11608 subsequent insn. This differs from a `REG_DEAD' note, which
11609 indicates that the value in an input will not be used subsequently.
11610 These two notes are independent; both may be present for the same
11614 The register OP is incremented (or decremented; at this level
11615 there is no distinction) by an embedded side effect inside this
11616 insn. This means it appears in a `post_inc', `pre_inc',
11617 `post_dec' or `pre_dec' expression.
11620 The register OP is known to have a nonnegative value when this
11621 insn is reached. This is used so that decrement and branch until
11622 zero instructions, such as the m68k dbra, can be matched.
11624 The `REG_NONNEG' note is added to insns only if the machine
11625 description has a `decrement_and_branch_until_zero' pattern.
11627 `REG_LABEL_OPERAND'
11628 This insn uses OP, a `code_label' or a `note' of type
11629 `NOTE_INSN_DELETED_LABEL', but is not a `jump_insn', or it is a
11630 `jump_insn' that refers to the operand as an ordinary operand.
11631 The label may still eventually be a jump target, but if so in an
11632 indirect jump in a subsequent insn. The presence of this note
11633 allows jump optimization to be aware that OP is, in fact, being
11634 used, and flow optimization to build an accurate flow graph.
11637 This insn is a `jump_insn' but not a `addr_vec' or
11638 `addr_diff_vec'. It uses OP, a `code_label' as a direct or
11639 indirect jump target. Its purpose is similar to that of
11640 `REG_LABEL_OPERAND'. This note is only present if the insn has
11641 multiple targets; the last label in the insn (in the highest
11642 numbered insn-field) goes into the `JUMP_LABEL' field and does not
11643 have a `REG_LABEL_TARGET' note. *Note JUMP_LABEL: Insns.
11645 `REG_CROSSING_JUMP'
11646 This insn is an branching instruction (either an unconditional
11647 jump or an indirect jump) which crosses between hot and cold
11648 sections, which could potentially be very far apart in the
11649 executable. The presence of this note indicates to other
11650 optimizations that this branching instruction should not be
11651 "collapsed" into a simpler branching construct. It is used when
11652 the optimization to partition basic blocks into hot and cold
11653 sections is turned on.
11656 Appears attached to each `CALL_INSN' to `setjmp' or a related
11659 The following notes describe attributes of outputs of an insn:
11663 This note is only valid on an insn that sets only one register and
11664 indicates that that register will be equal to OP at run time; the
11665 scope of this equivalence differs between the two types of notes.
11666 The value which the insn explicitly copies into the register may
11667 look different from OP, but they will be equal at run time. If the
11668 output of the single `set' is a `strict_low_part' expression, the
11669 note refers to the register that is contained in `SUBREG_REG' of
11670 the `subreg' expression.
11672 For `REG_EQUIV', the register is equivalent to OP throughout the
11673 entire function, and could validly be replaced in all its
11674 occurrences by OP. ("Validly" here refers to the data flow of the
11675 program; simple replacement may make some insns invalid.) For
11676 example, when a constant is loaded into a register that is never
11677 assigned any other value, this kind of note is used.
11679 When a parameter is copied into a pseudo-register at entry to a
11680 function, a note of this kind records that the register is
11681 equivalent to the stack slot where the parameter was passed.
11682 Although in this case the register may be set by other insns, it
11683 is still valid to replace the register by the stack slot
11684 throughout the function.
11686 A `REG_EQUIV' note is also used on an instruction which copies a
11687 register parameter into a pseudo-register at entry to a function,
11688 if there is a stack slot where that parameter could be stored.
11689 Although other insns may set the pseudo-register, it is valid for
11690 the compiler to replace the pseudo-register by stack slot
11691 throughout the function, provided the compiler ensures that the
11692 stack slot is properly initialized by making the replacement in
11693 the initial copy instruction as well. This is used on machines
11694 for which the calling convention allocates stack space for
11695 register parameters. See `REG_PARM_STACK_SPACE' in *Note Stack
11698 In the case of `REG_EQUAL', the register that is set by this insn
11699 will be equal to OP at run time at the end of this insn but not
11700 necessarily elsewhere in the function. In this case, OP is
11701 typically an arithmetic expression. For example, when a sequence
11702 of insns such as a library call is used to perform an arithmetic
11703 operation, this kind of note is attached to the insn that produces
11704 or copies the final value.
11706 These two notes are used in different ways by the compiler passes.
11707 `REG_EQUAL' is used by passes prior to register allocation (such as
11708 common subexpression elimination and loop optimization) to tell
11709 them how to think of that value. `REG_EQUIV' notes are used by
11710 register allocation to indicate that there is an available
11711 substitute expression (either a constant or a `mem' expression for
11712 the location of a parameter on the stack) that may be used in
11713 place of a register if insufficient registers are available.
11715 Except for stack homes for parameters, which are indicated by a
11716 `REG_EQUIV' note and are not useful to the early optimization
11717 passes and pseudo registers that are equivalent to a memory
11718 location throughout their entire life, which is not detected until
11719 later in the compilation, all equivalences are initially indicated
11720 by an attached `REG_EQUAL' note. In the early stages of register
11721 allocation, a `REG_EQUAL' note is changed into a `REG_EQUIV' note
11722 if OP is a constant and the insn represents the only set of its
11723 destination register.
11725 Thus, compiler passes prior to register allocation need only check
11726 for `REG_EQUAL' notes and passes subsequent to register allocation
11727 need only check for `REG_EQUIV' notes.
11729 These notes describe linkages between insns. They occur in pairs: one
11730 insn has one of a pair of notes that points to a second insn, which has
11731 the inverse note pointing back to the first insn.
11735 On machines that use `cc0', the insns which set and use `cc0' set
11736 and use `cc0' are adjacent. However, when branch delay slot
11737 filling is done, this may no longer be true. In this case a
11738 `REG_CC_USER' note will be placed on the insn setting `cc0' to
11739 point to the insn using `cc0' and a `REG_CC_SETTER' note will be
11740 placed on the insn using `cc0' to point to the insn setting `cc0'.
11742 These values are only used in the `LOG_LINKS' field, and indicate the
11743 type of dependency that each link represents. Links which indicate a
11744 data dependence (a read after write dependence) do not use any code,
11745 they simply have mode `VOIDmode', and are printed without any
11749 This indicates a true dependence (a read after write dependence).
11752 This indicates an output dependence (a write after write
11756 This indicates an anti dependence (a write after read dependence).
11759 These notes describe information gathered from gcov profile data. They
11760 are stored in the `REG_NOTES' field of an insn as an `expr_list'.
11763 This is used to specify the ratio of branches to non-branches of a
11764 branch insn according to the profile data. The value is stored as
11765 a value between 0 and REG_BR_PROB_BASE; larger values indicate a
11766 higher probability that the branch will be taken.
11769 These notes are found in JUMP insns after delayed branch scheduling
11770 has taken place. They indicate both the direction and the
11771 likelihood of the JUMP. The format is a bitmask of ATTR_FLAG_*
11774 `REG_FRAME_RELATED_EXPR'
11775 This is used on an RTX_FRAME_RELATED_P insn wherein the attached
11776 expression is used in place of the actual insn pattern. This is
11777 done in cases where the pattern is either complex or misleading.
11779 For convenience, the machine mode in an `insn_list' or `expr_list' is
11780 printed using these symbolic codes in debugging dumps.
11782 The only difference between the expression codes `insn_list' and
11783 `expr_list' is that the first operand of an `insn_list' is assumed to
11784 be an insn and is printed in debugging dumps as the insn's unique id;
11785 the first operand of an `expr_list' is printed in the ordinary way as
11789 File: gccint.info, Node: Calls, Next: Sharing, Prev: Insns, Up: RTL
11791 10.19 RTL Representation of Function-Call Insns
11792 ===============================================
11794 Insns that call subroutines have the RTL expression code `call_insn'.
11795 These insns must satisfy special rules, and their bodies must use a
11796 special RTL expression code, `call'.
11798 A `call' expression has two operands, as follows:
11800 (call (mem:FM ADDR) NBYTES)
11802 Here NBYTES is an operand that represents the number of bytes of
11803 argument data being passed to the subroutine, FM is a machine mode
11804 (which must equal as the definition of the `FUNCTION_MODE' macro in the
11805 machine description) and ADDR represents the address of the subroutine.
11807 For a subroutine that returns no value, the `call' expression as shown
11808 above is the entire body of the insn, except that the insn might also
11809 contain `use' or `clobber' expressions.
11811 For a subroutine that returns a value whose mode is not `BLKmode', the
11812 value is returned in a hard register. If this register's number is R,
11813 then the body of the call insn looks like this:
11816 (call (mem:FM ADDR) NBYTES))
11818 This RTL expression makes it clear (to the optimizer passes) that the
11819 appropriate register receives a useful value in this insn.
11821 When a subroutine returns a `BLKmode' value, it is handled by passing
11822 to the subroutine the address of a place to store the value. So the
11823 call insn itself does not "return" any value, and it has the same RTL
11824 form as a call that returns nothing.
11826 On some machines, the call instruction itself clobbers some register,
11827 for example to contain the return address. `call_insn' insns on these
11828 machines should have a body which is a `parallel' that contains both
11829 the `call' expression and `clobber' expressions that indicate which
11830 registers are destroyed. Similarly, if the call instruction requires
11831 some register other than the stack pointer that is not explicitly
11832 mentioned in its RTL, a `use' subexpression should mention that
11835 Functions that are called are assumed to modify all registers listed in
11836 the configuration macro `CALL_USED_REGISTERS' (*note Register Basics::)
11837 and, with the exception of `const' functions and library calls, to
11838 modify all of memory.
11840 Insns containing just `use' expressions directly precede the
11841 `call_insn' insn to indicate which registers contain inputs to the
11842 function. Similarly, if registers other than those in
11843 `CALL_USED_REGISTERS' are clobbered by the called function, insns
11844 containing a single `clobber' follow immediately after the call to
11845 indicate which registers.
11848 File: gccint.info, Node: Sharing, Next: Reading RTL, Prev: Calls, Up: RTL
11850 10.20 Structure Sharing Assumptions
11851 ===================================
11853 The compiler assumes that certain kinds of RTL expressions are unique;
11854 there do not exist two distinct objects representing the same value.
11855 In other cases, it makes an opposite assumption: that no RTL expression
11856 object of a certain kind appears in more than one place in the
11857 containing structure.
11859 These assumptions refer to a single function; except for the RTL
11860 objects that describe global variables and external functions, and a
11861 few standard objects such as small integer constants, no RTL objects
11862 are common to two functions.
11864 * Each pseudo-register has only a single `reg' object to represent
11865 it, and therefore only a single machine mode.
11867 * For any symbolic label, there is only one `symbol_ref' object
11870 * All `const_int' expressions with equal values are shared.
11872 * There is only one `pc' expression.
11874 * There is only one `cc0' expression.
11876 * There is only one `const_double' expression with value 0 for each
11877 floating point mode. Likewise for values 1 and 2.
11879 * There is only one `const_vector' expression with value 0 for each
11880 vector mode, be it an integer or a double constant vector.
11882 * No `label_ref' or `scratch' appears in more than one place in the
11883 RTL structure; in other words, it is safe to do a tree-walk of all
11884 the insns in the function and assume that each time a `label_ref'
11885 or `scratch' is seen it is distinct from all others that are seen.
11887 * Only one `mem' object is normally created for each static variable
11888 or stack slot, so these objects are frequently shared in all the
11889 places they appear. However, separate but equal objects for these
11890 variables are occasionally made.
11892 * When a single `asm' statement has multiple output operands, a
11893 distinct `asm_operands' expression is made for each output operand.
11894 However, these all share the vector which contains the sequence of
11895 input operands. This sharing is used later on to test whether two
11896 `asm_operands' expressions come from the same statement, so all
11897 optimizations must carefully preserve the sharing if they copy the
11900 * No RTL object appears in more than one place in the RTL structure
11901 except as described above. Many passes of the compiler rely on
11902 this by assuming that they can modify RTL objects in place without
11903 unwanted side-effects on other insns.
11905 * During initial RTL generation, shared structure is freely
11906 introduced. After all the RTL for a function has been generated,
11907 all shared structure is copied by `unshare_all_rtl' in
11908 `emit-rtl.c', after which the above rules are guaranteed to be
11911 * During the combiner pass, shared structure within an insn can exist
11912 temporarily. However, the shared structure is copied before the
11913 combiner is finished with the insn. This is done by calling
11914 `copy_rtx_if_shared', which is a subroutine of `unshare_all_rtl'.
11917 File: gccint.info, Node: Reading RTL, Prev: Sharing, Up: RTL
11922 To read an RTL object from a file, call `read_rtx'. It takes one
11923 argument, a stdio stream, and returns a single RTL object. This routine
11924 is defined in `read-rtl.c'. It is not available in the compiler
11925 itself, only the various programs that generate the compiler back end
11926 from the machine description.
11928 People frequently have the idea of using RTL stored as text in a file
11929 as an interface between a language front end and the bulk of GCC. This
11930 idea is not feasible.
11932 GCC was designed to use RTL internally only. Correct RTL for a given
11933 program is very dependent on the particular target machine. And the RTL
11934 does not contain all the information about the program.
11936 The proper way to interface GCC to a new language front end is with
11937 the "tree" data structure, described in the files `tree.h' and
11938 `tree.def'. The documentation for this structure (*note Trees::) is
11942 File: gccint.info, Node: GENERIC, Next: GIMPLE, Prev: RTL, Up: Top
11947 The purpose of GENERIC is simply to provide a language-independent way
11948 of representing an entire function in trees. To this end, it was
11949 necessary to add a few new tree codes to the back end, but most
11950 everything was already there. If you can express it with the codes in
11951 `gcc/tree.def', it's GENERIC.
11953 Early on, there was a great deal of debate about how to think about
11954 statements in a tree IL. In GENERIC, a statement is defined as any
11955 expression whose value, if any, is ignored. A statement will always
11956 have `TREE_SIDE_EFFECTS' set (or it will be discarded), but a
11957 non-statement expression may also have side effects. A `CALL_EXPR',
11960 It would be possible for some local optimizations to work on the
11961 GENERIC form of a function; indeed, the adapted tree inliner works fine
11962 on GENERIC, but the current compiler performs inlining after lowering
11963 to GIMPLE (a restricted form described in the next section). Indeed,
11964 currently the frontends perform this lowering before handing off to
11965 `tree_rest_of_compilation', but this seems inelegant.
11967 If necessary, a front end can use some language-dependent tree codes
11968 in its GENERIC representation, so long as it provides a hook for
11969 converting them to GIMPLE and doesn't expect them to work with any
11970 (hypothetical) optimizers that run before the conversion to GIMPLE. The
11971 intermediate representation used while parsing C and C++ looks very
11972 little like GENERIC, but the C and C++ gimplifier hooks are perfectly
11973 happy to take it as input and spit out GIMPLE.
11980 File: gccint.info, Node: Statements, Up: GENERIC
11985 Most statements in GIMPLE are assignment statements, represented by
11986 `GIMPLE_ASSIGN'. No other C expressions can appear at statement level;
11987 a reference to a volatile object is converted into a `GIMPLE_ASSIGN'.
11989 There are also several varieties of complex statements.
11994 * Statement Sequences::
11995 * Empty Statements::
12000 File: gccint.info, Node: Blocks, Next: Statement Sequences, Up: Statements
12005 Block scopes and the variables they declare in GENERIC are expressed
12006 using the `BIND_EXPR' code, which in previous versions of GCC was
12007 primarily used for the C statement-expression extension.
12009 Variables in a block are collected into `BIND_EXPR_VARS' in
12010 declaration order. Any runtime initialization is moved out of
12011 `DECL_INITIAL' and into a statement in the controlled block. When
12012 gimplifying from C or C++, this initialization replaces the `DECL_STMT'.
12014 Variable-length arrays (VLAs) complicate this process, as their size
12015 often refers to variables initialized earlier in the block. To handle
12016 this, we currently split the block at that point, and move the VLA into
12017 a new, inner `BIND_EXPR'. This strategy may change in the future.
12019 A C++ program will usually contain more `BIND_EXPR's than there are
12020 syntactic blocks in the source code, since several C++ constructs have
12021 implicit scopes associated with them. On the other hand, although the
12022 C++ front end uses pseudo-scopes to handle cleanups for objects with
12023 destructors, these don't translate into the GIMPLE form; multiple
12024 declarations at the same level use the same `BIND_EXPR'.
12027 File: gccint.info, Node: Statement Sequences, Next: Empty Statements, Prev: Blocks, Up: Statements
12029 11.1.2 Statement Sequences
12030 --------------------------
12032 Multiple statements at the same nesting level are collected into a
12033 `STATEMENT_LIST'. Statement lists are modified and traversed using the
12034 interface in `tree-iterator.h'.
12037 File: gccint.info, Node: Empty Statements, Next: Jumps, Prev: Statement Sequences, Up: Statements
12039 11.1.3 Empty Statements
12040 -----------------------
12042 Whenever possible, statements with no effect are discarded. But if
12043 they are nested within another construct which cannot be discarded for
12044 some reason, they are instead replaced with an empty statement,
12045 generated by `build_empty_stmt'. Initially, all empty statements were
12046 shared, after the pattern of the Java front end, but this caused a lot
12047 of trouble in practice.
12049 An empty statement is represented as `(void)0'.
12052 File: gccint.info, Node: Jumps, Next: Cleanups, Prev: Empty Statements, Up: Statements
12057 Other jumps are expressed by either `GOTO_EXPR' or `RETURN_EXPR'.
12059 The operand of a `GOTO_EXPR' must be either a label or a variable
12060 containing the address to jump to.
12062 The operand of a `RETURN_EXPR' is either `NULL_TREE', `RESULT_DECL',
12063 or a `MODIFY_EXPR' which sets the return value. It would be nice to
12064 move the `MODIFY_EXPR' into a separate statement, but the special
12065 return semantics in `expand_return' make that difficult. It may still
12066 happen in the future, perhaps by moving most of that logic into
12067 `expand_assignment'.
12070 File: gccint.info, Node: Cleanups, Prev: Jumps, Up: Statements
12075 Destructors for local C++ objects and similar dynamic cleanups are
12076 represented in GIMPLE by a `TRY_FINALLY_EXPR'. `TRY_FINALLY_EXPR' has
12077 two operands, both of which are a sequence of statements to execute.
12078 The first sequence is executed. When it completes the second sequence
12081 The first sequence may complete in the following ways:
12083 1. Execute the last statement in the sequence and fall off the end.
12085 2. Execute a goto statement (`GOTO_EXPR') to an ordinary label
12086 outside the sequence.
12088 3. Execute a return statement (`RETURN_EXPR').
12090 4. Throw an exception. This is currently not explicitly represented
12094 The second sequence is not executed if the first sequence completes by
12095 calling `setjmp' or `exit' or any other function that does not return.
12096 The second sequence is also not executed if the first sequence
12097 completes via a non-local goto or a computed goto (in general the
12098 compiler does not know whether such a goto statement exits the first
12099 sequence or not, so we assume that it doesn't).
12101 After the second sequence is executed, if it completes normally by
12102 falling off the end, execution continues wherever the first sequence
12103 would have continued, by falling off the end, or doing a goto, etc.
12105 `TRY_FINALLY_EXPR' complicates the flow graph, since the cleanup needs
12106 to appear on every edge out of the controlled block; this reduces the
12107 freedom to move code across these edges. Therefore, the EH lowering
12108 pass which runs before most of the optimization passes eliminates these
12109 expressions by explicitly adding the cleanup to each edge. Rethrowing
12110 the exception is represented using `RESX_EXPR'.
12113 File: gccint.info, Node: GIMPLE, Next: Tree SSA, Prev: GENERIC, Up: Top
12118 GIMPLE is a three-address representation derived from GENERIC by
12119 breaking down GENERIC expressions into tuples of no more than 3
12120 operands (with some exceptions like function calls). GIMPLE was
12121 heavily influenced by the SIMPLE IL used by the McCAT compiler project
12122 at McGill University, though we have made some different choices. For
12123 one thing, SIMPLE doesn't support `goto'.
12125 Temporaries are introduced to hold intermediate values needed to
12126 compute complex expressions. Additionally, all the control structures
12127 used in GENERIC are lowered into conditional jumps, lexical scopes are
12128 removed and exception regions are converted into an on the side
12129 exception region tree.
12131 The compiler pass which converts GENERIC into GIMPLE is referred to as
12132 the `gimplifier'. The gimplifier works recursively, generating GIMPLE
12133 tuples out of the original GENERIC expressions.
12135 One of the early implementation strategies used for the GIMPLE
12136 representation was to use the same internal data structures used by
12137 front ends to represent parse trees. This simplified implementation
12138 because we could leverage existing functionality and interfaces.
12139 However, GIMPLE is a much more restrictive representation than abstract
12140 syntax trees (AST), therefore it does not require the full structural
12141 complexity provided by the main tree data structure.
12143 The GENERIC representation of a function is stored in the
12144 `DECL_SAVED_TREE' field of the associated `FUNCTION_DECL' tree node.
12145 It is converted to GIMPLE by a call to `gimplify_function_tree'.
12147 If a front end wants to include language-specific tree codes in the
12148 tree representation which it provides to the back end, it must provide a
12149 definition of `LANG_HOOKS_GIMPLIFY_EXPR' which knows how to convert the
12150 front end trees to GIMPLE. Usually such a hook will involve much of
12151 the same code for expanding front end trees to RTL. This function can
12152 return fully lowered GIMPLE, or it can return GENERIC trees and let the
12153 main gimplifier lower them the rest of the way; this is often simpler.
12154 GIMPLE that is not fully lowered is known as "High GIMPLE" and consists
12155 of the IL before the pass `pass_lower_cf'. High GIMPLE contains some
12156 container statements like lexical scopes (represented by `GIMPLE_BIND')
12157 and nested expressions (e.g., `GIMPLE_TRY'), while "Low GIMPLE" exposes
12158 all of the implicit jumps for control and exception expressions
12159 directly in the IL and EH region trees.
12161 The C and C++ front ends currently convert directly from front end
12162 trees to GIMPLE, and hand that off to the back end rather than first
12163 converting to GENERIC. Their gimplifier hooks know about all the
12164 `_STMT' nodes and how to convert them to GENERIC forms. There was some
12165 work done on a genericization pass which would run first, but the
12166 existence of `STMT_EXPR' meant that in order to convert all of the C
12167 statements into GENERIC equivalents would involve walking the entire
12168 tree anyway, so it was simpler to lower all the way. This might change
12169 in the future if someone writes an optimization pass which would work
12170 better with higher-level trees, but currently the optimizers all expect
12173 You can request to dump a C-like representation of the GIMPLE form
12174 with the flag `-fdump-tree-gimple'.
12178 * Tuple representation::
12179 * GIMPLE instruction set::
12180 * GIMPLE Exception Handling::
12183 * Manipulating GIMPLE statements::
12184 * Tuple specific accessors::
12185 * GIMPLE sequences::
12186 * Sequence iterators::
12187 * Adding a new GIMPLE statement code::
12188 * Statement and operand traversals::
12191 File: gccint.info, Node: Tuple representation, Next: GIMPLE instruction set, Up: GIMPLE
12193 12.1 Tuple representation
12194 =========================
12196 GIMPLE instructions are tuples of variable size divided in two groups:
12197 a header describing the instruction and its locations, and a variable
12198 length body with all the operands. Tuples are organized into a
12199 hierarchy with 3 main classes of tuples.
12201 12.1.1 `gimple_statement_base' (gsbase)
12202 ---------------------------------------
12204 This is the root of the hierarchy, it holds basic information needed by
12205 most GIMPLE statements. There are some fields that may not be relevant
12206 to every GIMPLE statement, but those were moved into the base structure
12207 to take advantage of holes left by other fields (thus making the
12208 structure more compact). The structure takes 4 words (32 bytes) on 64
12216 `nontemporal_move' 1
12219 `has_volatile_ops' 1
12220 `references_memory_p' 1
12226 Total size 32 bytes
12228 * `code' Main identifier for a GIMPLE instruction.
12230 * `subcode' Used to distinguish different variants of the same basic
12231 instruction or provide flags applicable to a given code. The
12232 `subcode' flags field has different uses depending on the code of
12233 the instruction, but mostly it distinguishes instructions of the
12234 same family. The most prominent use of this field is in
12235 assignments, where subcode indicates the operation done on the RHS
12236 of the assignment. For example, a = b + c is encoded as
12237 `GIMPLE_ASSIGN <PLUS_EXPR, a, b, c>'.
12239 * `no_warning' Bitflag to indicate whether a warning has already
12240 been issued on this statement.
12242 * `visited' General purpose "visited" marker. Set and cleared by
12243 each pass when needed.
12245 * `nontemporal_move' Bitflag used in assignments that represent
12246 non-temporal moves. Although this bitflag is only used in
12247 assignments, it was moved into the base to take advantage of the
12248 bit holes left by the previous fields.
12250 * `plf' Pass Local Flags. This 2-bit mask can be used as general
12251 purpose markers by any pass. Passes are responsible for clearing
12252 and setting these two flags accordingly.
12254 * `modified' Bitflag to indicate whether the statement has been
12255 modified. Used mainly by the operand scanner to determine when to
12256 re-scan a statement for operands.
12258 * `has_volatile_ops' Bitflag to indicate whether this statement
12259 contains operands that have been marked volatile.
12261 * `references_memory_p' Bitflag to indicate whether this statement
12262 contains memory references (i.e., its operands are either global
12263 variables, or pointer dereferences or anything that must reside in
12266 * `uid' This is an unsigned integer used by passes that want to
12267 assign IDs to every statement. These IDs must be assigned and used
12270 * `location' This is a `location_t' identifier to specify source code
12271 location for this statement. It is inherited from the front end.
12273 * `num_ops' Number of operands that this statement has. This
12274 specifies the size of the operand vector embedded in the tuple.
12275 Only used in some tuples, but it is declared in the base tuple to
12276 take advantage of the 32-bit hole left by the previous fields.
12278 * `bb' Basic block holding the instruction.
12280 * `block' Lexical block holding this statement. Also used for debug
12281 information generation.
12283 12.1.2 `gimple_statement_with_ops'
12284 ----------------------------------
12286 This tuple is actually split in two: `gimple_statement_with_ops_base'
12287 and `gimple_statement_with_ops'. This is needed to accommodate the way
12288 the operand vector is allocated. The operand vector is defined to be an
12289 array of 1 element. So, to allocate a dynamic number of operands, the
12290 memory allocator (`gimple_alloc') simply allocates enough memory to
12291 hold the structure itself plus `N - 1' operands which run "off the end"
12292 of the structure. For example, to allocate space for a tuple with 3
12293 operands, `gimple_alloc' reserves `sizeof (struct
12294 gimple_statement_with_ops) + 2 * sizeof (tree)' bytes.
12296 On the other hand, several fields in this tuple need to be shared with
12297 the `gimple_statement_with_memory_ops' tuple. So, these common fields
12298 are placed in `gimple_statement_with_ops_base' which is then inherited
12299 from the other two tuples.
12302 `addresses_taken' 64
12305 `op' `num_ops' * 64
12306 Total size 56 + 8 * `num_ops' bytes
12308 * `gsbase' Inherited from `struct gimple_statement_base'.
12310 * `addresses_taken' Bitmap holding the UIDs of all the `VAR_DECL's
12311 whose addresses are taken by this statement. For example, a
12312 statement of the form `p = &b' will have the UID for symbol `b' in
12315 * `def_ops' Array of pointers into the operand array indicating all
12316 the slots that contain a variable written-to by the statement.
12317 This array is also used for immediate use chaining. Note that it
12318 would be possible to not rely on this array, but the changes
12319 required to implement this are pretty invasive.
12321 * `use_ops' Similar to `def_ops' but for variables read by the
12324 * `op' Array of trees with `num_ops' slots.
12326 12.1.3 `gimple_statement_with_memory_ops'
12327 -----------------------------------------
12329 This tuple is essentially identical to `gimple_statement_with_ops',
12330 except that it contains 4 additional fields to hold vectors related
12331 memory stores and loads. Similar to the previous case, the structure
12332 is split in two to accommodate for the operand vector
12333 (`gimple_statement_with_memory_ops_base' and
12334 `gimple_statement_with_memory_ops').
12338 `addresses_taken' 64
12345 `op' `num_ops' * 64
12346 Total size 88 + 8 * `num_ops' bytes
12348 * `vdef_ops' Similar to `def_ops' but for `VDEF' operators. There is
12349 one entry per memory symbol written by this statement. This is
12350 used to maintain the memory SSA use-def and def-def chains.
12352 * `vuse_ops' Similar to `use_ops' but for `VUSE' operators. There is
12353 one entry per memory symbol loaded by this statement. This is used
12354 to maintain the memory SSA use-def chains.
12356 * `stores' Bitset with all the UIDs for the symbols written-to by the
12357 statement. This is different than `vdef_ops' in that all the
12358 affected symbols are mentioned in this set. If memory
12359 partitioning is enabled, the `vdef_ops' vector will refer to memory
12360 partitions. Furthermore, no SSA information is stored in this set.
12362 * `loads' Similar to `stores', but for memory loads. (Note that there
12363 is some amount of redundancy here, it should be possible to reduce
12364 memory utilization further by removing these sets).
12366 All the other tuples are defined in terms of these three basic ones.
12367 Each tuple will add some fields. The main gimple type is defined to be
12368 the union of all these structures (`GTY' markers elided for clarity):
12370 union gimple_statement_d
12372 struct gimple_statement_base gsbase;
12373 struct gimple_statement_with_ops gsops;
12374 struct gimple_statement_with_memory_ops gsmem;
12375 struct gimple_statement_omp omp;
12376 struct gimple_statement_bind gimple_bind;
12377 struct gimple_statement_catch gimple_catch;
12378 struct gimple_statement_eh_filter gimple_eh_filter;
12379 struct gimple_statement_phi gimple_phi;
12380 struct gimple_statement_resx gimple_resx;
12381 struct gimple_statement_try gimple_try;
12382 struct gimple_statement_wce gimple_wce;
12383 struct gimple_statement_asm gimple_asm;
12384 struct gimple_statement_omp_critical gimple_omp_critical;
12385 struct gimple_statement_omp_for gimple_omp_for;
12386 struct gimple_statement_omp_parallel gimple_omp_parallel;
12387 struct gimple_statement_omp_task gimple_omp_task;
12388 struct gimple_statement_omp_sections gimple_omp_sections;
12389 struct gimple_statement_omp_single gimple_omp_single;
12390 struct gimple_statement_omp_continue gimple_omp_continue;
12391 struct gimple_statement_omp_atomic_load gimple_omp_atomic_load;
12392 struct gimple_statement_omp_atomic_store gimple_omp_atomic_store;
12396 File: gccint.info, Node: GIMPLE instruction set, Next: GIMPLE Exception Handling, Prev: Tuple representation, Up: GIMPLE
12398 12.2 GIMPLE instruction set
12399 ===========================
12401 The following table briefly describes the GIMPLE instruction set.
12403 Instruction High GIMPLE Low GIMPLE
12405 `GIMPLE_ASSIGN' x x
12409 `GIMPLE_CHANGE_DYNAMIC_TYPE' x x
12411 `GIMPLE_EH_FILTER' x
12415 `GIMPLE_OMP_ATOMIC_LOAD' x x
12416 `GIMPLE_OMP_ATOMIC_STORE' x x
12417 `GIMPLE_OMP_CONTINUE' x x
12418 `GIMPLE_OMP_CRITICAL' x x
12419 `GIMPLE_OMP_FOR' x x
12420 `GIMPLE_OMP_MASTER' x x
12421 `GIMPLE_OMP_ORDERED' x x
12422 `GIMPLE_OMP_PARALLEL' x x
12423 `GIMPLE_OMP_RETURN' x x
12424 `GIMPLE_OMP_SECTION' x x
12425 `GIMPLE_OMP_SECTIONS' x x
12426 `GIMPLE_OMP_SECTIONS_SWITCH' x x
12427 `GIMPLE_OMP_SINGLE' x x
12430 `GIMPLE_RETURN' x x
12431 `GIMPLE_SWITCH' x x
12435 File: gccint.info, Node: GIMPLE Exception Handling, Next: Temporaries, Prev: GIMPLE instruction set, Up: GIMPLE
12437 12.3 Exception Handling
12438 =======================
12440 Other exception handling constructs are represented using
12441 `GIMPLE_TRY_CATCH'. `GIMPLE_TRY_CATCH' has two operands. The first
12442 operand is a sequence of statements to execute. If executing these
12443 statements does not throw an exception, then the second operand is
12444 ignored. Otherwise, if an exception is thrown, then the second operand
12445 of the `GIMPLE_TRY_CATCH' is checked. The second operand may have the
12448 1. A sequence of statements to execute. When an exception occurs,
12449 these statements are executed, and then the exception is rethrown.
12451 2. A sequence of `GIMPLE_CATCH' statements. Each `GIMPLE_CATCH' has
12452 a list of applicable exception types and handler code. If the
12453 thrown exception matches one of the caught types, the associated
12454 handler code is executed. If the handler code falls off the
12455 bottom, execution continues after the original `GIMPLE_TRY_CATCH'.
12457 3. An `GIMPLE_EH_FILTER' statement. This has a list of permitted
12458 exception types, and code to handle a match failure. If the
12459 thrown exception does not match one of the allowed types, the
12460 associated match failure code is executed. If the thrown exception
12461 does match, it continues unwinding the stack looking for the next
12465 Currently throwing an exception is not directly represented in GIMPLE,
12466 since it is implemented by calling a function. At some point in the
12467 future we will want to add some way to express that the call will throw
12468 an exception of a known type.
12470 Just before running the optimizers, the compiler lowers the high-level
12471 EH constructs above into a set of `goto's, magic labels, and EH
12472 regions. Continuing to unwind at the end of a cleanup is represented
12473 with a `GIMPLE_RESX'.
12476 File: gccint.info, Node: Temporaries, Next: Operands, Prev: GIMPLE Exception Handling, Up: GIMPLE
12481 When gimplification encounters a subexpression that is too complex, it
12482 creates a new temporary variable to hold the value of the
12483 subexpression, and adds a new statement to initialize it before the
12484 current statement. These special temporaries are known as `expression
12485 temporaries', and are allocated using `get_formal_tmp_var'. The
12486 compiler tries to always evaluate identical expressions into the same
12487 temporary, to simplify elimination of redundant calculations.
12489 We can only use expression temporaries when we know that it will not
12490 be reevaluated before its value is used, and that it will not be
12491 otherwise modified(1). Other temporaries can be allocated using
12492 `get_initialized_tmp_var' or `create_tmp_var'.
12494 Currently, an expression like `a = b + 5' is not reduced any further.
12495 We tried converting it to something like
12498 but this bloated the representation for minimal benefit. However, a
12499 variable which must live in memory cannot appear in an expression; its
12500 value is explicitly loaded into a temporary first. Similarly, storing
12501 the value of an expression to a memory variable goes through a
12504 ---------- Footnotes ----------
12506 (1) These restrictions are derived from those in Morgan 4.8.
12509 File: gccint.info, Node: Operands, Next: Manipulating GIMPLE statements, Prev: Temporaries, Up: GIMPLE
12514 In general, expressions in GIMPLE consist of an operation and the
12515 appropriate number of simple operands; these operands must either be a
12516 GIMPLE rvalue (`is_gimple_val'), i.e. a constant or a register
12517 variable. More complex operands are factored out into temporaries, so
12524 The same rule holds for arguments to a `GIMPLE_CALL'.
12526 The target of an assignment is usually a variable, but can also be an
12527 `INDIRECT_REF' or a compound lvalue as described below.
12531 * Compound Expressions::
12532 * Compound Lvalues::
12533 * Conditional Expressions::
12534 * Logical Operators::
12537 File: gccint.info, Node: Compound Expressions, Next: Compound Lvalues, Up: Operands
12539 12.5.1 Compound Expressions
12540 ---------------------------
12542 The left-hand side of a C comma expression is simply moved into a
12543 separate statement.
12546 File: gccint.info, Node: Compound Lvalues, Next: Conditional Expressions, Prev: Compound Expressions, Up: Operands
12548 12.5.2 Compound Lvalues
12549 -----------------------
12551 Currently compound lvalues involving array and structure field
12552 references are not broken down; an expression like `a.b[2] = 42' is not
12553 reduced any further (though complex array subscripts are). This
12554 restriction is a workaround for limitations in later optimizers; if we
12555 were to convert this to
12560 alias analysis would not remember that the reference to `T1[2]' came
12561 by way of `a.b', so it would think that the assignment could alias
12562 another member of `a'; this broke `struct-alias-1.c'. Future optimizer
12563 improvements may make this limitation unnecessary.
12566 File: gccint.info, Node: Conditional Expressions, Next: Logical Operators, Prev: Compound Lvalues, Up: Operands
12568 12.5.3 Conditional Expressions
12569 ------------------------------
12571 A C `?:' expression is converted into an `if' statement with each
12572 branch assigning to the same temporary. So,
12582 The GIMPLE level if-conversion pass re-introduces `?:' expression, if
12583 appropriate. It is used to vectorize loops with conditions using vector
12584 conditional operations.
12586 Note that in GIMPLE, `if' statements are represented using
12587 `GIMPLE_COND', as described below.
12590 File: gccint.info, Node: Logical Operators, Prev: Conditional Expressions, Up: Operands
12592 12.5.4 Logical Operators
12593 ------------------------
12595 Except when they appear in the condition operand of a `GIMPLE_COND',
12596 logical `and' and `or' operators are simplified as follows: `a = b &&
12604 Note that `T1' in this example cannot be an expression temporary,
12605 because it has two different assignments.
12607 12.5.5 Manipulating operands
12608 ----------------------------
12610 All gimple operands are of type `tree'. But only certain types of
12611 trees are allowed to be used as operand tuples. Basic validation is
12612 controlled by the function `get_gimple_rhs_class', which given a tree
12613 code, returns an `enum' with the following values of type `enum
12616 * `GIMPLE_INVALID_RHS' The tree cannot be used as a GIMPLE operand.
12618 * `GIMPLE_BINARY_RHS' The tree is a valid GIMPLE binary operation.
12620 * `GIMPLE_UNARY_RHS' The tree is a valid GIMPLE unary operation.
12622 * `GIMPLE_SINGLE_RHS' The tree is a single object, that cannot be
12623 split into simpler operands (for instance, `SSA_NAME', `VAR_DECL',
12624 `COMPONENT_REF', etc).
12626 This operand class also acts as an escape hatch for tree nodes
12627 that may be flattened out into the operand vector, but would need
12628 more than two slots on the RHS. For instance, a `COND_EXPR'
12629 expression of the form `(a op b) ? x : y' could be flattened out
12630 on the operand vector using 4 slots, but it would also require
12631 additional processing to distinguish `c = a op b' from `c = a op b
12632 ? x : y'. Something similar occurs with `ASSERT_EXPR'. In time,
12633 these special case tree expressions should be flattened into the
12636 For tree nodes in the categories `GIMPLE_BINARY_RHS' and
12637 `GIMPLE_UNARY_RHS', they cannot be stored inside tuples directly. They
12638 first need to be flattened and separated into individual components.
12639 For instance, given the GENERIC expression
12643 its tree representation is:
12645 MODIFY_EXPR <VAR_DECL <a>, PLUS_EXPR <VAR_DECL <b>, VAR_DECL <c>>>
12647 In this case, the GIMPLE form for this statement is logically
12648 identical to its GENERIC form but in GIMPLE, the `PLUS_EXPR' on the RHS
12649 of the assignment is not represented as a tree, instead the two
12650 operands are taken out of the `PLUS_EXPR' sub-tree and flattened into
12651 the GIMPLE tuple as follows:
12653 GIMPLE_ASSIGN <PLUS_EXPR, VAR_DECL <a>, VAR_DECL <b>, VAR_DECL <c>>
12655 12.5.6 Operand vector allocation
12656 --------------------------------
12658 The operand vector is stored at the bottom of the three tuple
12659 structures that accept operands. This means, that depending on the code
12660 of a given statement, its operand vector will be at different offsets
12661 from the base of the structure. To access tuple operands use the
12662 following accessors
12664 -- GIMPLE function: unsigned gimple_num_ops (gimple g)
12665 Returns the number of operands in statement G.
12667 -- GIMPLE function: tree gimple_op (gimple g, unsigned i)
12668 Returns operand `I' from statement `G'.
12670 -- GIMPLE function: tree *gimple_ops (gimple g)
12671 Returns a pointer into the operand vector for statement `G'. This
12672 is computed using an internal table called `gimple_ops_offset_'[].
12673 This table is indexed by the gimple code of `G'.
12675 When the compiler is built, this table is filled-in using the
12676 sizes of the structures used by each statement code defined in
12677 gimple.def. Since the operand vector is at the bottom of the
12678 structure, for a gimple code `C' the offset is computed as sizeof
12679 (struct-of `C') - sizeof (tree).
12681 This mechanism adds one memory indirection to every access when
12682 using `gimple_op'(), if this becomes a bottleneck, a pass can
12683 choose to memoize the result from `gimple_ops'() and use that to
12684 access the operands.
12686 12.5.7 Operand validation
12687 -------------------------
12689 When adding a new operand to a gimple statement, the operand will be
12690 validated according to what each tuple accepts in its operand vector.
12691 These predicates are called by the `gimple_<name>_set_...()'. Each
12692 tuple will use one of the following predicates (Note, this list is not
12695 -- GIMPLE function: is_gimple_operand (tree t)
12696 This is the most permissive of the predicates. It essentially
12697 checks whether t has a `gimple_rhs_class' of `GIMPLE_SINGLE_RHS'.
12699 -- GIMPLE function: is_gimple_val (tree t)
12700 Returns true if t is a "GIMPLE value", which are all the
12701 non-addressable stack variables (variables for which
12702 `is_gimple_reg' returns true) and constants (expressions for which
12703 `is_gimple_min_invariant' returns true).
12705 -- GIMPLE function: is_gimple_addressable (tree t)
12706 Returns true if t is a symbol or memory reference whose address
12709 -- GIMPLE function: is_gimple_asm_val (tree t)
12710 Similar to `is_gimple_val' but it also accepts hard registers.
12712 -- GIMPLE function: is_gimple_call_addr (tree t)
12713 Return true if t is a valid expression to use as the function
12714 called by a `GIMPLE_CALL'.
12716 -- GIMPLE function: is_gimple_constant (tree t)
12717 Return true if t is a valid gimple constant.
12719 -- GIMPLE function: is_gimple_min_invariant (tree t)
12720 Return true if t is a valid minimal invariant. This is different
12721 from constants, in that the specific value of t may not be known
12722 at compile time, but it is known that it doesn't change (e.g., the
12723 address of a function local variable).
12725 -- GIMPLE function: is_gimple_min_invariant_address (tree t)
12726 Return true if t is an `ADDR_EXPR' that does not change once the
12727 program is running.
12729 12.5.8 Statement validation
12730 ---------------------------
12732 -- GIMPLE function: is_gimple_assign (gimple g)
12733 Return true if the code of g is `GIMPLE_ASSIGN'.
12735 -- GIMPLE function: is_gimple_call (gimple g)
12736 Return true if the code of g is `GIMPLE_CALL'
12738 -- GIMPLE function: gimple_assign_cast_p (gimple g)
12739 Return true if g is a `GIMPLE_ASSIGN' that performs a type cast
12743 File: gccint.info, Node: Manipulating GIMPLE statements, Next: Tuple specific accessors, Prev: Operands, Up: GIMPLE
12745 12.6 Manipulating GIMPLE statements
12746 ===================================
12748 This section documents all the functions available to handle each of
12749 the GIMPLE instructions.
12751 12.6.1 Common accessors
12752 -----------------------
12754 The following are common accessors for gimple statements.
12756 -- GIMPLE function: enum gimple_code gimple_code (gimple g)
12757 Return the code for statement `G'.
12759 -- GIMPLE function: basic_block gimple_bb (gimple g)
12760 Return the basic block to which statement `G' belongs to.
12762 -- GIMPLE function: tree gimple_block (gimple g)
12763 Return the lexical scope block holding statement `G'.
12765 -- GIMPLE function: tree gimple_expr_type (gimple stmt)
12766 Return the type of the main expression computed by `STMT'. Return
12767 `void_type_node' if `STMT' computes nothing. This will only return
12768 something meaningful for `GIMPLE_ASSIGN', `GIMPLE_COND' and
12769 `GIMPLE_CALL'. For all other tuple codes, it will return
12772 -- GIMPLE function: enum tree_code gimple_expr_code (gimple stmt)
12773 Return the tree code for the expression computed by `STMT'. This
12774 is only meaningful for `GIMPLE_CALL', `GIMPLE_ASSIGN' and
12775 `GIMPLE_COND'. If `STMT' is `GIMPLE_CALL', it will return
12776 `CALL_EXPR'. For `GIMPLE_COND', it returns the code of the
12777 comparison predicate. For `GIMPLE_ASSIGN' it returns the code of
12778 the operation performed by the `RHS' of the assignment.
12780 -- GIMPLE function: void gimple_set_block (gimple g, tree block)
12781 Set the lexical scope block of `G' to `BLOCK'.
12783 -- GIMPLE function: location_t gimple_locus (gimple g)
12784 Return locus information for statement `G'.
12786 -- GIMPLE function: void gimple_set_locus (gimple g, location_t locus)
12787 Set locus information for statement `G'.
12789 -- GIMPLE function: bool gimple_locus_empty_p (gimple g)
12790 Return true if `G' does not have locus information.
12792 -- GIMPLE function: bool gimple_no_warning_p (gimple stmt)
12793 Return true if no warnings should be emitted for statement `STMT'.
12795 -- GIMPLE function: void gimple_set_visited (gimple stmt, bool
12797 Set the visited status on statement `STMT' to `VISITED_P'.
12799 -- GIMPLE function: bool gimple_visited_p (gimple stmt)
12800 Return the visited status on statement `STMT'.
12802 -- GIMPLE function: void gimple_set_plf (gimple stmt, enum plf_mask
12804 Set pass local flag `PLF' on statement `STMT' to `VAL_P'.
12806 -- GIMPLE function: unsigned int gimple_plf (gimple stmt, enum
12808 Return the value of pass local flag `PLF' on statement `STMT'.
12810 -- GIMPLE function: bool gimple_has_ops (gimple g)
12811 Return true if statement `G' has register or memory operands.
12813 -- GIMPLE function: bool gimple_has_mem_ops (gimple g)
12814 Return true if statement `G' has memory operands.
12816 -- GIMPLE function: unsigned gimple_num_ops (gimple g)
12817 Return the number of operands for statement `G'.
12819 -- GIMPLE function: tree *gimple_ops (gimple g)
12820 Return the array of operands for statement `G'.
12822 -- GIMPLE function: tree gimple_op (gimple g, unsigned i)
12823 Return operand `I' for statement `G'.
12825 -- GIMPLE function: tree *gimple_op_ptr (gimple g, unsigned i)
12826 Return a pointer to operand `I' for statement `G'.
12828 -- GIMPLE function: void gimple_set_op (gimple g, unsigned i, tree op)
12829 Set operand `I' of statement `G' to `OP'.
12831 -- GIMPLE function: bitmap gimple_addresses_taken (gimple stmt)
12832 Return the set of symbols that have had their address taken by
12835 -- GIMPLE function: struct def_optype_d *gimple_def_ops (gimple g)
12836 Return the set of `DEF' operands for statement `G'.
12838 -- GIMPLE function: void gimple_set_def_ops (gimple g, struct
12840 Set `DEF' to be the set of `DEF' operands for statement `G'.
12842 -- GIMPLE function: struct use_optype_d *gimple_use_ops (gimple g)
12843 Return the set of `USE' operands for statement `G'.
12845 -- GIMPLE function: void gimple_set_use_ops (gimple g, struct
12847 Set `USE' to be the set of `USE' operands for statement `G'.
12849 -- GIMPLE function: struct voptype_d *gimple_vuse_ops (gimple g)
12850 Return the set of `VUSE' operands for statement `G'.
12852 -- GIMPLE function: void gimple_set_vuse_ops (gimple g, struct
12854 Set `OPS' to be the set of `VUSE' operands for statement `G'.
12856 -- GIMPLE function: struct voptype_d *gimple_vdef_ops (gimple g)
12857 Return the set of `VDEF' operands for statement `G'.
12859 -- GIMPLE function: void gimple_set_vdef_ops (gimple g, struct
12861 Set `OPS' to be the set of `VDEF' operands for statement `G'.
12863 -- GIMPLE function: bitmap gimple_loaded_syms (gimple g)
12864 Return the set of symbols loaded by statement `G'. Each element of
12865 the set is the `DECL_UID' of the corresponding symbol.
12867 -- GIMPLE function: bitmap gimple_stored_syms (gimple g)
12868 Return the set of symbols stored by statement `G'. Each element of
12869 the set is the `DECL_UID' of the corresponding symbol.
12871 -- GIMPLE function: bool gimple_modified_p (gimple g)
12872 Return true if statement `G' has operands and the modified field
12875 -- GIMPLE function: bool gimple_has_volatile_ops (gimple stmt)
12876 Return true if statement `STMT' contains volatile operands.
12878 -- GIMPLE function: void gimple_set_has_volatile_ops (gimple stmt,
12880 Return true if statement `STMT' contains volatile operands.
12882 -- GIMPLE function: void update_stmt (gimple s)
12883 Mark statement `S' as modified, and update it.
12885 -- GIMPLE function: void update_stmt_if_modified (gimple s)
12886 Update statement `S' if it has been marked modified.
12888 -- GIMPLE function: gimple gimple_copy (gimple stmt)
12889 Return a deep copy of statement `STMT'.
12892 File: gccint.info, Node: Tuple specific accessors, Next: GIMPLE sequences, Prev: Manipulating GIMPLE statements, Up: GIMPLE
12894 12.7 Tuple specific accessors
12895 =============================
12900 * `GIMPLE_ASSIGN'::
12904 * `GIMPLE_CHANGE_DYNAMIC_TYPE'::
12906 * `GIMPLE_EH_FILTER'::
12909 * `GIMPLE_OMP_ATOMIC_LOAD'::
12910 * `GIMPLE_OMP_ATOMIC_STORE'::
12911 * `GIMPLE_OMP_CONTINUE'::
12912 * `GIMPLE_OMP_CRITICAL'::
12913 * `GIMPLE_OMP_FOR'::
12914 * `GIMPLE_OMP_MASTER'::
12915 * `GIMPLE_OMP_ORDERED'::
12916 * `GIMPLE_OMP_PARALLEL'::
12917 * `GIMPLE_OMP_RETURN'::
12918 * `GIMPLE_OMP_SECTION'::
12919 * `GIMPLE_OMP_SECTIONS'::
12920 * `GIMPLE_OMP_SINGLE'::
12923 * `GIMPLE_RETURN'::
12924 * `GIMPLE_SWITCH'::
12926 * `GIMPLE_WITH_CLEANUP_EXPR'::
12929 File: gccint.info, Node: `GIMPLE_ASM', Next: `GIMPLE_ASSIGN', Up: Tuple specific accessors
12931 12.7.1 `GIMPLE_ASM'
12932 -------------------
12934 -- GIMPLE function: gimple gimple_build_asm (const char *string,
12935 ninputs, noutputs, nclobbers, ...)
12936 Build a `GIMPLE_ASM' statement. This statement is used for
12937 building in-line assembly constructs. `STRING' is the assembly
12938 code. `NINPUT' is the number of register inputs. `NOUTPUT' is the
12939 number of register outputs. `NCLOBBERS' is the number of clobbered
12940 registers. The rest of the arguments trees for each input,
12941 output, and clobbered registers.
12943 -- GIMPLE function: gimple gimple_build_asm_vec (const char *,
12944 VEC(tree,gc) *, VEC(tree,gc) *, VEC(tree,gc) *)
12945 Identical to gimple_build_asm, but the arguments are passed in
12948 -- GIMPLE function: gimple_asm_ninputs (gimple g)
12949 Return the number of input operands for `GIMPLE_ASM' `G'.
12951 -- GIMPLE function: gimple_asm_noutputs (gimple g)
12952 Return the number of output operands for `GIMPLE_ASM' `G'.
12954 -- GIMPLE function: gimple_asm_nclobbers (gimple g)
12955 Return the number of clobber operands for `GIMPLE_ASM' `G'.
12957 -- GIMPLE function: tree gimple_asm_input_op (gimple g, unsigned index)
12958 Return input operand `INDEX' of `GIMPLE_ASM' `G'.
12960 -- GIMPLE function: void gimple_asm_set_input_op (gimple g, unsigned
12962 Set `IN_OP' to be input operand `INDEX' in `GIMPLE_ASM' `G'.
12964 -- GIMPLE function: tree gimple_asm_output_op (gimple g, unsigned
12966 Return output operand `INDEX' of `GIMPLE_ASM' `G'.
12968 -- GIMPLE function: void gimple_asm_set_output_op (gimple g, unsigned
12969 index, tree out_op)
12970 Set `OUT_OP' to be output operand `INDEX' in `GIMPLE_ASM' `G'.
12972 -- GIMPLE function: tree gimple_asm_clobber_op (gimple g, unsigned
12974 Return clobber operand `INDEX' of `GIMPLE_ASM' `G'.
12976 -- GIMPLE function: void gimple_asm_set_clobber_op (gimple g, unsigned
12977 index, tree clobber_op)
12978 Set `CLOBBER_OP' to be clobber operand `INDEX' in `GIMPLE_ASM' `G'.
12980 -- GIMPLE function: const char *gimple_asm_string (gimple g)
12981 Return the string representing the assembly instruction in
12984 -- GIMPLE function: bool gimple_asm_volatile_p (gimple g)
12985 Return true if `G' is an asm statement marked volatile.
12987 -- GIMPLE function: void gimple_asm_set_volatile (gimple g)
12988 Mark asm statement `G' as volatile.
12990 -- GIMPLE function: void gimple_asm_clear_volatile (gimple g)
12991 Remove volatile marker from asm statement `G'.
12994 File: gccint.info, Node: `GIMPLE_ASSIGN', Next: `GIMPLE_BIND', Prev: `GIMPLE_ASM', Up: Tuple specific accessors
12996 12.7.2 `GIMPLE_ASSIGN'
12997 ----------------------
12999 -- GIMPLE function: gimple gimple_build_assign (tree lhs, tree rhs)
13000 Build a `GIMPLE_ASSIGN' statement. The left-hand side is an lvalue
13001 passed in lhs. The right-hand side can be either a unary or
13002 binary tree expression. The expression tree rhs will be flattened
13003 and its operands assigned to the corresponding operand slots in
13004 the new statement. This function is useful when you already have
13005 a tree expression that you want to convert into a tuple. However,
13006 try to avoid building expression trees for the sole purpose of
13007 calling this function. If you already have the operands in
13008 separate trees, it is better to use `gimple_build_assign_with_ops'.
13010 -- GIMPLE function: gimple gimplify_assign (tree dst, tree src,
13012 Build a new `GIMPLE_ASSIGN' tuple and append it to the end of
13015 `DST'/`SRC' are the destination and source respectively. You can pass
13016 ungimplified trees in `DST' or `SRC', in which case they will be
13017 converted to a gimple operand if necessary.
13019 This function returns the newly created `GIMPLE_ASSIGN' tuple.
13021 -- GIMPLE function: gimple gimple_build_assign_with_ops (enum
13022 tree_code subcode, tree lhs, tree op1, tree op2)
13023 This function is similar to `gimple_build_assign', but is used to
13024 build a `GIMPLE_ASSIGN' statement when the operands of the
13025 right-hand side of the assignment are already split into different
13028 The left-hand side is an lvalue passed in lhs. Subcode is the
13029 `tree_code' for the right-hand side of the assignment. Op1 and op2
13030 are the operands. If op2 is null, subcode must be a `tree_code'
13031 for a unary expression.
13033 -- GIMPLE function: enum tree_code gimple_assign_rhs_code (gimple g)
13034 Return the code of the expression computed on the `RHS' of
13035 assignment statement `G'.
13037 -- GIMPLE function: enum gimple_rhs_class gimple_assign_rhs_class
13039 Return the gimple rhs class of the code for the expression
13040 computed on the rhs of assignment statement `G'. This will never
13041 return `GIMPLE_INVALID_RHS'.
13043 -- GIMPLE function: tree gimple_assign_lhs (gimple g)
13044 Return the `LHS' of assignment statement `G'.
13046 -- GIMPLE function: tree *gimple_assign_lhs_ptr (gimple g)
13047 Return a pointer to the `LHS' of assignment statement `G'.
13049 -- GIMPLE function: tree gimple_assign_rhs1 (gimple g)
13050 Return the first operand on the `RHS' of assignment statement `G'.
13052 -- GIMPLE function: tree *gimple_assign_rhs1_ptr (gimple g)
13053 Return the address of the first operand on the `RHS' of assignment
13056 -- GIMPLE function: tree gimple_assign_rhs2 (gimple g)
13057 Return the second operand on the `RHS' of assignment statement `G'.
13059 -- GIMPLE function: tree *gimple_assign_rhs2_ptr (gimple g)
13060 Return the address of the second operand on the `RHS' of assignment
13063 -- GIMPLE function: void gimple_assign_set_lhs (gimple g, tree lhs)
13064 Set `LHS' to be the `LHS' operand of assignment statement `G'.
13066 -- GIMPLE function: void gimple_assign_set_rhs1 (gimple g, tree rhs)
13067 Set `RHS' to be the first operand on the `RHS' of assignment
13070 -- GIMPLE function: tree gimple_assign_rhs2 (gimple g)
13071 Return the second operand on the `RHS' of assignment statement `G'.
13073 -- GIMPLE function: tree *gimple_assign_rhs2_ptr (gimple g)
13074 Return a pointer to the second operand on the `RHS' of assignment
13077 -- GIMPLE function: void gimple_assign_set_rhs2 (gimple g, tree rhs)
13078 Set `RHS' to be the second operand on the `RHS' of assignment
13081 -- GIMPLE function: bool gimple_assign_cast_p (gimple s)
13082 Return true if `S' is an type-cast assignment.
13085 File: gccint.info, Node: `GIMPLE_BIND', Next: `GIMPLE_CALL', Prev: `GIMPLE_ASSIGN', Up: Tuple specific accessors
13087 12.7.3 `GIMPLE_BIND'
13088 --------------------
13090 -- GIMPLE function: gimple gimple_build_bind (tree vars, gimple_seq
13092 Build a `GIMPLE_BIND' statement with a list of variables in `VARS'
13093 and a body of statements in sequence `BODY'.
13095 -- GIMPLE function: tree gimple_bind_vars (gimple g)
13096 Return the variables declared in the `GIMPLE_BIND' statement `G'.
13098 -- GIMPLE function: void gimple_bind_set_vars (gimple g, tree vars)
13099 Set `VARS' to be the set of variables declared in the `GIMPLE_BIND'
13102 -- GIMPLE function: void gimple_bind_append_vars (gimple g, tree vars)
13103 Append `VARS' to the set of variables declared in the `GIMPLE_BIND'
13106 -- GIMPLE function: gimple_seq gimple_bind_body (gimple g)
13107 Return the GIMPLE sequence contained in the `GIMPLE_BIND' statement
13110 -- GIMPLE function: void gimple_bind_set_body (gimple g, gimple_seq
13112 Set `SEQ' to be sequence contained in the `GIMPLE_BIND' statement
13115 -- GIMPLE function: void gimple_bind_add_stmt (gimple gs, gimple stmt)
13116 Append a statement to the end of a `GIMPLE_BIND''s body.
13118 -- GIMPLE function: void gimple_bind_add_seq (gimple gs, gimple_seq
13120 Append a sequence of statements to the end of a `GIMPLE_BIND''s
13123 -- GIMPLE function: tree gimple_bind_block (gimple g)
13124 Return the `TREE_BLOCK' node associated with `GIMPLE_BIND'
13125 statement `G'. This is analogous to the `BIND_EXPR_BLOCK' field in
13128 -- GIMPLE function: void gimple_bind_set_block (gimple g, tree block)
13129 Set `BLOCK' to be the `TREE_BLOCK' node associated with
13130 `GIMPLE_BIND' statement `G'.
13133 File: gccint.info, Node: `GIMPLE_CALL', Next: `GIMPLE_CATCH', Prev: `GIMPLE_BIND', Up: Tuple specific accessors
13135 12.7.4 `GIMPLE_CALL'
13136 --------------------
13138 -- GIMPLE function: gimple gimple_build_call (tree fn, unsigned nargs,
13140 Build a `GIMPLE_CALL' statement to function `FN'. The argument
13141 `FN' must be either a `FUNCTION_DECL' or a gimple call address as
13142 determined by `is_gimple_call_addr'. `NARGS' are the number of
13143 arguments. The rest of the arguments follow the argument `NARGS',
13144 and must be trees that are valid as rvalues in gimple (i.e., each
13145 operand is validated with `is_gimple_operand').
13147 -- GIMPLE function: gimple gimple_build_call_from_tree (tree call_expr)
13148 Build a `GIMPLE_CALL' from a `CALL_EXPR' node. The arguments and
13149 the function are taken from the expression directly. This routine
13150 assumes that `call_expr' is already in GIMPLE form. That is, its
13151 operands are GIMPLE values and the function call needs no further
13152 simplification. All the call flags in `call_expr' are copied over
13153 to the new `GIMPLE_CALL'.
13155 -- GIMPLE function: gimple gimple_build_call_vec (tree fn, `VEC'(tree,
13157 Identical to `gimple_build_call' but the arguments are stored in a
13160 -- GIMPLE function: tree gimple_call_lhs (gimple g)
13161 Return the `LHS' of call statement `G'.
13163 -- GIMPLE function: tree *gimple_call_lhs_ptr (gimple g)
13164 Return a pointer to the `LHS' of call statement `G'.
13166 -- GIMPLE function: void gimple_call_set_lhs (gimple g, tree lhs)
13167 Set `LHS' to be the `LHS' operand of call statement `G'.
13169 -- GIMPLE function: tree gimple_call_fn (gimple g)
13170 Return the tree node representing the function called by call
13173 -- GIMPLE function: void gimple_call_set_fn (gimple g, tree fn)
13174 Set `FN' to be the function called by call statement `G'. This has
13175 to be a gimple value specifying the address of the called function.
13177 -- GIMPLE function: tree gimple_call_fndecl (gimple g)
13178 If a given `GIMPLE_CALL''s callee is a `FUNCTION_DECL', return it.
13179 Otherwise return `NULL'. This function is analogous to
13180 `get_callee_fndecl' in `GENERIC'.
13182 -- GIMPLE function: tree gimple_call_set_fndecl (gimple g, tree fndecl)
13183 Set the called function to `FNDECL'.
13185 -- GIMPLE function: tree gimple_call_return_type (gimple g)
13186 Return the type returned by call statement `G'.
13188 -- GIMPLE function: tree gimple_call_chain (gimple g)
13189 Return the static chain for call statement `G'.
13191 -- GIMPLE function: void gimple_call_set_chain (gimple g, tree chain)
13192 Set `CHAIN' to be the static chain for call statement `G'.
13194 -- GIMPLE function: gimple_call_num_args (gimple g)
13195 Return the number of arguments used by call statement `G'.
13197 -- GIMPLE function: tree gimple_call_arg (gimple g, unsigned index)
13198 Return the argument at position `INDEX' for call statement `G'.
13199 The first argument is 0.
13201 -- GIMPLE function: tree *gimple_call_arg_ptr (gimple g, unsigned
13203 Return a pointer to the argument at position `INDEX' for call
13206 -- GIMPLE function: void gimple_call_set_arg (gimple g, unsigned
13208 Set `ARG' to be the argument at position `INDEX' for call statement
13211 -- GIMPLE function: void gimple_call_set_tail (gimple s)
13212 Mark call statement `S' as being a tail call (i.e., a call just
13213 before the exit of a function). These calls are candidate for tail
13216 -- GIMPLE function: bool gimple_call_tail_p (gimple s)
13217 Return true if `GIMPLE_CALL' `S' is marked as a tail call.
13219 -- GIMPLE function: void gimple_call_mark_uninlinable (gimple s)
13220 Mark `GIMPLE_CALL' `S' as being uninlinable.
13222 -- GIMPLE function: bool gimple_call_cannot_inline_p (gimple s)
13223 Return true if `GIMPLE_CALL' `S' cannot be inlined.
13225 -- GIMPLE function: bool gimple_call_noreturn_p (gimple s)
13226 Return true if `S' is a noreturn call.
13228 -- GIMPLE function: gimple gimple_call_copy_skip_args (gimple stmt,
13229 bitmap args_to_skip)
13230 Build a `GIMPLE_CALL' identical to `STMT' but skipping the
13231 arguments in the positions marked by the set `ARGS_TO_SKIP'.
13234 File: gccint.info, Node: `GIMPLE_CATCH', Next: `GIMPLE_CHANGE_DYNAMIC_TYPE', Prev: `GIMPLE_CALL', Up: Tuple specific accessors
13236 12.7.5 `GIMPLE_CATCH'
13237 ---------------------
13239 -- GIMPLE function: gimple gimple_build_catch (tree types, gimple_seq
13241 Build a `GIMPLE_CATCH' statement. `TYPES' are the tree types this
13242 catch handles. `HANDLER' is a sequence of statements with the code
13245 -- GIMPLE function: tree gimple_catch_types (gimple g)
13246 Return the types handled by `GIMPLE_CATCH' statement `G'.
13248 -- GIMPLE function: tree *gimple_catch_types_ptr (gimple g)
13249 Return a pointer to the types handled by `GIMPLE_CATCH' statement
13252 -- GIMPLE function: gimple_seq gimple_catch_handler (gimple g)
13253 Return the GIMPLE sequence representing the body of the handler of
13254 `GIMPLE_CATCH' statement `G'.
13256 -- GIMPLE function: void gimple_catch_set_types (gimple g, tree t)
13257 Set `T' to be the set of types handled by `GIMPLE_CATCH' `G'.
13259 -- GIMPLE function: void gimple_catch_set_handler (gimple g,
13260 gimple_seq handler)
13261 Set `HANDLER' to be the body of `GIMPLE_CATCH' `G'.
13264 File: gccint.info, Node: `GIMPLE_CHANGE_DYNAMIC_TYPE', Next: `GIMPLE_COND', Prev: `GIMPLE_CATCH', Up: Tuple specific accessors
13266 12.7.6 `GIMPLE_CHANGE_DYNAMIC_TYPE'
13267 -----------------------------------
13269 -- GIMPLE function: gimple gimple_build_cdt (tree type, tree ptr)
13270 Build a `GIMPLE_CHANGE_DYNAMIC_TYPE' statement. `TYPE' is the new
13271 type for the location `PTR'.
13273 -- GIMPLE function: tree gimple_cdt_new_type (gimple g)
13274 Return the new type set by `GIMPLE_CHANGE_DYNAMIC_TYPE' statement
13277 -- GIMPLE function: tree *gimple_cdt_new_type_ptr (gimple g)
13278 Return a pointer to the new type set by
13279 `GIMPLE_CHANGE_DYNAMIC_TYPE' statement `G'.
13281 -- GIMPLE function: void gimple_cdt_set_new_type (gimple g, tree
13283 Set `NEW_TYPE' to be the type returned by
13284 `GIMPLE_CHANGE_DYNAMIC_TYPE' statement `G'.
13286 -- GIMPLE function: tree gimple_cdt_location (gimple g)
13287 Return the location affected by `GIMPLE_CHANGE_DYNAMIC_TYPE'
13290 -- GIMPLE function: tree *gimple_cdt_location_ptr (gimple g)
13291 Return a pointer to the location affected by
13292 `GIMPLE_CHANGE_DYNAMIC_TYPE' statement `G'.
13294 -- GIMPLE function: void gimple_cdt_set_location (gimple g, tree ptr)
13295 Set `PTR' to be the location affected by
13296 `GIMPLE_CHANGE_DYNAMIC_TYPE' statement `G'.
13299 File: gccint.info, Node: `GIMPLE_COND', Next: `GIMPLE_EH_FILTER', Prev: `GIMPLE_CHANGE_DYNAMIC_TYPE', Up: Tuple specific accessors
13301 12.7.7 `GIMPLE_COND'
13302 --------------------
13304 -- GIMPLE function: gimple gimple_build_cond (enum tree_code
13305 pred_code, tree lhs, tree rhs, tree t_label, tree f_label)
13306 Build a `GIMPLE_COND' statement. `A' `GIMPLE_COND' statement
13307 compares `LHS' and `RHS' and if the condition in `PRED_CODE' is
13308 true, jump to the label in `t_label', otherwise jump to the label
13309 in `f_label'. `PRED_CODE' are relational operator tree codes like
13310 `EQ_EXPR', `LT_EXPR', `LE_EXPR', `NE_EXPR', etc.
13312 -- GIMPLE function: gimple gimple_build_cond_from_tree (tree cond,
13313 tree t_label, tree f_label)
13314 Build a `GIMPLE_COND' statement from the conditional expression
13315 tree `COND'. `T_LABEL' and `F_LABEL' are as in
13316 `gimple_build_cond'.
13318 -- GIMPLE function: enum tree_code gimple_cond_code (gimple g)
13319 Return the code of the predicate computed by conditional statement
13322 -- GIMPLE function: void gimple_cond_set_code (gimple g, enum
13324 Set `CODE' to be the predicate code for the conditional statement
13327 -- GIMPLE function: tree gimple_cond_lhs (gimple g)
13328 Return the `LHS' of the predicate computed by conditional statement
13331 -- GIMPLE function: void gimple_cond_set_lhs (gimple g, tree lhs)
13332 Set `LHS' to be the `LHS' operand of the predicate computed by
13333 conditional statement `G'.
13335 -- GIMPLE function: tree gimple_cond_rhs (gimple g)
13336 Return the `RHS' operand of the predicate computed by conditional
13339 -- GIMPLE function: void gimple_cond_set_rhs (gimple g, tree rhs)
13340 Set `RHS' to be the `RHS' operand of the predicate computed by
13341 conditional statement `G'.
13343 -- GIMPLE function: tree gimple_cond_true_label (gimple g)
13344 Return the label used by conditional statement `G' when its
13345 predicate evaluates to true.
13347 -- GIMPLE function: void gimple_cond_set_true_label (gimple g, tree
13349 Set `LABEL' to be the label used by conditional statement `G' when
13350 its predicate evaluates to true.
13352 -- GIMPLE function: void gimple_cond_set_false_label (gimple g, tree
13354 Set `LABEL' to be the label used by conditional statement `G' when
13355 its predicate evaluates to false.
13357 -- GIMPLE function: tree gimple_cond_false_label (gimple g)
13358 Return the label used by conditional statement `G' when its
13359 predicate evaluates to false.
13361 -- GIMPLE function: void gimple_cond_make_false (gimple g)
13362 Set the conditional `COND_STMT' to be of the form 'if (1 == 0)'.
13364 -- GIMPLE function: void gimple_cond_make_true (gimple g)
13365 Set the conditional `COND_STMT' to be of the form 'if (1 == 1)'.
13368 File: gccint.info, Node: `GIMPLE_EH_FILTER', Next: `GIMPLE_LABEL', Prev: `GIMPLE_COND', Up: Tuple specific accessors
13370 12.7.8 `GIMPLE_EH_FILTER'
13371 -------------------------
13373 -- GIMPLE function: gimple gimple_build_eh_filter (tree types,
13374 gimple_seq failure)
13375 Build a `GIMPLE_EH_FILTER' statement. `TYPES' are the filter's
13376 types. `FAILURE' is a sequence with the filter's failure action.
13378 -- GIMPLE function: tree gimple_eh_filter_types (gimple g)
13379 Return the types handled by `GIMPLE_EH_FILTER' statement `G'.
13381 -- GIMPLE function: tree *gimple_eh_filter_types_ptr (gimple g)
13382 Return a pointer to the types handled by `GIMPLE_EH_FILTER'
13385 -- GIMPLE function: gimple_seq gimple_eh_filter_failure (gimple g)
13386 Return the sequence of statement to execute when `GIMPLE_EH_FILTER'
13389 -- GIMPLE function: void gimple_eh_filter_set_types (gimple g, tree
13391 Set `TYPES' to be the set of types handled by `GIMPLE_EH_FILTER'
13394 -- GIMPLE function: void gimple_eh_filter_set_failure (gimple g,
13395 gimple_seq failure)
13396 Set `FAILURE' to be the sequence of statements to execute on
13397 failure for `GIMPLE_EH_FILTER' `G'.
13399 -- GIMPLE function: bool gimple_eh_filter_must_not_throw (gimple g)
13400 Return the `EH_FILTER_MUST_NOT_THROW' flag.
13402 -- GIMPLE function: void gimple_eh_filter_set_must_not_throw (gimple
13404 Set the `EH_FILTER_MUST_NOT_THROW' flag.
13407 File: gccint.info, Node: `GIMPLE_LABEL', Next: `GIMPLE_NOP', Prev: `GIMPLE_EH_FILTER', Up: Tuple specific accessors
13409 12.7.9 `GIMPLE_LABEL'
13410 ---------------------
13412 -- GIMPLE function: gimple gimple_build_label (tree label)
13413 Build a `GIMPLE_LABEL' statement with corresponding to the tree
13416 -- GIMPLE function: tree gimple_label_label (gimple g)
13417 Return the `LABEL_DECL' node used by `GIMPLE_LABEL' statement `G'.
13419 -- GIMPLE function: void gimple_label_set_label (gimple g, tree label)
13420 Set `LABEL' to be the `LABEL_DECL' node used by `GIMPLE_LABEL'
13423 -- GIMPLE function: gimple gimple_build_goto (tree dest)
13424 Build a `GIMPLE_GOTO' statement to label `DEST'.
13426 -- GIMPLE function: tree gimple_goto_dest (gimple g)
13427 Return the destination of the unconditional jump `G'.
13429 -- GIMPLE function: void gimple_goto_set_dest (gimple g, tree dest)
13430 Set `DEST' to be the destination of the unconditional jump `G'.
13433 File: gccint.info, Node: `GIMPLE_NOP', Next: `GIMPLE_OMP_ATOMIC_LOAD', Prev: `GIMPLE_LABEL', Up: Tuple specific accessors
13435 12.7.10 `GIMPLE_NOP'
13436 --------------------
13438 -- GIMPLE function: gimple gimple_build_nop (void)
13439 Build a `GIMPLE_NOP' statement.
13441 -- GIMPLE function: bool gimple_nop_p (gimple g)
13442 Returns `TRUE' if statement `G' is a `GIMPLE_NOP'.
13445 File: gccint.info, Node: `GIMPLE_OMP_ATOMIC_LOAD', Next: `GIMPLE_OMP_ATOMIC_STORE', Prev: `GIMPLE_NOP', Up: Tuple specific accessors
13447 12.7.11 `GIMPLE_OMP_ATOMIC_LOAD'
13448 --------------------------------
13450 -- GIMPLE function: gimple gimple_build_omp_atomic_load (tree lhs,
13452 Build a `GIMPLE_OMP_ATOMIC_LOAD' statement. `LHS' is the left-hand
13453 side of the assignment. `RHS' is the right-hand side of the
13456 -- GIMPLE function: void gimple_omp_atomic_load_set_lhs (gimple g,
13458 Set the `LHS' of an atomic load.
13460 -- GIMPLE function: tree gimple_omp_atomic_load_lhs (gimple g)
13461 Get the `LHS' of an atomic load.
13463 -- GIMPLE function: void gimple_omp_atomic_load_set_rhs (gimple g,
13465 Set the `RHS' of an atomic set.
13467 -- GIMPLE function: tree gimple_omp_atomic_load_rhs (gimple g)
13468 Get the `RHS' of an atomic set.
13471 File: gccint.info, Node: `GIMPLE_OMP_ATOMIC_STORE', Next: `GIMPLE_OMP_CONTINUE', Prev: `GIMPLE_OMP_ATOMIC_LOAD', Up: Tuple specific accessors
13473 12.7.12 `GIMPLE_OMP_ATOMIC_STORE'
13474 ---------------------------------
13476 -- GIMPLE function: gimple gimple_build_omp_atomic_store (tree val)
13477 Build a `GIMPLE_OMP_ATOMIC_STORE' statement. `VAL' is the value to
13480 -- GIMPLE function: void gimple_omp_atomic_store_set_val (gimple g,
13482 Set the value being stored in an atomic store.
13484 -- GIMPLE function: tree gimple_omp_atomic_store_val (gimple g)
13485 Return the value being stored in an atomic store.
13488 File: gccint.info, Node: `GIMPLE_OMP_CONTINUE', Next: `GIMPLE_OMP_CRITICAL', Prev: `GIMPLE_OMP_ATOMIC_STORE', Up: Tuple specific accessors
13490 12.7.13 `GIMPLE_OMP_CONTINUE'
13491 -----------------------------
13493 -- GIMPLE function: gimple gimple_build_omp_continue (tree
13494 control_def, tree control_use)
13495 Build a `GIMPLE_OMP_CONTINUE' statement. `CONTROL_DEF' is the
13496 definition of the control variable. `CONTROL_USE' is the use of
13497 the control variable.
13499 -- GIMPLE function: tree gimple_omp_continue_control_def (gimple s)
13500 Return the definition of the control variable on a
13501 `GIMPLE_OMP_CONTINUE' in `S'.
13503 -- GIMPLE function: tree gimple_omp_continue_control_def_ptr (gimple s)
13504 Same as above, but return the pointer.
13506 -- GIMPLE function: tree gimple_omp_continue_set_control_def (gimple s)
13507 Set the control variable definition for a `GIMPLE_OMP_CONTINUE'
13510 -- GIMPLE function: tree gimple_omp_continue_control_use (gimple s)
13511 Return the use of the control variable on a `GIMPLE_OMP_CONTINUE'
13514 -- GIMPLE function: tree gimple_omp_continue_control_use_ptr (gimple s)
13515 Same as above, but return the pointer.
13517 -- GIMPLE function: tree gimple_omp_continue_set_control_use (gimple s)
13518 Set the control variable use for a `GIMPLE_OMP_CONTINUE' statement
13522 File: gccint.info, Node: `GIMPLE_OMP_CRITICAL', Next: `GIMPLE_OMP_FOR', Prev: `GIMPLE_OMP_CONTINUE', Up: Tuple specific accessors
13524 12.7.14 `GIMPLE_OMP_CRITICAL'
13525 -----------------------------
13527 -- GIMPLE function: gimple gimple_build_omp_critical (gimple_seq body,
13529 Build a `GIMPLE_OMP_CRITICAL' statement. `BODY' is the sequence of
13530 statements for which only one thread can execute. `NAME' is an
13531 optional identifier for this critical block.
13533 -- GIMPLE function: tree gimple_omp_critical_name (gimple g)
13534 Return the name associated with `OMP_CRITICAL' statement `G'.
13536 -- GIMPLE function: tree *gimple_omp_critical_name_ptr (gimple g)
13537 Return a pointer to the name associated with `OMP' critical
13540 -- GIMPLE function: void gimple_omp_critical_set_name (gimple g, tree
13542 Set `NAME' to be the name associated with `OMP' critical statement
13546 File: gccint.info, Node: `GIMPLE_OMP_FOR', Next: `GIMPLE_OMP_MASTER', Prev: `GIMPLE_OMP_CRITICAL', Up: Tuple specific accessors
13548 12.7.15 `GIMPLE_OMP_FOR'
13549 ------------------------
13551 -- GIMPLE function: gimple gimple_build_omp_for (gimple_seq body, tree
13552 clauses, tree index, tree initial, tree final, tree incr,
13553 gimple_seq pre_body, enum tree_code omp_for_cond)
13554 Build a `GIMPLE_OMP_FOR' statement. `BODY' is sequence of
13555 statements inside the for loop. `CLAUSES', are any of the `OMP'
13556 loop construct's clauses: private, firstprivate, lastprivate,
13557 reductions, ordered, schedule, and nowait. `PRE_BODY' is the
13558 sequence of statements that are loop invariant. `INDEX' is the
13559 index variable. `INITIAL' is the initial value of `INDEX'.
13560 `FINAL' is final value of `INDEX'. OMP_FOR_COND is the predicate
13561 used to compare `INDEX' and `FINAL'. `INCR' is the increment
13564 -- GIMPLE function: tree gimple_omp_for_clauses (gimple g)
13565 Return the clauses associated with `OMP_FOR' `G'.
13567 -- GIMPLE function: tree *gimple_omp_for_clauses_ptr (gimple g)
13568 Return a pointer to the `OMP_FOR' `G'.
13570 -- GIMPLE function: void gimple_omp_for_set_clauses (gimple g, tree
13572 Set `CLAUSES' to be the list of clauses associated with `OMP_FOR'
13575 -- GIMPLE function: tree gimple_omp_for_index (gimple g)
13576 Return the index variable for `OMP_FOR' `G'.
13578 -- GIMPLE function: tree *gimple_omp_for_index_ptr (gimple g)
13579 Return a pointer to the index variable for `OMP_FOR' `G'.
13581 -- GIMPLE function: void gimple_omp_for_set_index (gimple g, tree
13583 Set `INDEX' to be the index variable for `OMP_FOR' `G'.
13585 -- GIMPLE function: tree gimple_omp_for_initial (gimple g)
13586 Return the initial value for `OMP_FOR' `G'.
13588 -- GIMPLE function: tree *gimple_omp_for_initial_ptr (gimple g)
13589 Return a pointer to the initial value for `OMP_FOR' `G'.
13591 -- GIMPLE function: void gimple_omp_for_set_initial (gimple g, tree
13593 Set `INITIAL' to be the initial value for `OMP_FOR' `G'.
13595 -- GIMPLE function: tree gimple_omp_for_final (gimple g)
13596 Return the final value for `OMP_FOR' `G'.
13598 -- GIMPLE function: tree *gimple_omp_for_final_ptr (gimple g)
13599 turn a pointer to the final value for `OMP_FOR' `G'.
13601 -- GIMPLE function: void gimple_omp_for_set_final (gimple g, tree
13603 Set `FINAL' to be the final value for `OMP_FOR' `G'.
13605 -- GIMPLE function: tree gimple_omp_for_incr (gimple g)
13606 Return the increment value for `OMP_FOR' `G'.
13608 -- GIMPLE function: tree *gimple_omp_for_incr_ptr (gimple g)
13609 Return a pointer to the increment value for `OMP_FOR' `G'.
13611 -- GIMPLE function: void gimple_omp_for_set_incr (gimple g, tree incr)
13612 Set `INCR' to be the increment value for `OMP_FOR' `G'.
13614 -- GIMPLE function: gimple_seq gimple_omp_for_pre_body (gimple g)
13615 Return the sequence of statements to execute before the `OMP_FOR'
13616 statement `G' starts.
13618 -- GIMPLE function: void gimple_omp_for_set_pre_body (gimple g,
13619 gimple_seq pre_body)
13620 Set `PRE_BODY' to be the sequence of statements to execute before
13621 the `OMP_FOR' statement `G' starts.
13623 -- GIMPLE function: void gimple_omp_for_set_cond (gimple g, enum
13625 Set `COND' to be the condition code for `OMP_FOR' `G'.
13627 -- GIMPLE function: enum tree_code gimple_omp_for_cond (gimple g)
13628 Return the condition code associated with `OMP_FOR' `G'.
13631 File: gccint.info, Node: `GIMPLE_OMP_MASTER', Next: `GIMPLE_OMP_ORDERED', Prev: `GIMPLE_OMP_FOR', Up: Tuple specific accessors
13633 12.7.16 `GIMPLE_OMP_MASTER'
13634 ---------------------------
13636 -- GIMPLE function: gimple gimple_build_omp_master (gimple_seq body)
13637 Build a `GIMPLE_OMP_MASTER' statement. `BODY' is the sequence of
13638 statements to be executed by just the master.
13641 File: gccint.info, Node: `GIMPLE_OMP_ORDERED', Next: `GIMPLE_OMP_PARALLEL', Prev: `GIMPLE_OMP_MASTER', Up: Tuple specific accessors
13643 12.7.17 `GIMPLE_OMP_ORDERED'
13644 ----------------------------
13646 -- GIMPLE function: gimple gimple_build_omp_ordered (gimple_seq body)
13647 Build a `GIMPLE_OMP_ORDERED' statement.
13649 `BODY' is the sequence of statements inside a loop that will executed
13653 File: gccint.info, Node: `GIMPLE_OMP_PARALLEL', Next: `GIMPLE_OMP_RETURN', Prev: `GIMPLE_OMP_ORDERED', Up: Tuple specific accessors
13655 12.7.18 `GIMPLE_OMP_PARALLEL'
13656 -----------------------------
13658 -- GIMPLE function: gimple gimple_build_omp_parallel (gimple_seq body,
13659 tree clauses, tree child_fn, tree data_arg)
13660 Build a `GIMPLE_OMP_PARALLEL' statement.
13662 `BODY' is sequence of statements which are executed in parallel.
13663 `CLAUSES', are the `OMP' parallel construct's clauses. `CHILD_FN' is
13664 the function created for the parallel threads to execute. `DATA_ARG'
13665 are the shared data argument(s).
13667 -- GIMPLE function: bool gimple_omp_parallel_combined_p (gimple g)
13668 Return true if `OMP' parallel statement `G' has the
13669 `GF_OMP_PARALLEL_COMBINED' flag set.
13671 -- GIMPLE function: void gimple_omp_parallel_set_combined_p (gimple g)
13672 Set the `GF_OMP_PARALLEL_COMBINED' field in `OMP' parallel
13675 -- GIMPLE function: gimple_seq gimple_omp_body (gimple g)
13676 Return the body for the `OMP' statement `G'.
13678 -- GIMPLE function: void gimple_omp_set_body (gimple g, gimple_seq
13680 Set `BODY' to be the body for the `OMP' statement `G'.
13682 -- GIMPLE function: tree gimple_omp_parallel_clauses (gimple g)
13683 Return the clauses associated with `OMP_PARALLEL' `G'.
13685 -- GIMPLE function: tree *gimple_omp_parallel_clauses_ptr (gimple g)
13686 Return a pointer to the clauses associated with `OMP_PARALLEL' `G'.
13688 -- GIMPLE function: void gimple_omp_parallel_set_clauses (gimple g,
13690 Set `CLAUSES' to be the list of clauses associated with
13691 `OMP_PARALLEL' `G'.
13693 -- GIMPLE function: tree gimple_omp_parallel_child_fn (gimple g)
13694 Return the child function used to hold the body of `OMP_PARALLEL'
13697 -- GIMPLE function: tree *gimple_omp_parallel_child_fn_ptr (gimple g)
13698 Return a pointer to the child function used to hold the body of
13699 `OMP_PARALLEL' `G'.
13701 -- GIMPLE function: void gimple_omp_parallel_set_child_fn (gimple g,
13703 Set `CHILD_FN' to be the child function for `OMP_PARALLEL' `G'.
13705 -- GIMPLE function: tree gimple_omp_parallel_data_arg (gimple g)
13706 Return the artificial argument used to send variables and values
13707 from the parent to the children threads in `OMP_PARALLEL' `G'.
13709 -- GIMPLE function: tree *gimple_omp_parallel_data_arg_ptr (gimple g)
13710 Return a pointer to the data argument for `OMP_PARALLEL' `G'.
13712 -- GIMPLE function: void gimple_omp_parallel_set_data_arg (gimple g,
13714 Set `DATA_ARG' to be the data argument for `OMP_PARALLEL' `G'.
13716 -- GIMPLE function: bool is_gimple_omp (gimple stmt)
13717 Returns true when the gimple statement `STMT' is any of the OpenMP
13721 File: gccint.info, Node: `GIMPLE_OMP_RETURN', Next: `GIMPLE_OMP_SECTION', Prev: `GIMPLE_OMP_PARALLEL', Up: Tuple specific accessors
13723 12.7.19 `GIMPLE_OMP_RETURN'
13724 ---------------------------
13726 -- GIMPLE function: gimple gimple_build_omp_return (bool wait_p)
13727 Build a `GIMPLE_OMP_RETURN' statement. `WAIT_P' is true if this is
13728 a non-waiting return.
13730 -- GIMPLE function: void gimple_omp_return_set_nowait (gimple s)
13731 Set the nowait flag on `GIMPLE_OMP_RETURN' statement `S'.
13733 -- GIMPLE function: bool gimple_omp_return_nowait_p (gimple g)
13734 Return true if `OMP' return statement `G' has the
13735 `GF_OMP_RETURN_NOWAIT' flag set.
13738 File: gccint.info, Node: `GIMPLE_OMP_SECTION', Next: `GIMPLE_OMP_SECTIONS', Prev: `GIMPLE_OMP_RETURN', Up: Tuple specific accessors
13740 12.7.20 `GIMPLE_OMP_SECTION'
13741 ----------------------------
13743 -- GIMPLE function: gimple gimple_build_omp_section (gimple_seq body)
13744 Build a `GIMPLE_OMP_SECTION' statement for a sections statement.
13746 `BODY' is the sequence of statements in the section.
13748 -- GIMPLE function: bool gimple_omp_section_last_p (gimple g)
13749 Return true if `OMP' section statement `G' has the
13750 `GF_OMP_SECTION_LAST' flag set.
13752 -- GIMPLE function: void gimple_omp_section_set_last (gimple g)
13753 Set the `GF_OMP_SECTION_LAST' flag on `G'.
13756 File: gccint.info, Node: `GIMPLE_OMP_SECTIONS', Next: `GIMPLE_OMP_SINGLE', Prev: `GIMPLE_OMP_SECTION', Up: Tuple specific accessors
13758 12.7.21 `GIMPLE_OMP_SECTIONS'
13759 -----------------------------
13761 -- GIMPLE function: gimple gimple_build_omp_sections (gimple_seq body,
13763 Build a `GIMPLE_OMP_SECTIONS' statement. `BODY' is a sequence of
13764 section statements. `CLAUSES' are any of the `OMP' sections
13765 construct's clauses: private, firstprivate, lastprivate,
13766 reduction, and nowait.
13768 -- GIMPLE function: gimple gimple_build_omp_sections_switch (void)
13769 Build a `GIMPLE_OMP_SECTIONS_SWITCH' statement.
13771 -- GIMPLE function: tree gimple_omp_sections_control (gimple g)
13772 Return the control variable associated with the
13773 `GIMPLE_OMP_SECTIONS' in `G'.
13775 -- GIMPLE function: tree *gimple_omp_sections_control_ptr (gimple g)
13776 Return a pointer to the clauses associated with the
13777 `GIMPLE_OMP_SECTIONS' in `G'.
13779 -- GIMPLE function: void gimple_omp_sections_set_control (gimple g,
13781 Set `CONTROL' to be the set of clauses associated with the
13782 `GIMPLE_OMP_SECTIONS' in `G'.
13784 -- GIMPLE function: tree gimple_omp_sections_clauses (gimple g)
13785 Return the clauses associated with `OMP_SECTIONS' `G'.
13787 -- GIMPLE function: tree *gimple_omp_sections_clauses_ptr (gimple g)
13788 Return a pointer to the clauses associated with `OMP_SECTIONS' `G'.
13790 -- GIMPLE function: void gimple_omp_sections_set_clauses (gimple g,
13792 Set `CLAUSES' to be the set of clauses associated with
13793 `OMP_SECTIONS' `G'.
13796 File: gccint.info, Node: `GIMPLE_OMP_SINGLE', Next: `GIMPLE_PHI', Prev: `GIMPLE_OMP_SECTIONS', Up: Tuple specific accessors
13798 12.7.22 `GIMPLE_OMP_SINGLE'
13799 ---------------------------
13801 -- GIMPLE function: gimple gimple_build_omp_single (gimple_seq body,
13803 Build a `GIMPLE_OMP_SINGLE' statement. `BODY' is the sequence of
13804 statements that will be executed once. `CLAUSES' are any of the
13805 `OMP' single construct's clauses: private, firstprivate,
13806 copyprivate, nowait.
13808 -- GIMPLE function: tree gimple_omp_single_clauses (gimple g)
13809 Return the clauses associated with `OMP_SINGLE' `G'.
13811 -- GIMPLE function: tree *gimple_omp_single_clauses_ptr (gimple g)
13812 Return a pointer to the clauses associated with `OMP_SINGLE' `G'.
13814 -- GIMPLE function: void gimple_omp_single_set_clauses (gimple g, tree
13816 Set `CLAUSES' to be the clauses associated with `OMP_SINGLE' `G'.
13819 File: gccint.info, Node: `GIMPLE_PHI', Next: `GIMPLE_RESX', Prev: `GIMPLE_OMP_SINGLE', Up: Tuple specific accessors
13821 12.7.23 `GIMPLE_PHI'
13822 --------------------
13824 -- GIMPLE function: gimple make_phi_node (tree var, int len)
13825 Build a `PHI' node with len argument slots for variable var.
13827 -- GIMPLE function: unsigned gimple_phi_capacity (gimple g)
13828 Return the maximum number of arguments supported by `GIMPLE_PHI'
13831 -- GIMPLE function: unsigned gimple_phi_num_args (gimple g)
13832 Return the number of arguments in `GIMPLE_PHI' `G'. This must
13833 always be exactly the number of incoming edges for the basic block
13836 -- GIMPLE function: tree gimple_phi_result (gimple g)
13837 Return the `SSA' name created by `GIMPLE_PHI' `G'.
13839 -- GIMPLE function: tree *gimple_phi_result_ptr (gimple g)
13840 Return a pointer to the `SSA' name created by `GIMPLE_PHI' `G'.
13842 -- GIMPLE function: void gimple_phi_set_result (gimple g, tree result)
13843 Set `RESULT' to be the `SSA' name created by `GIMPLE_PHI' `G'.
13845 -- GIMPLE function: struct phi_arg_d *gimple_phi_arg (gimple g, index)
13846 Return the `PHI' argument corresponding to incoming edge `INDEX'
13847 for `GIMPLE_PHI' `G'.
13849 -- GIMPLE function: void gimple_phi_set_arg (gimple g, index, struct
13850 phi_arg_d * phiarg)
13851 Set `PHIARG' to be the argument corresponding to incoming edge
13852 `INDEX' for `GIMPLE_PHI' `G'.
13855 File: gccint.info, Node: `GIMPLE_RESX', Next: `GIMPLE_RETURN', Prev: `GIMPLE_PHI', Up: Tuple specific accessors
13857 12.7.24 `GIMPLE_RESX'
13858 ---------------------
13860 -- GIMPLE function: gimple gimple_build_resx (int region)
13861 Build a `GIMPLE_RESX' statement which is a statement. This
13862 statement is a placeholder for _Unwind_Resume before we know if a
13863 function call or a branch is needed. `REGION' is the exception
13864 region from which control is flowing.
13866 -- GIMPLE function: int gimple_resx_region (gimple g)
13867 Return the region number for `GIMPLE_RESX' `G'.
13869 -- GIMPLE function: void gimple_resx_set_region (gimple g, int region)
13870 Set `REGION' to be the region number for `GIMPLE_RESX' `G'.
13873 File: gccint.info, Node: `GIMPLE_RETURN', Next: `GIMPLE_SWITCH', Prev: `GIMPLE_RESX', Up: Tuple specific accessors
13875 12.7.25 `GIMPLE_RETURN'
13876 -----------------------
13878 -- GIMPLE function: gimple gimple_build_return (tree retval)
13879 Build a `GIMPLE_RETURN' statement whose return value is retval.
13881 -- GIMPLE function: tree gimple_return_retval (gimple g)
13882 Return the return value for `GIMPLE_RETURN' `G'.
13884 -- GIMPLE function: void gimple_return_set_retval (gimple g, tree
13886 Set `RETVAL' to be the return value for `GIMPLE_RETURN' `G'.
13889 File: gccint.info, Node: `GIMPLE_SWITCH', Next: `GIMPLE_TRY', Prev: `GIMPLE_RETURN', Up: Tuple specific accessors
13891 12.7.26 `GIMPLE_SWITCH'
13892 -----------------------
13894 -- GIMPLE function: gimple gimple_build_switch ( nlabels, tree index,
13895 tree default_label, ...)
13896 Build a `GIMPLE_SWITCH' statement. `NLABELS' are the number of
13897 labels excluding the default label. The default label is passed
13898 in `DEFAULT_LABEL'. The rest of the arguments are trees
13899 representing the labels. Each label is a tree of code
13902 -- GIMPLE function: gimple gimple_build_switch_vec (tree index, tree
13903 default_label, `VEC'(tree,heap) *args)
13904 This function is an alternate way of building `GIMPLE_SWITCH'
13905 statements. `INDEX' and `DEFAULT_LABEL' are as in
13906 gimple_build_switch. `ARGS' is a vector of `CASE_LABEL_EXPR' trees
13907 that contain the labels.
13909 -- GIMPLE function: unsigned gimple_switch_num_labels (gimple g)
13910 Return the number of labels associated with the switch statement
13913 -- GIMPLE function: void gimple_switch_set_num_labels (gimple g,
13915 Set `NLABELS' to be the number of labels for the switch statement
13918 -- GIMPLE function: tree gimple_switch_index (gimple g)
13919 Return the index variable used by the switch statement `G'.
13921 -- GIMPLE function: void gimple_switch_set_index (gimple g, tree index)
13922 Set `INDEX' to be the index variable for switch statement `G'.
13924 -- GIMPLE function: tree gimple_switch_label (gimple g, unsigned index)
13925 Return the label numbered `INDEX'. The default label is 0, followed
13926 by any labels in a switch statement.
13928 -- GIMPLE function: void gimple_switch_set_label (gimple g, unsigned
13930 Set the label number `INDEX' to `LABEL'. 0 is always the default
13933 -- GIMPLE function: tree gimple_switch_default_label (gimple g)
13934 Return the default label for a switch statement.
13936 -- GIMPLE function: void gimple_switch_set_default_label (gimple g,
13938 Set the default label for a switch statement.
13941 File: gccint.info, Node: `GIMPLE_TRY', Next: `GIMPLE_WITH_CLEANUP_EXPR', Prev: `GIMPLE_SWITCH', Up: Tuple specific accessors
13943 12.7.27 `GIMPLE_TRY'
13944 --------------------
13946 -- GIMPLE function: gimple gimple_build_try (gimple_seq eval,
13947 gimple_seq cleanup, unsigned int kind)
13948 Build a `GIMPLE_TRY' statement. `EVAL' is a sequence with the
13949 expression to evaluate. `CLEANUP' is a sequence of statements to
13950 run at clean-up time. `KIND' is the enumeration value
13951 `GIMPLE_TRY_CATCH' if this statement denotes a try/catch construct
13952 or `GIMPLE_TRY_FINALLY' if this statement denotes a try/finally
13955 -- GIMPLE function: enum gimple_try_flags gimple_try_kind (gimple g)
13956 Return the kind of try block represented by `GIMPLE_TRY' `G'. This
13957 is either `GIMPLE_TRY_CATCH' or `GIMPLE_TRY_FINALLY'.
13959 -- GIMPLE function: bool gimple_try_catch_is_cleanup (gimple g)
13960 Return the `GIMPLE_TRY_CATCH_IS_CLEANUP' flag.
13962 -- GIMPLE function: gimple_seq gimple_try_eval (gimple g)
13963 Return the sequence of statements used as the body for `GIMPLE_TRY'
13966 -- GIMPLE function: gimple_seq gimple_try_cleanup (gimple g)
13967 Return the sequence of statements used as the cleanup body for
13970 -- GIMPLE function: void gimple_try_set_catch_is_cleanup (gimple g,
13971 bool catch_is_cleanup)
13972 Set the `GIMPLE_TRY_CATCH_IS_CLEANUP' flag.
13974 -- GIMPLE function: void gimple_try_set_eval (gimple g, gimple_seq
13976 Set `EVAL' to be the sequence of statements to use as the body for
13979 -- GIMPLE function: void gimple_try_set_cleanup (gimple g, gimple_seq
13981 Set `CLEANUP' to be the sequence of statements to use as the
13982 cleanup body for `GIMPLE_TRY' `G'.
13985 File: gccint.info, Node: `GIMPLE_WITH_CLEANUP_EXPR', Prev: `GIMPLE_TRY', Up: Tuple specific accessors
13987 12.7.28 `GIMPLE_WITH_CLEANUP_EXPR'
13988 ----------------------------------
13990 -- GIMPLE function: gimple gimple_build_wce (gimple_seq cleanup)
13991 Build a `GIMPLE_WITH_CLEANUP_EXPR' statement. `CLEANUP' is the
13992 clean-up expression.
13994 -- GIMPLE function: gimple_seq gimple_wce_cleanup (gimple g)
13995 Return the cleanup sequence for cleanup statement `G'.
13997 -- GIMPLE function: void gimple_wce_set_cleanup (gimple g, gimple_seq
13999 Set `CLEANUP' to be the cleanup sequence for `G'.
14001 -- GIMPLE function: bool gimple_wce_cleanup_eh_only (gimple g)
14002 Return the `CLEANUP_EH_ONLY' flag for a `WCE' tuple.
14004 -- GIMPLE function: void gimple_wce_set_cleanup_eh_only (gimple g,
14006 Set the `CLEANUP_EH_ONLY' flag for a `WCE' tuple.
14009 File: gccint.info, Node: GIMPLE sequences, Next: Sequence iterators, Prev: Tuple specific accessors, Up: GIMPLE
14011 12.8 GIMPLE sequences
14012 =====================
14014 GIMPLE sequences are the tuple equivalent of `STATEMENT_LIST''s used in
14015 `GENERIC'. They are used to chain statements together, and when used
14016 in conjunction with sequence iterators, provide a framework for
14017 iterating through statements.
14019 GIMPLE sequences are of type struct `gimple_sequence', but are more
14020 commonly passed by reference to functions dealing with sequences. The
14021 type for a sequence pointer is `gimple_seq' which is the same as struct
14022 `gimple_sequence' *. When declaring a local sequence, you can define a
14023 local variable of type struct `gimple_sequence'. When declaring a
14024 sequence allocated on the garbage collected heap, use the function
14025 `gimple_seq_alloc' documented below.
14027 There are convenience functions for iterating through sequences in the
14028 section entitled Sequence Iterators.
14030 Below is a list of functions to manipulate and query sequences.
14032 -- GIMPLE function: void gimple_seq_add_stmt (gimple_seq *seq, gimple
14034 Link a gimple statement to the end of the sequence *`SEQ' if `G' is
14035 not `NULL'. If *`SEQ' is `NULL', allocate a sequence before
14038 -- GIMPLE function: void gimple_seq_add_seq (gimple_seq *dest,
14040 Append sequence `SRC' to the end of sequence *`DEST' if `SRC' is
14041 not `NULL'. If *`DEST' is `NULL', allocate a new sequence before
14044 -- GIMPLE function: gimple_seq gimple_seq_deep_copy (gimple_seq src)
14045 Perform a deep copy of sequence `SRC' and return the result.
14047 -- GIMPLE function: gimple_seq gimple_seq_reverse (gimple_seq seq)
14048 Reverse the order of the statements in the sequence `SEQ'. Return
14051 -- GIMPLE function: gimple gimple_seq_first (gimple_seq s)
14052 Return the first statement in sequence `S'.
14054 -- GIMPLE function: gimple gimple_seq_last (gimple_seq s)
14055 Return the last statement in sequence `S'.
14057 -- GIMPLE function: void gimple_seq_set_last (gimple_seq s, gimple
14059 Set the last statement in sequence `S' to the statement in `LAST'.
14061 -- GIMPLE function: void gimple_seq_set_first (gimple_seq s, gimple
14063 Set the first statement in sequence `S' to the statement in
14066 -- GIMPLE function: void gimple_seq_init (gimple_seq s)
14067 Initialize sequence `S' to an empty sequence.
14069 -- GIMPLE function: gimple_seq gimple_seq_alloc (void)
14070 Allocate a new sequence in the garbage collected store and return
14073 -- GIMPLE function: void gimple_seq_copy (gimple_seq dest, gimple_seq
14075 Copy the sequence `SRC' into the sequence `DEST'.
14077 -- GIMPLE function: bool gimple_seq_empty_p (gimple_seq s)
14078 Return true if the sequence `S' is empty.
14080 -- GIMPLE function: gimple_seq bb_seq (basic_block bb)
14081 Returns the sequence of statements in `BB'.
14083 -- GIMPLE function: void set_bb_seq (basic_block bb, gimple_seq seq)
14084 Sets the sequence of statements in `BB' to `SEQ'.
14086 -- GIMPLE function: bool gimple_seq_singleton_p (gimple_seq seq)
14087 Determine whether `SEQ' contains exactly one statement.
14090 File: gccint.info, Node: Sequence iterators, Next: Adding a new GIMPLE statement code, Prev: GIMPLE sequences, Up: GIMPLE
14092 12.9 Sequence iterators
14093 =======================
14095 Sequence iterators are convenience constructs for iterating through
14096 statements in a sequence. Given a sequence `SEQ', here is a typical
14097 use of gimple sequence iterators:
14099 gimple_stmt_iterator gsi;
14101 for (gsi = gsi_start (seq); !gsi_end_p (gsi); gsi_next (&gsi))
14103 gimple g = gsi_stmt (gsi);
14104 /* Do something with gimple statement `G'. */
14107 Backward iterations are possible:
14109 for (gsi = gsi_last (seq); !gsi_end_p (gsi); gsi_prev (&gsi))
14111 Forward and backward iterations on basic blocks are possible with
14112 `gsi_start_bb' and `gsi_last_bb'.
14114 In the documentation below we sometimes refer to enum
14115 `gsi_iterator_update'. The valid options for this enumeration are:
14117 * `GSI_NEW_STMT' Only valid when a single statement is added. Move
14118 the iterator to it.
14120 * `GSI_SAME_STMT' Leave the iterator at the same statement.
14122 * `GSI_CONTINUE_LINKING' Move iterator to whatever position is
14123 suitable for linking other statements in the same direction.
14125 Below is a list of the functions used to manipulate and use statement
14128 -- GIMPLE function: gimple_stmt_iterator gsi_start (gimple_seq seq)
14129 Return a new iterator pointing to the sequence `SEQ''s first
14130 statement. If `SEQ' is empty, the iterator's basic block is
14131 `NULL'. Use `gsi_start_bb' instead when the iterator needs to
14132 always have the correct basic block set.
14134 -- GIMPLE function: gimple_stmt_iterator gsi_start_bb (basic_block bb)
14135 Return a new iterator pointing to the first statement in basic
14138 -- GIMPLE function: gimple_stmt_iterator gsi_last (gimple_seq seq)
14139 Return a new iterator initially pointing to the last statement of
14140 sequence `SEQ'. If `SEQ' is empty, the iterator's basic block is
14141 `NULL'. Use `gsi_last_bb' instead when the iterator needs to
14142 always have the correct basic block set.
14144 -- GIMPLE function: gimple_stmt_iterator gsi_last_bb (basic_block bb)
14145 Return a new iterator pointing to the last statement in basic
14148 -- GIMPLE function: bool gsi_end_p (gimple_stmt_iterator i)
14149 Return `TRUE' if at the end of `I'.
14151 -- GIMPLE function: bool gsi_one_before_end_p (gimple_stmt_iterator i)
14152 Return `TRUE' if we're one statement before the end of `I'.
14154 -- GIMPLE function: void gsi_next (gimple_stmt_iterator *i)
14155 Advance the iterator to the next gimple statement.
14157 -- GIMPLE function: void gsi_prev (gimple_stmt_iterator *i)
14158 Advance the iterator to the previous gimple statement.
14160 -- GIMPLE function: gimple gsi_stmt (gimple_stmt_iterator i)
14161 Return the current stmt.
14163 -- GIMPLE function: gimple_stmt_iterator gsi_after_labels (basic_block
14165 Return a block statement iterator that points to the first
14166 non-label statement in block `BB'.
14168 -- GIMPLE function: gimple *gsi_stmt_ptr (gimple_stmt_iterator *i)
14169 Return a pointer to the current stmt.
14171 -- GIMPLE function: basic_block gsi_bb (gimple_stmt_iterator i)
14172 Return the basic block associated with this iterator.
14174 -- GIMPLE function: gimple_seq gsi_seq (gimple_stmt_iterator i)
14175 Return the sequence associated with this iterator.
14177 -- GIMPLE function: void gsi_remove (gimple_stmt_iterator *i, bool
14179 Remove the current stmt from the sequence. The iterator is
14180 updated to point to the next statement. When `REMOVE_EH_INFO' is
14181 true we remove the statement pointed to by iterator `I' from the
14182 `EH' tables. Otherwise we do not modify the `EH' tables.
14183 Generally, `REMOVE_EH_INFO' should be true when the statement is
14184 going to be removed from the `IL' and not reinserted elsewhere.
14186 -- GIMPLE function: void gsi_link_seq_before (gimple_stmt_iterator *i,
14187 gimple_seq seq, enum gsi_iterator_update mode)
14188 Links the sequence of statements `SEQ' before the statement pointed
14189 by iterator `I'. `MODE' indicates what to do with the iterator
14190 after insertion (see `enum gsi_iterator_update' above).
14192 -- GIMPLE function: void gsi_link_before (gimple_stmt_iterator *i,
14193 gimple g, enum gsi_iterator_update mode)
14194 Links statement `G' before the statement pointed-to by iterator
14195 `I'. Updates iterator `I' according to `MODE'.
14197 -- GIMPLE function: void gsi_link_seq_after (gimple_stmt_iterator *i,
14198 gimple_seq seq, enum gsi_iterator_update mode)
14199 Links sequence `SEQ' after the statement pointed-to by iterator
14200 `I'. `MODE' is as in `gsi_insert_after'.
14202 -- GIMPLE function: void gsi_link_after (gimple_stmt_iterator *i,
14203 gimple g, enum gsi_iterator_update mode)
14204 Links statement `G' after the statement pointed-to by iterator `I'.
14205 `MODE' is as in `gsi_insert_after'.
14207 -- GIMPLE function: gimple_seq gsi_split_seq_after
14208 (gimple_stmt_iterator i)
14209 Move all statements in the sequence after `I' to a new sequence.
14210 Return this new sequence.
14212 -- GIMPLE function: gimple_seq gsi_split_seq_before
14213 (gimple_stmt_iterator *i)
14214 Move all statements in the sequence before `I' to a new sequence.
14215 Return this new sequence.
14217 -- GIMPLE function: void gsi_replace (gimple_stmt_iterator *i, gimple
14218 stmt, bool update_eh_info)
14219 Replace the statement pointed-to by `I' to `STMT'. If
14220 `UPDATE_EH_INFO' is true, the exception handling information of
14221 the original statement is moved to the new statement.
14223 -- GIMPLE function: void gsi_insert_before (gimple_stmt_iterator *i,
14224 gimple stmt, enum gsi_iterator_update mode)
14225 Insert statement `STMT' before the statement pointed-to by iterator
14226 `I', update `STMT''s basic block and scan it for new operands.
14227 `MODE' specifies how to update iterator `I' after insertion (see
14228 enum `gsi_iterator_update').
14230 -- GIMPLE function: void gsi_insert_seq_before (gimple_stmt_iterator
14231 *i, gimple_seq seq, enum gsi_iterator_update mode)
14232 Like `gsi_insert_before', but for all the statements in `SEQ'.
14234 -- GIMPLE function: void gsi_insert_after (gimple_stmt_iterator *i,
14235 gimple stmt, enum gsi_iterator_update mode)
14236 Insert statement `STMT' after the statement pointed-to by iterator
14237 `I', update `STMT''s basic block and scan it for new operands.
14238 `MODE' specifies how to update iterator `I' after insertion (see
14239 enum `gsi_iterator_update').
14241 -- GIMPLE function: void gsi_insert_seq_after (gimple_stmt_iterator
14242 *i, gimple_seq seq, enum gsi_iterator_update mode)
14243 Like `gsi_insert_after', but for all the statements in `SEQ'.
14245 -- GIMPLE function: gimple_stmt_iterator gsi_for_stmt (gimple stmt)
14246 Finds iterator for `STMT'.
14248 -- GIMPLE function: void gsi_move_after (gimple_stmt_iterator *from,
14249 gimple_stmt_iterator *to)
14250 Move the statement at `FROM' so it comes right after the statement
14253 -- GIMPLE function: void gsi_move_before (gimple_stmt_iterator *from,
14254 gimple_stmt_iterator *to)
14255 Move the statement at `FROM' so it comes right before the statement
14258 -- GIMPLE function: void gsi_move_to_bb_end (gimple_stmt_iterator
14259 *from, basic_block bb)
14260 Move the statement at `FROM' to the end of basic block `BB'.
14262 -- GIMPLE function: void gsi_insert_on_edge (edge e, gimple stmt)
14263 Add `STMT' to the pending list of edge `E'. No actual insertion is
14264 made until a call to `gsi_commit_edge_inserts'() is made.
14266 -- GIMPLE function: void gsi_insert_seq_on_edge (edge e, gimple_seq
14268 Add the sequence of statements in `SEQ' to the pending list of edge
14269 `E'. No actual insertion is made until a call to
14270 `gsi_commit_edge_inserts'() is made.
14272 -- GIMPLE function: basic_block gsi_insert_on_edge_immediate (edge e,
14274 Similar to `gsi_insert_on_edge'+`gsi_commit_edge_inserts'. If a
14275 new block has to be created, it is returned.
14277 -- GIMPLE function: void gsi_commit_one_edge_insert (edge e,
14278 basic_block *new_bb)
14279 Commit insertions pending at edge `E'. If a new block is created,
14280 set `NEW_BB' to this block, otherwise set it to `NULL'.
14282 -- GIMPLE function: void gsi_commit_edge_inserts (void)
14283 This routine will commit all pending edge insertions, creating any
14284 new basic blocks which are necessary.
14287 File: gccint.info, Node: Adding a new GIMPLE statement code, Next: Statement and operand traversals, Prev: Sequence iterators, Up: GIMPLE
14289 12.10 Adding a new GIMPLE statement code
14290 ========================================
14292 The first step in adding a new GIMPLE statement code, is modifying the
14293 file `gimple.def', which contains all the GIMPLE codes. Then you must
14294 add a corresponding structure, and an entry in `union
14295 gimple_statement_d', both of which are located in `gimple.h'. This in
14296 turn, will require you to add a corresponding `GTY' tag in
14297 `gsstruct.def', and code to handle this tag in `gss_for_code' which is
14298 located in `gimple.c'.
14300 In order for the garbage collector to know the size of the structure
14301 you created in `gimple.h', you need to add a case to handle your new
14302 GIMPLE statement in `gimple_size' which is located in `gimple.c'.
14304 You will probably want to create a function to build the new gimple
14305 statement in `gimple.c'. The function should be called
14306 `gimple_build_<`NEW_TUPLE_NAME'>', and should return the new tuple of
14309 If your new statement requires accessors for any members or operands
14310 it may have, put simple inline accessors in `gimple.h' and any
14311 non-trivial accessors in `gimple.c' with a corresponding prototype in
14315 File: gccint.info, Node: Statement and operand traversals, Prev: Adding a new GIMPLE statement code, Up: GIMPLE
14317 12.11 Statement and operand traversals
14318 ======================================
14320 There are two functions available for walking statements and sequences:
14321 `walk_gimple_stmt' and `walk_gimple_seq', accordingly, and a third
14322 function for walking the operands in a statement: `walk_gimple_op'.
14324 -- GIMPLE function: tree walk_gimple_stmt (gimple_stmt_iterator *gsi,
14325 walk_stmt_fn callback_stmt, walk_tree_fn callback_op, struct
14326 walk_stmt_info *wi)
14327 This function is used to walk the current statement in `GSI',
14328 optionally using traversal state stored in `WI'. If `WI' is
14329 `NULL', no state is kept during the traversal.
14331 The callback `CALLBACK_STMT' is called. If `CALLBACK_STMT' returns
14332 true, it means that the callback function has handled all the
14333 operands of the statement and it is not necessary to walk its
14336 If `CALLBACK_STMT' is `NULL' or it returns false, `CALLBACK_OP' is
14337 called on each operand of the statement via `walk_gimple_op'. If
14338 `walk_gimple_op' returns non-`NULL' for any operand, the remaining
14339 operands are not scanned.
14341 The return value is that returned by the last call to
14342 `walk_gimple_op', or `NULL_TREE' if no `CALLBACK_OP' is specified.
14344 -- GIMPLE function: tree walk_gimple_op (gimple stmt, walk_tree_fn
14345 callback_op, struct walk_stmt_info *wi)
14346 Use this function to walk the operands of statement `STMT'. Every
14347 operand is walked via `walk_tree' with optional state information
14350 `CALLBACK_OP' is called on each operand of `STMT' via `walk_tree'.
14351 Additional parameters to `walk_tree' must be stored in `WI'. For
14352 each operand `OP', `walk_tree' is called as:
14354 walk_tree (&`OP', `CALLBACK_OP', `WI', `WI'- `PSET')
14356 If `CALLBACK_OP' returns non-`NULL' for an operand, the remaining
14357 operands are not scanned. The return value is that returned by
14358 the last call to `walk_tree', or `NULL_TREE' if no `CALLBACK_OP' is
14361 -- GIMPLE function: tree walk_gimple_seq (gimple_seq seq, walk_stmt_fn
14362 callback_stmt, walk_tree_fn callback_op, struct
14363 walk_stmt_info *wi)
14364 This function walks all the statements in the sequence `SEQ'
14365 calling `walk_gimple_stmt' on each one. `WI' is as in
14366 `walk_gimple_stmt'. If `walk_gimple_stmt' returns non-`NULL', the
14367 walk is stopped and the value returned. Otherwise, all the
14368 statements are walked and `NULL_TREE' returned.
14371 File: gccint.info, Node: Tree SSA, Next: Control Flow, Prev: GIMPLE, Up: Top
14373 13 Analysis and Optimization of GIMPLE tuples
14374 *********************************************
14376 GCC uses three main intermediate languages to represent the program
14377 during compilation: GENERIC, GIMPLE and RTL. GENERIC is a
14378 language-independent representation generated by each front end. It is
14379 used to serve as an interface between the parser and optimizer.
14380 GENERIC is a common representation that is able to represent programs
14381 written in all the languages supported by GCC.
14383 GIMPLE and RTL are used to optimize the program. GIMPLE is used for
14384 target and language independent optimizations (e.g., inlining, constant
14385 propagation, tail call elimination, redundancy elimination, etc). Much
14386 like GENERIC, GIMPLE is a language independent, tree based
14387 representation. However, it differs from GENERIC in that the GIMPLE
14388 grammar is more restrictive: expressions contain no more than 3
14389 operands (except function calls), it has no control flow structures and
14390 expressions with side-effects are only allowed on the right hand side
14391 of assignments. See the chapter describing GENERIC and GIMPLE for more
14394 This chapter describes the data structures and functions used in the
14395 GIMPLE optimizers (also known as "tree optimizers" or "middle end").
14396 In particular, it focuses on all the macros, data structures, functions
14397 and programming constructs needed to implement optimization passes for
14402 * Annotations:: Attributes for variables.
14403 * SSA Operands:: SSA names referenced by GIMPLE statements.
14404 * SSA:: Static Single Assignment representation.
14405 * Alias analysis:: Representing aliased loads and stores.
14408 File: gccint.info, Node: Annotations, Next: SSA Operands, Up: Tree SSA
14413 The optimizers need to associate attributes with variables during the
14414 optimization process. For instance, we need to know whether a variable
14415 has aliases. All these attributes are stored in data structures called
14416 annotations which are then linked to the field `ann' in `struct
14419 Presently, we define annotations for variables (`var_ann_t').
14420 Annotations are defined and documented in `tree-flow.h'.
14423 File: gccint.info, Node: SSA Operands, Next: SSA, Prev: Annotations, Up: Tree SSA
14428 Almost every GIMPLE statement will contain a reference to a variable or
14429 memory location. Since statements come in different shapes and sizes,
14430 their operands are going to be located at various spots inside the
14431 statement's tree. To facilitate access to the statement's operands,
14432 they are organized into lists associated inside each statement's
14433 annotation. Each element in an operand list is a pointer to a
14434 `VAR_DECL', `PARM_DECL' or `SSA_NAME' tree node. This provides a very
14435 convenient way of examining and replacing operands.
14437 Data flow analysis and optimization is done on all tree nodes
14438 representing variables. Any node for which `SSA_VAR_P' returns nonzero
14439 is considered when scanning statement operands. However, not all
14440 `SSA_VAR_P' variables are processed in the same way. For the purposes
14441 of optimization, we need to distinguish between references to local
14442 scalar variables and references to globals, statics, structures,
14443 arrays, aliased variables, etc. The reason is simple, the compiler can
14444 gather complete data flow information for a local scalar. On the other
14445 hand, a global variable may be modified by a function call, it may not
14446 be possible to keep track of all the elements of an array or the fields
14447 of a structure, etc.
14449 The operand scanner gathers two kinds of operands: "real" and
14450 "virtual". An operand for which `is_gimple_reg' returns true is
14451 considered real, otherwise it is a virtual operand. We also
14452 distinguish between uses and definitions. An operand is used if its
14453 value is loaded by the statement (e.g., the operand at the RHS of an
14454 assignment). If the statement assigns a new value to the operand, the
14455 operand is considered a definition (e.g., the operand at the LHS of an
14458 Virtual and real operands also have very different data flow
14459 properties. Real operands are unambiguous references to the full
14460 object that they represent. For instance, given
14467 Since `a' and `b' are non-aliased locals, the statement `a = b' will
14468 have one real definition and one real use because variable `b' is
14469 completely modified with the contents of variable `a'. Real definition
14470 are also known as "killing definitions". Similarly, the use of `a'
14471 reads all its bits.
14473 In contrast, virtual operands are used with variables that can have a
14474 partial or ambiguous reference. This includes structures, arrays,
14475 globals, and aliased variables. In these cases, we have two types of
14476 definitions. For globals, structures, and arrays, we can determine from
14477 a statement whether a variable of these types has a killing definition.
14478 If the variable does, then the statement is marked as having a "must
14479 definition" of that variable. However, if a statement is only defining
14480 a part of the variable (i.e. a field in a structure), or if we know
14481 that a statement might define the variable but we cannot say for sure,
14482 then we mark that statement as having a "may definition". For
14496 The assignment `*p = 5' may be a definition of `a' or `b'. If we
14497 cannot determine statically where `p' is pointing to at the time of the
14498 store operation, we create virtual definitions to mark that statement
14499 as a potential definition site for `a' and `b'. Memory loads are
14500 similarly marked with virtual use operands. Virtual operands are shown
14501 in tree dumps right before the statement that contains them. To
14502 request a tree dump with virtual operands, use the `-vops' option to
14521 Notice that `VDEF' operands have two copies of the referenced
14522 variable. This indicates that this is not a killing definition of that
14523 variable. In this case we refer to it as a "may definition" or
14524 "aliased store". The presence of the second copy of the variable in
14525 the `VDEF' operand will become important when the function is converted
14526 into SSA form. This will be used to link all the non-killing
14527 definitions to prevent optimizations from making incorrect assumptions
14530 Operands are updated as soon as the statement is finished via a call
14531 to `update_stmt'. If statement elements are changed via `SET_USE' or
14532 `SET_DEF', then no further action is required (i.e., those macros take
14533 care of updating the statement). If changes are made by manipulating
14534 the statement's tree directly, then a call must be made to
14535 `update_stmt' when complete. Calling one of the `bsi_insert' routines
14536 or `bsi_replace' performs an implicit call to `update_stmt'.
14538 13.2.1 Operand Iterators And Access Routines
14539 --------------------------------------------
14541 Operands are collected by `tree-ssa-operands.c'. They are stored
14542 inside each statement's annotation and can be accessed through either
14543 the operand iterators or an access routine.
14545 The following access routines are available for examining operands:
14547 1. `SINGLE_SSA_{USE,DEF,TREE}_OPERAND': These accessors will return
14548 NULL unless there is exactly one operand matching the specified
14549 flags. If there is exactly one operand, the operand is returned
14550 as either a `tree', `def_operand_p', or `use_operand_p'.
14552 tree t = SINGLE_SSA_TREE_OPERAND (stmt, flags);
14553 use_operand_p u = SINGLE_SSA_USE_OPERAND (stmt, SSA_ALL_VIRTUAL_USES);
14554 def_operand_p d = SINGLE_SSA_DEF_OPERAND (stmt, SSA_OP_ALL_DEFS);
14556 2. `ZERO_SSA_OPERANDS': This macro returns true if there are no
14557 operands matching the specified flags.
14559 if (ZERO_SSA_OPERANDS (stmt, SSA_OP_ALL_VIRTUALS))
14562 3. `NUM_SSA_OPERANDS': This macro Returns the number of operands
14563 matching 'flags'. This actually executes a loop to perform the
14564 count, so only use this if it is really needed.
14566 int count = NUM_SSA_OPERANDS (stmt, flags)
14568 If you wish to iterate over some or all operands, use the
14569 `FOR_EACH_SSA_{USE,DEF,TREE}_OPERAND' iterator. For example, to print
14570 all the operands for a statement:
14573 print_ops (tree stmt)
14578 FOR_EACH_SSA_TREE_OPERAND (var, stmt, iter, SSA_OP_ALL_OPERANDS)
14579 print_generic_expr (stderr, var, TDF_SLIM);
14582 How to choose the appropriate iterator:
14584 1. Determine whether you are need to see the operand pointers, or
14585 just the trees, and choose the appropriate macro:
14589 use_operand_p FOR_EACH_SSA_USE_OPERAND
14590 def_operand_p FOR_EACH_SSA_DEF_OPERAND
14591 tree FOR_EACH_SSA_TREE_OPERAND
14593 2. You need to declare a variable of the type you are interested in,
14594 and an ssa_op_iter structure which serves as the loop controlling
14597 3. Determine which operands you wish to use, and specify the flags of
14598 those you are interested in. They are documented in
14599 `tree-ssa-operands.h':
14601 #define SSA_OP_USE 0x01 /* Real USE operands. */
14602 #define SSA_OP_DEF 0x02 /* Real DEF operands. */
14603 #define SSA_OP_VUSE 0x04 /* VUSE operands. */
14604 #define SSA_OP_VMAYUSE 0x08 /* USE portion of VDEFS. */
14605 #define SSA_OP_VDEF 0x10 /* DEF portion of VDEFS. */
14607 /* These are commonly grouped operand flags. */
14608 #define SSA_OP_VIRTUAL_USES (SSA_OP_VUSE | SSA_OP_VMAYUSE)
14609 #define SSA_OP_VIRTUAL_DEFS (SSA_OP_VDEF)
14610 #define SSA_OP_ALL_USES (SSA_OP_VIRTUAL_USES | SSA_OP_USE)
14611 #define SSA_OP_ALL_DEFS (SSA_OP_VIRTUAL_DEFS | SSA_OP_DEF)
14612 #define SSA_OP_ALL_OPERANDS (SSA_OP_ALL_USES | SSA_OP_ALL_DEFS)
14614 So if you want to look at the use pointers for all the `USE' and
14615 `VUSE' operands, you would do something like:
14617 use_operand_p use_p;
14620 FOR_EACH_SSA_USE_OPERAND (use_p, stmt, iter, (SSA_OP_USE | SSA_OP_VUSE))
14622 process_use_ptr (use_p);
14625 The `TREE' macro is basically the same as the `USE' and `DEF' macros,
14626 only with the use or def dereferenced via `USE_FROM_PTR (use_p)' and
14627 `DEF_FROM_PTR (def_p)'. Since we aren't using operand pointers, use
14628 and defs flags can be mixed.
14633 FOR_EACH_SSA_TREE_OPERAND (var, stmt, iter, SSA_OP_VUSE)
14635 print_generic_expr (stderr, var, TDF_SLIM);
14638 `VDEF's are broken into two flags, one for the `DEF' portion
14639 (`SSA_OP_VDEF') and one for the USE portion (`SSA_OP_VMAYUSE'). If all
14640 you want to look at are the `VDEF's together, there is a fourth
14641 iterator macro for this, which returns both a def_operand_p and a
14642 use_operand_p for each `VDEF' in the statement. Note that you don't
14643 need any flags for this one.
14645 use_operand_p use_p;
14646 def_operand_p def_p;
14649 FOR_EACH_SSA_MAYDEF_OPERAND (def_p, use_p, stmt, iter)
14654 There are many examples in the code as well, as well as the
14655 documentation in `tree-ssa-operands.h'.
14657 There are also a couple of variants on the stmt iterators regarding PHI
14660 `FOR_EACH_PHI_ARG' Works exactly like `FOR_EACH_SSA_USE_OPERAND',
14661 except it works over `PHI' arguments instead of statement operands.
14663 /* Look at every virtual PHI use. */
14664 FOR_EACH_PHI_ARG (use_p, phi_stmt, iter, SSA_OP_VIRTUAL_USES)
14669 /* Look at every real PHI use. */
14670 FOR_EACH_PHI_ARG (use_p, phi_stmt, iter, SSA_OP_USES)
14673 /* Look at every PHI use. */
14674 FOR_EACH_PHI_ARG (use_p, phi_stmt, iter, SSA_OP_ALL_USES)
14677 `FOR_EACH_PHI_OR_STMT_{USE,DEF}' works exactly like
14678 `FOR_EACH_SSA_{USE,DEF}_OPERAND', except it will function on either a
14679 statement or a `PHI' node. These should be used when it is appropriate
14680 but they are not quite as efficient as the individual `FOR_EACH_PHI'
14681 and `FOR_EACH_SSA' routines.
14683 FOR_EACH_PHI_OR_STMT_USE (use_operand_p, stmt, iter, flags)
14688 FOR_EACH_PHI_OR_STMT_DEF (def_operand_p, phi, iter, flags)
14693 13.2.2 Immediate Uses
14694 ---------------------
14696 Immediate use information is now always available. Using the immediate
14697 use iterators, you may examine every use of any `SSA_NAME'. For
14698 instance, to change each use of `ssa_var' to `ssa_var2' and call
14699 fold_stmt on each stmt after that is done:
14701 use_operand_p imm_use_p;
14702 imm_use_iterator iterator;
14703 tree ssa_var, stmt;
14706 FOR_EACH_IMM_USE_STMT (stmt, iterator, ssa_var)
14708 FOR_EACH_IMM_USE_ON_STMT (imm_use_p, iterator)
14709 SET_USE (imm_use_p, ssa_var_2);
14713 There are 2 iterators which can be used. `FOR_EACH_IMM_USE_FAST' is
14714 used when the immediate uses are not changed, i.e., you are looking at
14715 the uses, but not setting them.
14717 If they do get changed, then care must be taken that things are not
14718 changed under the iterators, so use the `FOR_EACH_IMM_USE_STMT' and
14719 `FOR_EACH_IMM_USE_ON_STMT' iterators. They attempt to preserve the
14720 sanity of the use list by moving all the uses for a statement into a
14721 controlled position, and then iterating over those uses. Then the
14722 optimization can manipulate the stmt when all the uses have been
14723 processed. This is a little slower than the FAST version since it adds
14724 a placeholder element and must sort through the list a bit for each
14725 statement. This placeholder element must be also be removed if the
14726 loop is terminated early. The macro `BREAK_FROM_IMM_USE_SAFE' is
14727 provided to do this :
14729 FOR_EACH_IMM_USE_STMT (stmt, iterator, ssa_var)
14731 if (stmt == last_stmt)
14732 BREAK_FROM_SAFE_IMM_USE (iter);
14734 FOR_EACH_IMM_USE_ON_STMT (imm_use_p, iterator)
14735 SET_USE (imm_use_p, ssa_var_2);
14739 There are checks in `verify_ssa' which verify that the immediate use
14740 list is up to date, as well as checking that an optimization didn't
14741 break from the loop without using this macro. It is safe to simply
14742 'break'; from a `FOR_EACH_IMM_USE_FAST' traverse.
14744 Some useful functions and macros:
14745 1. `has_zero_uses (ssa_var)' : Returns true if there are no uses of
14748 2. `has_single_use (ssa_var)' : Returns true if there is only a
14749 single use of `ssa_var'.
14751 3. `single_imm_use (ssa_var, use_operand_p *ptr, tree *stmt)' :
14752 Returns true if there is only a single use of `ssa_var', and also
14753 returns the use pointer and statement it occurs in, in the second
14754 and third parameters.
14756 4. `num_imm_uses (ssa_var)' : Returns the number of immediate uses of
14757 `ssa_var'. It is better not to use this if possible since it simply
14758 utilizes a loop to count the uses.
14760 5. `PHI_ARG_INDEX_FROM_USE (use_p)' : Given a use within a `PHI'
14761 node, return the index number for the use. An assert is triggered
14762 if the use isn't located in a `PHI' node.
14764 6. `USE_STMT (use_p)' : Return the statement a use occurs in.
14766 Note that uses are not put into an immediate use list until their
14767 statement is actually inserted into the instruction stream via a
14770 It is also still possible to utilize lazy updating of statements, but
14771 this should be used only when absolutely required. Both alias analysis
14772 and the dominator optimizations currently do this.
14774 When lazy updating is being used, the immediate use information is out
14775 of date and cannot be used reliably. Lazy updating is achieved by
14776 simply marking statements modified via calls to `mark_stmt_modified'
14777 instead of `update_stmt'. When lazy updating is no longer required,
14778 all the modified statements must have `update_stmt' called in order to
14779 bring them up to date. This must be done before the optimization is
14780 finished, or `verify_ssa' will trigger an abort.
14782 This is done with a simple loop over the instruction stream:
14783 block_stmt_iterator bsi;
14787 for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
14788 update_stmt_if_modified (bsi_stmt (bsi));
14792 File: gccint.info, Node: SSA, Next: Alias analysis, Prev: SSA Operands, Up: Tree SSA
14794 13.3 Static Single Assignment
14795 =============================
14797 Most of the tree optimizers rely on the data flow information provided
14798 by the Static Single Assignment (SSA) form. We implement the SSA form
14799 as described in `R. Cytron, J. Ferrante, B. Rosen, M. Wegman, and K.
14800 Zadeck. Efficiently Computing Static Single Assignment Form and the
14801 Control Dependence Graph. ACM Transactions on Programming Languages
14802 and Systems, 13(4):451-490, October 1991'.
14804 The SSA form is based on the premise that program variables are
14805 assigned in exactly one location in the program. Multiple assignments
14806 to the same variable create new versions of that variable. Naturally,
14807 actual programs are seldom in SSA form initially because variables tend
14808 to be assigned multiple times. The compiler modifies the program
14809 representation so that every time a variable is assigned in the code, a
14810 new version of the variable is created. Different versions of the same
14811 variable are distinguished by subscripting the variable name with its
14812 version number. Variables used in the right-hand side of expressions
14813 are renamed so that their version number matches that of the most
14816 We represent variable versions using `SSA_NAME' nodes. The renaming
14817 process in `tree-ssa.c' wraps every real and virtual operand with an
14818 `SSA_NAME' node which contains the version number and the statement
14819 that created the `SSA_NAME'. Only definitions and virtual definitions
14820 may create new `SSA_NAME' nodes.
14822 Sometimes, flow of control makes it impossible to determine the most
14823 recent version of a variable. In these cases, the compiler inserts an
14824 artificial definition for that variable called "PHI function" or "PHI
14825 node". This new definition merges all the incoming versions of the
14826 variable to create a new name for it. For instance,
14835 # a_4 = PHI <a_1, a_2, a_3>
14838 Since it is not possible to determine which of the three branches will
14839 be taken at runtime, we don't know which of `a_1', `a_2' or `a_3' to
14840 use at the return statement. So, the SSA renamer creates a new version
14841 `a_4' which is assigned the result of "merging" `a_1', `a_2' and `a_3'.
14842 Hence, PHI nodes mean "one of these operands. I don't know which".
14844 The following macros can be used to examine PHI nodes
14846 -- Macro: PHI_RESULT (PHI)
14847 Returns the `SSA_NAME' created by PHI node PHI (i.e., PHI's LHS).
14849 -- Macro: PHI_NUM_ARGS (PHI)
14850 Returns the number of arguments in PHI. This number is exactly
14851 the number of incoming edges to the basic block holding PHI.
14853 -- Macro: PHI_ARG_ELT (PHI, I)
14854 Returns a tuple representing the Ith argument of PHI. Each
14855 element of this tuple contains an `SSA_NAME' VAR and the incoming
14856 edge through which VAR flows.
14858 -- Macro: PHI_ARG_EDGE (PHI, I)
14859 Returns the incoming edge for the Ith argument of PHI.
14861 -- Macro: PHI_ARG_DEF (PHI, I)
14862 Returns the `SSA_NAME' for the Ith argument of PHI.
14864 13.3.1 Preserving the SSA form
14865 ------------------------------
14867 Some optimization passes make changes to the function that invalidate
14868 the SSA property. This can happen when a pass has added new symbols or
14869 changed the program so that variables that were previously aliased
14870 aren't anymore. Whenever something like this happens, the affected
14871 symbols must be renamed into SSA form again. Transformations that emit
14872 new code or replicate existing statements will also need to update the
14875 Since GCC implements two different SSA forms for register and virtual
14876 variables, keeping the SSA form up to date depends on whether you are
14877 updating register or virtual names. In both cases, the general idea
14878 behind incremental SSA updates is similar: when new SSA names are
14879 created, they typically are meant to replace other existing names in
14882 For instance, given the following code:
14885 2 x_1 = PHI (0, x_5)
14896 Suppose that we insert new names `x_10' and `x_11' (lines `4' and `8').
14899 2 x_1 = PHI (0, x_5)
14912 We want to replace all the uses of `x_1' with the new definitions of
14913 `x_10' and `x_11'. Note that the only uses that should be replaced are
14914 those at lines `5', `9' and `11'. Also, the use of `x_7' at line `9'
14915 should _not_ be replaced (this is why we cannot just mark symbol `x' for
14918 Additionally, we may need to insert a PHI node at line `11' because
14919 that is a merge point for `x_10' and `x_11'. So the use of `x_1' at
14920 line `11' will be replaced with the new PHI node. The insertion of PHI
14921 nodes is optional. They are not strictly necessary to preserve the SSA
14922 form, and depending on what the caller inserted, they may not even be
14923 useful for the optimizers.
14925 Updating the SSA form is a two step process. First, the pass has to
14926 identify which names need to be updated and/or which symbols need to be
14927 renamed into SSA form for the first time. When new names are
14928 introduced to replace existing names in the program, the mapping
14929 between the old and the new names are registered by calling
14930 `register_new_name_mapping' (note that if your pass creates new code by
14931 duplicating basic blocks, the call to `tree_duplicate_bb' will set up
14932 the necessary mappings automatically). On the other hand, if your pass
14933 exposes a new symbol that should be put in SSA form for the first time,
14934 the new symbol should be registered with `mark_sym_for_renaming'.
14936 After the replacement mappings have been registered and new symbols
14937 marked for renaming, a call to `update_ssa' makes the registered
14938 changes. This can be done with an explicit call or by creating `TODO'
14939 flags in the `tree_opt_pass' structure for your pass. There are
14940 several `TODO' flags that control the behavior of `update_ssa':
14942 * `TODO_update_ssa'. Update the SSA form inserting PHI nodes for
14943 newly exposed symbols and virtual names marked for updating. When
14944 updating real names, only insert PHI nodes for a real name `O_j'
14945 in blocks reached by all the new and old definitions for `O_j'.
14946 If the iterated dominance frontier for `O_j' is not pruned, we may
14947 end up inserting PHI nodes in blocks that have one or more edges
14948 with no incoming definition for `O_j'. This would lead to
14949 uninitialized warnings for `O_j''s symbol.
14951 * `TODO_update_ssa_no_phi'. Update the SSA form without inserting
14952 any new PHI nodes at all. This is used by passes that have either
14953 inserted all the PHI nodes themselves or passes that need only to
14954 patch use-def and def-def chains for virtuals (e.g., DCE).
14956 * `TODO_update_ssa_full_phi'. Insert PHI nodes everywhere they are
14957 needed. No pruning of the IDF is done. This is used by passes
14958 that need the PHI nodes for `O_j' even if it means that some
14959 arguments will come from the default definition of `O_j''s symbol
14960 (e.g., `pass_linear_transform').
14962 WARNING: If you need to use this flag, chances are that your pass
14963 may be doing something wrong. Inserting PHI nodes for an old name
14964 where not all edges carry a new replacement may lead to silent
14965 codegen errors or spurious uninitialized warnings.
14967 * `TODO_update_ssa_only_virtuals'. Passes that update the SSA form
14968 on their own may want to delegate the updating of virtual names to
14969 the generic updater. Since FUD chains are easier to maintain,
14970 this simplifies the work they need to do. NOTE: If this flag is
14971 used, any OLD->NEW mappings for real names are explicitly
14972 destroyed and only the symbols marked for renaming are processed.
14974 13.3.2 Preserving the virtual SSA form
14975 --------------------------------------
14977 The virtual SSA form is harder to preserve than the non-virtual SSA form
14978 mainly because the set of virtual operands for a statement may change at
14979 what some would consider unexpected times. In general, statement
14980 modifications should be bracketed between calls to `push_stmt_changes'
14981 and `pop_stmt_changes'. For example,
14983 munge_stmt (tree stmt)
14985 push_stmt_changes (&stmt);
14986 ... rewrite STMT ...
14987 pop_stmt_changes (&stmt);
14990 The call to `push_stmt_changes' saves the current state of the
14991 statement operands and the call to `pop_stmt_changes' compares the
14992 saved state with the current one and does the appropriate symbol
14993 marking for the SSA renamer.
14995 It is possible to modify several statements at a time, provided that
14996 `push_stmt_changes' and `pop_stmt_changes' are called in LIFO order, as
14997 when processing a stack of statements.
14999 Additionally, if the pass discovers that it did not need to make
15000 changes to the statement after calling `push_stmt_changes', it can
15001 simply discard the topmost change buffer by calling
15002 `discard_stmt_changes'. This will avoid the expensive operand re-scan
15003 operation and the buffer comparison that determines if symbols need to
15004 be marked for renaming.
15006 13.3.3 Examining `SSA_NAME' nodes
15007 ---------------------------------
15009 The following macros can be used to examine `SSA_NAME' nodes
15011 -- Macro: SSA_NAME_DEF_STMT (VAR)
15012 Returns the statement S that creates the `SSA_NAME' VAR. If S is
15013 an empty statement (i.e., `IS_EMPTY_STMT (S)' returns `true'), it
15014 means that the first reference to this variable is a USE or a VUSE.
15016 -- Macro: SSA_NAME_VERSION (VAR)
15017 Returns the version number of the `SSA_NAME' object VAR.
15019 13.3.4 Walking use-def chains
15020 -----------------------------
15022 -- Tree SSA function: void walk_use_def_chains (VAR, FN, DATA)
15023 Walks use-def chains starting at the `SSA_NAME' node VAR. Calls
15024 function FN at each reaching definition found. Function FN takes
15025 three arguments: VAR, its defining statement (DEF_STMT) and a
15026 generic pointer to whatever state information that FN may want to
15027 maintain (DATA). Function FN is able to stop the walk by
15028 returning `true', otherwise in order to continue the walk, FN
15029 should return `false'.
15031 Note, that if DEF_STMT is a `PHI' node, the semantics are slightly
15032 different. For each argument ARG of the PHI node, this function
15035 1. Walk the use-def chains for ARG.
15037 2. Call `FN (ARG, PHI, DATA)'.
15039 Note how the first argument to FN is no longer the original
15040 variable VAR, but the PHI argument currently being examined. If
15041 FN wants to get at VAR, it should call `PHI_RESULT' (PHI).
15043 13.3.5 Walking the dominator tree
15044 ---------------------------------
15046 -- Tree SSA function: void walk_dominator_tree (WALK_DATA, BB)
15047 This function walks the dominator tree for the current CFG calling
15048 a set of callback functions defined in STRUCT DOM_WALK_DATA in
15049 `domwalk.h'. The call back functions you need to define give you
15050 hooks to execute custom code at various points during traversal:
15052 1. Once to initialize any local data needed while processing BB
15053 and its children. This local data is pushed into an internal
15054 stack which is automatically pushed and popped as the walker
15055 traverses the dominator tree.
15057 2. Once before traversing all the statements in the BB.
15059 3. Once for every statement inside BB.
15061 4. Once after traversing all the statements and before recursing
15062 into BB's dominator children.
15064 5. It then recurses into all the dominator children of BB.
15066 6. After recursing into all the dominator children of BB it can,
15067 optionally, traverse every statement in BB again (i.e.,
15068 repeating steps 2 and 3).
15070 7. Once after walking the statements in BB and BB's dominator
15071 children. At this stage, the block local data stack is
15075 File: gccint.info, Node: Alias analysis, Prev: SSA, Up: Tree SSA
15077 13.4 Alias analysis
15078 ===================
15080 Alias analysis proceeds in 4 main phases:
15082 1. Structural alias analysis.
15084 This phase walks the types for structure variables, and determines
15085 which of the fields can overlap using offset and size of each
15086 field. For each field, a "subvariable" called a "Structure field
15087 tag" (SFT) is created, which represents that field as a separate
15088 variable. All accesses that could possibly overlap with a given
15089 field will have virtual operands for the SFT of that field.
15099 int tmp1, tmp2, tmp3;
15100 SFT.0_2 = VDEF <SFT.0_1>
15102 SFT.1_4 = VDEF <SFT.1_3>
15110 tmp3_7 = tmp1_5 + tmp2_6;
15114 If you copy the symbol tag for a variable for some reason, you
15115 probably also want to copy the subvariables for that variable.
15117 2. Points-to and escape analysis.
15119 This phase walks the use-def chains in the SSA web looking for
15122 * Assignments of the form `P_i = &VAR'
15124 * Assignments of the form P_i = malloc()
15126 * Pointers and ADDR_EXPR that escape the current function.
15128 The concept of `escaping' is the same one used in the Java world.
15129 When a pointer or an ADDR_EXPR escapes, it means that it has been
15130 exposed outside of the current function. So, assignment to global
15131 variables, function arguments and returning a pointer are all
15134 This is where we are currently limited. Since not everything is
15135 renamed into SSA, we lose track of escape properties when a
15136 pointer is stashed inside a field in a structure, for instance.
15137 In those cases, we are assuming that the pointer does escape.
15139 We use escape analysis to determine whether a variable is
15140 call-clobbered. Simply put, if an ADDR_EXPR escapes, then the
15141 variable is call-clobbered. If a pointer P_i escapes, then all
15142 the variables pointed-to by P_i (and its memory tag) also escape.
15144 3. Compute flow-sensitive aliases
15146 We have two classes of memory tags. Memory tags associated with
15147 the pointed-to data type of the pointers in the program. These
15148 tags are called "symbol memory tag" (SMT). The other class are
15149 those associated with SSA_NAMEs, called "name memory tag" (NMT).
15150 The basic idea is that when adding operands for an INDIRECT_REF
15151 *P_i, we will first check whether P_i has a name tag, if it does
15152 we use it, because that will have more precise aliasing
15153 information. Otherwise, we use the standard symbol tag.
15155 In this phase, we go through all the pointers we found in
15156 points-to analysis and create alias sets for the name memory tags
15157 associated with each pointer P_i. If P_i escapes, we mark
15158 call-clobbered the variables it points to and its tag.
15160 4. Compute flow-insensitive aliases
15162 This pass will compare the alias set of every symbol memory tag and
15163 every addressable variable found in the program. Given a symbol
15164 memory tag SMT and an addressable variable V. If the alias sets
15165 of SMT and V conflict (as computed by may_alias_p), then V is
15166 marked as an alias tag and added to the alias set of SMT.
15168 Every language that wishes to perform language-specific alias
15169 analysis should define a function that computes, given a `tree'
15170 node, an alias set for the node. Nodes in different alias sets
15171 are not allowed to alias. For an example, see the C front-end
15172 function `c_get_alias_set'.
15174 For instance, consider the following function:
15191 After aliasing analysis has finished, the symbol memory tag for
15192 pointer `p' will have two aliases, namely variables `a' and `b'. Every
15193 time pointer `p' is dereferenced, we want to mark the operation as a
15194 potential reference to `a' and `b'.
15204 # p_1 = PHI <p_4(1), p_6(2)>;
15206 # a_7 = VDEF <a_3>;
15207 # b_8 = VDEF <b_5>;
15219 In certain cases, the list of may aliases for a pointer may grow too
15220 large. This may cause an explosion in the number of virtual operands
15221 inserted in the code. Resulting in increased memory consumption and
15224 When the number of virtual operands needed to represent aliased loads
15225 and stores grows too large (configurable with `--param
15226 max-aliased-vops'), alias sets are grouped to avoid severe compile-time
15227 slow downs and memory consumption. The alias grouping heuristic
15228 proceeds as follows:
15230 1. Sort the list of pointers in decreasing number of contributed
15233 2. Take the first pointer from the list and reverse the role of the
15234 memory tag and its aliases. Usually, whenever an aliased variable
15235 Vi is found to alias with a memory tag T, we add Vi to the
15236 may-aliases set for T. Meaning that after alias analysis, we will
15239 may-aliases(T) = { V1, V2, V3, ..., Vn }
15241 This means that every statement that references T, will get `n'
15242 virtual operands for each of the Vi tags. But, when alias
15243 grouping is enabled, we make T an alias tag and add it to the
15244 alias set of all the Vi variables:
15246 may-aliases(V1) = { T }
15247 may-aliases(V2) = { T }
15249 may-aliases(Vn) = { T }
15251 This has two effects: (a) statements referencing T will only get a
15252 single virtual operand, and, (b) all the variables Vi will now
15253 appear to alias each other. So, we lose alias precision to
15254 improve compile time. But, in theory, a program with such a high
15255 level of aliasing should not be very optimizable in the first
15258 3. Since variables may be in the alias set of more than one memory
15259 tag, the grouping done in step (2) needs to be extended to all the
15260 memory tags that have a non-empty intersection with the
15261 may-aliases set of tag T. For instance, if we originally had
15262 these may-aliases sets:
15264 may-aliases(T) = { V1, V2, V3 }
15265 may-aliases(R) = { V2, V4 }
15267 In step (2) we would have reverted the aliases for T as:
15269 may-aliases(V1) = { T }
15270 may-aliases(V2) = { T }
15271 may-aliases(V3) = { T }
15273 But note that now V2 is no longer aliased with R. We could add R
15274 to may-aliases(V2), but we are in the process of grouping aliases
15275 to reduce virtual operands so what we do is add V4 to the grouping
15278 may-aliases(V1) = { T }
15279 may-aliases(V2) = { T }
15280 may-aliases(V3) = { T }
15281 may-aliases(V4) = { T }
15283 4. If the total number of virtual operands due to aliasing is still
15284 above the threshold set by max-alias-vops, go back to (2).
15287 File: gccint.info, Node: Loop Analysis and Representation, Next: Machine Desc, Prev: Control Flow, Up: Top
15289 14 Analysis and Representation of Loops
15290 ***************************************
15292 GCC provides extensive infrastructure for work with natural loops, i.e.,
15293 strongly connected components of CFG with only one entry block. This
15294 chapter describes representation of loops in GCC, both on GIMPLE and in
15295 RTL, as well as the interfaces to loop-related analyses (induction
15296 variable analysis and number of iterations analysis).
15300 * Loop representation:: Representation and analysis of loops.
15301 * Loop querying:: Getting information about loops.
15302 * Loop manipulation:: Loop manipulation functions.
15303 * LCSSA:: Loop-closed SSA form.
15304 * Scalar evolutions:: Induction variables on GIMPLE.
15305 * loop-iv:: Induction variables on RTL.
15306 * Number of iterations:: Number of iterations analysis.
15307 * Dependency analysis:: Data dependency analysis.
15308 * Lambda:: Linear loop transformations framework.
15309 * Omega:: A solver for linear programming problems.
15312 File: gccint.info, Node: Loop representation, Next: Loop querying, Up: Loop Analysis and Representation
15314 14.1 Loop representation
15315 ========================
15317 This chapter describes the representation of loops in GCC, and functions
15318 that can be used to build, modify and analyze this representation. Most
15319 of the interfaces and data structures are declared in `cfgloop.h'. At
15320 the moment, loop structures are analyzed and this information is
15321 updated only by the optimization passes that deal with loops, but some
15322 efforts are being made to make it available throughout most of the
15323 optimization passes.
15325 In general, a natural loop has one entry block (header) and possibly
15326 several back edges (latches) leading to the header from the inside of
15327 the loop. Loops with several latches may appear if several loops share
15328 a single header, or if there is a branching in the middle of the loop.
15329 The representation of loops in GCC however allows only loops with a
15330 single latch. During loop analysis, headers of such loops are split and
15331 forwarder blocks are created in order to disambiguate their structures.
15332 Heuristic based on profile information and structure of the induction
15333 variables in the loops is used to determine whether the latches
15334 correspond to sub-loops or to control flow in a single loop. This means
15335 that the analysis sometimes changes the CFG, and if you run it in the
15336 middle of an optimization pass, you must be able to deal with the new
15337 blocks. You may avoid CFG changes by passing
15338 `LOOPS_MAY_HAVE_MULTIPLE_LATCHES' flag to the loop discovery, note
15339 however that most other loop manipulation functions will not work
15340 correctly for loops with multiple latch edges (the functions that only
15341 query membership of blocks to loops and subloop relationships, or
15342 enumerate and test loop exits, can be expected to work).
15344 Body of the loop is the set of blocks that are dominated by its header,
15345 and reachable from its latch against the direction of edges in CFG. The
15346 loops are organized in a containment hierarchy (tree) such that all the
15347 loops immediately contained inside loop L are the children of L in the
15348 tree. This tree is represented by the `struct loops' structure. The
15349 root of this tree is a fake loop that contains all blocks in the
15350 function. Each of the loops is represented in a `struct loop'
15351 structure. Each loop is assigned an index (`num' field of the `struct
15352 loop' structure), and the pointer to the loop is stored in the
15353 corresponding field of the `larray' vector in the loops structure. The
15354 indices do not have to be continuous, there may be empty (`NULL')
15355 entries in the `larray' created by deleting loops. Also, there is no
15356 guarantee on the relative order of a loop and its subloops in the
15357 numbering. The index of a loop never changes.
15359 The entries of the `larray' field should not be accessed directly.
15360 The function `get_loop' returns the loop description for a loop with
15361 the given index. `number_of_loops' function returns number of loops in
15362 the function. To traverse all loops, use `FOR_EACH_LOOP' macro. The
15363 `flags' argument of the macro is used to determine the direction of
15364 traversal and the set of loops visited. Each loop is guaranteed to be
15365 visited exactly once, regardless of the changes to the loop tree, and
15366 the loops may be removed during the traversal. The newly created loops
15367 are never traversed, if they need to be visited, this must be done
15368 separately after their creation. The `FOR_EACH_LOOP' macro allocates
15369 temporary variables. If the `FOR_EACH_LOOP' loop were ended using
15370 break or goto, they would not be released; `FOR_EACH_LOOP_BREAK' macro
15371 must be used instead.
15373 Each basic block contains the reference to the innermost loop it
15374 belongs to (`loop_father'). For this reason, it is only possible to
15375 have one `struct loops' structure initialized at the same time for each
15376 CFG. The global variable `current_loops' contains the `struct loops'
15377 structure. Many of the loop manipulation functions assume that
15378 dominance information is up-to-date.
15380 The loops are analyzed through `loop_optimizer_init' function. The
15381 argument of this function is a set of flags represented in an integer
15382 bitmask. These flags specify what other properties of the loop
15383 structures should be calculated/enforced and preserved later:
15385 * `LOOPS_MAY_HAVE_MULTIPLE_LATCHES': If this flag is set, no changes
15386 to CFG will be performed in the loop analysis, in particular,
15387 loops with multiple latch edges will not be disambiguated. If a
15388 loop has multiple latches, its latch block is set to NULL. Most of
15389 the loop manipulation functions will not work for loops in this
15390 shape. No other flags that require CFG changes can be passed to
15391 loop_optimizer_init.
15393 * `LOOPS_HAVE_PREHEADERS': Forwarder blocks are created in such a
15394 way that each loop has only one entry edge, and additionally, the
15395 source block of this entry edge has only one successor. This
15396 creates a natural place where the code can be moved out of the
15397 loop, and ensures that the entry edge of the loop leads from its
15398 immediate super-loop.
15400 * `LOOPS_HAVE_SIMPLE_LATCHES': Forwarder blocks are created to force
15401 the latch block of each loop to have only one successor. This
15402 ensures that the latch of the loop does not belong to any of its
15403 sub-loops, and makes manipulation with the loops significantly
15404 easier. Most of the loop manipulation functions assume that the
15405 loops are in this shape. Note that with this flag, the "normal"
15406 loop without any control flow inside and with one exit consists of
15409 * `LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS': Basic blocks and edges in
15410 the strongly connected components that are not natural loops (have
15411 more than one entry block) are marked with `BB_IRREDUCIBLE_LOOP'
15412 and `EDGE_IRREDUCIBLE_LOOP' flags. The flag is not set for blocks
15413 and edges that belong to natural loops that are in such an
15414 irreducible region (but it is set for the entry and exit edges of
15415 such a loop, if they lead to/from this region).
15417 * `LOOPS_HAVE_RECORDED_EXITS': The lists of exits are recorded and
15418 updated for each loop. This makes some functions (e.g.,
15419 `get_loop_exit_edges') more efficient. Some functions (e.g.,
15420 `single_exit') can be used only if the lists of exits are recorded.
15422 These properties may also be computed/enforced later, using functions
15423 `create_preheaders', `force_single_succ_latches',
15424 `mark_irreducible_loops' and `record_loop_exits'.
15426 The memory occupied by the loops structures should be freed with
15427 `loop_optimizer_finalize' function.
15429 The CFG manipulation functions in general do not update loop
15430 structures. Specialized versions that additionally do so are provided
15431 for the most common tasks. On GIMPLE, `cleanup_tree_cfg_loop' function
15432 can be used to cleanup CFG while updating the loops structures if
15433 `current_loops' is set.
15436 File: gccint.info, Node: Loop querying, Next: Loop manipulation, Prev: Loop representation, Up: Loop Analysis and Representation
15441 The functions to query the information about loops are declared in
15442 `cfgloop.h'. Some of the information can be taken directly from the
15443 structures. `loop_father' field of each basic block contains the
15444 innermost loop to that the block belongs. The most useful fields of
15445 loop structure (that are kept up-to-date at all times) are:
15447 * `header', `latch': Header and latch basic blocks of the loop.
15449 * `num_nodes': Number of basic blocks in the loop (including the
15450 basic blocks of the sub-loops).
15452 * `depth': The depth of the loop in the loops tree, i.e., the number
15453 of super-loops of the loop.
15455 * `outer', `inner', `next': The super-loop, the first sub-loop, and
15456 the sibling of the loop in the loops tree.
15458 There are other fields in the loop structures, many of them used only
15459 by some of the passes, or not updated during CFG changes; in general,
15460 they should not be accessed directly.
15462 The most important functions to query loop structures are:
15464 * `flow_loops_dump': Dumps the information about loops to a file.
15466 * `verify_loop_structure': Checks consistency of the loop structures.
15468 * `loop_latch_edge': Returns the latch edge of a loop.
15470 * `loop_preheader_edge': If loops have preheaders, returns the
15471 preheader edge of a loop.
15473 * `flow_loop_nested_p': Tests whether loop is a sub-loop of another
15476 * `flow_bb_inside_loop_p': Tests whether a basic block belongs to a
15477 loop (including its sub-loops).
15479 * `find_common_loop': Finds the common super-loop of two loops.
15481 * `superloop_at_depth': Returns the super-loop of a loop with the
15484 * `tree_num_loop_insns', `num_loop_insns': Estimates the number of
15485 insns in the loop, on GIMPLE and on RTL.
15487 * `loop_exit_edge_p': Tests whether edge is an exit from a loop.
15489 * `mark_loop_exit_edges': Marks all exit edges of all loops with
15490 `EDGE_LOOP_EXIT' flag.
15492 * `get_loop_body', `get_loop_body_in_dom_order',
15493 `get_loop_body_in_bfs_order': Enumerates the basic blocks in the
15494 loop in depth-first search order in reversed CFG, ordered by
15495 dominance relation, and breath-first search order, respectively.
15497 * `single_exit': Returns the single exit edge of the loop, or `NULL'
15498 if the loop has more than one exit. You can only use this
15499 function if LOOPS_HAVE_MARKED_SINGLE_EXITS property is used.
15501 * `get_loop_exit_edges': Enumerates the exit edges of a loop.
15503 * `just_once_each_iteration_p': Returns true if the basic block is
15504 executed exactly once during each iteration of a loop (that is, it
15505 does not belong to a sub-loop, and it dominates the latch of the
15509 File: gccint.info, Node: Loop manipulation, Next: LCSSA, Prev: Loop querying, Up: Loop Analysis and Representation
15511 14.3 Loop manipulation
15512 ======================
15514 The loops tree can be manipulated using the following functions:
15516 * `flow_loop_tree_node_add': Adds a node to the tree.
15518 * `flow_loop_tree_node_remove': Removes a node from the tree.
15520 * `add_bb_to_loop': Adds a basic block to a loop.
15522 * `remove_bb_from_loops': Removes a basic block from loops.
15524 Most low-level CFG functions update loops automatically. The following
15525 functions handle some more complicated cases of CFG manipulations:
15527 * `remove_path': Removes an edge and all blocks it dominates.
15529 * `split_loop_exit_edge': Splits exit edge of the loop, ensuring
15530 that PHI node arguments remain in the loop (this ensures that
15531 loop-closed SSA form is preserved). Only useful on GIMPLE.
15533 Finally, there are some higher-level loop transformations implemented.
15534 While some of them are written so that they should work on non-innermost
15535 loops, they are mostly untested in that case, and at the moment, they
15536 are only reliable for the innermost loops:
15538 * `create_iv': Creates a new induction variable. Only works on
15539 GIMPLE. `standard_iv_increment_position' can be used to find a
15540 suitable place for the iv increment.
15542 * `duplicate_loop_to_header_edge',
15543 `tree_duplicate_loop_to_header_edge': These functions (on RTL and
15544 on GIMPLE) duplicate the body of the loop prescribed number of
15545 times on one of the edges entering loop header, thus performing
15546 either loop unrolling or loop peeling. `can_duplicate_loop_p'
15547 (`can_unroll_loop_p' on GIMPLE) must be true for the duplicated
15550 * `loop_version', `tree_ssa_loop_version': These function create a
15551 copy of a loop, and a branch before them that selects one of them
15552 depending on the prescribed condition. This is useful for
15553 optimizations that need to verify some assumptions in runtime (one
15554 of the copies of the loop is usually left unchanged, while the
15555 other one is transformed in some way).
15557 * `tree_unroll_loop': Unrolls the loop, including peeling the extra
15558 iterations to make the number of iterations divisible by unroll
15559 factor, updating the exit condition, and removing the exits that
15560 now cannot be taken. Works only on GIMPLE.
15563 File: gccint.info, Node: LCSSA, Next: Scalar evolutions, Prev: Loop manipulation, Up: Loop Analysis and Representation
15565 14.4 Loop-closed SSA form
15566 =========================
15568 Throughout the loop optimizations on tree level, one extra condition is
15569 enforced on the SSA form: No SSA name is used outside of the loop in
15570 that it is defined. The SSA form satisfying this condition is called
15571 "loop-closed SSA form" - LCSSA. To enforce LCSSA, PHI nodes must be
15572 created at the exits of the loops for the SSA names that are used
15573 outside of them. Only the real operands (not virtual SSA names) are
15574 held in LCSSA, in order to save memory.
15576 There are various benefits of LCSSA:
15578 * Many optimizations (value range analysis, final value replacement)
15579 are interested in the values that are defined in the loop and used
15580 outside of it, i.e., exactly those for that we create new PHI
15583 * In induction variable analysis, it is not necessary to specify the
15584 loop in that the analysis should be performed - the scalar
15585 evolution analysis always returns the results with respect to the
15586 loop in that the SSA name is defined.
15588 * It makes updating of SSA form during loop transformations simpler.
15589 Without LCSSA, operations like loop unrolling may force creation
15590 of PHI nodes arbitrarily far from the loop, while in LCSSA, the
15591 SSA form can be updated locally. However, since we only keep real
15592 operands in LCSSA, we cannot use this advantage (we could have
15593 local updating of real operands, but it is not much more efficient
15594 than to use generic SSA form updating for it as well; the amount
15595 of changes to SSA is the same).
15597 However, it also means LCSSA must be updated. This is usually
15598 straightforward, unless you create a new value in loop and use it
15599 outside, or unless you manipulate loop exit edges (functions are
15600 provided to make these manipulations simple).
15601 `rewrite_into_loop_closed_ssa' is used to rewrite SSA form to LCSSA,
15602 and `verify_loop_closed_ssa' to check that the invariant of LCSSA is
15606 File: gccint.info, Node: Scalar evolutions, Next: loop-iv, Prev: LCSSA, Up: Loop Analysis and Representation
15608 14.5 Scalar evolutions
15609 ======================
15611 Scalar evolutions (SCEV) are used to represent results of induction
15612 variable analysis on GIMPLE. They enable us to represent variables with
15613 complicated behavior in a simple and consistent way (we only use it to
15614 express values of polynomial induction variables, but it is possible to
15615 extend it). The interfaces to SCEV analysis are declared in
15616 `tree-scalar-evolution.h'. To use scalar evolutions analysis,
15617 `scev_initialize' must be used. To stop using SCEV, `scev_finalize'
15618 should be used. SCEV analysis caches results in order to save time and
15619 memory. This cache however is made invalid by most of the loop
15620 transformations, including removal of code. If such a transformation
15621 is performed, `scev_reset' must be called to clean the caches.
15623 Given an SSA name, its behavior in loops can be analyzed using the
15624 `analyze_scalar_evolution' function. The returned SCEV however does
15625 not have to be fully analyzed and it may contain references to other
15626 SSA names defined in the loop. To resolve these (potentially
15627 recursive) references, `instantiate_parameters' or `resolve_mixers'
15628 functions must be used. `instantiate_parameters' is useful when you
15629 use the results of SCEV only for some analysis, and when you work with
15630 whole nest of loops at once. It will try replacing all SSA names by
15631 their SCEV in all loops, including the super-loops of the current loop,
15632 thus providing a complete information about the behavior of the
15633 variable in the loop nest. `resolve_mixers' is useful if you work with
15634 only one loop at a time, and if you possibly need to create code based
15635 on the value of the induction variable. It will only resolve the SSA
15636 names defined in the current loop, leaving the SSA names defined
15637 outside unchanged, even if their evolution in the outer loops is known.
15639 The SCEV is a normal tree expression, except for the fact that it may
15640 contain several special tree nodes. One of them is `SCEV_NOT_KNOWN',
15641 used for SSA names whose value cannot be expressed. The other one is
15642 `POLYNOMIAL_CHREC'. Polynomial chrec has three arguments - base, step
15643 and loop (both base and step may contain further polynomial chrecs).
15644 Type of the expression and of base and step must be the same. A
15645 variable has evolution `POLYNOMIAL_CHREC(base, step, loop)' if it is
15646 (in the specified loop) equivalent to `x_1' in the following example
15650 x_1 = phi (base, x_2);
15654 Note that this includes the language restrictions on the operations.
15655 For example, if we compile C code and `x' has signed type, then the
15656 overflow in addition would cause undefined behavior, and we may assume
15657 that this does not happen. Hence, the value with this SCEV cannot
15658 overflow (which restricts the number of iterations of such a loop).
15660 In many cases, one wants to restrict the attention just to affine
15661 induction variables. In this case, the extra expressive power of SCEV
15662 is not useful, and may complicate the optimizations. In this case,
15663 `simple_iv' function may be used to analyze a value - the result is a
15664 loop-invariant base and step.
15667 File: gccint.info, Node: loop-iv, Next: Number of iterations, Prev: Scalar evolutions, Up: Loop Analysis and Representation
15669 14.6 IV analysis on RTL
15670 =======================
15672 The induction variable on RTL is simple and only allows analysis of
15673 affine induction variables, and only in one loop at once. The interface
15674 is declared in `cfgloop.h'. Before analyzing induction variables in a
15675 loop L, `iv_analysis_loop_init' function must be called on L. After
15676 the analysis (possibly calling `iv_analysis_loop_init' for several
15677 loops) is finished, `iv_analysis_done' should be called. The following
15678 functions can be used to access the results of the analysis:
15680 * `iv_analyze': Analyzes a single register used in the given insn.
15681 If no use of the register in this insn is found, the following
15682 insns are scanned, so that this function can be called on the insn
15683 returned by get_condition.
15685 * `iv_analyze_result': Analyzes result of the assignment in the
15688 * `iv_analyze_expr': Analyzes a more complicated expression. All
15689 its operands are analyzed by `iv_analyze', and hence they must be
15690 used in the specified insn or one of the following insns.
15692 The description of the induction variable is provided in `struct
15693 rtx_iv'. In order to handle subregs, the representation is a bit
15694 complicated; if the value of the `extend' field is not `UNKNOWN', the
15695 value of the induction variable in the i-th iteration is
15697 delta + mult * extend_{extend_mode} (subreg_{mode} (base + i * step)),
15699 with the following exception: if `first_special' is true, then the
15700 value in the first iteration (when `i' is zero) is `delta + mult *
15701 base'. However, if `extend' is equal to `UNKNOWN', then
15702 `first_special' must be false, `delta' 0, `mult' 1 and the value in the
15705 subreg_{mode} (base + i * step)
15707 The function `get_iv_value' can be used to perform these calculations.
15710 File: gccint.info, Node: Number of iterations, Next: Dependency analysis, Prev: loop-iv, Up: Loop Analysis and Representation
15712 14.7 Number of iterations analysis
15713 ==================================
15715 Both on GIMPLE and on RTL, there are functions available to determine
15716 the number of iterations of a loop, with a similar interface. The
15717 number of iterations of a loop in GCC is defined as the number of
15718 executions of the loop latch. In many cases, it is not possible to
15719 determine the number of iterations unconditionally - the determined
15720 number is correct only if some assumptions are satisfied. The analysis
15721 tries to verify these conditions using the information contained in the
15722 program; if it fails, the conditions are returned together with the
15723 result. The following information and conditions are provided by the
15726 * `assumptions': If this condition is false, the rest of the
15727 information is invalid.
15729 * `noloop_assumptions' on RTL, `may_be_zero' on GIMPLE: If this
15730 condition is true, the loop exits in the first iteration.
15732 * `infinite': If this condition is true, the loop is infinite. This
15733 condition is only available on RTL. On GIMPLE, conditions for
15734 finiteness of the loop are included in `assumptions'.
15736 * `niter_expr' on RTL, `niter' on GIMPLE: The expression that gives
15737 number of iterations. The number of iterations is defined as the
15738 number of executions of the loop latch.
15740 Both on GIMPLE and on RTL, it necessary for the induction variable
15741 analysis framework to be initialized (SCEV on GIMPLE, loop-iv on RTL).
15742 On GIMPLE, the results are stored to `struct tree_niter_desc'
15743 structure. Number of iterations before the loop is exited through a
15744 given exit can be determined using `number_of_iterations_exit'
15745 function. On RTL, the results are returned in `struct niter_desc'
15746 structure. The corresponding function is named `check_simple_exit'.
15747 There are also functions that pass through all the exits of a loop and
15748 try to find one with easy to determine number of iterations -
15749 `find_loop_niter' on GIMPLE and `find_simple_exit' on RTL. Finally,
15750 there are functions that provide the same information, but additionally
15751 cache it, so that repeated calls to number of iterations are not so
15752 costly - `number_of_latch_executions' on GIMPLE and
15753 `get_simple_loop_desc' on RTL.
15755 Note that some of these functions may behave slightly differently than
15756 others - some of them return only the expression for the number of
15757 iterations, and fail if there are some assumptions. The function
15758 `number_of_latch_executions' works only for single-exit loops. The
15759 function `number_of_cond_exit_executions' can be used to determine
15760 number of executions of the exit condition of a single-exit loop (i.e.,
15761 the `number_of_latch_executions' increased by one).
15764 File: gccint.info, Node: Dependency analysis, Next: Lambda, Prev: Number of iterations, Up: Loop Analysis and Representation
15766 14.8 Data Dependency Analysis
15767 =============================
15769 The code for the data dependence analysis can be found in
15770 `tree-data-ref.c' and its interface and data structures are described
15771 in `tree-data-ref.h'. The function that computes the data dependences
15772 for all the array and pointer references for a given loop is
15773 `compute_data_dependences_for_loop'. This function is currently used
15774 by the linear loop transform and the vectorization passes. Before
15775 calling this function, one has to allocate two vectors: a first vector
15776 will contain the set of data references that are contained in the
15777 analyzed loop body, and the second vector will contain the dependence
15778 relations between the data references. Thus if the vector of data
15779 references is of size `n', the vector containing the dependence
15780 relations will contain `n*n' elements. However if the analyzed loop
15781 contains side effects, such as calls that potentially can interfere
15782 with the data references in the current analyzed loop, the analysis
15783 stops while scanning the loop body for data references, and inserts a
15784 single `chrec_dont_know' in the dependence relation array.
15786 The data references are discovered in a particular order during the
15787 scanning of the loop body: the loop body is analyzed in execution order,
15788 and the data references of each statement are pushed at the end of the
15789 data reference array. Two data references syntactically occur in the
15790 program in the same order as in the array of data references. This
15791 syntactic order is important in some classical data dependence tests,
15792 and mapping this order to the elements of this array avoids costly
15793 queries to the loop body representation.
15795 Three types of data references are currently handled: ARRAY_REF,
15796 INDIRECT_REF and COMPONENT_REF. The data structure for the data
15797 reference is `data_reference', where `data_reference_p' is a name of a
15798 pointer to the data reference structure. The structure contains the
15799 following elements:
15801 * `base_object_info': Provides information about the base object of
15802 the data reference and its access functions. These access functions
15803 represent the evolution of the data reference in the loop relative
15804 to its base, in keeping with the classical meaning of the data
15805 reference access function for the support of arrays. For example,
15806 for a reference `a.b[i][j]', the base object is `a.b' and the
15807 access functions, one for each array subscript, are: `{i_init, +
15808 i_step}_1, {j_init, +, j_step}_2'.
15810 * `first_location_in_loop': Provides information about the first
15811 location accessed by the data reference in the loop and about the
15812 access function used to represent evolution relative to this
15813 location. This data is used to support pointers, and is not used
15814 for arrays (for which we have base objects). Pointer accesses are
15815 represented as a one-dimensional access that starts from the first
15816 location accessed in the loop. For example:
15820 *((int *)p + i + j) = a[i][j];
15822 The access function of the pointer access is `{0, + 4B}_for2'
15823 relative to `p + i'. The access functions of the array are
15824 `{i_init, + i_step}_for1' and `{j_init, +, j_step}_for2' relative
15827 Usually, the object the pointer refers to is either unknown, or we
15828 can't prove that the access is confined to the boundaries of a
15831 Two data references can be compared only if at least one of these
15832 two representations has all its fields filled for both data
15835 The current strategy for data dependence tests is as follows: If
15836 both `a' and `b' are represented as arrays, compare
15837 `a.base_object' and `b.base_object'; if they are equal, apply
15838 dependence tests (use access functions based on base_objects).
15839 Else if both `a' and `b' are represented as pointers, compare
15840 `a.first_location' and `b.first_location'; if they are equal,
15841 apply dependence tests (use access functions based on first
15842 location). However, if `a' and `b' are represented differently,
15843 only try to prove that the bases are definitely different.
15845 * Aliasing information.
15847 * Alignment information.
15849 The structure describing the relation between two data references is
15850 `data_dependence_relation' and the shorter name for a pointer to such a
15851 structure is `ddr_p'. This structure contains:
15853 * a pointer to each data reference,
15855 * a tree node `are_dependent' that is set to `chrec_known' if the
15856 analysis has proved that there is no dependence between these two
15857 data references, `chrec_dont_know' if the analysis was not able to
15858 determine any useful result and potentially there could exist a
15859 dependence between these data references, and `are_dependent' is
15860 set to `NULL_TREE' if there exist a dependence relation between the
15861 data references, and the description of this dependence relation is
15862 given in the `subscripts', `dir_vects', and `dist_vects' arrays,
15864 * a boolean that determines whether the dependence relation can be
15865 represented by a classical distance vector,
15867 * an array `subscripts' that contains a description of each
15868 subscript of the data references. Given two array accesses a
15869 subscript is the tuple composed of the access functions for a given
15870 dimension. For example, given `A[f1][f2][f3]' and
15871 `B[g1][g2][g3]', there are three subscripts: `(f1, g1), (f2, g2),
15874 * two arrays `dir_vects' and `dist_vects' that contain classical
15875 representations of the data dependences under the form of
15876 direction and distance dependence vectors,
15878 * an array of loops `loop_nest' that contains the loops to which the
15879 distance and direction vectors refer to.
15881 Several functions for pretty printing the information extracted by the
15882 data dependence analysis are available: `dump_ddrs' prints with a
15883 maximum verbosity the details of a data dependence relations array,
15884 `dump_dist_dir_vectors' prints only the classical distance and
15885 direction vectors for a data dependence relations array, and
15886 `dump_data_references' prints the details of the data references
15887 contained in a data reference array.
15890 File: gccint.info, Node: Lambda, Next: Omega, Prev: Dependency analysis, Up: Loop Analysis and Representation
15892 14.9 Linear loop transformations framework
15893 ==========================================
15895 Lambda is a framework that allows transformations of loops using
15896 non-singular matrix based transformations of the iteration space and
15897 loop bounds. This allows compositions of skewing, scaling, interchange,
15898 and reversal transformations. These transformations are often used to
15899 improve cache behavior or remove inner loop dependencies to allow
15900 parallelization and vectorization to take place.
15902 To perform these transformations, Lambda requires that the loopnest be
15903 converted into a internal form that can be matrix transformed easily.
15904 To do this conversion, the function `gcc_loopnest_to_lambda_loopnest'
15905 is provided. If the loop cannot be transformed using lambda, this
15906 function will return NULL.
15908 Once a `lambda_loopnest' is obtained from the conversion function, it
15909 can be transformed by using `lambda_loopnest_transform', which takes a
15910 transformation matrix to apply. Note that it is up to the caller to
15911 verify that the transformation matrix is legal to apply to the loop
15912 (dependence respecting, etc). Lambda simply applies whatever matrix it
15913 is told to provide. It can be extended to make legal matrices out of
15914 any non-singular matrix, but this is not currently implemented.
15915 Legality of a matrix for a given loopnest can be verified using
15916 `lambda_transform_legal_p'.
15918 Given a transformed loopnest, conversion back into gcc IR is done by
15919 `lambda_loopnest_to_gcc_loopnest'. This function will modify the loops
15920 so that they match the transformed loopnest.
15923 File: gccint.info, Node: Omega, Prev: Lambda, Up: Loop Analysis and Representation
15925 14.10 Omega a solver for linear programming problems
15926 ====================================================
15928 The data dependence analysis contains several solvers triggered
15929 sequentially from the less complex ones to the more sophisticated. For
15930 ensuring the consistency of the results of these solvers, a data
15931 dependence check pass has been implemented based on two different
15932 solvers. The second method that has been integrated to GCC is based on
15933 the Omega dependence solver, written in the 1990's by William Pugh and
15934 David Wonnacott. Data dependence tests can be formulated using a
15935 subset of the Presburger arithmetics that can be translated to linear
15936 constraint systems. These linear constraint systems can then be solved
15937 using the Omega solver.
15939 The Omega solver is using Fourier-Motzkin's algorithm for variable
15940 elimination: a linear constraint system containing `n' variables is
15941 reduced to a linear constraint system with `n-1' variables. The Omega
15942 solver can also be used for solving other problems that can be
15943 expressed under the form of a system of linear equalities and
15944 inequalities. The Omega solver is known to have an exponential worst
15945 case, also known under the name of "omega nightmare" in the literature,
15946 but in practice, the omega test is known to be efficient for the common
15947 data dependence tests.
15949 The interface used by the Omega solver for describing the linear
15950 programming problems is described in `omega.h', and the solver is
15951 `omega_solve_problem'.
15954 File: gccint.info, Node: Control Flow, Next: Loop Analysis and Representation, Prev: Tree SSA, Up: Top
15956 15 Control Flow Graph
15957 *********************
15959 A control flow graph (CFG) is a data structure built on top of the
15960 intermediate code representation (the RTL or `tree' instruction stream)
15961 abstracting the control flow behavior of a function that is being
15962 compiled. The CFG is a directed graph where the vertices represent
15963 basic blocks and edges represent possible transfer of control flow from
15964 one basic block to another. The data structures used to represent the
15965 control flow graph are defined in `basic-block.h'.
15969 * Basic Blocks:: The definition and representation of basic blocks.
15970 * Edges:: Types of edges and their representation.
15971 * Profile information:: Representation of frequencies and probabilities.
15972 * Maintaining the CFG:: Keeping the control flow graph and up to date.
15973 * Liveness information:: Using and maintaining liveness information.
15976 File: gccint.info, Node: Basic Blocks, Next: Edges, Up: Control Flow
15981 A basic block is a straight-line sequence of code with only one entry
15982 point and only one exit. In GCC, basic blocks are represented using
15983 the `basic_block' data type.
15985 Two pointer members of the `basic_block' structure are the pointers
15986 `next_bb' and `prev_bb'. These are used to keep doubly linked chain of
15987 basic blocks in the same order as the underlying instruction stream.
15988 The chain of basic blocks is updated transparently by the provided API
15989 for manipulating the CFG. The macro `FOR_EACH_BB' can be used to visit
15990 all the basic blocks in lexicographical order. Dominator traversals
15991 are also possible using `walk_dominator_tree'. Given two basic blocks
15992 A and B, block A dominates block B if A is _always_ executed before B.
15994 The `BASIC_BLOCK' array contains all basic blocks in an unspecified
15995 order. Each `basic_block' structure has a field that holds a unique
15996 integer identifier `index' that is the index of the block in the
15997 `BASIC_BLOCK' array. The total number of basic blocks in the function
15998 is `n_basic_blocks'. Both the basic block indices and the total number
15999 of basic blocks may vary during the compilation process, as passes
16000 reorder, create, duplicate, and destroy basic blocks. The index for
16001 any block should never be greater than `last_basic_block'.
16003 Special basic blocks represent possible entry and exit points of a
16004 function. These blocks are called `ENTRY_BLOCK_PTR' and
16005 `EXIT_BLOCK_PTR'. These blocks do not contain any code, and are not
16006 elements of the `BASIC_BLOCK' array. Therefore they have been assigned
16007 unique, negative index numbers.
16009 Each `basic_block' also contains pointers to the first instruction
16010 (the "head") and the last instruction (the "tail") or "end" of the
16011 instruction stream contained in a basic block. In fact, since the
16012 `basic_block' data type is used to represent blocks in both major
16013 intermediate representations of GCC (`tree' and RTL), there are
16014 pointers to the head and end of a basic block for both representations.
16016 For RTL, these pointers are `rtx head, end'. In the RTL function
16017 representation, the head pointer always points either to a
16018 `NOTE_INSN_BASIC_BLOCK' or to a `CODE_LABEL', if present. In the RTL
16019 representation of a function, the instruction stream contains not only
16020 the "real" instructions, but also "notes". Any function that moves or
16021 duplicates the basic blocks needs to take care of updating of these
16022 notes. Many of these notes expect that the instruction stream consists
16023 of linear regions, making such updates difficult. The
16024 `NOTE_INSN_BASIC_BLOCK' note is the only kind of note that may appear
16025 in the instruction stream contained in a basic block. The instruction
16026 stream of a basic block always follows a `NOTE_INSN_BASIC_BLOCK', but
16027 zero or more `CODE_LABEL' nodes can precede the block note. A basic
16028 block ends by control flow instruction or last instruction before
16029 following `CODE_LABEL' or `NOTE_INSN_BASIC_BLOCK'. A `CODE_LABEL'
16030 cannot appear in the instruction stream of a basic block.
16032 In addition to notes, the jump table vectors are also represented as
16033 "pseudo-instructions" inside the insn stream. These vectors never
16034 appear in the basic block and should always be placed just after the
16035 table jump instructions referencing them. After removing the
16036 table-jump it is often difficult to eliminate the code computing the
16037 address and referencing the vector, so cleaning up these vectors is
16038 postponed until after liveness analysis. Thus the jump table vectors
16039 may appear in the insn stream unreferenced and without any purpose.
16040 Before any edge is made "fall-thru", the existence of such construct in
16041 the way needs to be checked by calling `can_fallthru' function.
16043 For the `tree' representation, the head and end of the basic block are
16044 being pointed to by the `stmt_list' field, but this special `tree'
16045 should never be referenced directly. Instead, at the tree level
16046 abstract containers and iterators are used to access statements and
16047 expressions in basic blocks. These iterators are called "block
16048 statement iterators" (BSIs). Grep for `^bsi' in the various `tree-*'
16049 files. The following snippet will pretty-print all the statements of
16050 the program in the GIMPLE representation.
16054 block_stmt_iterator si;
16056 for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si))
16058 tree stmt = bsi_stmt (si);
16059 print_generic_stmt (stderr, stmt, 0);
16064 File: gccint.info, Node: Edges, Next: Profile information, Prev: Basic Blocks, Up: Control Flow
16069 Edges represent possible control flow transfers from the end of some
16070 basic block A to the head of another basic block B. We say that A is a
16071 predecessor of B, and B is a successor of A. Edges are represented in
16072 GCC with the `edge' data type. Each `edge' acts as a link between two
16073 basic blocks: the `src' member of an edge points to the predecessor
16074 basic block of the `dest' basic block. The members `preds' and `succs'
16075 of the `basic_block' data type point to type-safe vectors of edges to
16076 the predecessors and successors of the block.
16078 When walking the edges in an edge vector, "edge iterators" should be
16079 used. Edge iterators are constructed using the `edge_iterator' data
16080 structure and several methods are available to operate on them:
16083 This function initializes an `edge_iterator' that points to the
16084 first edge in a vector of edges.
16087 This function initializes an `edge_iterator' that points to the
16088 last edge in a vector of edges.
16091 This predicate is `true' if an `edge_iterator' represents the last
16092 edge in an edge vector.
16094 `ei_one_before_end_p'
16095 This predicate is `true' if an `edge_iterator' represents the
16096 second last edge in an edge vector.
16099 This function takes a pointer to an `edge_iterator' and makes it
16100 point to the next edge in the sequence.
16103 This function takes a pointer to an `edge_iterator' and makes it
16104 point to the previous edge in the sequence.
16107 This function returns the `edge' currently pointed to by an
16111 This function returns the `edge' currently pointed to by an
16112 `edge_iterator', but returns `NULL' if the iterator is pointing at
16113 the end of the sequence. This function has been provided for
16114 existing code makes the assumption that a `NULL' edge indicates
16115 the end of the sequence.
16118 The convenience macro `FOR_EACH_EDGE' can be used to visit all of the
16119 edges in a sequence of predecessor or successor edges. It must not be
16120 used when an element might be removed during the traversal, otherwise
16121 elements will be missed. Here is an example of how to use the macro:
16126 FOR_EACH_EDGE (e, ei, bb->succs)
16128 if (e->flags & EDGE_FALLTHRU)
16132 There are various reasons why control flow may transfer from one block
16133 to another. One possibility is that some instruction, for example a
16134 `CODE_LABEL', in a linearized instruction stream just always starts a
16135 new basic block. In this case a "fall-thru" edge links the basic block
16136 to the first following basic block. But there are several other
16137 reasons why edges may be created. The `flags' field of the `edge' data
16138 type is used to store information about the type of edge we are dealing
16139 with. Each edge is of one of the following types:
16142 No type flags are set for edges corresponding to jump instructions.
16143 These edges are used for unconditional or conditional jumps and in
16144 RTL also for table jumps. They are the easiest to manipulate as
16145 they may be freely redirected when the flow graph is not in SSA
16149 Fall-thru edges are present in case where the basic block may
16150 continue execution to the following one without branching. These
16151 edges have the `EDGE_FALLTHRU' flag set. Unlike other types of
16152 edges, these edges must come into the basic block immediately
16153 following in the instruction stream. The function
16154 `force_nonfallthru' is available to insert an unconditional jump
16155 in the case that redirection is needed. Note that this may
16156 require creation of a new basic block.
16158 _exception handling_
16159 Exception handling edges represent possible control transfers from
16160 a trapping instruction to an exception handler. The definition of
16161 "trapping" varies. In C++, only function calls can throw, but for
16162 Java, exceptions like division by zero or segmentation fault are
16163 defined and thus each instruction possibly throwing this kind of
16164 exception needs to be handled as control flow instruction.
16165 Exception edges have the `EDGE_ABNORMAL' and `EDGE_EH' flags set.
16167 When updating the instruction stream it is easy to change possibly
16168 trapping instruction to non-trapping, by simply removing the
16169 exception edge. The opposite conversion is difficult, but should
16170 not happen anyway. The edges can be eliminated via
16171 `purge_dead_edges' call.
16173 In the RTL representation, the destination of an exception edge is
16174 specified by `REG_EH_REGION' note attached to the insn. In case
16175 of a trapping call the `EDGE_ABNORMAL_CALL' flag is set too. In
16176 the `tree' representation, this extra flag is not set.
16178 In the RTL representation, the predicate `may_trap_p' may be used
16179 to check whether instruction still may trap or not. For the tree
16180 representation, the `tree_could_trap_p' predicate is available,
16181 but this predicate only checks for possible memory traps, as in
16182 dereferencing an invalid pointer location.
16185 Sibling calls or tail calls terminate the function in a
16186 non-standard way and thus an edge to the exit must be present.
16187 `EDGE_SIBCALL' and `EDGE_ABNORMAL' are set in such case. These
16188 edges only exist in the RTL representation.
16191 Computed jumps contain edges to all labels in the function
16192 referenced from the code. All those edges have `EDGE_ABNORMAL'
16193 flag set. The edges used to represent computed jumps often cause
16194 compile time performance problems, since functions consisting of
16195 many taken labels and many computed jumps may have _very_ dense
16196 flow graphs, so these edges need to be handled with special care.
16197 During the earlier stages of the compilation process, GCC tries to
16198 avoid such dense flow graphs by factoring computed jumps. For
16199 example, given the following series of jumps,
16210 factoring the computed jumps results in the following code sequence
16211 which has a much simpler flow graph:
16225 However, the classic problem with this transformation is that it
16226 has a runtime cost in there resulting code: An extra jump.
16227 Therefore, the computed jumps are un-factored in the later passes
16228 of the compiler. Be aware of that when you work on passes in that
16229 area. There have been numerous examples already where the compile
16230 time for code with unfactored computed jumps caused some serious
16233 _nonlocal goto handlers_
16234 GCC allows nested functions to return into caller using a `goto'
16235 to a label passed to as an argument to the callee. The labels
16236 passed to nested functions contain special code to cleanup after
16237 function call. Such sections of code are referred to as "nonlocal
16238 goto receivers". If a function contains such nonlocal goto
16239 receivers, an edge from the call to the label is created with the
16240 `EDGE_ABNORMAL' and `EDGE_ABNORMAL_CALL' flags set.
16242 _function entry points_
16243 By definition, execution of function starts at basic block 0, so
16244 there is always an edge from the `ENTRY_BLOCK_PTR' to basic block
16245 0. There is no `tree' representation for alternate entry points at
16246 this moment. In RTL, alternate entry points are specified by
16247 `CODE_LABEL' with `LABEL_ALTERNATE_NAME' defined. This feature is
16248 currently used for multiple entry point prologues and is limited
16249 to post-reload passes only. This can be used by back-ends to emit
16250 alternate prologues for functions called from different contexts.
16251 In future full support for multiple entry functions defined by
16252 Fortran 90 needs to be implemented.
16255 In the pre-reload representation a function terminates after the
16256 last instruction in the insn chain and no explicit return
16257 instructions are used. This corresponds to the fall-thru edge
16258 into exit block. After reload, optimal RTL epilogues are used
16259 that use explicit (conditional) return instructions that are
16260 represented by edges with no flags set.
16264 File: gccint.info, Node: Profile information, Next: Maintaining the CFG, Prev: Edges, Up: Control Flow
16266 15.3 Profile information
16267 ========================
16269 In many cases a compiler must make a choice whether to trade speed in
16270 one part of code for speed in another, or to trade code size for code
16271 speed. In such cases it is useful to know information about how often
16272 some given block will be executed. That is the purpose for maintaining
16273 profile within the flow graph. GCC can handle profile information
16274 obtained through "profile feedback", but it can also estimate branch
16275 probabilities based on statics and heuristics.
16277 The feedback based profile is produced by compiling the program with
16278 instrumentation, executing it on a train run and reading the numbers of
16279 executions of basic blocks and edges back to the compiler while
16280 re-compiling the program to produce the final executable. This method
16281 provides very accurate information about where a program spends most of
16282 its time on the train run. Whether it matches the average run of
16283 course depends on the choice of train data set, but several studies
16284 have shown that the behavior of a program usually changes just
16285 marginally over different data sets.
16287 When profile feedback is not available, the compiler may be asked to
16288 attempt to predict the behavior of each branch in the program using a
16289 set of heuristics (see `predict.def' for details) and compute estimated
16290 frequencies of each basic block by propagating the probabilities over
16293 Each `basic_block' contains two integer fields to represent profile
16294 information: `frequency' and `count'. The `frequency' is an estimation
16295 how often is basic block executed within a function. It is represented
16296 as an integer scaled in the range from 0 to `BB_FREQ_BASE'. The most
16297 frequently executed basic block in function is initially set to
16298 `BB_FREQ_BASE' and the rest of frequencies are scaled accordingly.
16299 During optimization, the frequency of the most frequent basic block can
16300 both decrease (for instance by loop unrolling) or grow (for instance by
16301 cross-jumping optimization), so scaling sometimes has to be performed
16304 The `count' contains hard-counted numbers of execution measured during
16305 training runs and is nonzero only when profile feedback is available.
16306 This value is represented as the host's widest integer (typically a 64
16307 bit integer) of the special type `gcov_type'.
16309 Most optimization passes can use only the frequency information of a
16310 basic block, but a few passes may want to know hard execution counts.
16311 The frequencies should always match the counts after scaling, however
16312 during updating of the profile information numerical error may
16313 accumulate into quite large errors.
16315 Each edge also contains a branch probability field: an integer in the
16316 range from 0 to `REG_BR_PROB_BASE'. It represents probability of
16317 passing control from the end of the `src' basic block to the `dest'
16318 basic block, i.e. the probability that control will flow along this
16319 edge. The `EDGE_FREQUENCY' macro is available to compute how
16320 frequently a given edge is taken. There is a `count' field for each
16321 edge as well, representing same information as for a basic block.
16323 The basic block frequencies are not represented in the instruction
16324 stream, but in the RTL representation the edge frequencies are
16325 represented for conditional jumps (via the `REG_BR_PROB' macro) since
16326 they are used when instructions are output to the assembly file and the
16327 flow graph is no longer maintained.
16329 The probability that control flow arrives via a given edge to its
16330 destination basic block is called "reverse probability" and is not
16331 directly represented, but it may be easily computed from frequencies of
16334 Updating profile information is a delicate task that can unfortunately
16335 not be easily integrated with the CFG manipulation API. Many of the
16336 functions and hooks to modify the CFG, such as
16337 `redirect_edge_and_branch', do not have enough information to easily
16338 update the profile, so updating it is in the majority of cases left up
16339 to the caller. It is difficult to uncover bugs in the profile updating
16340 code, because they manifest themselves only by producing worse code,
16341 and checking profile consistency is not possible because of numeric
16342 error accumulation. Hence special attention needs to be given to this
16343 issue in each pass that modifies the CFG.
16345 It is important to point out that `REG_BR_PROB_BASE' and
16346 `BB_FREQ_BASE' are both set low enough to be possible to compute second
16347 power of any frequency or probability in the flow graph, it is not
16348 possible to even square the `count' field, as modern CPUs are fast
16349 enough to execute $2^32$ operations quickly.
16352 File: gccint.info, Node: Maintaining the CFG, Next: Liveness information, Prev: Profile information, Up: Control Flow
16354 15.4 Maintaining the CFG
16355 ========================
16357 An important task of each compiler pass is to keep both the control
16358 flow graph and all profile information up-to-date. Reconstruction of
16359 the control flow graph after each pass is not an option, since it may be
16360 very expensive and lost profile information cannot be reconstructed at
16363 GCC has two major intermediate representations, and both use the
16364 `basic_block' and `edge' data types to represent control flow. Both
16365 representations share as much of the CFG maintenance code as possible.
16366 For each representation, a set of "hooks" is defined so that each
16367 representation can provide its own implementation of CFG manipulation
16368 routines when necessary. These hooks are defined in `cfghooks.h'.
16369 There are hooks for almost all common CFG manipulations, including
16370 block splitting and merging, edge redirection and creating and deleting
16371 basic blocks. These hooks should provide everything you need to
16372 maintain and manipulate the CFG in both the RTL and `tree'
16375 At the moment, the basic block boundaries are maintained transparently
16376 when modifying instructions, so there rarely is a need to move them
16377 manually (such as in case someone wants to output instruction outside
16378 basic block explicitly). Often the CFG may be better viewed as
16379 integral part of instruction chain, than structure built on the top of
16380 it. However, in principle the control flow graph for the `tree'
16381 representation is _not_ an integral part of the representation, in that
16382 a function tree may be expanded without first building a flow graph
16383 for the `tree' representation at all. This happens when compiling
16384 without any `tree' optimization enabled. When the `tree' optimizations
16385 are enabled and the instruction stream is rewritten in SSA form, the
16386 CFG is very tightly coupled with the instruction stream. In
16387 particular, statement insertion and removal has to be done with care.
16388 In fact, the whole `tree' representation can not be easily used or
16389 maintained without proper maintenance of the CFG simultaneously.
16391 In the RTL representation, each instruction has a `BLOCK_FOR_INSN'
16392 value that represents pointer to the basic block that contains the
16393 instruction. In the `tree' representation, the function `bb_for_stmt'
16394 returns a pointer to the basic block containing the queried statement.
16396 When changes need to be applied to a function in its `tree'
16397 representation, "block statement iterators" should be used. These
16398 iterators provide an integrated abstraction of the flow graph and the
16399 instruction stream. Block statement iterators are constructed using
16400 the `block_stmt_iterator' data structure and several modifier are
16401 available, including the following:
16404 This function initializes a `block_stmt_iterator' that points to
16405 the first non-empty statement in a basic block.
16408 This function initializes a `block_stmt_iterator' that points to
16409 the last statement in a basic block.
16412 This predicate is `true' if a `block_stmt_iterator' represents the
16413 end of a basic block.
16416 This function takes a `block_stmt_iterator' and makes it point to
16420 This function takes a `block_stmt_iterator' and makes it point to
16424 This function inserts a statement after the `block_stmt_iterator'
16425 passed in. The final parameter determines whether the statement
16426 iterator is updated to point to the newly inserted statement, or
16427 left pointing to the original statement.
16429 `bsi_insert_before'
16430 This function inserts a statement before the `block_stmt_iterator'
16431 passed in. The final parameter determines whether the statement
16432 iterator is updated to point to the newly inserted statement, or
16433 left pointing to the original statement.
16436 This function removes the `block_stmt_iterator' passed in and
16437 rechains the remaining statements in a basic block, if any.
16439 In the RTL representation, the macros `BB_HEAD' and `BB_END' may be
16440 used to get the head and end `rtx' of a basic block. No abstract
16441 iterators are defined for traversing the insn chain, but you can just
16442 use `NEXT_INSN' and `PREV_INSN' instead. See *Note Insns::.
16444 Usually a code manipulating pass simplifies the instruction stream and
16445 the flow of control, possibly eliminating some edges. This may for
16446 example happen when a conditional jump is replaced with an
16447 unconditional jump, but also when simplifying possibly trapping
16448 instruction to non-trapping while compiling Java. Updating of edges is
16449 not transparent and each optimization pass is required to do so
16450 manually. However only few cases occur in practice. The pass may call
16451 `purge_dead_edges' on a given basic block to remove superfluous edges,
16454 Another common scenario is redirection of branch instructions, but
16455 this is best modeled as redirection of edges in the control flow graph
16456 and thus use of `redirect_edge_and_branch' is preferred over more low
16457 level functions, such as `redirect_jump' that operate on RTL chain
16458 only. The CFG hooks defined in `cfghooks.h' should provide the
16459 complete API required for manipulating and maintaining the CFG.
16461 It is also possible that a pass has to insert control flow instruction
16462 into the middle of a basic block, thus creating an entry point in the
16463 middle of the basic block, which is impossible by definition: The block
16464 must be split to make sure it only has one entry point, i.e. the head
16465 of the basic block. The CFG hook `split_block' may be used when an
16466 instruction in the middle of a basic block has to become the target of
16467 a jump or branch instruction.
16469 For a global optimizer, a common operation is to split edges in the
16470 flow graph and insert instructions on them. In the RTL representation,
16471 this can be easily done using the `insert_insn_on_edge' function that
16472 emits an instruction "on the edge", caching it for a later
16473 `commit_edge_insertions' call that will take care of moving the
16474 inserted instructions off the edge into the instruction stream
16475 contained in a basic block. This includes the creation of new basic
16476 blocks where needed. In the `tree' representation, the equivalent
16477 functions are `bsi_insert_on_edge' which inserts a block statement
16478 iterator on an edge, and `bsi_commit_edge_inserts' which flushes the
16479 instruction to actual instruction stream.
16481 While debugging the optimization pass, an `verify_flow_info' function
16482 may be useful to find bugs in the control flow graph updating code.
16484 Note that at present, the representation of control flow in the `tree'
16485 representation is discarded before expanding to RTL. Long term the CFG
16486 should be maintained and "expanded" to the RTL representation along
16487 with the function `tree' itself.
16490 File: gccint.info, Node: Liveness information, Prev: Maintaining the CFG, Up: Control Flow
16492 15.5 Liveness information
16493 =========================
16495 Liveness information is useful to determine whether some register is
16496 "live" at given point of program, i.e. that it contains a value that
16497 may be used at a later point in the program. This information is used,
16498 for instance, during register allocation, as the pseudo registers only
16499 need to be assigned to a unique hard register or to a stack slot if
16500 they are live. The hard registers and stack slots may be freely reused
16501 for other values when a register is dead.
16503 Liveness information is available in the back end starting with
16504 `pass_df_initialize' and ending with `pass_df_finish'. Three flavors
16505 of live analysis are available: With `LR', it is possible to determine
16506 at any point `P' in the function if the register may be used on some
16507 path from `P' to the end of the function. With `UR', it is possible to
16508 determine if there is a path from the beginning of the function to `P'
16509 that defines the variable. `LIVE' is the intersection of the `LR' and
16510 `UR' and a variable is live at `P' if there is both an assignment that
16511 reaches it from the beginning of the function and a uses that can be
16512 reached on some path from `P' to the end of the function.
16514 In general `LIVE' is the most useful of the three. The macros
16515 `DF_[LR,UR,LIVE]_[IN,OUT]' can be used to access this information. The
16516 macros take a basic block number and return a bitmap that is indexed by
16517 the register number. This information is only guaranteed to be up to
16518 date after calls are made to `df_analyze'. See the file `df-core.c'
16519 for details on using the dataflow.
16521 The liveness information is stored partly in the RTL instruction stream
16522 and partly in the flow graph. Local information is stored in the
16523 instruction stream: Each instruction may contain `REG_DEAD' notes
16524 representing that the value of a given register is no longer needed, or
16525 `REG_UNUSED' notes representing that the value computed by the
16526 instruction is never used. The second is useful for instructions
16527 computing multiple values at once.
16530 File: gccint.info, Node: Machine Desc, Next: Target Macros, Prev: Loop Analysis and Representation, Up: Top
16532 16 Machine Descriptions
16533 ***********************
16535 A machine description has two parts: a file of instruction patterns
16536 (`.md' file) and a C header file of macro definitions.
16538 The `.md' file for a target machine contains a pattern for each
16539 instruction that the target machine supports (or at least each
16540 instruction that is worth telling the compiler about). It may also
16541 contain comments. A semicolon causes the rest of the line to be a
16542 comment, unless the semicolon is inside a quoted string.
16544 See the next chapter for information on the C header file.
16548 * Overview:: How the machine description is used.
16549 * Patterns:: How to write instruction patterns.
16550 * Example:: An explained example of a `define_insn' pattern.
16551 * RTL Template:: The RTL template defines what insns match a pattern.
16552 * Output Template:: The output template says how to make assembler code
16554 * Output Statement:: For more generality, write C code to output
16555 the assembler code.
16556 * Predicates:: Controlling what kinds of operands can be used
16558 * Constraints:: Fine-tuning operand selection.
16559 * Standard Names:: Names mark patterns to use for code generation.
16560 * Pattern Ordering:: When the order of patterns makes a difference.
16561 * Dependent Patterns:: Having one pattern may make you need another.
16562 * Jump Patterns:: Special considerations for patterns for jump insns.
16563 * Looping Patterns:: How to define patterns for special looping insns.
16564 * Insn Canonicalizations::Canonicalization of Instructions
16565 * Expander Definitions::Generating a sequence of several RTL insns
16566 for a standard operation.
16567 * Insn Splitting:: Splitting Instructions into Multiple Instructions.
16568 * Including Patterns:: Including Patterns in Machine Descriptions.
16569 * Peephole Definitions::Defining machine-specific peephole optimizations.
16570 * Insn Attributes:: Specifying the value of attributes for generated insns.
16571 * Conditional Execution::Generating `define_insn' patterns for
16573 * Constant Definitions::Defining symbolic constants that can be used in the
16575 * Iterators:: Using iterators to generate patterns from a template.
16578 File: gccint.info, Node: Overview, Next: Patterns, Up: Machine Desc
16580 16.1 Overview of How the Machine Description is Used
16581 ====================================================
16583 There are three main conversions that happen in the compiler:
16585 1. The front end reads the source code and builds a parse tree.
16587 2. The parse tree is used to generate an RTL insn list based on named
16588 instruction patterns.
16590 3. The insn list is matched against the RTL templates to produce
16594 For the generate pass, only the names of the insns matter, from either
16595 a named `define_insn' or a `define_expand'. The compiler will choose
16596 the pattern with the right name and apply the operands according to the
16597 documentation later in this chapter, without regard for the RTL
16598 template or operand constraints. Note that the names the compiler looks
16599 for are hard-coded in the compiler--it will ignore unnamed patterns and
16600 patterns with names it doesn't know about, but if you don't provide a
16601 named pattern it needs, it will abort.
16603 If a `define_insn' is used, the template given is inserted into the
16604 insn list. If a `define_expand' is used, one of three things happens,
16605 based on the condition logic. The condition logic may manually create
16606 new insns for the insn list, say via `emit_insn()', and invoke `DONE'.
16607 For certain named patterns, it may invoke `FAIL' to tell the compiler
16608 to use an alternate way of performing that task. If it invokes neither
16609 `DONE' nor `FAIL', the template given in the pattern is inserted, as if
16610 the `define_expand' were a `define_insn'.
16612 Once the insn list is generated, various optimization passes convert,
16613 replace, and rearrange the insns in the insn list. This is where the
16614 `define_split' and `define_peephole' patterns get used, for example.
16616 Finally, the insn list's RTL is matched up with the RTL templates in
16617 the `define_insn' patterns, and those patterns are used to emit the
16618 final assembly code. For this purpose, each named `define_insn' acts
16619 like it's unnamed, since the names are ignored.
16622 File: gccint.info, Node: Patterns, Next: Example, Prev: Overview, Up: Machine Desc
16624 16.2 Everything about Instruction Patterns
16625 ==========================================
16627 Each instruction pattern contains an incomplete RTL expression, with
16628 pieces to be filled in later, operand constraints that restrict how the
16629 pieces can be filled in, and an output pattern or C code to generate
16630 the assembler output, all wrapped up in a `define_insn' expression.
16632 A `define_insn' is an RTL expression containing four or five operands:
16634 1. An optional name. The presence of a name indicate that this
16635 instruction pattern can perform a certain standard job for the
16636 RTL-generation pass of the compiler. This pass knows certain
16637 names and will use the instruction patterns with those names, if
16638 the names are defined in the machine description.
16640 The absence of a name is indicated by writing an empty string
16641 where the name should go. Nameless instruction patterns are never
16642 used for generating RTL code, but they may permit several simpler
16643 insns to be combined later on.
16645 Names that are not thus known and used in RTL-generation have no
16646 effect; they are equivalent to no name at all.
16648 For the purpose of debugging the compiler, you may also specify a
16649 name beginning with the `*' character. Such a name is used only
16650 for identifying the instruction in RTL dumps; it is entirely
16651 equivalent to having a nameless pattern for all other purposes.
16653 2. The "RTL template" (*note RTL Template::) is a vector of incomplete
16654 RTL expressions which show what the instruction should look like.
16655 It is incomplete because it may contain `match_operand',
16656 `match_operator', and `match_dup' expressions that stand for
16657 operands of the instruction.
16659 If the vector has only one element, that element is the template
16660 for the instruction pattern. If the vector has multiple elements,
16661 then the instruction pattern is a `parallel' expression containing
16662 the elements described.
16664 3. A condition. This is a string which contains a C expression that
16665 is the final test to decide whether an insn body matches this
16668 For a named pattern, the condition (if present) may not depend on
16669 the data in the insn being matched, but only the
16670 target-machine-type flags. The compiler needs to test these
16671 conditions during initialization in order to learn exactly which
16672 named instructions are available in a particular run.
16674 For nameless patterns, the condition is applied only when matching
16675 an individual insn, and only after the insn has matched the
16676 pattern's recognition template. The insn's operands may be found
16677 in the vector `operands'. For an insn where the condition has
16678 once matched, it can't be used to control register allocation, for
16679 example by excluding certain hard registers or hard register
16682 4. The "output template": a string that says how to output matching
16683 insns as assembler code. `%' in this string specifies where to
16684 substitute the value of an operand. *Note Output Template::.
16686 When simple substitution isn't general enough, you can specify a
16687 piece of C code to compute the output. *Note Output Statement::.
16689 5. Optionally, a vector containing the values of attributes for insns
16690 matching this pattern. *Note Insn Attributes::.
16693 File: gccint.info, Node: Example, Next: RTL Template, Prev: Patterns, Up: Machine Desc
16695 16.3 Example of `define_insn'
16696 =============================
16698 Here is an actual example of an instruction pattern, for the
16701 (define_insn "tstsi"
16703 (match_operand:SI 0 "general_operand" "rm"))]
16707 if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
16708 return \"tstl %0\";
16709 return \"cmpl #0,%0\";
16712 This can also be written using braced strings:
16714 (define_insn "tstsi"
16716 (match_operand:SI 0 "general_operand" "rm"))]
16719 if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
16721 return "cmpl #0,%0";
16724 This is an instruction that sets the condition codes based on the
16725 value of a general operand. It has no condition, so any insn whose RTL
16726 description has the form shown may be handled according to this
16727 pattern. The name `tstsi' means "test a `SImode' value" and tells the
16728 RTL generation pass that, when it is necessary to test such a value, an
16729 insn to do so can be constructed using this pattern.
16731 The output control string is a piece of C code which chooses which
16732 output template to return based on the kind of operand and the specific
16733 type of CPU for which code is being generated.
16735 `"rm"' is an operand constraint. Its meaning is explained below.
16738 File: gccint.info, Node: RTL Template, Next: Output Template, Prev: Example, Up: Machine Desc
16743 The RTL template is used to define which insns match the particular
16744 pattern and how to find their operands. For named patterns, the RTL
16745 template also says how to construct an insn from specified operands.
16747 Construction involves substituting specified operands into a copy of
16748 the template. Matching involves determining the values that serve as
16749 the operands in the insn being matched. Both of these activities are
16750 controlled by special expression types that direct matching and
16751 substitution of the operands.
16753 `(match_operand:M N PREDICATE CONSTRAINT)'
16754 This expression is a placeholder for operand number N of the insn.
16755 When constructing an insn, operand number N will be substituted
16756 at this point. When matching an insn, whatever appears at this
16757 position in the insn will be taken as operand number N; but it
16758 must satisfy PREDICATE or this instruction pattern will not match
16761 Operand numbers must be chosen consecutively counting from zero in
16762 each instruction pattern. There may be only one `match_operand'
16763 expression in the pattern for each operand number. Usually
16764 operands are numbered in the order of appearance in `match_operand'
16765 expressions. In the case of a `define_expand', any operand numbers
16766 used only in `match_dup' expressions have higher values than all
16767 other operand numbers.
16769 PREDICATE is a string that is the name of a function that accepts
16770 two arguments, an expression and a machine mode. *Note
16771 Predicates::. During matching, the function will be called with
16772 the putative operand as the expression and M as the mode argument
16773 (if M is not specified, `VOIDmode' will be used, which normally
16774 causes PREDICATE to accept any mode). If it returns zero, this
16775 instruction pattern fails to match. PREDICATE may be an empty
16776 string; then it means no test is to be done on the operand, so
16777 anything which occurs in this position is valid.
16779 Most of the time, PREDICATE will reject modes other than M--but
16780 not always. For example, the predicate `address_operand' uses M
16781 as the mode of memory ref that the address should be valid for.
16782 Many predicates accept `const_int' nodes even though their mode is
16785 CONSTRAINT controls reloading and the choice of the best register
16786 class to use for a value, as explained later (*note Constraints::).
16787 If the constraint would be an empty string, it can be omitted.
16789 People are often unclear on the difference between the constraint
16790 and the predicate. The predicate helps decide whether a given
16791 insn matches the pattern. The constraint plays no role in this
16792 decision; instead, it controls various decisions in the case of an
16793 insn which does match.
16795 `(match_scratch:M N CONSTRAINT)'
16796 This expression is also a placeholder for operand number N and
16797 indicates that operand must be a `scratch' or `reg' expression.
16799 When matching patterns, this is equivalent to
16801 (match_operand:M N "scratch_operand" PRED)
16803 but, when generating RTL, it produces a (`scratch':M) expression.
16805 If the last few expressions in a `parallel' are `clobber'
16806 expressions whose operands are either a hard register or
16807 `match_scratch', the combiner can add or delete them when
16808 necessary. *Note Side Effects::.
16811 This expression is also a placeholder for operand number N. It is
16812 used when the operand needs to appear more than once in the insn.
16814 In construction, `match_dup' acts just like `match_operand': the
16815 operand is substituted into the insn being constructed. But in
16816 matching, `match_dup' behaves differently. It assumes that operand
16817 number N has already been determined by a `match_operand'
16818 appearing earlier in the recognition template, and it matches only
16819 an identical-looking expression.
16821 Note that `match_dup' should not be used to tell the compiler that
16822 a particular register is being used for two operands (example:
16823 `add' that adds one register to another; the second register is
16824 both an input operand and the output operand). Use a matching
16825 constraint (*note Simple Constraints::) for those. `match_dup' is
16826 for the cases where one operand is used in two places in the
16827 template, such as an instruction that computes both a quotient and
16828 a remainder, where the opcode takes two input operands but the RTL
16829 template has to refer to each of those twice; once for the
16830 quotient pattern and once for the remainder pattern.
16832 `(match_operator:M N PREDICATE [OPERANDS...])'
16833 This pattern is a kind of placeholder for a variable RTL expression
16836 When constructing an insn, it stands for an RTL expression whose
16837 expression code is taken from that of operand N, and whose
16838 operands are constructed from the patterns OPERANDS.
16840 When matching an expression, it matches an expression if the
16841 function PREDICATE returns nonzero on that expression _and_ the
16842 patterns OPERANDS match the operands of the expression.
16844 Suppose that the function `commutative_operator' is defined as
16845 follows, to match any expression whose operator is one of the
16846 commutative arithmetic operators of RTL and whose mode is MODE:
16849 commutative_integer_operator (x, mode)
16851 enum machine_mode mode;
16853 enum rtx_code code = GET_CODE (x);
16854 if (GET_MODE (x) != mode)
16856 return (GET_RTX_CLASS (code) == RTX_COMM_ARITH
16857 || code == EQ || code == NE);
16860 Then the following pattern will match any RTL expression consisting
16861 of a commutative operator applied to two general operands:
16863 (match_operator:SI 3 "commutative_operator"
16864 [(match_operand:SI 1 "general_operand" "g")
16865 (match_operand:SI 2 "general_operand" "g")])
16867 Here the vector `[OPERANDS...]' contains two patterns because the
16868 expressions to be matched all contain two operands.
16870 When this pattern does match, the two operands of the commutative
16871 operator are recorded as operands 1 and 2 of the insn. (This is
16872 done by the two instances of `match_operand'.) Operand 3 of the
16873 insn will be the entire commutative expression: use `GET_CODE
16874 (operands[3])' to see which commutative operator was used.
16876 The machine mode M of `match_operator' works like that of
16877 `match_operand': it is passed as the second argument to the
16878 predicate function, and that function is solely responsible for
16879 deciding whether the expression to be matched "has" that mode.
16881 When constructing an insn, argument 3 of the gen-function will
16882 specify the operation (i.e. the expression code) for the
16883 expression to be made. It should be an RTL expression, whose
16884 expression code is copied into a new expression whose operands are
16885 arguments 1 and 2 of the gen-function. The subexpressions of
16886 argument 3 are not used; only its expression code matters.
16888 When `match_operator' is used in a pattern for matching an insn,
16889 it usually best if the operand number of the `match_operator' is
16890 higher than that of the actual operands of the insn. This improves
16891 register allocation because the register allocator often looks at
16892 operands 1 and 2 of insns to see if it can do register tying.
16894 There is no way to specify constraints in `match_operator'. The
16895 operand of the insn which corresponds to the `match_operator'
16896 never has any constraints because it is never reloaded as a whole.
16897 However, if parts of its OPERANDS are matched by `match_operand'
16898 patterns, those parts may have constraints of their own.
16900 `(match_op_dup:M N[OPERANDS...])'
16901 Like `match_dup', except that it applies to operators instead of
16902 operands. When constructing an insn, operand number N will be
16903 substituted at this point. But in matching, `match_op_dup' behaves
16904 differently. It assumes that operand number N has already been
16905 determined by a `match_operator' appearing earlier in the
16906 recognition template, and it matches only an identical-looking
16909 `(match_parallel N PREDICATE [SUBPAT...])'
16910 This pattern is a placeholder for an insn that consists of a
16911 `parallel' expression with a variable number of elements. This
16912 expression should only appear at the top level of an insn pattern.
16914 When constructing an insn, operand number N will be substituted at
16915 this point. When matching an insn, it matches if the body of the
16916 insn is a `parallel' expression with at least as many elements as
16917 the vector of SUBPAT expressions in the `match_parallel', if each
16918 SUBPAT matches the corresponding element of the `parallel', _and_
16919 the function PREDICATE returns nonzero on the `parallel' that is
16920 the body of the insn. It is the responsibility of the predicate
16921 to validate elements of the `parallel' beyond those listed in the
16924 A typical use of `match_parallel' is to match load and store
16925 multiple expressions, which can contain a variable number of
16926 elements in a `parallel'. For example,
16929 [(match_parallel 0 "load_multiple_operation"
16930 [(set (match_operand:SI 1 "gpc_reg_operand" "=r")
16931 (match_operand:SI 2 "memory_operand" "m"))
16933 (clobber (reg:SI 179))])]
16937 This example comes from `a29k.md'. The function
16938 `load_multiple_operation' is defined in `a29k.c' and checks that
16939 subsequent elements in the `parallel' are the same as the `set' in
16940 the pattern, except that they are referencing subsequent registers
16941 and memory locations.
16943 An insn that matches this pattern might look like:
16946 [(set (reg:SI 20) (mem:SI (reg:SI 100)))
16948 (clobber (reg:SI 179))
16950 (mem:SI (plus:SI (reg:SI 100)
16953 (mem:SI (plus:SI (reg:SI 100)
16956 `(match_par_dup N [SUBPAT...])'
16957 Like `match_op_dup', but for `match_parallel' instead of
16962 File: gccint.info, Node: Output Template, Next: Output Statement, Prev: RTL Template, Up: Machine Desc
16964 16.5 Output Templates and Operand Substitution
16965 ==============================================
16967 The "output template" is a string which specifies how to output the
16968 assembler code for an instruction pattern. Most of the template is a
16969 fixed string which is output literally. The character `%' is used to
16970 specify where to substitute an operand; it can also be used to identify
16971 places where different variants of the assembler require different
16974 In the simplest case, a `%' followed by a digit N says to output
16975 operand N at that point in the string.
16977 `%' followed by a letter and a digit says to output an operand in an
16978 alternate fashion. Four letters have standard, built-in meanings
16979 described below. The machine description macro `PRINT_OPERAND' can
16980 define additional letters with nonstandard meanings.
16982 `%cDIGIT' can be used to substitute an operand that is a constant
16983 value without the syntax that normally indicates an immediate operand.
16985 `%nDIGIT' is like `%cDIGIT' except that the value of the constant is
16986 negated before printing.
16988 `%aDIGIT' can be used to substitute an operand as if it were a memory
16989 reference, with the actual operand treated as the address. This may be
16990 useful when outputting a "load address" instruction, because often the
16991 assembler syntax for such an instruction requires you to write the
16992 operand as if it were a memory reference.
16994 `%lDIGIT' is used to substitute a `label_ref' into a jump instruction.
16996 `%=' outputs a number which is unique to each instruction in the
16997 entire compilation. This is useful for making local labels to be
16998 referred to more than once in a single template that generates multiple
16999 assembler instructions.
17001 `%' followed by a punctuation character specifies a substitution that
17002 does not use an operand. Only one case is standard: `%%' outputs a `%'
17003 into the assembler code. Other nonstandard cases can be defined in the
17004 `PRINT_OPERAND' macro. You must also define which punctuation
17005 characters are valid with the `PRINT_OPERAND_PUNCT_VALID_P' macro.
17007 The template may generate multiple assembler instructions. Write the
17008 text for the instructions, with `\;' between them.
17010 When the RTL contains two operands which are required by constraint to
17011 match each other, the output template must refer only to the
17012 lower-numbered operand. Matching operands are not always identical,
17013 and the rest of the compiler arranges to put the proper RTL expression
17014 for printing into the lower-numbered operand.
17016 One use of nonstandard letters or punctuation following `%' is to
17017 distinguish between different assembler languages for the same machine;
17018 for example, Motorola syntax versus MIT syntax for the 68000. Motorola
17019 syntax requires periods in most opcode names, while MIT syntax does
17020 not. For example, the opcode `movel' in MIT syntax is `move.l' in
17021 Motorola syntax. The same file of patterns is used for both kinds of
17022 output syntax, but the character sequence `%.' is used in each place
17023 where Motorola syntax wants a period. The `PRINT_OPERAND' macro for
17024 Motorola syntax defines the sequence to output a period; the macro for
17025 MIT syntax defines it to do nothing.
17027 As a special case, a template consisting of the single character `#'
17028 instructs the compiler to first split the insn, and then output the
17029 resulting instructions separately. This helps eliminate redundancy in
17030 the output templates. If you have a `define_insn' that needs to emit
17031 multiple assembler instructions, and there is an matching `define_split'
17032 already defined, then you can simply use `#' as the output template
17033 instead of writing an output template that emits the multiple assembler
17036 If the macro `ASSEMBLER_DIALECT' is defined, you can use construct of
17037 the form `{option0|option1|option2}' in the templates. These describe
17038 multiple variants of assembler language syntax. *Note Instruction
17042 File: gccint.info, Node: Output Statement, Next: Predicates, Prev: Output Template, Up: Machine Desc
17044 16.6 C Statements for Assembler Output
17045 ======================================
17047 Often a single fixed template string cannot produce correct and
17048 efficient assembler code for all the cases that are recognized by a
17049 single instruction pattern. For example, the opcodes may depend on the
17050 kinds of operands; or some unfortunate combinations of operands may
17051 require extra machine instructions.
17053 If the output control string starts with a `@', then it is actually a
17054 series of templates, each on a separate line. (Blank lines and leading
17055 spaces and tabs are ignored.) The templates correspond to the
17056 pattern's constraint alternatives (*note Multi-Alternative::). For
17057 example, if a target machine has a two-address add instruction `addr'
17058 to add into a register and another `addm' to add a register to memory,
17059 you might write this pattern:
17061 (define_insn "addsi3"
17062 [(set (match_operand:SI 0 "general_operand" "=r,m")
17063 (plus:SI (match_operand:SI 1 "general_operand" "0,0")
17064 (match_operand:SI 2 "general_operand" "g,r")))]
17070 If the output control string starts with a `*', then it is not an
17071 output template but rather a piece of C program that should compute a
17072 template. It should execute a `return' statement to return the
17073 template-string you want. Most such templates use C string literals,
17074 which require doublequote characters to delimit them. To include these
17075 doublequote characters in the string, prefix each one with `\'.
17077 If the output control string is written as a brace block instead of a
17078 double-quoted string, it is automatically assumed to be C code. In that
17079 case, it is not necessary to put in a leading asterisk, or to escape the
17080 doublequotes surrounding C string literals.
17082 The operands may be found in the array `operands', whose C data type
17085 It is very common to select different ways of generating assembler code
17086 based on whether an immediate operand is within a certain range. Be
17087 careful when doing this, because the result of `INTVAL' is an integer
17088 on the host machine. If the host machine has more bits in an `int'
17089 than the target machine has in the mode in which the constant will be
17090 used, then some of the bits you get from `INTVAL' will be superfluous.
17091 For proper results, you must carefully disregard the values of those
17094 It is possible to output an assembler instruction and then go on to
17095 output or compute more of them, using the subroutine `output_asm_insn'.
17096 This receives two arguments: a template-string and a vector of
17097 operands. The vector may be `operands', or it may be another array of
17098 `rtx' that you declare locally and initialize yourself.
17100 When an insn pattern has multiple alternatives in its constraints,
17101 often the appearance of the assembler code is determined mostly by
17102 which alternative was matched. When this is so, the C code can test
17103 the variable `which_alternative', which is the ordinal number of the
17104 alternative that was actually satisfied (0 for the first, 1 for the
17105 second alternative, etc.).
17107 For example, suppose there are two opcodes for storing zero, `clrreg'
17108 for registers and `clrmem' for memory locations. Here is how a pattern
17109 could use `which_alternative' to choose between them:
17112 [(set (match_operand:SI 0 "general_operand" "=r,m")
17116 return (which_alternative == 0
17117 ? "clrreg %0" : "clrmem %0");
17120 The example above, where the assembler code to generate was _solely_
17121 determined by the alternative, could also have been specified as
17122 follows, having the output control string start with a `@':
17125 [(set (match_operand:SI 0 "general_operand" "=r,m")
17133 File: gccint.info, Node: Predicates, Next: Constraints, Prev: Output Statement, Up: Machine Desc
17138 A predicate determines whether a `match_operand' or `match_operator'
17139 expression matches, and therefore whether the surrounding instruction
17140 pattern will be used for that combination of operands. GCC has a
17141 number of machine-independent predicates, and you can define
17142 machine-specific predicates as needed. By convention, predicates used
17143 with `match_operand' have names that end in `_operand', and those used
17144 with `match_operator' have names that end in `_operator'.
17146 All predicates are Boolean functions (in the mathematical sense) of
17147 two arguments: the RTL expression that is being considered at that
17148 position in the instruction pattern, and the machine mode that the
17149 `match_operand' or `match_operator' specifies. In this section, the
17150 first argument is called OP and the second argument MODE. Predicates
17151 can be called from C as ordinary two-argument functions; this can be
17152 useful in output templates or other machine-specific code.
17154 Operand predicates can allow operands that are not actually acceptable
17155 to the hardware, as long as the constraints give reload the ability to
17156 fix them up (*note Constraints::). However, GCC will usually generate
17157 better code if the predicates specify the requirements of the machine
17158 instructions as closely as possible. Reload cannot fix up operands
17159 that must be constants ("immediate operands"); you must use a predicate
17160 that allows only constants, or else enforce the requirement in the
17163 Most predicates handle their MODE argument in a uniform manner. If
17164 MODE is `VOIDmode' (unspecified), then OP can have any mode. If MODE
17165 is anything else, then OP must have the same mode, unless OP is a
17166 `CONST_INT' or integer `CONST_DOUBLE'. These RTL expressions always
17167 have `VOIDmode', so it would be counterproductive to check that their
17168 mode matches. Instead, predicates that accept `CONST_INT' and/or
17169 integer `CONST_DOUBLE' check that the value stored in the constant will
17170 fit in the requested mode.
17172 Predicates with this behavior are called "normal". `genrecog' can
17173 optimize the instruction recognizer based on knowledge of how normal
17174 predicates treat modes. It can also diagnose certain kinds of common
17175 errors in the use of normal predicates; for instance, it is almost
17176 always an error to use a normal predicate without specifying a mode.
17178 Predicates that do something different with their MODE argument are
17179 called "special". The generic predicates `address_operand' and
17180 `pmode_register_operand' are special predicates. `genrecog' does not
17181 do any optimizations or diagnosis when special predicates are used.
17185 * Machine-Independent Predicates:: Predicates available to all back ends.
17186 * Defining Predicates:: How to write machine-specific predicate
17190 File: gccint.info, Node: Machine-Independent Predicates, Next: Defining Predicates, Up: Predicates
17192 16.7.1 Machine-Independent Predicates
17193 -------------------------------------
17195 These are the generic predicates available to all back ends. They are
17196 defined in `recog.c'. The first category of predicates allow only
17197 constant, or "immediate", operands.
17199 -- Function: immediate_operand
17200 This predicate allows any sort of constant that fits in MODE. It
17201 is an appropriate choice for instructions that take operands that
17204 -- Function: const_int_operand
17205 This predicate allows any `CONST_INT' expression that fits in
17206 MODE. It is an appropriate choice for an immediate operand that
17207 does not allow a symbol or label.
17209 -- Function: const_double_operand
17210 This predicate accepts any `CONST_DOUBLE' expression that has
17211 exactly MODE. If MODE is `VOIDmode', it will also accept
17212 `CONST_INT'. It is intended for immediate floating point
17215 The second category of predicates allow only some kind of machine
17218 -- Function: register_operand
17219 This predicate allows any `REG' or `SUBREG' expression that is
17220 valid for MODE. It is often suitable for arithmetic instruction
17221 operands on a RISC machine.
17223 -- Function: pmode_register_operand
17224 This is a slight variant on `register_operand' which works around
17225 a limitation in the machine-description reader.
17227 (match_operand N "pmode_register_operand" CONSTRAINT)
17231 (match_operand:P N "register_operand" CONSTRAINT)
17233 would mean, if the machine-description reader accepted `:P' mode
17234 suffixes. Unfortunately, it cannot, because `Pmode' is an alias
17235 for some other mode, and might vary with machine-specific options.
17238 -- Function: scratch_operand
17239 This predicate allows hard registers and `SCRATCH' expressions,
17240 but not pseudo-registers. It is used internally by
17241 `match_scratch'; it should not be used directly.
17243 The third category of predicates allow only some kind of memory
17246 -- Function: memory_operand
17247 This predicate allows any valid reference to a quantity of mode
17248 MODE in memory, as determined by the weak form of
17249 `GO_IF_LEGITIMATE_ADDRESS' (*note Addressing Modes::).
17251 -- Function: address_operand
17252 This predicate is a little unusual; it allows any operand that is a
17253 valid expression for the _address_ of a quantity of mode MODE,
17254 again determined by the weak form of `GO_IF_LEGITIMATE_ADDRESS'.
17255 To first order, if `(mem:MODE (EXP))' is acceptable to
17256 `memory_operand', then EXP is acceptable to `address_operand'.
17257 Note that EXP does not necessarily have the mode MODE.
17259 -- Function: indirect_operand
17260 This is a stricter form of `memory_operand' which allows only
17261 memory references with a `general_operand' as the address
17262 expression. New uses of this predicate are discouraged, because
17263 `general_operand' is very permissive, so it's hard to tell what an
17264 `indirect_operand' does or does not allow. If a target has
17265 different requirements for memory operands for different
17266 instructions, it is better to define target-specific predicates
17267 which enforce the hardware's requirements explicitly.
17269 -- Function: push_operand
17270 This predicate allows a memory reference suitable for pushing a
17271 value onto the stack. This will be a `MEM' which refers to
17272 `stack_pointer_rtx', with a side-effect in its address expression
17273 (*note Incdec::); which one is determined by the `STACK_PUSH_CODE'
17274 macro (*note Frame Layout::).
17276 -- Function: pop_operand
17277 This predicate allows a memory reference suitable for popping a
17278 value off the stack. Again, this will be a `MEM' referring to
17279 `stack_pointer_rtx', with a side-effect in its address expression.
17280 However, this time `STACK_POP_CODE' is expected.
17282 The fourth category of predicates allow some combination of the above
17285 -- Function: nonmemory_operand
17286 This predicate allows any immediate or register operand valid for
17289 -- Function: nonimmediate_operand
17290 This predicate allows any register or memory operand valid for
17293 -- Function: general_operand
17294 This predicate allows any immediate, register, or memory operand
17297 Finally, there is one generic operator predicate.
17299 -- Function: comparison_operator
17300 This predicate matches any expression which performs an arithmetic
17301 comparison in MODE; that is, `COMPARISON_P' is true for the
17305 File: gccint.info, Node: Defining Predicates, Prev: Machine-Independent Predicates, Up: Predicates
17307 16.7.2 Defining Machine-Specific Predicates
17308 -------------------------------------------
17310 Many machines have requirements for their operands that cannot be
17311 expressed precisely using the generic predicates. You can define
17312 additional predicates using `define_predicate' and
17313 `define_special_predicate' expressions. These expressions have three
17316 * The name of the predicate, as it will be referred to in
17317 `match_operand' or `match_operator' expressions.
17319 * An RTL expression which evaluates to true if the predicate allows
17320 the operand OP, false if it does not. This expression can only use
17321 the following RTL codes:
17324 When written inside a predicate expression, a `MATCH_OPERAND'
17325 expression evaluates to true if the predicate it names would
17326 allow OP. The operand number and constraint are ignored.
17327 Due to limitations in `genrecog', you can only refer to
17328 generic predicates and predicates that have already been
17332 This expression evaluates to true if OP or a specified
17333 subexpression of OP has one of a given list of RTX codes.
17335 The first operand of this expression is a string constant
17336 containing a comma-separated list of RTX code names (in lower
17337 case). These are the codes for which the `MATCH_CODE' will
17340 The second operand is a string constant which indicates what
17341 subexpression of OP to examine. If it is absent or the empty
17342 string, OP itself is examined. Otherwise, the string constant
17343 must be a sequence of digits and/or lowercase letters. Each
17344 character indicates a subexpression to extract from the
17345 current expression; for the first character this is OP, for
17346 the second and subsequent characters it is the result of the
17347 previous character. A digit N extracts `XEXP (E, N)'; a
17348 letter L extracts `XVECEXP (E, 0, N)' where N is the
17349 alphabetic ordinal of L (0 for `a', 1 for 'b', and so on).
17350 The `MATCH_CODE' then examines the RTX code of the
17351 subexpression extracted by the complete string. It is not
17352 possible to extract components of an `rtvec' that is not at
17353 position 0 within its RTX object.
17356 This expression has one operand, a string constant containing
17357 a C expression. The predicate's arguments, OP and MODE, are
17358 available with those names in the C expression. The
17359 `MATCH_TEST' evaluates to true if the C expression evaluates
17360 to a nonzero value. `MATCH_TEST' expressions must not have
17367 The basic `MATCH_' expressions can be combined using these
17368 logical operators, which have the semantics of the C operators
17369 `&&', `||', `!', and `? :' respectively. As in Common Lisp,
17370 you may give an `AND' or `IOR' expression an arbitrary number
17371 of arguments; this has exactly the same effect as writing a
17372 chain of two-argument `AND' or `IOR' expressions.
17374 * An optional block of C code, which should execute `return true' if
17375 the predicate is found to match and `return false' if it does not.
17376 It must not have any side effects. The predicate arguments, OP
17377 and MODE, are available with those names.
17379 If a code block is present in a predicate definition, then the RTL
17380 expression must evaluate to true _and_ the code block must execute
17381 `return true' for the predicate to allow the operand. The RTL
17382 expression is evaluated first; do not re-check anything in the
17383 code block that was checked in the RTL expression.
17385 The program `genrecog' scans `define_predicate' and
17386 `define_special_predicate' expressions to determine which RTX codes are
17387 possibly allowed. You should always make this explicit in the RTL
17388 predicate expression, using `MATCH_OPERAND' and `MATCH_CODE'.
17390 Here is an example of a simple predicate definition, from the IA64
17391 machine description:
17393 ;; True if OP is a `SYMBOL_REF' which refers to the sdata section.
17394 (define_predicate "small_addr_symbolic_operand"
17395 (and (match_code "symbol_ref")
17396 (match_test "SYMBOL_REF_SMALL_ADDR_P (op)")))
17398 And here is another, showing the use of the C block.
17400 ;; True if OP is a register operand that is (or could be) a GR reg.
17401 (define_predicate "gr_register_operand"
17402 (match_operand 0 "register_operand")
17404 unsigned int regno;
17405 if (GET_CODE (op) == SUBREG)
17406 op = SUBREG_REG (op);
17408 regno = REGNO (op);
17409 return (regno >= FIRST_PSEUDO_REGISTER || GENERAL_REGNO_P (regno));
17412 Predicates written with `define_predicate' automatically include a
17413 test that MODE is `VOIDmode', or OP has the same mode as MODE, or OP is
17414 a `CONST_INT' or `CONST_DOUBLE'. They do _not_ check specifically for
17415 integer `CONST_DOUBLE', nor do they test that the value of either kind
17416 of constant fits in the requested mode. This is because
17417 target-specific predicates that take constants usually have to do more
17418 stringent value checks anyway. If you need the exact same treatment of
17419 `CONST_INT' or `CONST_DOUBLE' that the generic predicates provide, use
17420 a `MATCH_OPERAND' subexpression to call `const_int_operand',
17421 `const_double_operand', or `immediate_operand'.
17423 Predicates written with `define_special_predicate' do not get any
17424 automatic mode checks, and are treated as having special mode handling
17427 The program `genpreds' is responsible for generating code to test
17428 predicates. It also writes a header file containing function
17429 declarations for all machine-specific predicates. It is not necessary
17430 to declare these predicates in `CPU-protos.h'.
17433 File: gccint.info, Node: Constraints, Next: Standard Names, Prev: Predicates, Up: Machine Desc
17435 16.8 Operand Constraints
17436 ========================
17438 Each `match_operand' in an instruction pattern can specify constraints
17439 for the operands allowed. The constraints allow you to fine-tune
17440 matching within the set of operands allowed by the predicate.
17442 Constraints can say whether an operand may be in a register, and which
17443 kinds of register; whether the operand can be a memory reference, and
17444 which kinds of address; whether the operand may be an immediate
17445 constant, and which possible values it may have. Constraints can also
17446 require two operands to match.
17450 * Simple Constraints:: Basic use of constraints.
17451 * Multi-Alternative:: When an insn has two alternative constraint-patterns.
17452 * Class Preferences:: Constraints guide which hard register to put things in.
17453 * Modifiers:: More precise control over effects of constraints.
17454 * Disable Insn Alternatives:: Disable insn alternatives using the `enabled' attribute.
17455 * Machine Constraints:: Existing constraints for some particular machines.
17456 * Define Constraints:: How to define machine-specific constraints.
17457 * C Constraint Interface:: How to test constraints from C code.
17460 File: gccint.info, Node: Simple Constraints, Next: Multi-Alternative, Up: Constraints
17462 16.8.1 Simple Constraints
17463 -------------------------
17465 The simplest kind of constraint is a string full of letters, each of
17466 which describes one kind of operand that is permitted. Here are the
17467 letters that are allowed:
17470 Whitespace characters are ignored and can be inserted at any
17471 position except the first. This enables each alternative for
17472 different operands to be visually aligned in the machine
17473 description even if they have different number of constraints and
17477 A memory operand is allowed, with any kind of address that the
17478 machine supports in general. Note that the letter used for the
17479 general memory constraint can be re-defined by a back end using
17480 the `TARGET_MEM_CONSTRAINT' macro.
17483 A memory operand is allowed, but only if the address is
17484 "offsettable". This means that adding a small integer (actually,
17485 the width in bytes of the operand, as determined by its machine
17486 mode) may be added to the address and the result is also a valid
17489 For example, an address which is constant is offsettable; so is an
17490 address that is the sum of a register and a constant (as long as a
17491 slightly larger constant is also within the range of
17492 address-offsets supported by the machine); but an autoincrement or
17493 autodecrement address is not offsettable. More complicated
17494 indirect/indexed addresses may or may not be offsettable depending
17495 on the other addressing modes that the machine supports.
17497 Note that in an output operand which can be matched by another
17498 operand, the constraint letter `o' is valid only when accompanied
17499 by both `<' (if the target machine has predecrement addressing)
17500 and `>' (if the target machine has preincrement addressing).
17503 A memory operand that is not offsettable. In other words,
17504 anything that would fit the `m' constraint but not the `o'
17508 A memory operand with autodecrement addressing (either
17509 predecrement or postdecrement) is allowed.
17512 A memory operand with autoincrement addressing (either
17513 preincrement or postincrement) is allowed.
17516 A register operand is allowed provided that it is in a general
17520 An immediate integer operand (one with constant value) is allowed.
17521 This includes symbolic constants whose values will be known only at
17522 assembly time or later.
17525 An immediate integer operand with a known numeric value is allowed.
17526 Many systems cannot support assembly-time constants for operands
17527 less than a word wide. Constraints for these operands should use
17528 `n' rather than `i'.
17530 `I', `J', `K', ... `P'
17531 Other letters in the range `I' through `P' may be defined in a
17532 machine-dependent fashion to permit immediate integer operands with
17533 explicit integer values in specified ranges. For example, on the
17534 68000, `I' is defined to stand for the range of values 1 to 8.
17535 This is the range permitted as a shift count in the shift
17539 An immediate floating operand (expression code `const_double') is
17540 allowed, but only if the target floating point format is the same
17541 as that of the host machine (on which the compiler is running).
17544 An immediate floating operand (expression code `const_double' or
17545 `const_vector') is allowed.
17548 `G' and `H' may be defined in a machine-dependent fashion to
17549 permit immediate floating operands in particular ranges of values.
17552 An immediate integer operand whose value is not an explicit
17553 integer is allowed.
17555 This might appear strange; if an insn allows a constant operand
17556 with a value not known at compile time, it certainly must allow
17557 any known value. So why use `s' instead of `i'? Sometimes it
17558 allows better code to be generated.
17560 For example, on the 68000 in a fullword instruction it is possible
17561 to use an immediate operand; but if the immediate value is between
17562 -128 and 127, better code results from loading the value into a
17563 register and using the register. This is because the load into
17564 the register can be done with a `moveq' instruction. We arrange
17565 for this to happen by defining the letter `K' to mean "any integer
17566 outside the range -128 to 127", and then specifying `Ks' in the
17567 operand constraints.
17570 Any register, memory or immediate integer operand is allowed,
17571 except for registers that are not general registers.
17574 Any operand whatsoever is allowed, even if it does not satisfy
17575 `general_operand'. This is normally used in the constraint of a
17576 `match_scratch' when certain alternatives will not actually
17577 require a scratch register.
17579 `0', `1', `2', ... `9'
17580 An operand that matches the specified operand number is allowed.
17581 If a digit is used together with letters within the same
17582 alternative, the digit should come last.
17584 This number is allowed to be more than a single digit. If multiple
17585 digits are encountered consecutively, they are interpreted as a
17586 single decimal integer. There is scant chance for ambiguity,
17587 since to-date it has never been desirable that `10' be interpreted
17588 as matching either operand 1 _or_ operand 0. Should this be
17589 desired, one can use multiple alternatives instead.
17591 This is called a "matching constraint" and what it really means is
17592 that the assembler has only a single operand that fills two roles
17593 considered separate in the RTL insn. For example, an add insn has
17594 two input operands and one output operand in the RTL, but on most
17595 CISC machines an add instruction really has only two operands, one
17596 of them an input-output operand:
17600 Matching constraints are used in these circumstances. More
17601 precisely, the two operands that match must include one input-only
17602 operand and one output-only operand. Moreover, the digit must be a
17603 smaller number than the number of the operand that uses it in the
17606 For operands to match in a particular case usually means that they
17607 are identical-looking RTL expressions. But in a few special cases
17608 specific kinds of dissimilarity are allowed. For example, `*x' as
17609 an input operand will match `*x++' as an output operand. For
17610 proper results in such cases, the output template should always
17611 use the output-operand's number when printing the operand.
17614 An operand that is a valid memory address is allowed. This is for
17615 "load address" and "push address" instructions.
17617 `p' in the constraint must be accompanied by `address_operand' as
17618 the predicate in the `match_operand'. This predicate interprets
17619 the mode specified in the `match_operand' as the mode of the memory
17620 reference for which the address would be valid.
17623 Other letters can be defined in machine-dependent fashion to stand
17624 for particular classes of registers or other arbitrary operand
17625 types. `d', `a' and `f' are defined on the 68000/68020 to stand
17626 for data, address and floating point registers.
17628 In order to have valid assembler code, each operand must satisfy its
17629 constraint. But a failure to do so does not prevent the pattern from
17630 applying to an insn. Instead, it directs the compiler to modify the
17631 code so that the constraint will be satisfied. Usually this is done by
17632 copying an operand into a register.
17634 Contrast, therefore, the two instruction patterns that follow:
17637 [(set (match_operand:SI 0 "general_operand" "=r")
17638 (plus:SI (match_dup 0)
17639 (match_operand:SI 1 "general_operand" "r")))]
17643 which has two operands, one of which must appear in two places, and
17646 [(set (match_operand:SI 0 "general_operand" "=r")
17647 (plus:SI (match_operand:SI 1 "general_operand" "0")
17648 (match_operand:SI 2 "general_operand" "r")))]
17652 which has three operands, two of which are required by a constraint to
17653 be identical. If we are considering an insn of the form
17657 (plus:SI (reg:SI 6) (reg:SI 109)))
17660 the first pattern would not apply at all, because this insn does not
17661 contain two identical subexpressions in the right place. The pattern
17662 would say, "That does not look like an add instruction; try other
17663 patterns". The second pattern would say, "Yes, that's an add
17664 instruction, but there is something wrong with it". It would direct
17665 the reload pass of the compiler to generate additional insns to make
17666 the constraint true. The results might look like this:
17669 (set (reg:SI 3) (reg:SI 6))
17674 (plus:SI (reg:SI 3) (reg:SI 109)))
17677 It is up to you to make sure that each operand, in each pattern, has
17678 constraints that can handle any RTL expression that could be present for
17679 that operand. (When multiple alternatives are in use, each pattern
17680 must, for each possible combination of operand expressions, have at
17681 least one alternative which can handle that combination of operands.)
17682 The constraints don't need to _allow_ any possible operand--when this is
17683 the case, they do not constrain--but they must at least point the way to
17684 reloading any possible operand so that it will fit.
17686 * If the constraint accepts whatever operands the predicate permits,
17687 there is no problem: reloading is never necessary for this operand.
17689 For example, an operand whose constraints permit everything except
17690 registers is safe provided its predicate rejects registers.
17692 An operand whose predicate accepts only constant values is safe
17693 provided its constraints include the letter `i'. If any possible
17694 constant value is accepted, then nothing less than `i' will do; if
17695 the predicate is more selective, then the constraints may also be
17698 * Any operand expression can be reloaded by copying it into a
17699 register. So if an operand's constraints allow some kind of
17700 register, it is certain to be safe. It need not permit all
17701 classes of registers; the compiler knows how to copy a register
17702 into another register of the proper class in order to make an
17705 * A nonoffsettable memory reference can be reloaded by copying the
17706 address into a register. So if the constraint uses the letter
17707 `o', all memory references are taken care of.
17709 * A constant operand can be reloaded by allocating space in memory to
17710 hold it as preinitialized data. Then the memory reference can be
17711 used in place of the constant. So if the constraint uses the
17712 letters `o' or `m', constant operands are not a problem.
17714 * If the constraint permits a constant and a pseudo register used in
17715 an insn was not allocated to a hard register and is equivalent to
17716 a constant, the register will be replaced with the constant. If
17717 the predicate does not permit a constant and the insn is
17718 re-recognized for some reason, the compiler will crash. Thus the
17719 predicate must always recognize any objects allowed by the
17722 If the operand's predicate can recognize registers, but the constraint
17723 does not permit them, it can make the compiler crash. When this
17724 operand happens to be a register, the reload pass will be stymied,
17725 because it does not know how to copy a register temporarily into memory.
17727 If the predicate accepts a unary operator, the constraint applies to
17728 the operand. For example, the MIPS processor at ISA level 3 supports an
17729 instruction which adds two registers in `SImode' to produce a `DImode'
17730 result, but only if the registers are correctly sign extended. This
17731 predicate for the input operands accepts a `sign_extend' of an `SImode'
17732 register. Write the constraint to indicate the type of register that
17733 is required for the operand of the `sign_extend'.
17736 File: gccint.info, Node: Multi-Alternative, Next: Class Preferences, Prev: Simple Constraints, Up: Constraints
17738 16.8.2 Multiple Alternative Constraints
17739 ---------------------------------------
17741 Sometimes a single instruction has multiple alternative sets of possible
17742 operands. For example, on the 68000, a logical-or instruction can
17743 combine register or an immediate value into memory, or it can combine
17744 any kind of operand into a register; but it cannot combine one memory
17745 location into another.
17747 These constraints are represented as multiple alternatives. An
17748 alternative can be described by a series of letters for each operand.
17749 The overall constraint for an operand is made from the letters for this
17750 operand from the first alternative, a comma, the letters for this
17751 operand from the second alternative, a comma, and so on until the last
17752 alternative. Here is how it is done for fullword logical-or on the
17755 (define_insn "iorsi3"
17756 [(set (match_operand:SI 0 "general_operand" "=m,d")
17757 (ior:SI (match_operand:SI 1 "general_operand" "%0,0")
17758 (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
17761 The first alternative has `m' (memory) for operand 0, `0' for operand
17762 1 (meaning it must match operand 0), and `dKs' for operand 2. The
17763 second alternative has `d' (data register) for operand 0, `0' for
17764 operand 1, and `dmKs' for operand 2. The `=' and `%' in the
17765 constraints apply to all the alternatives; their meaning is explained
17766 in the next section (*note Class Preferences::).
17768 If all the operands fit any one alternative, the instruction is valid.
17769 Otherwise, for each alternative, the compiler counts how many
17770 instructions must be added to copy the operands so that that
17771 alternative applies. The alternative requiring the least copying is
17772 chosen. If two alternatives need the same amount of copying, the one
17773 that comes first is chosen. These choices can be altered with the `?'
17774 and `!' characters:
17777 Disparage slightly the alternative that the `?' appears in, as a
17778 choice when no alternative applies exactly. The compiler regards
17779 this alternative as one unit more costly for each `?' that appears
17783 Disparage severely the alternative that the `!' appears in. This
17784 alternative can still be used if it fits without reloading, but if
17785 reloading is needed, some other alternative will be used.
17787 When an insn pattern has multiple alternatives in its constraints,
17788 often the appearance of the assembler code is determined mostly by which
17789 alternative was matched. When this is so, the C code for writing the
17790 assembler code can use the variable `which_alternative', which is the
17791 ordinal number of the alternative that was actually satisfied (0 for
17792 the first, 1 for the second alternative, etc.). *Note Output
17796 File: gccint.info, Node: Class Preferences, Next: Modifiers, Prev: Multi-Alternative, Up: Constraints
17798 16.8.3 Register Class Preferences
17799 ---------------------------------
17801 The operand constraints have another function: they enable the compiler
17802 to decide which kind of hardware register a pseudo register is best
17803 allocated to. The compiler examines the constraints that apply to the
17804 insns that use the pseudo register, looking for the machine-dependent
17805 letters such as `d' and `a' that specify classes of registers. The
17806 pseudo register is put in whichever class gets the most "votes". The
17807 constraint letters `g' and `r' also vote: they vote in favor of a
17808 general register. The machine description says which registers are
17809 considered general.
17811 Of course, on some machines all registers are equivalent, and no
17812 register classes are defined. Then none of this complexity is relevant.
17815 File: gccint.info, Node: Modifiers, Next: Disable Insn Alternatives, Prev: Class Preferences, Up: Constraints
17817 16.8.4 Constraint Modifier Characters
17818 -------------------------------------
17820 Here are constraint modifier characters.
17823 Means that this operand is write-only for this instruction: the
17824 previous value is discarded and replaced by output data.
17827 Means that this operand is both read and written by the
17830 When the compiler fixes up the operands to satisfy the constraints,
17831 it needs to know which operands are inputs to the instruction and
17832 which are outputs from it. `=' identifies an output; `+'
17833 identifies an operand that is both input and output; all other
17834 operands are assumed to be input only.
17836 If you specify `=' or `+' in a constraint, you put it in the first
17837 character of the constraint string.
17840 Means (in a particular alternative) that this operand is an
17841 "earlyclobber" operand, which is modified before the instruction is
17842 finished using the input operands. Therefore, this operand may
17843 not lie in a register that is used as an input operand or as part
17844 of any memory address.
17846 `&' applies only to the alternative in which it is written. In
17847 constraints with multiple alternatives, sometimes one alternative
17848 requires `&' while others do not. See, for example, the `movdf'
17851 An input operand can be tied to an earlyclobber operand if its only
17852 use as an input occurs before the early result is written. Adding
17853 alternatives of this form often allows GCC to produce better code
17854 when only some of the inputs can be affected by the earlyclobber.
17855 See, for example, the `mulsi3' insn of the ARM.
17857 `&' does not obviate the need to write `='.
17860 Declares the instruction to be commutative for this operand and the
17861 following operand. This means that the compiler may interchange
17862 the two operands if that is the cheapest way to make all operands
17863 fit the constraints. This is often used in patterns for addition
17864 instructions that really have only two operands: the result must
17865 go in one of the arguments. Here for example, is how the 68000
17866 halfword-add instruction is defined:
17868 (define_insn "addhi3"
17869 [(set (match_operand:HI 0 "general_operand" "=m,r")
17870 (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
17871 (match_operand:HI 2 "general_operand" "di,g")))]
17873 GCC can only handle one commutative pair in an asm; if you use
17874 more, the compiler may fail. Note that you need not use the
17875 modifier if the two alternatives are strictly identical; this
17876 would only waste time in the reload pass. The modifier is not
17877 operational after register allocation, so the result of
17878 `define_peephole2' and `define_split's performed after reload
17879 cannot rely on `%' to make the intended insn match.
17882 Says that all following characters, up to the next comma, are to be
17883 ignored as a constraint. They are significant only for choosing
17884 register preferences.
17887 Says that the following character should be ignored when choosing
17888 register preferences. `*' has no effect on the meaning of the
17889 constraint as a constraint, and no effect on reloading.
17891 Here is an example: the 68000 has an instruction to sign-extend a
17892 halfword in a data register, and can also sign-extend a value by
17893 copying it into an address register. While either kind of
17894 register is acceptable, the constraints on an address-register
17895 destination are less strict, so it is best if register allocation
17896 makes an address register its goal. Therefore, `*' is used so
17897 that the `d' constraint letter (for data register) is ignored when
17898 computing register preferences.
17900 (define_insn "extendhisi2"
17901 [(set (match_operand:SI 0 "general_operand" "=*d,a")
17903 (match_operand:HI 1 "general_operand" "0,g")))]
17907 File: gccint.info, Node: Machine Constraints, Next: Define Constraints, Prev: Disable Insn Alternatives, Up: Constraints
17909 16.8.5 Constraints for Particular Machines
17910 ------------------------------------------
17912 Whenever possible, you should use the general-purpose constraint letters
17913 in `asm' arguments, since they will convey meaning more readily to
17914 people reading your code. Failing that, use the constraint letters
17915 that usually have very similar meanings across architectures. The most
17916 commonly used constraints are `m' and `r' (for memory and
17917 general-purpose registers respectively; *note Simple Constraints::), and
17918 `I', usually the letter indicating the most common immediate-constant
17921 Each architecture defines additional constraints. These constraints
17922 are used by the compiler itself for instruction generation, as well as
17923 for `asm' statements; therefore, some of the constraints are not
17924 particularly useful for `asm'. Here is a summary of some of the
17925 machine-dependent constraints available on some particular machines; it
17926 includes both constraints that are useful for `asm' and constraints
17927 that aren't. The compiler source file mentioned in the table heading
17928 for each architecture is the definitive reference for the meanings of
17929 that architecture's constraints.
17931 _ARM family--`config/arm/arm.h'_
17934 Floating-point register
17937 VFP floating-point register
17940 One of the floating-point constants 0.0, 0.5, 1.0, 2.0, 3.0,
17944 Floating-point constant that would satisfy the constraint `F'
17948 Integer that is valid as an immediate operand in a data
17949 processing instruction. That is, an integer in the range 0
17950 to 255 rotated by a multiple of 2
17953 Integer in the range -4095 to 4095
17956 Integer that satisfies constraint `I' when inverted (ones
17960 Integer that satisfies constraint `I' when negated (twos
17964 Integer in the range 0 to 32
17967 A memory reference where the exact address is in a single
17968 register (``m'' is preferable for `asm' statements)
17971 An item in the constant pool
17974 A symbol in the text segment of the current file
17977 A memory reference suitable for VFP load/store insns
17978 (reg+constant offset)
17981 A memory reference suitable for iWMMXt load/store
17985 A memory reference suitable for the ARMv4 ldrsb instruction.
17987 _AVR family--`config/avr/constraints.md'_
17990 Registers from r0 to r15
17993 Registers from r16 to r23
17996 Registers from r16 to r31
17999 Registers from r24 to r31. These registers can be used in
18003 Pointer register (r26-r31)
18006 Base pointer register (r28-r31)
18009 Stack pointer register (SPH:SPL)
18012 Temporary register r0
18015 Register pair X (r27:r26)
18018 Register pair Y (r29:r28)
18021 Register pair Z (r31:r30)
18024 Constant greater than -1, less than 64
18027 Constant greater than -64, less than 1
18036 Constant that fits in 8 bits
18039 Constant integer -1
18042 Constant integer 8, 16, or 24
18048 A floating point constant 0.0
18051 Integer constant in the range -6 ... 5.
18054 A memory address based on Y or Z pointer with displacement.
18056 _CRX Architecture--`config/crx/crx.h'_
18059 Registers from r0 to r14 (registers without stack pointer)
18062 Register r16 (64-bit accumulator lo register)
18065 Register r17 (64-bit accumulator hi register)
18068 Register pair r16-r17. (64-bit accumulator lo-hi pair)
18071 Constant that fits in 3 bits
18074 Constant that fits in 4 bits
18077 Constant that fits in 5 bits
18080 Constant that is one of -1, 4, -4, 7, 8, 12, 16, 20, 32, 48
18083 Floating point constant that is legal for store immediate
18085 _Hewlett-Packard PA-RISC--`config/pa/pa.h'_
18091 Floating point register
18094 Shift amount register
18097 Floating point register (deprecated)
18100 Upper floating point register (32-bit), floating point
18107 Signed 11-bit integer constant
18110 Signed 14-bit integer constant
18113 Integer constant that can be deposited with a `zdepi'
18117 Signed 5-bit integer constant
18123 Integer constant that can be loaded with a `ldil' instruction
18126 Integer constant whose value plus one is a power of 2
18129 Integer constant that can be used for `and' operations in
18130 `depi' and `extru' instructions
18133 Integer constant 31
18136 Integer constant 63
18139 Floating-point constant 0.0
18142 A `lo_sum' data-linkage-table memory operand
18145 A memory operand that can be used as the destination operand
18146 of an integer store instruction
18149 A scaled or unscaled indexed memory operand
18152 A memory operand for floating-point loads and stores
18155 A register indirect memory operand
18157 _picoChip family--`picochip.h'_
18163 Pointer register. A register which can be used to access
18164 memory without supplying an offset. Any other register can
18165 be used to access memory, but will need a constant offset.
18166 In the case of the offset being zero, it is more efficient to
18167 use a pointer register, since this reduces code size.
18170 A twin register. A register which may be paired with an
18171 adjacent register to create a 32-bit register.
18174 Any absolute memory address (e.g., symbolic constant, symbolic
18175 constant + offset).
18178 4-bit signed integer.
18181 4-bit unsigned integer.
18184 8-bit signed integer.
18187 Any constant whose absolute value is no greater than 4-bits.
18190 10-bit signed integer
18193 16-bit signed integer.
18196 _PowerPC and IBM RS6000--`config/rs6000/rs6000.h'_
18199 Address base register
18202 Floating point register
18208 `MQ', `CTR', or `LINK' register
18220 `CR' register (condition register) number 0
18223 `CR' register (condition register)
18226 `FPMEM' stack memory for FPR-GPR transfers
18229 Signed 16-bit constant
18232 Unsigned 16-bit constant shifted left 16 bits (use `L'
18233 instead for `SImode' constants)
18236 Unsigned 16-bit constant
18239 Signed 16-bit constant shifted left 16 bits
18242 Constant larger than 31
18251 Constant whose negation is a signed 16-bit constant
18254 Floating point constant that can be loaded into a register
18255 with one instruction per word
18258 Integer/Floating point constant that can be loaded into a
18259 register using three instructions
18262 Memory operand that is an offset from a register (`m' is
18263 preferable for `asm' statements)
18266 Memory operand that is an indexed or indirect from a register
18267 (`m' is preferable for `asm' statements)
18273 Address operand that is an indexed or indirect from a
18274 register (`p' is preferable for `asm' statements)
18277 Constant suitable as a 64-bit mask operand
18280 Constant suitable as a 32-bit mask operand
18283 System V Release 4 small data area reference
18286 AND masks that can be performed by two rldic{l, r}
18290 Vector constant that does not require memory
18293 _Intel 386--`config/i386/constraints.md'_
18296 Legacy register--the eight integer registers available on all
18297 i386 processors (`a', `b', `c', `d', `si', `di', `bp', `sp').
18300 Any register accessible as `Rl'. In 32-bit mode, `a', `b',
18301 `c', and `d'; in 64-bit mode, any integer register.
18304 Any register accessible as `Rh': `a', `b', `c', and `d'.
18307 Any register that can be used as the index in a base+index
18308 memory access: that is, any general register except the stack
18330 The `a' and `d' registers, as a pair (for instructions that
18331 return half the result in one and half in the other).
18334 Any 80387 floating-point (stack) register.
18337 Top of 80387 floating-point stack (`%st(0)').
18340 Second from top of 80387 floating-point stack (`%st(1)').
18349 First SSE register (`%xmm0').
18352 Any SSE register, when SSE2 is enabled.
18355 Any SSE register, when SSE2 and inter-unit moves are enabled.
18358 Any MMX register, when inter-unit moves are enabled.
18361 Integer constant in the range 0 ... 31, for 32-bit shifts.
18364 Integer constant in the range 0 ... 63, for 64-bit shifts.
18367 Signed 8-bit integer constant.
18370 `0xFF' or `0xFFFF', for andsi as a zero-extending move.
18373 0, 1, 2, or 3 (shifts for the `lea' instruction).
18376 Unsigned 8-bit integer constant (for `in' and `out'
18380 Integer constant in the range 0 ... 127, for 128-bit shifts.
18383 Standard 80387 floating point constant.
18386 Standard SSE floating point constant.
18389 32-bit signed integer constant, or a symbolic reference known
18390 to fit that range (for immediate operands in sign-extending
18391 x86-64 instructions).
18394 32-bit unsigned integer constant, or a symbolic reference
18395 known to fit that range (for immediate operands in
18396 zero-extending x86-64 instructions).
18399 _Intel IA-64--`config/ia64/ia64.h'_
18402 General register `r0' to `r3' for `addl' instruction
18408 Predicate register (`c' as in "conditional")
18411 Application register residing in M-unit
18414 Application register residing in I-unit
18417 Floating-point register
18420 Memory operand. Remember that `m' allows postincrement and
18421 postdecrement which require printing with `%Pn' on IA-64.
18422 Use `S' to disallow postincrement and postdecrement.
18425 Floating-point constant 0.0 or 1.0
18428 14-bit signed integer constant
18431 22-bit signed integer constant
18434 8-bit signed integer constant for logical instructions
18437 8-bit adjusted signed integer constant for compare pseudo-ops
18440 6-bit unsigned integer constant for shift counts
18443 9-bit signed integer constant for load and store
18450 0 or -1 for `dep' instruction
18453 Non-volatile memory for floating-point loads and stores
18456 Integer constant in the range 1 to 4 for `shladd' instruction
18459 Memory operand except postincrement and postdecrement
18461 _FRV--`config/frv/frv.h'_
18464 Register in the class `ACC_REGS' (`acc0' to `acc7').
18467 Register in the class `EVEN_ACC_REGS' (`acc0' to `acc7').
18470 Register in the class `CC_REGS' (`fcc0' to `fcc3' and `icc0'
18474 Register in the class `GPR_REGS' (`gr0' to `gr63').
18477 Register in the class `EVEN_REGS' (`gr0' to `gr63'). Odd
18478 registers are excluded not in the class but through the use
18479 of a machine mode larger than 4 bytes.
18482 Register in the class `FPR_REGS' (`fr0' to `fr63').
18485 Register in the class `FEVEN_REGS' (`fr0' to `fr63'). Odd
18486 registers are excluded not in the class but through the use
18487 of a machine mode larger than 4 bytes.
18490 Register in the class `LR_REG' (the `lr' register).
18493 Register in the class `QUAD_REGS' (`gr2' to `gr63').
18494 Register numbers not divisible by 4 are excluded not in the
18495 class but through the use of a machine mode larger than 8
18499 Register in the class `ICC_REGS' (`icc0' to `icc3').
18502 Register in the class `FCC_REGS' (`fcc0' to `fcc3').
18505 Register in the class `ICR_REGS' (`cc4' to `cc7').
18508 Register in the class `FCR_REGS' (`cc0' to `cc3').
18511 Register in the class `QUAD_FPR_REGS' (`fr0' to `fr63').
18512 Register numbers not divisible by 4 are excluded not in the
18513 class but through the use of a machine mode larger than 8
18517 Register in the class `SPR_REGS' (`lcr' and `lr').
18520 Register in the class `QUAD_ACC_REGS' (`acc0' to `acc7').
18523 Register in the class `ACCG_REGS' (`accg0' to `accg7').
18526 Register in the class `CR_REGS' (`cc0' to `cc7').
18529 Floating point constant zero
18532 6-bit signed integer constant
18535 10-bit signed integer constant
18538 16-bit signed integer constant
18541 16-bit unsigned integer constant
18544 12-bit signed integer constant that is negative--i.e. in the
18545 range of -2048 to -1
18551 12-bit signed integer constant that is greater than
18552 zero--i.e. in the range of 1 to 2047.
18555 _Blackfin family--`config/bfin/constraints.md'_
18564 A call clobbered P register.
18567 A single register. If N is in the range 0 to 7, the
18568 corresponding D register. If it is `A', then the register P0.
18571 Even-numbered D register
18574 Odd-numbered D register
18577 Accumulator register.
18580 Even-numbered accumulator register.
18583 Odd-numbered accumulator register.
18595 Registers used for circular buffering, i.e. I, B, or L
18611 Any D, P, B, M, I or L register.
18614 Additional registers typically used only in prologues and
18615 epilogues: RETS, RETN, RETI, RETX, RETE, ASTAT, SEQSTAT and
18619 Any register except accumulators or CC.
18622 Signed 16 bit integer (in the range -32768 to 32767)
18625 Unsigned 16 bit integer (in the range 0 to 65535)
18628 Signed 7 bit integer (in the range -64 to 63)
18631 Unsigned 7 bit integer (in the range 0 to 127)
18634 Unsigned 5 bit integer (in the range 0 to 31)
18637 Signed 4 bit integer (in the range -8 to 7)
18640 Signed 3 bit integer (in the range -3 to 4)
18643 Unsigned 3 bit integer (in the range 0 to 7)
18646 Constant N, where N is a single-digit constant in the range 0
18650 An integer equal to one of the MACFLAG_XXX constants that is
18651 suitable for use with either accumulator.
18654 An integer equal to one of the MACFLAG_XXX constants that is
18655 suitable for use only with accumulator A1.
18664 An integer constant with exactly a single bit set.
18667 An integer constant with all bits set except exactly one.
18674 _M32C--`config/m32c/m32c.c'_
18679 `$sp', `$fb', `$sb'.
18682 Any control register, when they're 16 bits wide (nothing if
18683 control registers are 24 bits wide)
18686 Any control register, when they're 24 bits wide.
18692 $r0, $r1, $r2, $r3.
18695 $r0 or $r2, or $r2r0 for 32 bit values.
18698 $r1 or $r3, or $r3r1 for 32 bit values.
18701 A register that can hold a 64 bit value.
18704 $r0 or $r1 (registers with addressable high/low bytes)
18713 Address registers when they're 16 bits wide.
18716 Address registers when they're 24 bits wide.
18719 Registers that can hold QI values.
18722 Registers that can be used with displacements ($a0, $a1, $sb).
18725 Registers that can hold 32 bit values.
18728 Registers that can hold 16 bit values.
18731 Registers chat can hold 16 bit values, including all control
18735 $r0 through R1, plus $a0 and $a1.
18738 The flags register.
18741 The memory-based pseudo-registers $mem0 through $mem15.
18744 Registers that can hold pointers (16 bit registers for r8c,
18745 m16c; 24 bit registers for m32cm, m32c).
18748 Matches multiple registers in a PARALLEL to form a larger
18749 register. Used to match function return values.
18764 -8 ... -1 or 1 ... 8
18767 -16 ... -1 or 1 ... 16
18770 -32 ... -1 or 1 ... 32
18776 An 8 bit value with exactly one bit set.
18779 A 16 bit value with exactly one bit set.
18782 The common src/dest memory addressing modes.
18785 Memory addressed using $a0 or $a1.
18788 Memory addressed with immediate addresses.
18791 Memory addressed using the stack pointer ($sp).
18794 Memory addressed using the frame base register ($fb).
18797 Memory addressed using the small base register ($sb).
18802 _MIPS--`config/mips/constraints.md'_
18805 An address register. This is equivalent to `r' unless
18806 generating MIPS16 code.
18809 A floating-point register (if available).
18812 Formerly the `hi' register. This constraint is no longer
18816 The `lo' register. Use this register to store values that are
18817 no bigger than a word.
18820 The concatenated `hi' and `lo' registers. Use this register
18821 to store doubleword values.
18824 A register suitable for use in an indirect jump. This will
18825 always be `$25' for `-mabicalls'.
18828 Register `$3'. Do not use this constraint in new code; it is
18829 retained only for compatibility with glibc.
18832 Equivalent to `r'; retained for backwards compatibility.
18835 A floating-point condition code register.
18838 A signed 16-bit constant (for arithmetic instructions).
18844 An unsigned 16-bit constant (for logic instructions).
18847 A signed 32-bit constant in which the lower 16 bits are zero.
18848 Such constants can be loaded using `lui'.
18851 A constant that cannot be loaded using `lui', `addiu' or
18855 A constant in the range -65535 to -1 (inclusive).
18858 A signed 15-bit constant.
18861 A constant in the range 1 to 65535 (inclusive).
18864 Floating-point zero.
18867 An address that can be used in a non-macro load or store.
18869 _Motorola 680x0--`config/m68k/constraints.md'_
18878 68881 floating-point register, if available
18881 Integer in the range 1 to 8
18884 16-bit signed number
18887 Signed number whose magnitude is greater than 0x80
18890 Integer in the range -8 to -1
18893 Signed number whose magnitude is greater than 0x100
18896 Range 24 to 31, rotatert:SI 8 to 1 expressed as rotate
18899 16 (for rotate using swap)
18902 Range 8 to 15, rotatert:HI 8 to 1 expressed as rotate
18905 Numbers that mov3q can handle
18908 Floating point constant that is not a 68881 constant
18911 Operands that satisfy 'm' when -mpcrel is in effect
18914 Operands that satisfy 's' when -mpcrel is not in effect
18917 Address register indirect addressing mode
18920 Register offset addressing
18926 symbol_ref or const
18935 Range of signed numbers that don't fit in 16 bits
18938 Integers valid for mvq
18941 Integers valid for a moveq followed by a swap
18944 Integers valid for mvz
18947 Integers valid for mvs
18953 Non-register operands allowed in clr
18956 _Motorola 68HC11 & 68HC12 families--`config/m68hc11/m68hc11.h'_
18971 Temporary soft register _.tmp
18974 A soft register _.d1 to _.d31
18977 Stack pointer register
18986 Pseudo register `z' (replaced by `x' or `y' at the end)
18989 An address register: x, y or z
18992 An address register: x or y
18995 Register pair (x:d) to form a 32-bit value
18998 Constants in the range -65536 to 65535
19001 Constants whose 16-bit low part is zero
19004 Constant integer 1 or -1
19007 Constant integer 16
19010 Constants in the range -8 to 2
19013 _SPARC--`config/sparc/sparc.h'_
19016 Floating-point register on the SPARC-V8 architecture and
19017 lower floating-point register on the SPARC-V9 architecture.
19020 Floating-point register. It is equivalent to `f' on the
19021 SPARC-V8 architecture and contains both lower and upper
19022 floating-point registers on the SPARC-V9 architecture.
19025 Floating-point condition code register.
19028 Lower floating-point register. It is only valid on the
19029 SPARC-V9 architecture when the Visual Instruction Set is
19033 Floating-point register. It is only valid on the SPARC-V9
19034 architecture when the Visual Instruction Set is available.
19037 64-bit global or out register for the SPARC-V8+ architecture.
19043 Signed 13-bit constant
19049 32-bit constant with the low 12 bits clear (a constant that
19050 can be loaded with the `sethi' instruction)
19053 A constant in the range supported by `movcc' instructions
19056 A constant in the range supported by `movrcc' instructions
19059 Same as `K', except that it verifies that bits that are not
19060 in the lower 32-bit range are all zero. Must be used instead
19061 of `K' for modes wider than `SImode'
19067 Floating-point zero
19070 Signed 13-bit constant, sign-extended to 32 or 64 bits
19073 Floating-point constant whose integral representation can be
19074 moved into an integer register using a single sethi
19078 Floating-point constant whose integral representation can be
19079 moved into an integer register using a single mov instruction
19082 Floating-point constant whose integral representation can be
19083 moved into an integer register using a high/lo_sum
19084 instruction sequence
19087 Memory address aligned to an 8-byte boundary
19093 Memory address for `e' constraint registers
19099 _SPU--`config/spu/spu.h'_
19102 An immediate which can be loaded with the il/ila/ilh/ilhu
19103 instructions. const_int is treated as a 64 bit value.
19106 An immediate for and/xor/or instructions. const_int is
19107 treated as a 64 bit value.
19110 An immediate for the `iohl' instruction. const_int is
19111 treated as a 64 bit value.
19114 An immediate which can be loaded with `fsmbi'.
19117 An immediate which can be loaded with the il/ila/ilh/ilhu
19118 instructions. const_int is treated as a 32 bit value.
19121 An immediate for most arithmetic instructions. const_int is
19122 treated as a 32 bit value.
19125 An immediate for and/xor/or instructions. const_int is
19126 treated as a 32 bit value.
19129 An immediate for the `iohl' instruction. const_int is
19130 treated as a 32 bit value.
19133 A constant in the range [-64, 63] for shift/rotate
19137 An unsigned 7-bit constant for conversion/nop/channel
19141 A signed 10-bit constant for most arithmetic instructions.
19144 A signed 16 bit immediate for `stop'.
19147 An unsigned 16-bit constant for `iohl' and `fsmbi'.
19150 An unsigned 7-bit constant whose 3 least significant bits are
19154 An unsigned 3-bit constant for 16-byte rotates and shifts
19157 Call operand, reg, for indirect calls
19160 Call operand, symbol, for relative calls.
19163 Call operand, const_int, for absolute calls.
19166 An immediate which can be loaded with the il/ila/ilh/ilhu
19167 instructions. const_int is sign extended to 128 bit.
19170 An immediate for shift and rotate instructions. const_int is
19171 treated as a 32 bit value.
19174 An immediate for and/xor/or instructions. const_int is sign
19175 extended as a 128 bit.
19178 An immediate for the `iohl' instruction. const_int is sign
19179 extended to 128 bit.
19182 _S/390 and zSeries--`config/s390/s390.h'_
19185 Address register (general purpose register except r0)
19188 Condition code register
19191 Data register (arbitrary general purpose register)
19194 Floating-point register
19197 Unsigned 8-bit constant (0-255)
19200 Unsigned 12-bit constant (0-4095)
19203 Signed 16-bit constant (-32768-32767)
19206 Value appropriate as displacement.
19208 for short displacement
19210 `(-524288..524287)'
19211 for long displacement
19214 Constant integer with a value of 0x7fffffff.
19217 Multiple letter constraint followed by 4 parameter letters.
19219 number of the part counting from most to least
19226 mode of the containing operand
19229 value of the other parts (F--all bits set)
19230 The constraint matches if the specified part of a constant
19231 has a value different from its other parts.
19234 Memory reference without index register and with short
19238 Memory reference with index register and short displacement.
19241 Memory reference without index register but with long
19245 Memory reference with index register and long displacement.
19248 Pointer with short displacement.
19251 Pointer with long displacement.
19254 Shift count operand.
19257 _Score family--`config/score/score.h'_
19260 Registers from r0 to r32.
19263 Registers from r0 to r16.
19266 r8--r11 or r22--r27 registers.
19287 cnt + lcb + scb register.
19290 cr0--cr15 register.
19302 cp1 + cp2 + cp3 registers.
19305 High 16-bit constant (32-bit constant with 16 LSBs zero).
19308 Unsigned 5 bit integer (in the range 0 to 31).
19311 Unsigned 16 bit integer (in the range 0 to 65535).
19314 Signed 16 bit integer (in the range -32768 to 32767).
19317 Unsigned 14 bit integer (in the range 0 to 16383).
19320 Signed 14 bit integer (in the range -8192 to 8191).
19325 _Xstormy16--`config/stormy16/stormy16.h'_
19340 Registers r0 through r7.
19343 Registers r0 and r1.
19346 The carry register.
19349 Registers r8 and r9.
19352 A constant between 0 and 3 inclusive.
19355 A constant that has exactly one bit set.
19358 A constant that has exactly one bit clear.
19361 A constant between 0 and 255 inclusive.
19364 A constant between -255 and 0 inclusive.
19367 A constant between -3 and 0 inclusive.
19370 A constant between 1 and 4 inclusive.
19373 A constant between -4 and -1 inclusive.
19376 A memory reference that is a stack push.
19379 A memory reference that is a stack pop.
19382 A memory reference that refers to a constant address of known
19386 The register indicated by Rx (not implemented yet).
19389 A constant that is not between 2 and 15 inclusive.
19395 _Xtensa--`config/xtensa/constraints.md'_
19398 General-purpose 32-bit register
19401 One-bit boolean register
19404 MAC16 40-bit accumulator register
19407 Signed 12-bit integer constant, for use in MOVI instructions
19410 Signed 8-bit integer constant, for use in ADDI instructions
19413 Integer constant valid for BccI instructions
19416 Unsigned constant valid for BccUI instructions
19421 File: gccint.info, Node: Disable Insn Alternatives, Next: Machine Constraints, Prev: Modifiers, Up: Constraints
19423 16.8.6 Disable insn alternatives using the `enabled' attribute
19424 --------------------------------------------------------------
19426 The `enabled' insn attribute may be used to disable certain insn
19427 alternatives for machine-specific reasons. This is useful when adding
19428 new instructions to an existing pattern which are only available for
19429 certain cpu architecture levels as specified with the `-march=' option.
19431 If an insn alternative is disabled, then it will never be used. The
19432 compiler treats the constraints for the disabled alternative as
19435 In order to make use of the `enabled' attribute a back end has to add
19436 in the machine description files:
19438 1. A definition of the `enabled' insn attribute. The attribute is
19439 defined as usual using the `define_attr' command. This definition
19440 should be based on other insn attributes and/or target flags. The
19441 `enabled' attribute is a numeric attribute and should evaluate to
19442 `(const_int 1)' for an enabled alternative and to `(const_int 0)'
19445 2. A definition of another insn attribute used to describe for what
19446 reason an insn alternative might be available or not. E.g.
19447 `cpu_facility' as in the example below.
19449 3. An assignment for the second attribute to each insn definition
19450 combining instructions which are not all available under the same
19451 circumstances. (Note: It obviously only makes sense for
19452 definitions with more than one alternative. Otherwise the insn
19453 pattern should be disabled or enabled using the insn condition.)
19455 E.g. the following two patterns could easily be merged using the
19456 `enabled' attribute:
19459 (define_insn "*movdi_old"
19460 [(set (match_operand:DI 0 "register_operand" "=d")
19461 (match_operand:DI 1 "register_operand" " d"))]
19465 (define_insn "*movdi_new"
19466 [(set (match_operand:DI 0 "register_operand" "=d,f,d")
19467 (match_operand:DI 1 "register_operand" " d,d,f"))]
19477 (define_insn "*movdi_combined"
19478 [(set (match_operand:DI 0 "register_operand" "=d,f,d")
19479 (match_operand:DI 1 "register_operand" " d,d,f"))]
19485 [(set_attr "cpu_facility" "*,new,new")])
19487 with the `enabled' attribute defined like this:
19490 (define_attr "cpu_facility" "standard,new" (const_string "standard"))
19492 (define_attr "enabled" ""
19493 (cond [(eq_attr "cpu_facility" "standard") (const_int 1)
19494 (and (eq_attr "cpu_facility" "new")
19495 (ne (symbol_ref "TARGET_NEW") (const_int 0)))
19500 File: gccint.info, Node: Define Constraints, Next: C Constraint Interface, Prev: Machine Constraints, Up: Constraints
19502 16.8.7 Defining Machine-Specific Constraints
19503 --------------------------------------------
19505 Machine-specific constraints fall into two categories: register and
19506 non-register constraints. Within the latter category, constraints
19507 which allow subsets of all possible memory or address operands should
19508 be specially marked, to give `reload' more information.
19510 Machine-specific constraints can be given names of arbitrary length,
19511 but they must be entirely composed of letters, digits, underscores
19512 (`_'), and angle brackets (`< >'). Like C identifiers, they must begin
19513 with a letter or underscore.
19515 In order to avoid ambiguity in operand constraint strings, no
19516 constraint can have a name that begins with any other constraint's
19517 name. For example, if `x' is defined as a constraint name, `xy' may
19518 not be, and vice versa. As a consequence of this rule, no constraint
19519 may begin with one of the generic constraint letters: `E F V X g i m n
19522 Register constraints correspond directly to register classes. *Note
19523 Register Classes::. There is thus not much flexibility in their
19526 -- MD Expression: define_register_constraint name regclass docstring
19527 All three arguments are string constants. NAME is the name of the
19528 constraint, as it will appear in `match_operand' expressions. If
19529 NAME is a multi-letter constraint its length shall be the same for
19530 all constraints starting with the same letter. REGCLASS can be
19531 either the name of the corresponding register class (*note
19532 Register Classes::), or a C expression which evaluates to the
19533 appropriate register class. If it is an expression, it must have
19534 no side effects, and it cannot look at the operand. The usual use
19535 of expressions is to map some register constraints to `NO_REGS'
19536 when the register class is not available on a given
19539 DOCSTRING is a sentence documenting the meaning of the constraint.
19540 Docstrings are explained further below.
19542 Non-register constraints are more like predicates: the constraint
19543 definition gives a Boolean expression which indicates whether the
19544 constraint matches.
19546 -- MD Expression: define_constraint name docstring exp
19547 The NAME and DOCSTRING arguments are the same as for
19548 `define_register_constraint', but note that the docstring comes
19549 immediately after the name for these expressions. EXP is an RTL
19550 expression, obeying the same rules as the RTL expressions in
19551 predicate definitions. *Note Defining Predicates::, for details.
19552 If it evaluates true, the constraint matches; if it evaluates
19553 false, it doesn't. Constraint expressions should indicate which
19554 RTL codes they might match, just like predicate expressions.
19556 `match_test' C expressions have access to the following variables:
19559 The RTL object defining the operand.
19562 The machine mode of OP.
19565 `INTVAL (OP)', if OP is a `const_int'.
19568 `CONST_DOUBLE_HIGH (OP)', if OP is an integer `const_double'.
19571 `CONST_DOUBLE_LOW (OP)', if OP is an integer `const_double'.
19574 `CONST_DOUBLE_REAL_VALUE (OP)', if OP is a floating-point
19577 The *VAL variables should only be used once another piece of the
19578 expression has verified that OP is the appropriate kind of RTL
19581 Most non-register constraints should be defined with
19582 `define_constraint'. The remaining two definition expressions are only
19583 appropriate for constraints that should be handled specially by
19584 `reload' if they fail to match.
19586 -- MD Expression: define_memory_constraint name docstring exp
19587 Use this expression for constraints that match a subset of all
19588 memory operands: that is, `reload' can make them match by
19589 converting the operand to the form `(mem (reg X))', where X is a
19590 base register (from the register class specified by
19591 `BASE_REG_CLASS', *note Register Classes::).
19593 For example, on the S/390, some instructions do not accept
19594 arbitrary memory references, but only those that do not make use
19595 of an index register. The constraint letter `Q' is defined to
19596 represent a memory address of this type. If `Q' is defined with
19597 `define_memory_constraint', a `Q' constraint can handle any memory
19598 operand, because `reload' knows it can simply copy the memory
19599 address into a base register if required. This is analogous to
19600 the way a `o' constraint can handle any memory operand.
19602 The syntax and semantics are otherwise identical to
19603 `define_constraint'.
19605 -- MD Expression: define_address_constraint name docstring exp
19606 Use this expression for constraints that match a subset of all
19607 address operands: that is, `reload' can make the constraint match
19608 by converting the operand to the form `(reg X)', again with X a
19611 Constraints defined with `define_address_constraint' can only be
19612 used with the `address_operand' predicate, or machine-specific
19613 predicates that work the same way. They are treated analogously to
19614 the generic `p' constraint.
19616 The syntax and semantics are otherwise identical to
19617 `define_constraint'.
19619 For historical reasons, names beginning with the letters `G H' are
19620 reserved for constraints that match only `const_double's, and names
19621 beginning with the letters `I J K L M N O P' are reserved for
19622 constraints that match only `const_int's. This may change in the
19623 future. For the time being, constraints with these names must be
19624 written in a stylized form, so that `genpreds' can tell you did it
19627 (define_constraint "[GHIJKLMNOP]..."
19629 (and (match_code "const_int") ; `const_double' for G/H
19630 CONDITION...)) ; usually a `match_test'
19632 It is fine to use names beginning with other letters for constraints
19633 that match `const_double's or `const_int's.
19635 Each docstring in a constraint definition should be one or more
19636 complete sentences, marked up in Texinfo format. _They are currently
19637 unused._ In the future they will be copied into the GCC manual, in
19638 *Note Machine Constraints::, replacing the hand-maintained tables
19639 currently found in that section. Also, in the future the compiler may
19640 use this to give more helpful diagnostics when poor choice of `asm'
19641 constraints causes a reload failure.
19643 If you put the pseudo-Texinfo directive `@internal' at the beginning
19644 of a docstring, then (in the future) it will appear only in the
19645 internals manual's version of the machine-specific constraint tables.
19646 Use this for constraints that should not appear in `asm' statements.
19649 File: gccint.info, Node: C Constraint Interface, Prev: Define Constraints, Up: Constraints
19651 16.8.8 Testing constraints from C
19652 ---------------------------------
19654 It is occasionally useful to test a constraint from C code rather than
19655 implicitly via the constraint string in a `match_operand'. The
19656 generated file `tm_p.h' declares a few interfaces for working with
19657 machine-specific constraints. None of these interfaces work with the
19658 generic constraints described in *Note Simple Constraints::. This may
19659 change in the future.
19661 *Warning:* `tm_p.h' may declare other functions that operate on
19662 constraints, besides the ones documented here. Do not use those
19663 functions from machine-dependent code. They exist to implement the old
19664 constraint interface that machine-independent components of the
19665 compiler still expect. They will change or disappear in the future.
19667 Some valid constraint names are not valid C identifiers, so there is a
19668 mangling scheme for referring to them from C. Constraint names that do
19669 not contain angle brackets or underscores are left unchanged.
19670 Underscores are doubled, each `<' is replaced with `_l', and each `>'
19671 with `_g'. Here are some examples:
19673 *Original* *Mangled*
19681 Throughout this section, the variable C is either a constraint in the
19682 abstract sense, or a constant from `enum constraint_num'; the variable
19683 M is a mangled constraint name (usually as part of a larger identifier).
19685 -- Enum: constraint_num
19686 For each machine-specific constraint, there is a corresponding
19687 enumeration constant: `CONSTRAINT_' plus the mangled name of the
19688 constraint. Functions that take an `enum constraint_num' as an
19689 argument expect one of these constants.
19691 Machine-independent constraints do not have associated constants.
19692 This may change in the future.
19694 -- Function: inline bool satisfies_constraint_M (rtx EXP)
19695 For each machine-specific, non-register constraint M, there is one
19696 of these functions; it returns `true' if EXP satisfies the
19697 constraint. These functions are only visible if `rtl.h' was
19698 included before `tm_p.h'.
19700 -- Function: bool constraint_satisfied_p (rtx EXP, enum constraint_num
19702 Like the `satisfies_constraint_M' functions, but the constraint to
19703 test is given as an argument, C. If C specifies a register
19704 constraint, this function will always return `false'.
19706 -- Function: enum reg_class regclass_for_constraint (enum
19708 Returns the register class associated with C. If C is not a
19709 register constraint, or those registers are not available for the
19710 currently selected subtarget, returns `NO_REGS'.
19712 Here is an example use of `satisfies_constraint_M'. In peephole
19713 optimizations (*note Peephole Definitions::), operand constraint
19714 strings are ignored, so if there are relevant constraints, they must be
19715 tested in the C condition. In the example, the optimization is applied
19716 if operand 2 does _not_ satisfy the `K' constraint. (This is a
19717 simplified version of a peephole definition from the i386 machine
19721 [(match_scratch:SI 3 "r")
19722 (set (match_operand:SI 0 "register_operand" "")
19723 (mult:SI (match_operand:SI 1 "memory_operand" "")
19724 (match_operand:SI 2 "immediate_operand" "")))]
19726 "!satisfies_constraint_K (operands[2])"
19728 [(set (match_dup 3) (match_dup 1))
19729 (set (match_dup 0) (mult:SI (match_dup 3) (match_dup 2)))]
19734 File: gccint.info, Node: Standard Names, Next: Pattern Ordering, Prev: Constraints, Up: Machine Desc
19736 16.9 Standard Pattern Names For Generation
19737 ==========================================
19739 Here is a table of the instruction names that are meaningful in the RTL
19740 generation pass of the compiler. Giving one of these names to an
19741 instruction pattern tells the RTL generation pass that it can use the
19742 pattern to accomplish a certain task.
19745 Here M stands for a two-letter machine mode name, in lowercase.
19746 This instruction pattern moves data with that machine mode from
19747 operand 1 to operand 0. For example, `movsi' moves full-word data.
19749 If operand 0 is a `subreg' with mode M of a register whose own
19750 mode is wider than M, the effect of this instruction is to store
19751 the specified value in the part of the register that corresponds
19752 to mode M. Bits outside of M, but which are within the same
19753 target word as the `subreg' are undefined. Bits which are outside
19754 the target word are left unchanged.
19756 This class of patterns is special in several ways. First of all,
19757 each of these names up to and including full word size _must_ be
19758 defined, because there is no other way to copy a datum from one
19759 place to another. If there are patterns accepting operands in
19760 larger modes, `movM' must be defined for integer modes of those
19763 Second, these patterns are not used solely in the RTL generation
19764 pass. Even the reload pass can generate move insns to copy values
19765 from stack slots into temporary registers. When it does so, one
19766 of the operands is a hard register and the other is an operand
19767 that can need to be reloaded into a register.
19769 Therefore, when given such a pair of operands, the pattern must
19770 generate RTL which needs no reloading and needs no temporary
19771 registers--no registers other than the operands. For example, if
19772 you support the pattern with a `define_expand', then in such a
19773 case the `define_expand' mustn't call `force_reg' or any other such
19774 function which might generate new pseudo registers.
19776 This requirement exists even for subword modes on a RISC machine
19777 where fetching those modes from memory normally requires several
19778 insns and some temporary registers.
19780 During reload a memory reference with an invalid address may be
19781 passed as an operand. Such an address will be replaced with a
19782 valid address later in the reload pass. In this case, nothing may
19783 be done with the address except to use it as it stands. If it is
19784 copied, it will not be replaced with a valid address. No attempt
19785 should be made to make such an address into a valid address and no
19786 routine (such as `change_address') that will do so may be called.
19787 Note that `general_operand' will fail when applied to such an
19790 The global variable `reload_in_progress' (which must be explicitly
19791 declared if required) can be used to determine whether such special
19792 handling is required.
19794 The variety of operands that have reloads depends on the rest of
19795 the machine description, but typically on a RISC machine these can
19796 only be pseudo registers that did not get hard registers, while on
19797 other machines explicit memory references will get optional
19800 If a scratch register is required to move an object to or from
19801 memory, it can be allocated using `gen_reg_rtx' prior to life
19804 If there are cases which need scratch registers during or after
19805 reload, you must provide an appropriate secondary_reload target
19808 The macro `can_create_pseudo_p' can be used to determine if it is
19809 unsafe to create new pseudo registers. If this variable is
19810 nonzero, then it is unsafe to call `gen_reg_rtx' to allocate a new
19813 The constraints on a `movM' must permit moving any hard register
19814 to any other hard register provided that `HARD_REGNO_MODE_OK'
19815 permits mode M in both registers and `REGISTER_MOVE_COST' applied
19816 to their classes returns a value of 2.
19818 It is obligatory to support floating point `movM' instructions
19819 into and out of any registers that can hold fixed point values,
19820 because unions and structures (which have modes `SImode' or
19821 `DImode') can be in those registers and they may have floating
19824 There may also be a need to support fixed point `movM'
19825 instructions in and out of floating point registers.
19826 Unfortunately, I have forgotten why this was so, and I don't know
19827 whether it is still true. If `HARD_REGNO_MODE_OK' rejects fixed
19828 point values in floating point registers, then the constraints of
19829 the fixed point `movM' instructions must be designed to avoid ever
19830 trying to reload into a floating point register.
19834 These named patterns have been obsoleted by the target hook
19835 `secondary_reload'.
19837 Like `movM', but used when a scratch register is required to move
19838 between operand 0 and operand 1. Operand 2 describes the scratch
19839 register. See the discussion of the `SECONDARY_RELOAD_CLASS'
19840 macro in *note Register Classes::.
19842 There are special restrictions on the form of the `match_operand's
19843 used in these patterns. First, only the predicate for the reload
19844 operand is examined, i.e., `reload_in' examines operand 1, but not
19845 the predicates for operand 0 or 2. Second, there may be only one
19846 alternative in the constraints. Third, only a single register
19847 class letter may be used for the constraint; subsequent constraint
19848 letters are ignored. As a special exception, an empty constraint
19849 string matches the `ALL_REGS' register class. This may relieve
19850 ports of the burden of defining an `ALL_REGS' constraint letter
19851 just for these patterns.
19854 Like `movM' except that if operand 0 is a `subreg' with mode M of
19855 a register whose natural mode is wider, the `movstrictM'
19856 instruction is guaranteed not to alter any of the register except
19857 the part which belongs to mode M.
19860 This variant of a move pattern is designed to load or store a value
19861 from a memory address that is not naturally aligned for its mode.
19862 For a store, the memory will be in operand 0; for a load, the
19863 memory will be in operand 1. The other operand is guaranteed not
19864 to be a memory, so that it's easy to tell whether this is a load
19867 This pattern is used by the autovectorizer, and when expanding a
19868 `MISALIGNED_INDIRECT_REF' expression.
19871 Load several consecutive memory locations into consecutive
19872 registers. Operand 0 is the first of the consecutive registers,
19873 operand 1 is the first memory location, and operand 2 is a
19874 constant: the number of consecutive registers.
19876 Define this only if the target machine really has such an
19877 instruction; do not define this if the most efficient way of
19878 loading consecutive registers from memory is to do them one at a
19881 On some machines, there are restrictions as to which consecutive
19882 registers can be stored into memory, such as particular starting or
19883 ending register numbers or only a range of valid counts. For those
19884 machines, use a `define_expand' (*note Expander Definitions::) and
19885 make the pattern fail if the restrictions are not met.
19887 Write the generated insn as a `parallel' with elements being a
19888 `set' of one register from the appropriate memory location (you may
19889 also need `use' or `clobber' elements). Use a `match_parallel'
19890 (*note RTL Template::) to recognize the insn. See `rs6000.md' for
19891 examples of the use of this insn pattern.
19894 Similar to `load_multiple', but store several consecutive registers
19895 into consecutive memory locations. Operand 0 is the first of the
19896 consecutive memory locations, operand 1 is the first register, and
19897 operand 2 is a constant: the number of consecutive registers.
19900 Set given field in the vector value. Operand 0 is the vector to
19901 modify, operand 1 is new value of field and operand 2 specify the
19905 Extract given field from the vector value. Operand 1 is the
19906 vector, operand 2 specify field index and operand 0 place to store
19909 `vec_extract_evenM'
19910 Extract even elements from the input vectors (operand 1 and
19911 operand 2). The even elements of operand 2 are concatenated to
19912 the even elements of operand 1 in their original order. The result
19913 is stored in operand 0. The output and input vectors should have
19917 Extract odd elements from the input vectors (operand 1 and operand
19918 2). The odd elements of operand 2 are concatenated to the odd
19919 elements of operand 1 in their original order. The result is
19920 stored in operand 0. The output and input vectors should have the
19923 `vec_interleave_highM'
19924 Merge high elements of the two input vectors into the output
19925 vector. The output and input vectors should have the same modes
19926 (`N' elements). The high `N/2' elements of the first input vector
19927 are interleaved with the high `N/2' elements of the second input
19930 `vec_interleave_lowM'
19931 Merge low elements of the two input vectors into the output
19932 vector. The output and input vectors should have the same modes
19933 (`N' elements). The low `N/2' elements of the first input vector
19934 are interleaved with the low `N/2' elements of the second input
19938 Initialize the vector to given values. Operand 0 is the vector to
19939 initialize and operand 1 is parallel containing values for
19943 Output a push instruction. Operand 0 is value to push. Used only
19944 when `PUSH_ROUNDING' is defined. For historical reason, this
19945 pattern may be missing and in such case an `mov' expander is used
19946 instead, with a `MEM' expression forming the push operation. The
19947 `mov' expander method is deprecated.
19950 Add operand 2 and operand 1, storing the result in operand 0. All
19951 operands must have mode M. This can be used even on two-address
19952 machines, by means of constraints requiring operands 1 and 0 to be
19955 `ssaddM3', `usaddM3'
19957 `subM3', `sssubM3', `ussubM3'
19959 `mulM3', `ssmulM3', `usmulM3'
19961 `udivM3', `usdivM3'
19964 `andM3', `iorM3', `xorM3'
19965 Similar, for other arithmetic operations.
19968 Signed minimum and maximum operations. When used with floating
19969 point, if both operands are zeros, or if either operand is `NaN',
19970 then it is unspecified which of the two operands is returned as
19973 `reduc_smin_M', `reduc_smax_M'
19974 Find the signed minimum/maximum of the elements of a vector. The
19975 vector is operand 1, and the scalar result is stored in the least
19976 significant bits of operand 0 (also a vector). The output and
19977 input vector should have the same modes.
19979 `reduc_umin_M', `reduc_umax_M'
19980 Find the unsigned minimum/maximum of the elements of a vector. The
19981 vector is operand 1, and the scalar result is stored in the least
19982 significant bits of operand 0 (also a vector). The output and
19983 input vector should have the same modes.
19986 Compute the sum of the signed elements of a vector. The vector is
19987 operand 1, and the scalar result is stored in the least
19988 significant bits of operand 0 (also a vector). The output and
19989 input vector should have the same modes.
19992 Compute the sum of the unsigned elements of a vector. The vector
19993 is operand 1, and the scalar result is stored in the least
19994 significant bits of operand 0 (also a vector). The output and
19995 input vector should have the same modes.
20000 Compute the sum of the products of two signed/unsigned elements.
20001 Operand 1 and operand 2 are of the same mode. Their product, which
20002 is of a wider mode, is computed and added to operand 3. Operand 3
20003 is of a mode equal or wider than the mode of the product. The
20004 result is placed in operand 0, which is of the same mode as
20010 Operands 0 and 2 are of the same mode, which is wider than the
20011 mode of operand 1. Add operand 1 to operand 2 and place the
20012 widened result in operand 0. (This is used express accumulation of
20013 elements into an accumulator of a wider mode.)
20015 `vec_shl_M', `vec_shr_M'
20016 Whole vector left/right shift in bits. Operand 1 is a vector to
20017 be shifted. Operand 2 is an integer shift amount in bits.
20018 Operand 0 is where the resulting shifted vector is stored. The
20019 output and input vectors should have the same modes.
20022 Narrow (demote) and merge the elements of two vectors. Operands 1
20023 and 2 are vectors of the same mode having N integral or floating
20024 point elements of size S. Operand 0 is the resulting vector in
20025 which 2*N elements of size N/2 are concatenated after narrowing
20026 them down using truncation.
20028 `vec_pack_ssat_M', `vec_pack_usat_M'
20029 Narrow (demote) and merge the elements of two vectors. Operands 1
20030 and 2 are vectors of the same mode having N integral elements of
20031 size S. Operand 0 is the resulting vector in which the elements
20032 of the two input vectors are concatenated after narrowing them
20033 down using signed/unsigned saturating arithmetic.
20035 `vec_pack_sfix_trunc_M', `vec_pack_ufix_trunc_M'
20036 Narrow, convert to signed/unsigned integral type and merge the
20037 elements of two vectors. Operands 1 and 2 are vectors of the same
20038 mode having N floating point elements of size S. Operand 0 is the
20039 resulting vector in which 2*N elements of size N/2 are
20042 `vec_unpacks_hi_M', `vec_unpacks_lo_M'
20043 Extract and widen (promote) the high/low part of a vector of signed
20044 integral or floating point elements. The input vector (operand 1)
20045 has N elements of size S. Widen (promote) the high/low elements
20046 of the vector using signed or floating point extension and place
20047 the resulting N/2 values of size 2*S in the output vector (operand
20050 `vec_unpacku_hi_M', `vec_unpacku_lo_M'
20051 Extract and widen (promote) the high/low part of a vector of
20052 unsigned integral elements. The input vector (operand 1) has N
20053 elements of size S. Widen (promote) the high/low elements of the
20054 vector using zero extension and place the resulting N/2 values of
20055 size 2*S in the output vector (operand 0).
20057 `vec_unpacks_float_hi_M', `vec_unpacks_float_lo_M'
20058 `vec_unpacku_float_hi_M', `vec_unpacku_float_lo_M'
20059 Extract, convert to floating point type and widen the high/low
20060 part of a vector of signed/unsigned integral elements. The input
20061 vector (operand 1) has N elements of size S. Convert the high/low
20062 elements of the vector using floating point conversion and place
20063 the resulting N/2 values of size 2*S in the output vector (operand
20066 `vec_widen_umult_hi_M', `vec_widen_umult_lo_M'
20067 `vec_widen_smult_hi_M', `vec_widen_smult_lo_M'
20068 Signed/Unsigned widening multiplication. The two inputs (operands
20069 1 and 2) are vectors with N signed/unsigned elements of size S.
20070 Multiply the high/low elements of the two vectors, and put the N/2
20071 products of size 2*S in the output vector (operand 0).
20074 Multiply operands 1 and 2, which have mode `HImode', and store a
20075 `SImode' product in operand 0.
20077 `mulqihi3', `mulsidi3'
20078 Similar widening-multiplication instructions of other widths.
20080 `umulqihi3', `umulhisi3', `umulsidi3'
20081 Similar widening-multiplication instructions that do unsigned
20084 `usmulqihi3', `usmulhisi3', `usmulsidi3'
20085 Similar widening-multiplication instructions that interpret the
20086 first operand as unsigned and the second operand as signed, then
20087 do a signed multiplication.
20090 Perform a signed multiplication of operands 1 and 2, which have
20091 mode M, and store the most significant half of the product in
20092 operand 0. The least significant half of the product is discarded.
20095 Similar, but the multiplication is unsigned.
20098 Multiply operands 1 and 2, sign-extend them to mode N, add operand
20099 3, and store the result in operand 0. Operands 1 and 2 have mode
20100 M and operands 0 and 3 have mode N. Both modes must be integer or
20101 fixed-point modes and N must be twice the size of M.
20103 In other words, `maddMN4' is like `mulMN3' except that it also
20106 These instructions are not allowed to `FAIL'.
20109 Like `maddMN4', but zero-extend the multiplication operands
20110 instead of sign-extending them.
20113 Like `maddMN4', but all involved operations must be
20117 Like `umaddMN4', but all involved operations must be
20118 unsigned-saturating.
20121 Multiply operands 1 and 2, sign-extend them to mode N, subtract the
20122 result from operand 3, and store the result in operand 0.
20123 Operands 1 and 2 have mode M and operands 0 and 3 have mode N.
20124 Both modes must be integer or fixed-point modes and N must be twice
20127 In other words, `msubMN4' is like `mulMN3' except that it also
20128 subtracts the result from operand 3.
20130 These instructions are not allowed to `FAIL'.
20133 Like `msubMN4', but zero-extend the multiplication operands
20134 instead of sign-extending them.
20137 Like `msubMN4', but all involved operations must be
20141 Like `umsubMN4', but all involved operations must be
20142 unsigned-saturating.
20145 Signed division that produces both a quotient and a remainder.
20146 Operand 1 is divided by operand 2 to produce a quotient stored in
20147 operand 0 and a remainder stored in operand 3.
20149 For machines with an instruction that produces both a quotient and
20150 a remainder, provide a pattern for `divmodM4' but do not provide
20151 patterns for `divM3' and `modM3'. This allows optimization in the
20152 relatively common case when both the quotient and remainder are
20155 If an instruction that just produces a quotient or just a remainder
20156 exists and is more efficient than the instruction that produces
20157 both, write the output routine of `divmodM4' to call
20158 `find_reg_note' and look for a `REG_UNUSED' note on the quotient
20159 or remainder and generate the appropriate instruction.
20162 Similar, but does unsigned division.
20164 `ashlM3', `ssashlM3', `usashlM3'
20165 Arithmetic-shift operand 1 left by a number of bits specified by
20166 operand 2, and store the result in operand 0. Here M is the mode
20167 of operand 0 and operand 1; operand 2's mode is specified by the
20168 instruction pattern, and the compiler will convert the operand to
20169 that mode before generating the instruction. The meaning of
20170 out-of-range shift counts can optionally be specified by
20171 `TARGET_SHIFT_TRUNCATION_MASK'. *Note
20172 TARGET_SHIFT_TRUNCATION_MASK::. Operand 2 is always a scalar type.
20174 `ashrM3', `lshrM3', `rotlM3', `rotrM3'
20175 Other shift and rotate instructions, analogous to the `ashlM3'
20176 instructions. Operand 2 is always a scalar type.
20178 `vashlM3', `vashrM3', `vlshrM3', `vrotlM3', `vrotrM3'
20179 Vector shift and rotate instructions that take vectors as operand 2
20180 instead of a scalar type.
20182 `negM2', `ssnegM2', `usnegM2'
20183 Negate operand 1 and store the result in operand 0.
20186 Store the absolute value of operand 1 into operand 0.
20189 Store the square root of operand 1 into operand 0.
20191 The `sqrt' built-in function of C always uses the mode which
20192 corresponds to the C data type `double' and the `sqrtf' built-in
20193 function uses the mode which corresponds to the C data type
20197 Store the remainder of dividing operand 1 by operand 2 into
20198 operand 0, rounded towards zero to an integer.
20200 The `fmod' built-in function of C always uses the mode which
20201 corresponds to the C data type `double' and the `fmodf' built-in
20202 function uses the mode which corresponds to the C data type
20206 Store the remainder of dividing operand 1 by operand 2 into
20207 operand 0, rounded to the nearest integer.
20209 The `remainder' built-in function of C always uses the mode which
20210 corresponds to the C data type `double' and the `remainderf'
20211 built-in function uses the mode which corresponds to the C data
20215 Store the cosine of operand 1 into operand 0.
20217 The `cos' built-in function of C always uses the mode which
20218 corresponds to the C data type `double' and the `cosf' built-in
20219 function uses the mode which corresponds to the C data type
20223 Store the sine of operand 1 into operand 0.
20225 The `sin' built-in function of C always uses the mode which
20226 corresponds to the C data type `double' and the `sinf' built-in
20227 function uses the mode which corresponds to the C data type
20231 Store the exponential of operand 1 into operand 0.
20233 The `exp' built-in function of C always uses the mode which
20234 corresponds to the C data type `double' and the `expf' built-in
20235 function uses the mode which corresponds to the C data type
20239 Store the natural logarithm of operand 1 into operand 0.
20241 The `log' built-in function of C always uses the mode which
20242 corresponds to the C data type `double' and the `logf' built-in
20243 function uses the mode which corresponds to the C data type
20247 Store the value of operand 1 raised to the exponent operand 2 into
20250 The `pow' built-in function of C always uses the mode which
20251 corresponds to the C data type `double' and the `powf' built-in
20252 function uses the mode which corresponds to the C data type
20256 Store the arc tangent (inverse tangent) of operand 1 divided by
20257 operand 2 into operand 0, using the signs of both arguments to
20258 determine the quadrant of the result.
20260 The `atan2' built-in function of C always uses the mode which
20261 corresponds to the C data type `double' and the `atan2f' built-in
20262 function uses the mode which corresponds to the C data type
20266 Store the largest integral value not greater than argument.
20268 The `floor' built-in function of C always uses the mode which
20269 corresponds to the C data type `double' and the `floorf' built-in
20270 function uses the mode which corresponds to the C data type
20274 Store the argument rounded to integer towards zero.
20276 The `trunc' built-in function of C always uses the mode which
20277 corresponds to the C data type `double' and the `truncf' built-in
20278 function uses the mode which corresponds to the C data type
20282 Store the argument rounded to integer away from zero.
20284 The `round' built-in function of C always uses the mode which
20285 corresponds to the C data type `double' and the `roundf' built-in
20286 function uses the mode which corresponds to the C data type
20290 Store the argument rounded to integer away from zero.
20292 The `ceil' built-in function of C always uses the mode which
20293 corresponds to the C data type `double' and the `ceilf' built-in
20294 function uses the mode which corresponds to the C data type
20298 Store the argument rounded according to the default rounding mode
20300 The `nearbyint' built-in function of C always uses the mode which
20301 corresponds to the C data type `double' and the `nearbyintf'
20302 built-in function uses the mode which corresponds to the C data
20306 Store the argument rounded according to the default rounding mode
20307 and raise the inexact exception when the result differs in value
20310 The `rint' built-in function of C always uses the mode which
20311 corresponds to the C data type `double' and the `rintf' built-in
20312 function uses the mode which corresponds to the C data type
20316 Convert operand 1 (valid for floating point mode M) to fixed point
20317 mode N as a signed number according to the current rounding mode
20318 and store in operand 0 (which has mode N).
20321 Convert operand 1 (valid for floating point mode M) to fixed point
20322 mode N as a signed number rounding to nearest and away from zero
20323 and store in operand 0 (which has mode N).
20326 Convert operand 1 (valid for floating point mode M) to fixed point
20327 mode N as a signed number rounding down and store in operand 0
20328 (which has mode N).
20331 Convert operand 1 (valid for floating point mode M) to fixed point
20332 mode N as a signed number rounding up and store in operand 0
20333 (which has mode N).
20336 Store a value with the magnitude of operand 1 and the sign of
20337 operand 2 into operand 0.
20339 The `copysign' built-in function of C always uses the mode which
20340 corresponds to the C data type `double' and the `copysignf'
20341 built-in function uses the mode which corresponds to the C data
20345 Store into operand 0 one plus the index of the least significant
20346 1-bit of operand 1. If operand 1 is zero, store zero. M is the
20347 mode of operand 0; operand 1's mode is specified by the instruction
20348 pattern, and the compiler will convert the operand to that mode
20349 before generating the instruction.
20351 The `ffs' built-in function of C always uses the mode which
20352 corresponds to the C data type `int'.
20355 Store into operand 0 the number of leading 0-bits in X, starting
20356 at the most significant bit position. If X is 0, the
20357 `CLZ_DEFINED_VALUE_AT_ZERO' (*note Misc::) macro defines if the
20358 result is undefined or has a useful value. M is the mode of
20359 operand 0; operand 1's mode is specified by the instruction
20360 pattern, and the compiler will convert the operand to that mode
20361 before generating the instruction.
20364 Store into operand 0 the number of trailing 0-bits in X, starting
20365 at the least significant bit position. If X is 0, the
20366 `CTZ_DEFINED_VALUE_AT_ZERO' (*note Misc::) macro defines if the
20367 result is undefined or has a useful value. M is the mode of
20368 operand 0; operand 1's mode is specified by the instruction
20369 pattern, and the compiler will convert the operand to that mode
20370 before generating the instruction.
20373 Store into operand 0 the number of 1-bits in X. M is the mode of
20374 operand 0; operand 1's mode is specified by the instruction
20375 pattern, and the compiler will convert the operand to that mode
20376 before generating the instruction.
20379 Store into operand 0 the parity of X, i.e. the number of 1-bits in
20380 X modulo 2. M is the mode of operand 0; operand 1's mode is
20381 specified by the instruction pattern, and the compiler will convert
20382 the operand to that mode before generating the instruction.
20385 Store the bitwise-complement of operand 1 into operand 0.
20388 Compare operand 0 and operand 1, and set the condition codes. The
20389 RTL pattern should look like this:
20391 (set (cc0) (compare (match_operand:M 0 ...)
20392 (match_operand:M 1 ...)))
20395 Compare operand 0 against zero, and set the condition codes. The
20396 RTL pattern should look like this:
20398 (set (cc0) (match_operand:M 0 ...))
20400 `tstM' patterns should not be defined for machines that do not use
20401 `(cc0)'. Doing so would confuse the optimizer since it would no
20402 longer be clear which `set' operations were comparisons. The
20403 `cmpM' patterns should be used instead.
20406 Block move instruction. The destination and source blocks of
20407 memory are the first two operands, and both are `mem:BLK's with an
20408 address in mode `Pmode'.
20410 The number of bytes to move is the third operand, in mode M.
20411 Usually, you specify `word_mode' for M. However, if you can
20412 generate better code knowing the range of valid lengths is smaller
20413 than those representable in a full word, you should provide a
20414 pattern with a mode corresponding to the range of values you can
20415 handle efficiently (e.g., `QImode' for values in the range 0-127;
20416 note we avoid numbers that appear negative) and also a pattern
20419 The fourth operand is the known shared alignment of the source and
20420 destination, in the form of a `const_int' rtx. Thus, if the
20421 compiler knows that both source and destination are word-aligned,
20422 it may provide the value 4 for this operand.
20424 Optional operands 5 and 6 specify expected alignment and size of
20425 block respectively. The expected alignment differs from alignment
20426 in operand 4 in a way that the blocks are not required to be
20427 aligned according to it in all cases. This expected alignment is
20428 also in bytes, just like operand 4. Expected size, when unknown,
20429 is set to `(const_int -1)'.
20431 Descriptions of multiple `movmemM' patterns can only be beneficial
20432 if the patterns for smaller modes have fewer restrictions on their
20433 first, second and fourth operands. Note that the mode M in
20434 `movmemM' does not impose any restriction on the mode of
20435 individually moved data units in the block.
20437 These patterns need not give special consideration to the
20438 possibility that the source and destination strings might overlap.
20441 String copy instruction, with `stpcpy' semantics. Operand 0 is an
20442 output operand in mode `Pmode'. The addresses of the destination
20443 and source strings are operands 1 and 2, and both are `mem:BLK's
20444 with addresses in mode `Pmode'. The execution of the expansion of
20445 this pattern should store in operand 0 the address in which the
20446 `NUL' terminator was stored in the destination string.
20449 Block set instruction. The destination string is the first
20450 operand, given as a `mem:BLK' whose address is in mode `Pmode'.
20451 The number of bytes to set is the second operand, in mode M. The
20452 value to initialize the memory with is the third operand. Targets
20453 that only support the clearing of memory should reject any value
20454 that is not the constant 0. See `movmemM' for a discussion of the
20457 The fourth operand is the known alignment of the destination, in
20458 the form of a `const_int' rtx. Thus, if the compiler knows that
20459 the destination is word-aligned, it may provide the value 4 for
20462 Optional operands 5 and 6 specify expected alignment and size of
20463 block respectively. The expected alignment differs from alignment
20464 in operand 4 in a way that the blocks are not required to be
20465 aligned according to it in all cases. This expected alignment is
20466 also in bytes, just like operand 4. Expected size, when unknown,
20467 is set to `(const_int -1)'.
20469 The use for multiple `setmemM' is as for `movmemM'.
20472 String compare instruction, with five operands. Operand 0 is the
20473 output; it has mode M. The remaining four operands are like the
20474 operands of `movmemM'. The two memory blocks specified are
20475 compared byte by byte in lexicographic order starting at the
20476 beginning of each string. The instruction is not allowed to
20477 prefetch more than one byte at a time since either string may end
20478 in the first byte and reading past that may access an invalid page
20479 or segment and cause a fault. The effect of the instruction is to
20480 store a value in operand 0 whose sign indicates the result of the
20484 String compare instruction, without known maximum length. Operand
20485 0 is the output; it has mode M. The second and third operand are
20486 the blocks of memory to be compared; both are `mem:BLK' with an
20487 address in mode `Pmode'.
20489 The fourth operand is the known shared alignment of the source and
20490 destination, in the form of a `const_int' rtx. Thus, if the
20491 compiler knows that both source and destination are word-aligned,
20492 it may provide the value 4 for this operand.
20494 The two memory blocks specified are compared byte by byte in
20495 lexicographic order starting at the beginning of each string. The
20496 instruction is not allowed to prefetch more than one byte at a
20497 time since either string may end in the first byte and reading
20498 past that may access an invalid page or segment and cause a fault.
20499 The effect of the instruction is to store a value in operand 0
20500 whose sign indicates the result of the comparison.
20503 Block compare instruction, with five operands like the operands of
20504 `cmpstrM'. The two memory blocks specified are compared byte by
20505 byte in lexicographic order starting at the beginning of each
20506 block. Unlike `cmpstrM' the instruction can prefetch any bytes in
20507 the two memory blocks. The effect of the instruction is to store
20508 a value in operand 0 whose sign indicates the result of the
20512 Compute the length of a string, with three operands. Operand 0 is
20513 the result (of mode M), operand 1 is a `mem' referring to the
20514 first character of the string, operand 2 is the character to
20515 search for (normally zero), and operand 3 is a constant describing
20516 the known alignment of the beginning of the string.
20519 Convert signed integer operand 1 (valid for fixed point mode M) to
20520 floating point mode N and store in operand 0 (which has mode N).
20523 Convert unsigned integer operand 1 (valid for fixed point mode M)
20524 to floating point mode N and store in operand 0 (which has mode N).
20527 Convert operand 1 (valid for floating point mode M) to fixed point
20528 mode N as a signed number and store in operand 0 (which has mode
20529 N). This instruction's result is defined only when the value of
20530 operand 1 is an integer.
20532 If the machine description defines this pattern, it also needs to
20533 define the `ftrunc' pattern.
20536 Convert operand 1 (valid for floating point mode M) to fixed point
20537 mode N as an unsigned number and store in operand 0 (which has
20538 mode N). This instruction's result is defined only when the value
20539 of operand 1 is an integer.
20542 Convert operand 1 (valid for floating point mode M) to an integer
20543 value, still represented in floating point mode M, and store it in
20544 operand 0 (valid for floating point mode M).
20547 Like `fixMN2' but works for any floating point value of mode M by
20548 converting the value to an integer.
20551 Like `fixunsMN2' but works for any floating point value of mode M
20552 by converting the value to an integer.
20555 Truncate operand 1 (valid for mode M) to mode N and store in
20556 operand 0 (which has mode N). Both modes must be fixed point or
20557 both floating point.
20560 Sign-extend operand 1 (valid for mode M) to mode N and store in
20561 operand 0 (which has mode N). Both modes must be fixed point or
20562 both floating point.
20565 Zero-extend operand 1 (valid for mode M) to mode N and store in
20566 operand 0 (which has mode N). Both modes must be fixed point.
20569 Convert operand 1 of mode M to mode N and store in operand 0
20570 (which has mode N). Mode M and mode N could be fixed-point to
20571 fixed-point, signed integer to fixed-point, fixed-point to signed
20572 integer, floating-point to fixed-point, or fixed-point to
20573 floating-point. When overflows or underflows happen, the results
20577 Convert operand 1 of mode M to mode N and store in operand 0
20578 (which has mode N). Mode M and mode N could be fixed-point to
20579 fixed-point, signed integer to fixed-point, or floating-point to
20580 fixed-point. When overflows or underflows happen, the instruction
20581 saturates the results to the maximum or the minimum.
20584 Convert operand 1 of mode M to mode N and store in operand 0
20585 (which has mode N). Mode M and mode N could be unsigned integer
20586 to fixed-point, or fixed-point to unsigned integer. When
20587 overflows or underflows happen, the results are undefined.
20590 Convert unsigned integer operand 1 of mode M to fixed-point mode N
20591 and store in operand 0 (which has mode N). When overflows or
20592 underflows happen, the instruction saturates the results to the
20593 maximum or the minimum.
20596 Extract a bit-field from operand 1 (a register or memory operand),
20597 where operand 2 specifies the width in bits and operand 3 the
20598 starting bit, and store it in operand 0. Operand 0 must have mode
20599 `word_mode'. Operand 1 may have mode `byte_mode' or `word_mode';
20600 often `word_mode' is allowed only for registers. Operands 2 and 3
20601 must be valid for `word_mode'.
20603 The RTL generation pass generates this instruction only with
20604 constants for operands 2 and 3 and the constant is never zero for
20607 The bit-field value is sign-extended to a full word integer before
20608 it is stored in operand 0.
20611 Like `extv' except that the bit-field value is zero-extended.
20614 Store operand 3 (which must be valid for `word_mode') into a
20615 bit-field in operand 0, where operand 1 specifies the width in
20616 bits and operand 2 the starting bit. Operand 0 may have mode
20617 `byte_mode' or `word_mode'; often `word_mode' is allowed only for
20618 registers. Operands 1 and 2 must be valid for `word_mode'.
20620 The RTL generation pass generates this instruction only with
20621 constants for operands 1 and 2 and the constant is never zero for
20625 Conditionally move operand 2 or operand 3 into operand 0 according
20626 to the comparison in operand 1. If the comparison is true,
20627 operand 2 is moved into operand 0, otherwise operand 3 is moved.
20629 The mode of the operands being compared need not be the same as
20630 the operands being moved. Some machines, sparc64 for example,
20631 have instructions that conditionally move an integer value based
20632 on the floating point condition codes and vice versa.
20634 If the machine does not have conditional move instructions, do not
20635 define these patterns.
20638 Similar to `movMODEcc' but for conditional addition. Conditionally
20639 move operand 2 or (operands 2 + operand 3) into operand 0
20640 according to the comparison in operand 1. If the comparison is
20641 true, operand 2 is moved into operand 0, otherwise (operand 2 +
20642 operand 3) is moved.
20645 Store zero or nonzero in the operand according to the condition
20646 codes. Value stored is nonzero iff the condition COND is true.
20647 COND is the name of a comparison operation expression code, such
20648 as `eq', `lt' or `leu'.
20650 You specify the mode that the operand must have when you write the
20651 `match_operand' expression. The compiler automatically sees which
20652 mode you have used and supplies an operand of that mode.
20654 The value stored for a true condition must have 1 as its low bit,
20655 or else must be negative. Otherwise the instruction is not
20656 suitable and you should omit it from the machine description. You
20657 describe to the compiler exactly which value is stored by defining
20658 the macro `STORE_FLAG_VALUE' (*note Misc::). If a description
20659 cannot be found that can be used for all the `sCOND' patterns, you
20660 should omit those operations from the machine description.
20662 These operations may fail, but should do so only in relatively
20663 uncommon cases; if they would fail for common cases involving
20664 integer comparisons, it is best to omit these patterns.
20666 If these operations are omitted, the compiler will usually
20667 generate code that copies the constant one to the target and
20668 branches around an assignment of zero to the target. If this code
20669 is more efficient than the potential instructions used for the
20670 `sCOND' pattern followed by those required to convert the result
20671 into a 1 or a zero in `SImode', you should omit the `sCOND'
20672 operations from the machine description.
20675 Conditional branch instruction. Operand 0 is a `label_ref' that
20676 refers to the label to jump to. Jump if the condition codes meet
20679 Some machines do not follow the model assumed here where a
20680 comparison instruction is followed by a conditional branch
20681 instruction. In that case, the `cmpM' (and `tstM') patterns should
20682 simply store the operands away and generate all the required insns
20683 in a `define_expand' (*note Expander Definitions::) for the
20684 conditional branch operations. All calls to expand `bCOND'
20685 patterns are immediately preceded by calls to expand either a
20686 `cmpM' pattern or a `tstM' pattern.
20688 Machines that use a pseudo register for the condition code value,
20689 or where the mode used for the comparison depends on the condition
20690 being tested, should also use the above mechanism. *Note Jump
20693 The above discussion also applies to the `movMODEcc' and `sCOND'
20697 Conditional branch instruction combined with a compare instruction.
20698 Operand 0 is a comparison operator. Operand 1 and operand 2 are
20699 the first and second operands of the comparison, respectively.
20700 Operand 3 is a `label_ref' that refers to the label to jump to.
20703 A jump inside a function; an unconditional branch. Operand 0 is
20704 the `label_ref' of the label to jump to. This pattern name is
20705 mandatory on all machines.
20708 Subroutine call instruction returning no value. Operand 0 is the
20709 function to call; operand 1 is the number of bytes of arguments
20710 pushed as a `const_int'; operand 2 is the number of registers used
20713 On most machines, operand 2 is not actually stored into the RTL
20714 pattern. It is supplied for the sake of some RISC machines which
20715 need to put this information into the assembler code; they can put
20716 it in the RTL instead of operand 1.
20718 Operand 0 should be a `mem' RTX whose address is the address of the
20719 function. Note, however, that this address can be a `symbol_ref'
20720 expression even if it would not be a legitimate memory address on
20721 the target machine. If it is also not a valid argument for a call
20722 instruction, the pattern for this operation should be a
20723 `define_expand' (*note Expander Definitions::) that places the
20724 address into a register and uses that register in the call
20728 Subroutine call instruction returning a value. Operand 0 is the
20729 hard register in which the value is returned. There are three more
20730 operands, the same as the three operands of the `call' instruction
20731 (but with numbers increased by one).
20733 Subroutines that return `BLKmode' objects use the `call' insn.
20735 `call_pop', `call_value_pop'
20736 Similar to `call' and `call_value', except used if defined and if
20737 `RETURN_POPS_ARGS' is nonzero. They should emit a `parallel' that
20738 contains both the function call and a `set' to indicate the
20739 adjustment made to the frame pointer.
20741 For machines where `RETURN_POPS_ARGS' can be nonzero, the use of
20742 these patterns increases the number of functions for which the
20743 frame pointer can be eliminated, if desired.
20746 Subroutine call instruction returning a value of any type.
20747 Operand 0 is the function to call; operand 1 is a memory location
20748 where the result of calling the function is to be stored; operand
20749 2 is a `parallel' expression where each element is a `set'
20750 expression that indicates the saving of a function return value
20751 into the result block.
20753 This instruction pattern should be defined to support
20754 `__builtin_apply' on machines where special instructions are needed
20755 to call a subroutine with arbitrary arguments or to save the value
20756 returned. This instruction pattern is required on machines that
20757 have multiple registers that can hold a return value (i.e.
20758 `FUNCTION_VALUE_REGNO_P' is true for more than one register).
20761 Subroutine return instruction. This instruction pattern name
20762 should be defined only if a single instruction can do all the work
20763 of returning from a function.
20765 Like the `movM' patterns, this pattern is also used after the RTL
20766 generation phase. In this case it is to support machines where
20767 multiple instructions are usually needed to return from a
20768 function, but some class of functions only requires one
20769 instruction to implement a return. Normally, the applicable
20770 functions are those which do not need to save any registers or
20771 allocate stack space.
20773 For such machines, the condition specified in this pattern should
20774 only be true when `reload_completed' is nonzero and the function's
20775 epilogue would only be a single instruction. For machines with
20776 register windows, the routine `leaf_function_p' may be used to
20777 determine if a register window push is required.
20779 Machines that have conditional return instructions should define
20784 (if_then_else (match_operator
20785 0 "comparison_operator"
20786 [(cc0) (const_int 0)])
20792 where CONDITION would normally be the same condition specified on
20793 the named `return' pattern.
20796 Untyped subroutine return instruction. This instruction pattern
20797 should be defined to support `__builtin_return' on machines where
20798 special instructions are needed to return a value of any type.
20800 Operand 0 is a memory location where the result of calling a
20801 function with `__builtin_apply' is stored; operand 1 is a
20802 `parallel' expression where each element is a `set' expression
20803 that indicates the restoring of a function return value from the
20807 No-op instruction. This instruction pattern name should always be
20808 defined to output a no-op in assembler code. `(const_int 0)' will
20809 do as an RTL pattern.
20812 An instruction to jump to an address which is operand zero. This
20813 pattern name is mandatory on all machines.
20816 Instruction to jump through a dispatch table, including bounds
20817 checking. This instruction takes five operands:
20819 1. The index to dispatch on, which has mode `SImode'.
20821 2. The lower bound for indices in the table, an integer constant.
20823 3. The total range of indices in the table--the largest index
20824 minus the smallest one (both inclusive).
20826 4. A label that precedes the table itself.
20828 5. A label to jump to if the index has a value outside the
20831 The table is a `addr_vec' or `addr_diff_vec' inside of a
20832 `jump_insn'. The number of elements in the table is one plus the
20833 difference between the upper bound and the lower bound.
20836 Instruction to jump to a variable address. This is a low-level
20837 capability which can be used to implement a dispatch table when
20838 there is no `casesi' pattern.
20840 This pattern requires two operands: the address or offset, and a
20841 label which should immediately precede the jump table. If the
20842 macro `CASE_VECTOR_PC_RELATIVE' evaluates to a nonzero value then
20843 the first operand is an offset which counts from the address of
20844 the table; otherwise, it is an absolute address to jump to. In
20845 either case, the first operand has mode `Pmode'.
20847 The `tablejump' insn is always the last insn before the jump table
20848 it uses. Its assembler code normally has no need to use the
20849 second operand, but you should incorporate it in the RTL pattern so
20850 that the jump optimizer will not delete the table as unreachable
20853 `decrement_and_branch_until_zero'
20854 Conditional branch instruction that decrements a register and
20855 jumps if the register is nonzero. Operand 0 is the register to
20856 decrement and test; operand 1 is the label to jump to if the
20857 register is nonzero. *Note Looping Patterns::.
20859 This optional instruction pattern is only used by the combiner,
20860 typically for loops reversed by the loop optimizer when strength
20861 reduction is enabled.
20864 Conditional branch instruction that decrements a register and
20865 jumps if the register is nonzero. This instruction takes five
20866 operands: Operand 0 is the register to decrement and test; operand
20867 1 is the number of loop iterations as a `const_int' or
20868 `const0_rtx' if this cannot be determined until run-time; operand
20869 2 is the actual or estimated maximum number of iterations as a
20870 `const_int'; operand 3 is the number of enclosed loops as a
20871 `const_int' (an innermost loop has a value of 1); operand 4 is the
20872 label to jump to if the register is nonzero. *Note Looping
20875 This optional instruction pattern should be defined for machines
20876 with low-overhead looping instructions as the loop optimizer will
20877 try to modify suitable loops to utilize it. If nested
20878 low-overhead looping is not supported, use a `define_expand'
20879 (*note Expander Definitions::) and make the pattern fail if
20880 operand 3 is not `const1_rtx'. Similarly, if the actual or
20881 estimated maximum number of iterations is too large for this
20882 instruction, make it fail.
20885 Companion instruction to `doloop_end' required for machines that
20886 need to perform some initialization, such as loading special
20887 registers used by a low-overhead looping instruction. If
20888 initialization insns do not always need to be emitted, use a
20889 `define_expand' (*note Expander Definitions::) and make it fail.
20891 `canonicalize_funcptr_for_compare'
20892 Canonicalize the function pointer in operand 1 and store the result
20895 Operand 0 is always a `reg' and has mode `Pmode'; operand 1 may be
20896 a `reg', `mem', `symbol_ref', `const_int', etc and also has mode
20899 Canonicalization of a function pointer usually involves computing
20900 the address of the function which would be called if the function
20901 pointer were used in an indirect call.
20903 Only define this pattern if function pointers on the target machine
20904 can have different values but still call the same function when
20905 used in an indirect call.
20908 `save_stack_function'
20909 `save_stack_nonlocal'
20910 `restore_stack_block'
20911 `restore_stack_function'
20912 `restore_stack_nonlocal'
20913 Most machines save and restore the stack pointer by copying it to
20914 or from an object of mode `Pmode'. Do not define these patterns on
20917 Some machines require special handling for stack pointer saves and
20918 restores. On those machines, define the patterns corresponding to
20919 the non-standard cases by using a `define_expand' (*note Expander
20920 Definitions::) that produces the required insns. The three types
20921 of saves and restores are:
20923 1. `save_stack_block' saves the stack pointer at the start of a
20924 block that allocates a variable-sized object, and
20925 `restore_stack_block' restores the stack pointer when the
20928 2. `save_stack_function' and `restore_stack_function' do a
20929 similar job for the outermost block of a function and are
20930 used when the function allocates variable-sized objects or
20931 calls `alloca'. Only the epilogue uses the restored stack
20932 pointer, allowing a simpler save or restore sequence on some
20935 3. `save_stack_nonlocal' is used in functions that contain labels
20936 branched to by nested functions. It saves the stack pointer
20937 in such a way that the inner function can use
20938 `restore_stack_nonlocal' to restore the stack pointer. The
20939 compiler generates code to restore the frame and argument
20940 pointer registers, but some machines require saving and
20941 restoring additional data such as register window information
20942 or stack backchains. Place insns in these patterns to save
20943 and restore any such required data.
20945 When saving the stack pointer, operand 0 is the save area and
20946 operand 1 is the stack pointer. The mode used to allocate the
20947 save area defaults to `Pmode' but you can override that choice by
20948 defining the `STACK_SAVEAREA_MODE' macro (*note Storage Layout::).
20949 You must specify an integral mode, or `VOIDmode' if no save area
20950 is needed for a particular type of save (either because no save is
20951 needed or because a machine-specific save area can be used).
20952 Operand 0 is the stack pointer and operand 1 is the save area for
20953 restore operations. If `save_stack_block' is defined, operand 0
20954 must not be `VOIDmode' since these saves can be arbitrarily nested.
20956 A save area is a `mem' that is at a constant offset from
20957 `virtual_stack_vars_rtx' when the stack pointer is saved for use by
20958 nonlocal gotos and a `reg' in the other two cases.
20961 Subtract (or add if `STACK_GROWS_DOWNWARD' is undefined) operand 1
20962 from the stack pointer to create space for dynamically allocated
20965 Store the resultant pointer to this space into operand 0. If you
20966 are allocating space from the main stack, do this by emitting a
20967 move insn to copy `virtual_stack_dynamic_rtx' to operand 0. If
20968 you are allocating the space elsewhere, generate code to copy the
20969 location of the space to operand 0. In the latter case, you must
20970 ensure this space gets freed when the corresponding space on the
20971 main stack is free.
20973 Do not define this pattern if all that must be done is the
20974 subtraction. Some machines require other operations such as stack
20975 probes or maintaining the back chain. Define this pattern to emit
20976 those operations in addition to updating the stack pointer.
20979 If stack checking cannot be done on your system by probing the
20980 stack with a load or store instruction (*note Stack Checking::),
20981 define this pattern to perform the needed check and signaling an
20982 error if the stack has overflowed. The single operand is the
20983 location in the stack furthest from the current stack pointer that
20984 you need to validate. Normally, on machines where this pattern is
20985 needed, you would obtain the stack limit from a global or
20986 thread-specific variable or register.
20989 Emit code to generate a non-local goto, e.g., a jump from one
20990 function to a label in an outer function. This pattern has four
20991 arguments, each representing a value to be used in the jump. The
20992 first argument is to be loaded into the frame pointer, the second
20993 is the address to branch to (code to dispatch to the actual label),
20994 the third is the address of a location where the stack is saved,
20995 and the last is the address of the label, to be placed in the
20996 location for the incoming static chain.
20998 On most machines you need not define this pattern, since GCC will
20999 already generate the correct code, which is to load the frame
21000 pointer and static chain, restore the stack (using the
21001 `restore_stack_nonlocal' pattern, if defined), and jump indirectly
21002 to the dispatcher. You need only define this pattern if this code
21003 will not work on your machine.
21005 `nonlocal_goto_receiver'
21006 This pattern, if defined, contains code needed at the target of a
21007 nonlocal goto after the code already generated by GCC. You will
21008 not normally need to define this pattern. A typical reason why
21009 you might need this pattern is if some value, such as a pointer to
21010 a global table, must be restored when the frame pointer is
21011 restored. Note that a nonlocal goto only occurs within a
21012 unit-of-translation, so a global table pointer that is shared by
21013 all functions of a given module need not be restored. There are
21016 `exception_receiver'
21017 This pattern, if defined, contains code needed at the site of an
21018 exception handler that isn't needed at the site of a nonlocal
21019 goto. You will not normally need to define this pattern. A
21020 typical reason why you might need this pattern is if some value,
21021 such as a pointer to a global table, must be restored after
21022 control flow is branched to the handler of an exception. There
21025 `builtin_setjmp_setup'
21026 This pattern, if defined, contains additional code needed to
21027 initialize the `jmp_buf'. You will not normally need to define
21028 this pattern. A typical reason why you might need this pattern is
21029 if some value, such as a pointer to a global table, must be
21030 restored. Though it is preferred that the pointer value be
21031 recalculated if possible (given the address of a label for
21032 instance). The single argument is a pointer to the `jmp_buf'.
21033 Note that the buffer is five words long and that the first three
21034 are normally used by the generic mechanism.
21036 `builtin_setjmp_receiver'
21037 This pattern, if defined, contains code needed at the site of an
21038 built-in setjmp that isn't needed at the site of a nonlocal goto.
21039 You will not normally need to define this pattern. A typical
21040 reason why you might need this pattern is if some value, such as a
21041 pointer to a global table, must be restored. It takes one
21042 argument, which is the label to which builtin_longjmp transfered
21043 control; this pattern may be emitted at a small offset from that
21047 This pattern, if defined, performs the entire action of the
21048 longjmp. You will not normally need to define this pattern unless
21049 you also define `builtin_setjmp_setup'. The single argument is a
21050 pointer to the `jmp_buf'.
21053 This pattern, if defined, affects the way `__builtin_eh_return',
21054 and thence the call frame exception handling library routines, are
21055 built. It is intended to handle non-trivial actions needed along
21056 the abnormal return path.
21058 The address of the exception handler to which the function should
21059 return is passed as operand to this pattern. It will normally
21060 need to copied by the pattern to some special register or memory
21061 location. If the pattern needs to determine the location of the
21062 target call frame in order to do so, it may use
21063 `EH_RETURN_STACKADJ_RTX', if defined; it will have already been
21066 If this pattern is not defined, the default action will be to
21067 simply copy the return address to `EH_RETURN_HANDLER_RTX'. Either
21068 that macro or this pattern needs to be defined if call frame
21069 exception handling is to be used.
21072 This pattern, if defined, emits RTL for entry to a function. The
21073 function entry is responsible for setting up the stack frame,
21074 initializing the frame pointer register, saving callee saved
21077 Using a prologue pattern is generally preferred over defining
21078 `TARGET_ASM_FUNCTION_PROLOGUE' to emit assembly code for the
21081 The `prologue' pattern is particularly useful for targets which
21082 perform instruction scheduling.
21085 This pattern emits RTL for exit from a function. The function
21086 exit is responsible for deallocating the stack frame, restoring
21087 callee saved registers and emitting the return instruction.
21089 Using an epilogue pattern is generally preferred over defining
21090 `TARGET_ASM_FUNCTION_EPILOGUE' to emit assembly code for the
21093 The `epilogue' pattern is particularly useful for targets which
21094 perform instruction scheduling or which have delay slots for their
21095 return instruction.
21098 This pattern, if defined, emits RTL for exit from a function
21099 without the final branch back to the calling function. This
21100 pattern will be emitted before any sibling call (aka tail call)
21103 The `sibcall_epilogue' pattern must not clobber any arguments used
21104 for parameter passing or any stack slots for arguments passed to
21105 the current function.
21108 This pattern, if defined, signals an error, typically by causing
21109 some kind of signal to be raised. Among other places, it is used
21110 by the Java front end to signal `invalid array index' exceptions.
21113 Conditional trap instruction. Operand 0 is a piece of RTL which
21114 performs a comparison. Operand 1 is the trap code, an integer.
21116 A typical `conditional_trap' pattern looks like
21118 (define_insn "conditional_trap"
21119 [(trap_if (match_operator 0 "trap_operator"
21120 [(cc0) (const_int 0)])
21121 (match_operand 1 "const_int_operand" "i"))]
21126 This pattern, if defined, emits code for a non-faulting data
21127 prefetch instruction. Operand 0 is the address of the memory to
21128 prefetch. Operand 1 is a constant 1 if the prefetch is preparing
21129 for a write to the memory address, or a constant 0 otherwise.
21130 Operand 2 is the expected degree of temporal locality of the data
21131 and is a value between 0 and 3, inclusive; 0 means that the data
21132 has no temporal locality, so it need not be left in the cache
21133 after the access; 3 means that the data has a high degree of
21134 temporal locality and should be left in all levels of cache
21135 possible; 1 and 2 mean, respectively, a low or moderate degree of
21138 Targets that do not support write prefetches or locality hints can
21139 ignore the values of operands 1 and 2.
21142 This pattern defines a pseudo insn that prevents the instruction
21143 scheduler from moving instructions across the boundary defined by
21144 the blockage insn. Normally an UNSPEC_VOLATILE pattern.
21147 If the target memory model is not fully synchronous, then this
21148 pattern should be defined to an instruction that orders both loads
21149 and stores before the instruction with respect to loads and stores
21150 after the instruction. This pattern has no operands.
21152 `sync_compare_and_swapMODE'
21153 This pattern, if defined, emits code for an atomic compare-and-swap
21154 operation. Operand 1 is the memory on which the atomic operation
21155 is performed. Operand 2 is the "old" value to be compared against
21156 the current contents of the memory location. Operand 3 is the
21157 "new" value to store in the memory if the compare succeeds.
21158 Operand 0 is the result of the operation; it should contain the
21159 contents of the memory before the operation. If the compare
21160 succeeds, this should obviously be a copy of operand 2.
21162 This pattern must show that both operand 0 and operand 1 are
21165 This pattern must issue any memory barrier instructions such that
21166 all memory operations before the atomic operation occur before the
21167 atomic operation and all memory operations after the atomic
21168 operation occur after the atomic operation.
21170 `sync_compare_and_swap_ccMODE'
21171 This pattern is just like `sync_compare_and_swapMODE', except it
21172 should act as if compare part of the compare-and-swap were issued
21173 via `cmpM'. This comparison will only be used with `EQ' and `NE'
21174 branches and `setcc' operations.
21176 Some targets do expose the success or failure of the
21177 compare-and-swap operation via the status flags. Ideally we
21178 wouldn't need a separate named pattern in order to take advantage
21179 of this, but the combine pass does not handle patterns with
21180 multiple sets, which is required by definition for
21181 `sync_compare_and_swapMODE'.
21183 `sync_addMODE', `sync_subMODE'
21184 `sync_iorMODE', `sync_andMODE'
21185 `sync_xorMODE', `sync_nandMODE'
21186 These patterns emit code for an atomic operation on memory.
21187 Operand 0 is the memory on which the atomic operation is performed.
21188 Operand 1 is the second operand to the binary operator.
21190 The "nand" operation is `~op0 & op1'.
21192 This pattern must issue any memory barrier instructions such that
21193 all memory operations before the atomic operation occur before the
21194 atomic operation and all memory operations after the atomic
21195 operation occur after the atomic operation.
21197 If these patterns are not defined, the operation will be
21198 constructed from a compare-and-swap operation, if defined.
21200 `sync_old_addMODE', `sync_old_subMODE'
21201 `sync_old_iorMODE', `sync_old_andMODE'
21202 `sync_old_xorMODE', `sync_old_nandMODE'
21203 These patterns are emit code for an atomic operation on memory,
21204 and return the value that the memory contained before the
21205 operation. Operand 0 is the result value, operand 1 is the memory
21206 on which the atomic operation is performed, and operand 2 is the
21207 second operand to the binary operator.
21209 This pattern must issue any memory barrier instructions such that
21210 all memory operations before the atomic operation occur before the
21211 atomic operation and all memory operations after the atomic
21212 operation occur after the atomic operation.
21214 If these patterns are not defined, the operation will be
21215 constructed from a compare-and-swap operation, if defined.
21217 `sync_new_addMODE', `sync_new_subMODE'
21218 `sync_new_iorMODE', `sync_new_andMODE'
21219 `sync_new_xorMODE', `sync_new_nandMODE'
21220 These patterns are like their `sync_old_OP' counterparts, except
21221 that they return the value that exists in the memory location
21222 after the operation, rather than before the operation.
21224 `sync_lock_test_and_setMODE'
21225 This pattern takes two forms, based on the capabilities of the
21226 target. In either case, operand 0 is the result of the operand,
21227 operand 1 is the memory on which the atomic operation is
21228 performed, and operand 2 is the value to set in the lock.
21230 In the ideal case, this operation is an atomic exchange operation,
21231 in which the previous value in memory operand is copied into the
21232 result operand, and the value operand is stored in the memory
21235 For less capable targets, any value operand that is not the
21236 constant 1 should be rejected with `FAIL'. In this case the
21237 target may use an atomic test-and-set bit operation. The result
21238 operand should contain 1 if the bit was previously set and 0 if
21239 the bit was previously clear. The true contents of the memory
21240 operand are implementation defined.
21242 This pattern must issue any memory barrier instructions such that
21243 the pattern as a whole acts as an acquire barrier, that is all
21244 memory operations after the pattern do not occur until the lock is
21247 If this pattern is not defined, the operation will be constructed
21248 from a compare-and-swap operation, if defined.
21250 `sync_lock_releaseMODE'
21251 This pattern, if defined, releases a lock set by
21252 `sync_lock_test_and_setMODE'. Operand 0 is the memory that
21253 contains the lock; operand 1 is the value to store in the lock.
21255 If the target doesn't implement full semantics for
21256 `sync_lock_test_and_setMODE', any value operand which is not the
21257 constant 0 should be rejected with `FAIL', and the true contents
21258 of the memory operand are implementation defined.
21260 This pattern must issue any memory barrier instructions such that
21261 the pattern as a whole acts as a release barrier, that is the lock
21262 is released only after all previous memory operations have
21265 If this pattern is not defined, then a `memory_barrier' pattern
21266 will be emitted, followed by a store of the value to the memory
21269 `stack_protect_set'
21270 This pattern, if defined, moves a `Pmode' value from the memory in
21271 operand 1 to the memory in operand 0 without leaving the value in
21272 a register afterward. This is to avoid leaking the value some
21273 place that an attacker might use to rewrite the stack guard slot
21274 after having clobbered it.
21276 If this pattern is not defined, then a plain move pattern is
21279 `stack_protect_test'
21280 This pattern, if defined, compares a `Pmode' value from the memory
21281 in operand 1 with the memory in operand 0 without leaving the
21282 value in a register afterward and branches to operand 2 if the
21283 values weren't equal.
21285 If this pattern is not defined, then a plain compare pattern and
21286 conditional branch pattern is used.
21289 This pattern, if defined, flushes the instruction cache for a
21290 region of memory. The region is bounded to by the Pmode pointers
21291 in operand 0 inclusive and operand 1 exclusive.
21293 If this pattern is not defined, a call to the library function
21294 `__clear_cache' is used.
21298 File: gccint.info, Node: Pattern Ordering, Next: Dependent Patterns, Prev: Standard Names, Up: Machine Desc
21300 16.10 When the Order of Patterns Matters
21301 ========================================
21303 Sometimes an insn can match more than one instruction pattern. Then the
21304 pattern that appears first in the machine description is the one used.
21305 Therefore, more specific patterns (patterns that will match fewer
21306 things) and faster instructions (those that will produce better code
21307 when they do match) should usually go first in the description.
21309 In some cases the effect of ordering the patterns can be used to hide
21310 a pattern when it is not valid. For example, the 68000 has an
21311 instruction for converting a fullword to floating point and another for
21312 converting a byte to floating point. An instruction converting an
21313 integer to floating point could match either one. We put the pattern
21314 to convert the fullword first to make sure that one will be used rather
21315 than the other. (Otherwise a large integer might be generated as a
21316 single-byte immediate quantity, which would not work.) Instead of
21317 using this pattern ordering it would be possible to make the pattern
21318 for convert-a-byte smart enough to deal properly with any constant
21322 File: gccint.info, Node: Dependent Patterns, Next: Jump Patterns, Prev: Pattern Ordering, Up: Machine Desc
21324 16.11 Interdependence of Patterns
21325 =================================
21327 Every machine description must have a named pattern for each of the
21328 conditional branch names `bCOND'. The recognition template must always
21332 (if_then_else (COND (cc0) (const_int 0))
21333 (label_ref (match_operand 0 "" ""))
21336 In addition, every machine description must have an anonymous pattern
21337 for each of the possible reverse-conditional branches. Their templates
21341 (if_then_else (COND (cc0) (const_int 0))
21343 (label_ref (match_operand 0 "" ""))))
21345 They are necessary because jump optimization can turn direct-conditional
21346 branches into reverse-conditional branches.
21348 It is often convenient to use the `match_operator' construct to reduce
21349 the number of patterns that must be specified for branches. For
21354 (if_then_else (match_operator 0 "comparison_operator"
21355 [(cc0) (const_int 0)])
21357 (label_ref (match_operand 1 "" ""))))]
21361 In some cases machines support instructions identical except for the
21362 machine mode of one or more operands. For example, there may be
21363 "sign-extend halfword" and "sign-extend byte" instructions whose
21366 (set (match_operand:SI 0 ...)
21367 (extend:SI (match_operand:HI 1 ...)))
21369 (set (match_operand:SI 0 ...)
21370 (extend:SI (match_operand:QI 1 ...)))
21372 Constant integers do not specify a machine mode, so an instruction to
21373 extend a constant value could match either pattern. The pattern it
21374 actually will match is the one that appears first in the file. For
21375 correct results, this must be the one for the widest possible mode
21376 (`HImode', here). If the pattern matches the `QImode' instruction, the
21377 results will be incorrect if the constant value does not actually fit
21380 Such instructions to extend constants are rarely generated because
21381 they are optimized away, but they do occasionally happen in nonoptimized
21384 If a constraint in a pattern allows a constant, the reload pass may
21385 replace a register with a constant permitted by the constraint in some
21386 cases. Similarly for memory references. Because of this substitution,
21387 you should not provide separate patterns for increment and decrement
21388 instructions. Instead, they should be generated from the same pattern
21389 that supports register-register add insns by examining the operands and
21390 generating the appropriate machine instruction.
21393 File: gccint.info, Node: Jump Patterns, Next: Looping Patterns, Prev: Dependent Patterns, Up: Machine Desc
21395 16.12 Defining Jump Instruction Patterns
21396 ========================================
21398 For most machines, GCC assumes that the machine has a condition code.
21399 A comparison insn sets the condition code, recording the results of both
21400 signed and unsigned comparison of the given operands. A separate branch
21401 insn tests the condition code and branches or not according its value.
21402 The branch insns come in distinct signed and unsigned flavors. Many
21403 common machines, such as the VAX, the 68000 and the 32000, work this
21406 Some machines have distinct signed and unsigned compare instructions,
21407 and only one set of conditional branch instructions. The easiest way
21408 to handle these machines is to treat them just like the others until
21409 the final stage where assembly code is written. At this time, when
21410 outputting code for the compare instruction, peek ahead at the
21411 following branch using `next_cc0_user (insn)'. (The variable `insn'
21412 refers to the insn being output, in the output-writing code in an
21413 instruction pattern.) If the RTL says that is an unsigned branch,
21414 output an unsigned compare; otherwise output a signed compare. When
21415 the branch itself is output, you can treat signed and unsigned branches
21418 The reason you can do this is that GCC always generates a pair of
21419 consecutive RTL insns, possibly separated by `note' insns, one to set
21420 the condition code and one to test it, and keeps the pair inviolate
21423 To go with this technique, you must define the machine-description
21424 macro `NOTICE_UPDATE_CC' to do `CC_STATUS_INIT'; in other words, no
21425 compare instruction is superfluous.
21427 Some machines have compare-and-branch instructions and no condition
21428 code. A similar technique works for them. When it is time to "output"
21429 a compare instruction, record its operands in two static variables.
21430 When outputting the branch-on-condition-code instruction that follows,
21431 actually output a compare-and-branch instruction that uses the
21432 remembered operands.
21434 It also works to define patterns for compare-and-branch instructions.
21435 In optimizing compilation, the pair of compare and branch instructions
21436 will be combined according to these patterns. But this does not happen
21437 if optimization is not requested. So you must use one of the solutions
21438 above in addition to any special patterns you define.
21440 In many RISC machines, most instructions do not affect the condition
21441 code and there may not even be a separate condition code register. On
21442 these machines, the restriction that the definition and use of the
21443 condition code be adjacent insns is not necessary and can prevent
21444 important optimizations. For example, on the IBM RS/6000, there is a
21445 delay for taken branches unless the condition code register is set three
21446 instructions earlier than the conditional branch. The instruction
21447 scheduler cannot perform this optimization if it is not permitted to
21448 separate the definition and use of the condition code register.
21450 On these machines, do not use `(cc0)', but instead use a register to
21451 represent the condition code. If there is a specific condition code
21452 register in the machine, use a hard register. If the condition code or
21453 comparison result can be placed in any general register, or if there are
21454 multiple condition registers, use a pseudo register.
21456 On some machines, the type of branch instruction generated may depend
21457 on the way the condition code was produced; for example, on the 68k and
21458 SPARC, setting the condition code directly from an add or subtract
21459 instruction does not clear the overflow bit the way that a test
21460 instruction does, so a different branch instruction must be used for
21461 some conditional branches. For machines that use `(cc0)', the set and
21462 use of the condition code must be adjacent (separated only by `note'
21463 insns) allowing flags in `cc_status' to be used. (*Note Condition
21464 Code::.) Also, the comparison and branch insns can be located from
21465 each other by using the functions `prev_cc0_setter' and `next_cc0_user'.
21467 However, this is not true on machines that do not use `(cc0)'. On
21468 those machines, no assumptions can be made about the adjacency of the
21469 compare and branch insns and the above methods cannot be used. Instead,
21470 we use the machine mode of the condition code register to record
21471 different formats of the condition code register.
21473 Registers used to store the condition code value should have a mode
21474 that is in class `MODE_CC'. Normally, it will be `CCmode'. If
21475 additional modes are required (as for the add example mentioned above in
21476 the SPARC), define them in `MACHINE-modes.def' (*note Condition
21477 Code::). Also define `SELECT_CC_MODE' to choose a mode given an
21478 operand of a compare.
21480 If it is known during RTL generation that a different mode will be
21481 required (for example, if the machine has separate compare instructions
21482 for signed and unsigned quantities, like most IBM processors), they can
21483 be specified at that time.
21485 If the cases that require different modes would be made by instruction
21486 combination, the macro `SELECT_CC_MODE' determines which machine mode
21487 should be used for the comparison result. The patterns should be
21488 written using that mode. To support the case of the add on the SPARC
21489 discussed above, we have the pattern
21492 [(set (reg:CC_NOOV 0)
21494 (plus:SI (match_operand:SI 0 "register_operand" "%r")
21495 (match_operand:SI 1 "arith_operand" "rI"))
21500 The `SELECT_CC_MODE' macro on the SPARC returns `CC_NOOVmode' for
21501 comparisons whose argument is a `plus'.
21504 File: gccint.info, Node: Looping Patterns, Next: Insn Canonicalizations, Prev: Jump Patterns, Up: Machine Desc
21506 16.13 Defining Looping Instruction Patterns
21507 ===========================================
21509 Some machines have special jump instructions that can be utilized to
21510 make loops more efficient. A common example is the 68000 `dbra'
21511 instruction which performs a decrement of a register and a branch if the
21512 result was greater than zero. Other machines, in particular digital
21513 signal processors (DSPs), have special block repeat instructions to
21514 provide low-overhead loop support. For example, the TI TMS320C3x/C4x
21515 DSPs have a block repeat instruction that loads special registers to
21516 mark the top and end of a loop and to count the number of loop
21517 iterations. This avoids the need for fetching and executing a
21518 `dbra'-like instruction and avoids pipeline stalls associated with the
21521 GCC has three special named patterns to support low overhead looping.
21522 They are `decrement_and_branch_until_zero', `doloop_begin', and
21523 `doloop_end'. The first pattern, `decrement_and_branch_until_zero', is
21524 not emitted during RTL generation but may be emitted during the
21525 instruction combination phase. This requires the assistance of the
21526 loop optimizer, using information collected during strength reduction,
21527 to reverse a loop to count down to zero. Some targets also require the
21528 loop optimizer to add a `REG_NONNEG' note to indicate that the
21529 iteration count is always positive. This is needed if the target
21530 performs a signed loop termination test. For example, the 68000 uses a
21531 pattern similar to the following for its `dbra' instruction:
21533 (define_insn "decrement_and_branch_until_zero"
21536 (ge (plus:SI (match_operand:SI 0 "general_operand" "+d*am")
21539 (label_ref (match_operand 1 "" ""))
21542 (plus:SI (match_dup 0)
21544 "find_reg_note (insn, REG_NONNEG, 0)"
21547 Note that since the insn is both a jump insn and has an output, it must
21548 deal with its own reloads, hence the `m' constraints. Also note that
21549 since this insn is generated by the instruction combination phase
21550 combining two sequential insns together into an implicit parallel insn,
21551 the iteration counter needs to be biased by the same amount as the
21552 decrement operation, in this case -1. Note that the following similar
21553 pattern will not be matched by the combiner.
21555 (define_insn "decrement_and_branch_until_zero"
21558 (ge (match_operand:SI 0 "general_operand" "+d*am")
21560 (label_ref (match_operand 1 "" ""))
21563 (plus:SI (match_dup 0)
21565 "find_reg_note (insn, REG_NONNEG, 0)"
21568 The other two special looping patterns, `doloop_begin' and
21569 `doloop_end', are emitted by the loop optimizer for certain
21570 well-behaved loops with a finite number of loop iterations using
21571 information collected during strength reduction.
21573 The `doloop_end' pattern describes the actual looping instruction (or
21574 the implicit looping operation) and the `doloop_begin' pattern is an
21575 optional companion pattern that can be used for initialization needed
21576 for some low-overhead looping instructions.
21578 Note that some machines require the actual looping instruction to be
21579 emitted at the top of the loop (e.g., the TMS320C3x/C4x DSPs). Emitting
21580 the true RTL for a looping instruction at the top of the loop can cause
21581 problems with flow analysis. So instead, a dummy `doloop' insn is
21582 emitted at the end of the loop. The machine dependent reorg pass checks
21583 for the presence of this `doloop' insn and then searches back to the
21584 top of the loop, where it inserts the true looping insn (provided there
21585 are no instructions in the loop which would cause problems). Any
21586 additional labels can be emitted at this point. In addition, if the
21587 desired special iteration counter register was not allocated, this
21588 machine dependent reorg pass could emit a traditional compare and jump
21591 The essential difference between the `decrement_and_branch_until_zero'
21592 and the `doloop_end' patterns is that the loop optimizer allocates an
21593 additional pseudo register for the latter as an iteration counter.
21594 This pseudo register cannot be used within the loop (i.e., general
21595 induction variables cannot be derived from it), however, in many cases
21596 the loop induction variable may become redundant and removed by the
21600 File: gccint.info, Node: Insn Canonicalizations, Next: Expander Definitions, Prev: Looping Patterns, Up: Machine Desc
21602 16.14 Canonicalization of Instructions
21603 ======================================
21605 There are often cases where multiple RTL expressions could represent an
21606 operation performed by a single machine instruction. This situation is
21607 most commonly encountered with logical, branch, and multiply-accumulate
21608 instructions. In such cases, the compiler attempts to convert these
21609 multiple RTL expressions into a single canonical form to reduce the
21610 number of insn patterns required.
21612 In addition to algebraic simplifications, following canonicalizations
21615 * For commutative and comparison operators, a constant is always
21616 made the second operand. If a machine only supports a constant as
21617 the second operand, only patterns that match a constant in the
21618 second operand need be supplied.
21620 * For associative operators, a sequence of operators will always
21621 chain to the left; for instance, only the left operand of an
21622 integer `plus' can itself be a `plus'. `and', `ior', `xor',
21623 `plus', `mult', `smin', `smax', `umin', and `umax' are associative
21624 when applied to integers, and sometimes to floating-point.
21626 * For these operators, if only one operand is a `neg', `not',
21627 `mult', `plus', or `minus' expression, it will be the first
21630 * In combinations of `neg', `mult', `plus', and `minus', the `neg'
21631 operations (if any) will be moved inside the operations as far as
21632 possible. For instance, `(neg (mult A B))' is canonicalized as
21633 `(mult (neg A) B)', but `(plus (mult (neg A) B) C)' is
21634 canonicalized as `(minus A (mult B C))'.
21636 * For the `compare' operator, a constant is always the second operand
21637 on machines where `cc0' is used (*note Jump Patterns::). On other
21638 machines, there are rare cases where the compiler might want to
21639 construct a `compare' with a constant as the first operand.
21640 However, these cases are not common enough for it to be worthwhile
21641 to provide a pattern matching a constant as the first operand
21642 unless the machine actually has such an instruction.
21644 An operand of `neg', `not', `mult', `plus', or `minus' is made the
21645 first operand under the same conditions as above.
21647 * `(ltu (plus A B) B)' is converted to `(ltu (plus A B) A)'.
21648 Likewise with `geu' instead of `ltu'.
21650 * `(minus X (const_int N))' is converted to `(plus X (const_int
21653 * Within address computations (i.e., inside `mem'), a left shift is
21654 converted into the appropriate multiplication by a power of two.
21656 * De Morgan's Law is used to move bitwise negation inside a bitwise
21657 logical-and or logical-or operation. If this results in only one
21658 operand being a `not' expression, it will be the first one.
21660 A machine that has an instruction that performs a bitwise
21661 logical-and of one operand with the bitwise negation of the other
21662 should specify the pattern for that instruction as
21665 [(set (match_operand:M 0 ...)
21666 (and:M (not:M (match_operand:M 1 ...))
21667 (match_operand:M 2 ...)))]
21671 Similarly, a pattern for a "NAND" instruction should be written
21674 [(set (match_operand:M 0 ...)
21675 (ior:M (not:M (match_operand:M 1 ...))
21676 (not:M (match_operand:M 2 ...))))]
21680 In both cases, it is not necessary to include patterns for the many
21681 logically equivalent RTL expressions.
21683 * The only possible RTL expressions involving both bitwise
21684 exclusive-or and bitwise negation are `(xor:M X Y)' and `(not:M
21687 * The sum of three items, one of which is a constant, will only
21690 (plus:M (plus:M X Y) CONSTANT)
21692 * On machines that do not use `cc0', `(compare X (const_int 0))'
21693 will be converted to X.
21695 * Equality comparisons of a group of bits (usually a single bit)
21696 with zero will be written using `zero_extract' rather than the
21697 equivalent `and' or `sign_extract' operations.
21700 Further canonicalization rules are defined in the function
21701 `commutative_operand_precedence' in `gcc/rtlanal.c'.
21704 File: gccint.info, Node: Expander Definitions, Next: Insn Splitting, Prev: Insn Canonicalizations, Up: Machine Desc
21706 16.15 Defining RTL Sequences for Code Generation
21707 ================================================
21709 On some target machines, some standard pattern names for RTL generation
21710 cannot be handled with single insn, but a sequence of RTL insns can
21711 represent them. For these target machines, you can write a
21712 `define_expand' to specify how to generate the sequence of RTL.
21714 A `define_expand' is an RTL expression that looks almost like a
21715 `define_insn'; but, unlike the latter, a `define_expand' is used only
21716 for RTL generation and it can produce more than one RTL insn.
21718 A `define_expand' RTX has four operands:
21720 * The name. Each `define_expand' must have a name, since the only
21721 use for it is to refer to it by name.
21723 * The RTL template. This is a vector of RTL expressions representing
21724 a sequence of separate instructions. Unlike `define_insn', there
21725 is no implicit surrounding `PARALLEL'.
21727 * The condition, a string containing a C expression. This
21728 expression is used to express how the availability of this pattern
21729 depends on subclasses of target machine, selected by command-line
21730 options when GCC is run. This is just like the condition of a
21731 `define_insn' that has a standard name. Therefore, the condition
21732 (if present) may not depend on the data in the insn being matched,
21733 but only the target-machine-type flags. The compiler needs to
21734 test these conditions during initialization in order to learn
21735 exactly which named instructions are available in a particular run.
21737 * The preparation statements, a string containing zero or more C
21738 statements which are to be executed before RTL code is generated
21739 from the RTL template.
21741 Usually these statements prepare temporary registers for use as
21742 internal operands in the RTL template, but they can also generate
21743 RTL insns directly by calling routines such as `emit_insn', etc.
21744 Any such insns precede the ones that come from the RTL template.
21746 Every RTL insn emitted by a `define_expand' must match some
21747 `define_insn' in the machine description. Otherwise, the compiler will
21748 crash when trying to generate code for the insn or trying to optimize
21751 The RTL template, in addition to controlling generation of RTL insns,
21752 also describes the operands that need to be specified when this pattern
21753 is used. In particular, it gives a predicate for each operand.
21755 A true operand, which needs to be specified in order to generate RTL
21756 from the pattern, should be described with a `match_operand' in its
21757 first occurrence in the RTL template. This enters information on the
21758 operand's predicate into the tables that record such things. GCC uses
21759 the information to preload the operand into a register if that is
21760 required for valid RTL code. If the operand is referred to more than
21761 once, subsequent references should use `match_dup'.
21763 The RTL template may also refer to internal "operands" which are
21764 temporary registers or labels used only within the sequence made by the
21765 `define_expand'. Internal operands are substituted into the RTL
21766 template with `match_dup', never with `match_operand'. The values of
21767 the internal operands are not passed in as arguments by the compiler
21768 when it requests use of this pattern. Instead, they are computed
21769 within the pattern, in the preparation statements. These statements
21770 compute the values and store them into the appropriate elements of
21771 `operands' so that `match_dup' can find them.
21773 There are two special macros defined for use in the preparation
21774 statements: `DONE' and `FAIL'. Use them with a following semicolon, as
21778 Use the `DONE' macro to end RTL generation for the pattern. The
21779 only RTL insns resulting from the pattern on this occasion will be
21780 those already emitted by explicit calls to `emit_insn' within the
21781 preparation statements; the RTL template will not be generated.
21784 Make the pattern fail on this occasion. When a pattern fails, it
21785 means that the pattern was not truly available. The calling
21786 routines in the compiler will try other strategies for code
21787 generation using other patterns.
21789 Failure is currently supported only for binary (addition,
21790 multiplication, shifting, etc.) and bit-field (`extv', `extzv',
21791 and `insv') operations.
21793 If the preparation falls through (invokes neither `DONE' nor `FAIL'),
21794 then the `define_expand' acts like a `define_insn' in that the RTL
21795 template is used to generate the insn.
21797 The RTL template is not used for matching, only for generating the
21798 initial insn list. If the preparation statement always invokes `DONE'
21799 or `FAIL', the RTL template may be reduced to a simple list of
21800 operands, such as this example:
21802 (define_expand "addsi3"
21803 [(match_operand:SI 0 "register_operand" "")
21804 (match_operand:SI 1 "register_operand" "")
21805 (match_operand:SI 2 "register_operand" "")]
21809 handle_add (operands[0], operands[1], operands[2]);
21813 Here is an example, the definition of left-shift for the SPUR chip:
21815 (define_expand "ashlsi3"
21816 [(set (match_operand:SI 0 "register_operand" "")
21818 (match_operand:SI 1 "register_operand" "")
21819 (match_operand:SI 2 "nonmemory_operand" "")))]
21824 if (GET_CODE (operands[2]) != CONST_INT
21825 || (unsigned) INTVAL (operands[2]) > 3)
21829 This example uses `define_expand' so that it can generate an RTL insn
21830 for shifting when the shift-count is in the supported range of 0 to 3
21831 but fail in other cases where machine insns aren't available. When it
21832 fails, the compiler tries another strategy using different patterns
21833 (such as, a library call).
21835 If the compiler were able to handle nontrivial condition-strings in
21836 patterns with names, then it would be possible to use a `define_insn'
21837 in that case. Here is another case (zero-extension on the 68000) which
21838 makes more use of the power of `define_expand':
21840 (define_expand "zero_extendhisi2"
21841 [(set (match_operand:SI 0 "general_operand" "")
21843 (set (strict_low_part
21847 (match_operand:HI 1 "general_operand" ""))]
21849 "operands[1] = make_safe_from (operands[1], operands[0]);")
21851 Here two RTL insns are generated, one to clear the entire output operand
21852 and the other to copy the input operand into its low half. This
21853 sequence is incorrect if the input operand refers to [the old value of]
21854 the output operand, so the preparation statement makes sure this isn't
21855 so. The function `make_safe_from' copies the `operands[1]' into a
21856 temporary register if it refers to `operands[0]'. It does this by
21857 emitting another RTL insn.
21859 Finally, a third example shows the use of an internal operand.
21860 Zero-extension on the SPUR chip is done by `and'-ing the result against
21861 a halfword mask. But this mask cannot be represented by a `const_int'
21862 because the constant value is too large to be legitimate on this
21863 machine. So it must be copied into a register with `force_reg' and
21864 then the register used in the `and'.
21866 (define_expand "zero_extendhisi2"
21867 [(set (match_operand:SI 0 "register_operand" "")
21869 (match_operand:HI 1 "register_operand" "")
21874 = force_reg (SImode, GEN_INT (65535)); ")
21876 _Note:_ If the `define_expand' is used to serve a standard binary or
21877 unary arithmetic operation or a bit-field operation, then the last insn
21878 it generates must not be a `code_label', `barrier' or `note'. It must
21879 be an `insn', `jump_insn' or `call_insn'. If you don't need a real insn
21880 at the end, emit an insn to copy the result of the operation into
21881 itself. Such an insn will generate no code, but it can avoid problems
21885 File: gccint.info, Node: Insn Splitting, Next: Including Patterns, Prev: Expander Definitions, Up: Machine Desc
21887 16.16 Defining How to Split Instructions
21888 ========================================
21890 There are two cases where you should specify how to split a pattern
21891 into multiple insns. On machines that have instructions requiring
21892 delay slots (*note Delay Slots::) or that have instructions whose
21893 output is not available for multiple cycles (*note Processor pipeline
21894 description::), the compiler phases that optimize these cases need to
21895 be able to move insns into one-instruction delay slots. However, some
21896 insns may generate more than one machine instruction. These insns
21897 cannot be placed into a delay slot.
21899 Often you can rewrite the single insn as a list of individual insns,
21900 each corresponding to one machine instruction. The disadvantage of
21901 doing so is that it will cause the compilation to be slower and require
21902 more space. If the resulting insns are too complex, it may also
21903 suppress some optimizations. The compiler splits the insn if there is a
21904 reason to believe that it might improve instruction or delay slot
21907 The insn combiner phase also splits putative insns. If three insns are
21908 merged into one insn with a complex expression that cannot be matched by
21909 some `define_insn' pattern, the combiner phase attempts to split the
21910 complex pattern into two insns that are recognized. Usually it can
21911 break the complex pattern into two patterns by splitting out some
21912 subexpression. However, in some other cases, such as performing an
21913 addition of a large constant in two insns on a RISC machine, the way to
21914 split the addition into two insns is machine-dependent.
21916 The `define_split' definition tells the compiler how to split a
21917 complex insn into several simpler insns. It looks like this:
21922 [NEW-INSN-PATTERN-1
21925 "PREPARATION-STATEMENTS")
21927 INSN-PATTERN is a pattern that needs to be split and CONDITION is the
21928 final condition to be tested, as in a `define_insn'. When an insn
21929 matching INSN-PATTERN and satisfying CONDITION is found, it is replaced
21930 in the insn list with the insns given by NEW-INSN-PATTERN-1,
21931 NEW-INSN-PATTERN-2, etc.
21933 The PREPARATION-STATEMENTS are similar to those statements that are
21934 specified for `define_expand' (*note Expander Definitions::) and are
21935 executed before the new RTL is generated to prepare for the generated
21936 code or emit some insns whose pattern is not fixed. Unlike those in
21937 `define_expand', however, these statements must not generate any new
21938 pseudo-registers. Once reload has completed, they also must not
21939 allocate any space in the stack frame.
21941 Patterns are matched against INSN-PATTERN in two different
21942 circumstances. If an insn needs to be split for delay slot scheduling
21943 or insn scheduling, the insn is already known to be valid, which means
21944 that it must have been matched by some `define_insn' and, if
21945 `reload_completed' is nonzero, is known to satisfy the constraints of
21946 that `define_insn'. In that case, the new insn patterns must also be
21947 insns that are matched by some `define_insn' and, if `reload_completed'
21948 is nonzero, must also satisfy the constraints of those definitions.
21950 As an example of this usage of `define_split', consider the following
21951 example from `a29k.md', which splits a `sign_extend' from `HImode' to
21952 `SImode' into a pair of shift insns:
21955 [(set (match_operand:SI 0 "gen_reg_operand" "")
21956 (sign_extend:SI (match_operand:HI 1 "gen_reg_operand" "")))]
21958 [(set (match_dup 0)
21959 (ashift:SI (match_dup 1)
21962 (ashiftrt:SI (match_dup 0)
21965 { operands[1] = gen_lowpart (SImode, operands[1]); }")
21967 When the combiner phase tries to split an insn pattern, it is always
21968 the case that the pattern is _not_ matched by any `define_insn'. The
21969 combiner pass first tries to split a single `set' expression and then
21970 the same `set' expression inside a `parallel', but followed by a
21971 `clobber' of a pseudo-reg to use as a scratch register. In these
21972 cases, the combiner expects exactly two new insn patterns to be
21973 generated. It will verify that these patterns match some `define_insn'
21974 definitions, so you need not do this test in the `define_split' (of
21975 course, there is no point in writing a `define_split' that will never
21976 produce insns that match).
21978 Here is an example of this use of `define_split', taken from
21982 [(set (match_operand:SI 0 "gen_reg_operand" "")
21983 (plus:SI (match_operand:SI 1 "gen_reg_operand" "")
21984 (match_operand:SI 2 "non_add_cint_operand" "")))]
21986 [(set (match_dup 0) (plus:SI (match_dup 1) (match_dup 3)))
21987 (set (match_dup 0) (plus:SI (match_dup 0) (match_dup 4)))]
21990 int low = INTVAL (operands[2]) & 0xffff;
21991 int high = (unsigned) INTVAL (operands[2]) >> 16;
21994 high++, low |= 0xffff0000;
21996 operands[3] = GEN_INT (high << 16);
21997 operands[4] = GEN_INT (low);
22000 Here the predicate `non_add_cint_operand' matches any `const_int' that
22001 is _not_ a valid operand of a single add insn. The add with the
22002 smaller displacement is written so that it can be substituted into the
22003 address of a subsequent operation.
22005 An example that uses a scratch register, from the same file, generates
22006 an equality comparison of a register and a large constant:
22009 [(set (match_operand:CC 0 "cc_reg_operand" "")
22010 (compare:CC (match_operand:SI 1 "gen_reg_operand" "")
22011 (match_operand:SI 2 "non_short_cint_operand" "")))
22012 (clobber (match_operand:SI 3 "gen_reg_operand" ""))]
22013 "find_single_use (operands[0], insn, 0)
22014 && (GET_CODE (*find_single_use (operands[0], insn, 0)) == EQ
22015 || GET_CODE (*find_single_use (operands[0], insn, 0)) == NE)"
22016 [(set (match_dup 3) (xor:SI (match_dup 1) (match_dup 4)))
22017 (set (match_dup 0) (compare:CC (match_dup 3) (match_dup 5)))]
22020 /* Get the constant we are comparing against, C, and see what it
22021 looks like sign-extended to 16 bits. Then see what constant
22022 could be XOR'ed with C to get the sign-extended value. */
22024 int c = INTVAL (operands[2]);
22025 int sextc = (c << 16) >> 16;
22026 int xorv = c ^ sextc;
22028 operands[4] = GEN_INT (xorv);
22029 operands[5] = GEN_INT (sextc);
22032 To avoid confusion, don't write a single `define_split' that accepts
22033 some insns that match some `define_insn' as well as some insns that
22034 don't. Instead, write two separate `define_split' definitions, one for
22035 the insns that are valid and one for the insns that are not valid.
22037 The splitter is allowed to split jump instructions into sequence of
22038 jumps or create new jumps in while splitting non-jump instructions. As
22039 the central flowgraph and branch prediction information needs to be
22040 updated, several restriction apply.
22042 Splitting of jump instruction into sequence that over by another jump
22043 instruction is always valid, as compiler expect identical behavior of
22044 new jump. When new sequence contains multiple jump instructions or new
22045 labels, more assistance is needed. Splitter is required to create only
22046 unconditional jumps, or simple conditional jump instructions.
22047 Additionally it must attach a `REG_BR_PROB' note to each conditional
22048 jump. A global variable `split_branch_probability' holds the
22049 probability of the original branch in case it was an simple conditional
22050 jump, -1 otherwise. To simplify recomputing of edge frequencies, the
22051 new sequence is required to have only forward jumps to the newly
22054 For the common case where the pattern of a define_split exactly
22055 matches the pattern of a define_insn, use `define_insn_and_split'. It
22058 (define_insn_and_split
22063 [NEW-INSN-PATTERN-1
22066 "PREPARATION-STATEMENTS"
22069 INSN-PATTERN, CONDITION, OUTPUT-TEMPLATE, and INSN-ATTRIBUTES are used
22070 as in `define_insn'. The NEW-INSN-PATTERN vector and the
22071 PREPARATION-STATEMENTS are used as in a `define_split'. The
22072 SPLIT-CONDITION is also used as in `define_split', with the additional
22073 behavior that if the condition starts with `&&', the condition used for
22074 the split will be the constructed as a logical "and" of the split
22075 condition with the insn condition. For example, from i386.md:
22077 (define_insn_and_split "zero_extendhisi2_and"
22078 [(set (match_operand:SI 0 "register_operand" "=r")
22079 (zero_extend:SI (match_operand:HI 1 "register_operand" "0")))
22080 (clobber (reg:CC 17))]
22081 "TARGET_ZERO_EXTEND_WITH_AND && !optimize_size"
22083 "&& reload_completed"
22084 [(parallel [(set (match_dup 0)
22085 (and:SI (match_dup 0) (const_int 65535)))
22086 (clobber (reg:CC 17))])]
22088 [(set_attr "type" "alu1")])
22090 In this case, the actual split condition will be
22091 `TARGET_ZERO_EXTEND_WITH_AND && !optimize_size && reload_completed'.
22093 The `define_insn_and_split' construction provides exactly the same
22094 functionality as two separate `define_insn' and `define_split'
22095 patterns. It exists for compactness, and as a maintenance tool to
22096 prevent having to ensure the two patterns' templates match.
22099 File: gccint.info, Node: Including Patterns, Next: Peephole Definitions, Prev: Insn Splitting, Up: Machine Desc
22101 16.17 Including Patterns in Machine Descriptions.
22102 =================================================
22104 The `include' pattern tells the compiler tools where to look for
22105 patterns that are in files other than in the file `.md'. This is used
22106 only at build time and there is no preprocessing allowed.
22117 (include "filestuff")
22119 Where PATHNAME is a string that specifies the location of the file,
22120 specifies the include file to be in `gcc/config/target/filestuff'. The
22121 directory `gcc/config/target' is regarded as the default directory.
22123 Machine descriptions may be split up into smaller more manageable
22124 subsections and placed into subdirectories.
22129 (include "BOGUS/filestuff")
22131 the include file is specified to be in
22132 `gcc/config/TARGET/BOGUS/filestuff'.
22134 Specifying an absolute path for the include file such as;
22136 (include "/u2/BOGUS/filestuff")
22137 is permitted but is not encouraged.
22139 16.17.1 RTL Generation Tool Options for Directory Search
22140 --------------------------------------------------------
22142 The `-IDIR' option specifies directories to search for machine
22143 descriptions. For example:
22146 genrecog -I/p1/abc/proc1 -I/p2/abcd/pro2 target.md
22148 Add the directory DIR to the head of the list of directories to be
22149 searched for header files. This can be used to override a system
22150 machine definition file, substituting your own version, since these
22151 directories are searched before the default machine description file
22152 directories. If you use more than one `-I' option, the directories are
22153 scanned in left-to-right order; the standard default directory come
22157 File: gccint.info, Node: Peephole Definitions, Next: Insn Attributes, Prev: Including Patterns, Up: Machine Desc
22159 16.18 Machine-Specific Peephole Optimizers
22160 ==========================================
22162 In addition to instruction patterns the `md' file may contain
22163 definitions of machine-specific peephole optimizations.
22165 The combiner does not notice certain peephole optimizations when the
22166 data flow in the program does not suggest that it should try them. For
22167 example, sometimes two consecutive insns related in purpose can be
22168 combined even though the second one does not appear to use a register
22169 computed in the first one. A machine-specific peephole optimizer can
22170 detect such opportunities.
22172 There are two forms of peephole definitions that may be used. The
22173 original `define_peephole' is run at assembly output time to match
22174 insns and substitute assembly text. Use of `define_peephole' is
22177 A newer `define_peephole2' matches insns and substitutes new insns.
22178 The `peephole2' pass is run after register allocation but before
22179 scheduling, which may result in much better code for targets that do
22184 * define_peephole:: RTL to Text Peephole Optimizers
22185 * define_peephole2:: RTL to RTL Peephole Optimizers
22188 File: gccint.info, Node: define_peephole, Next: define_peephole2, Up: Peephole Definitions
22190 16.18.1 RTL to Text Peephole Optimizers
22191 ---------------------------------------
22193 A definition looks like this:
22201 "OPTIONAL-INSN-ATTRIBUTES")
22203 The last string operand may be omitted if you are not using any
22204 machine-specific information in this machine description. If present,
22205 it must obey the same rules as in a `define_insn'.
22207 In this skeleton, INSN-PATTERN-1 and so on are patterns to match
22208 consecutive insns. The optimization applies to a sequence of insns when
22209 INSN-PATTERN-1 matches the first one, INSN-PATTERN-2 matches the next,
22212 Each of the insns matched by a peephole must also match a
22213 `define_insn'. Peepholes are checked only at the last stage just
22214 before code generation, and only optionally. Therefore, any insn which
22215 would match a peephole but no `define_insn' will cause a crash in code
22216 generation in an unoptimized compilation, or at various optimization
22219 The operands of the insns are matched with `match_operands',
22220 `match_operator', and `match_dup', as usual. What is not usual is that
22221 the operand numbers apply to all the insn patterns in the definition.
22222 So, you can check for identical operands in two insns by using
22223 `match_operand' in one insn and `match_dup' in the other.
22225 The operand constraints used in `match_operand' patterns do not have
22226 any direct effect on the applicability of the peephole, but they will
22227 be validated afterward, so make sure your constraints are general enough
22228 to apply whenever the peephole matches. If the peephole matches but
22229 the constraints are not satisfied, the compiler will crash.
22231 It is safe to omit constraints in all the operands of the peephole; or
22232 you can write constraints which serve as a double-check on the criteria
22235 Once a sequence of insns matches the patterns, the CONDITION is
22236 checked. This is a C expression which makes the final decision whether
22237 to perform the optimization (we do so if the expression is nonzero). If
22238 CONDITION is omitted (in other words, the string is empty) then the
22239 optimization is applied to every sequence of insns that matches the
22242 The defined peephole optimizations are applied after register
22243 allocation is complete. Therefore, the peephole definition can check
22244 which operands have ended up in which kinds of registers, just by
22245 looking at the operands.
22247 The way to refer to the operands in CONDITION is to write
22248 `operands[I]' for operand number I (as matched by `(match_operand I
22249 ...)'). Use the variable `insn' to refer to the last of the insns
22250 being matched; use `prev_active_insn' to find the preceding insns.
22252 When optimizing computations with intermediate results, you can use
22253 CONDITION to match only when the intermediate results are not used
22254 elsewhere. Use the C expression `dead_or_set_p (INSN, OP)', where INSN
22255 is the insn in which you expect the value to be used for the last time
22256 (from the value of `insn', together with use of `prev_nonnote_insn'),
22257 and OP is the intermediate value (from `operands[I]').
22259 Applying the optimization means replacing the sequence of insns with
22260 one new insn. The TEMPLATE controls ultimate output of assembler code
22261 for this combined insn. It works exactly like the template of a
22262 `define_insn'. Operand numbers in this template are the same ones used
22263 in matching the original sequence of insns.
22265 The result of a defined peephole optimizer does not need to match any
22266 of the insn patterns in the machine description; it does not even have
22267 an opportunity to match them. The peephole optimizer definition itself
22268 serves as the insn pattern to control how the insn is output.
22270 Defined peephole optimizers are run as assembler code is being output,
22271 so the insns they produce are never combined or rearranged in any way.
22273 Here is an example, taken from the 68000 machine description:
22276 [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
22277 (set (match_operand:DF 0 "register_operand" "=f")
22278 (match_operand:DF 1 "register_operand" "ad"))]
22279 "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
22282 xoperands[1] = gen_rtx_REG (SImode, REGNO (operands[1]) + 1);
22284 output_asm_insn ("move.l %1,(sp)", xoperands);
22285 output_asm_insn ("move.l %1,-(sp)", operands);
22286 return "fmove.d (sp)+,%0";
22288 output_asm_insn ("movel %1,sp@", xoperands);
22289 output_asm_insn ("movel %1,sp@-", operands);
22290 return "fmoved sp@+,%0";
22294 The effect of this optimization is to change
22309 INSN-PATTERN-1 and so on look _almost_ like the second operand of
22310 `define_insn'. There is one important difference: the second operand
22311 of `define_insn' consists of one or more RTX's enclosed in square
22312 brackets. Usually, there is only one: then the same action can be
22313 written as an element of a `define_peephole'. But when there are
22314 multiple actions in a `define_insn', they are implicitly enclosed in a
22315 `parallel'. Then you must explicitly write the `parallel', and the
22316 square brackets within it, in the `define_peephole'. Thus, if an insn
22317 pattern looks like this,
22319 (define_insn "divmodsi4"
22320 [(set (match_operand:SI 0 "general_operand" "=d")
22321 (div:SI (match_operand:SI 1 "general_operand" "0")
22322 (match_operand:SI 2 "general_operand" "dmsK")))
22323 (set (match_operand:SI 3 "general_operand" "=d")
22324 (mod:SI (match_dup 1) (match_dup 2)))]
22326 "divsl%.l %2,%3:%0")
22328 then the way to mention this insn in a peephole is as follows:
22333 [(set (match_operand:SI 0 "general_operand" "=d")
22334 (div:SI (match_operand:SI 1 "general_operand" "0")
22335 (match_operand:SI 2 "general_operand" "dmsK")))
22336 (set (match_operand:SI 3 "general_operand" "=d")
22337 (mod:SI (match_dup 1) (match_dup 2)))])
22342 File: gccint.info, Node: define_peephole2, Prev: define_peephole, Up: Peephole Definitions
22344 16.18.2 RTL to RTL Peephole Optimizers
22345 --------------------------------------
22347 The `define_peephole2' definition tells the compiler how to substitute
22348 one sequence of instructions for another sequence, what additional
22349 scratch registers may be needed and what their lifetimes must be.
22356 [NEW-INSN-PATTERN-1
22359 "PREPARATION-STATEMENTS")
22361 The definition is almost identical to `define_split' (*note Insn
22362 Splitting::) except that the pattern to match is not a single
22363 instruction, but a sequence of instructions.
22365 It is possible to request additional scratch registers for use in the
22366 output template. If appropriate registers are not free, the pattern
22367 will simply not match.
22369 Scratch registers are requested with a `match_scratch' pattern at the
22370 top level of the input pattern. The allocated register (initially) will
22371 be dead at the point requested within the original sequence. If the
22372 scratch is used at more than a single point, a `match_dup' pattern at
22373 the top level of the input pattern marks the last position in the input
22374 sequence at which the register must be available.
22376 Here is an example from the IA-32 machine description:
22379 [(match_scratch:SI 2 "r")
22380 (parallel [(set (match_operand:SI 0 "register_operand" "")
22381 (match_operator:SI 3 "arith_or_logical_operator"
22383 (match_operand:SI 1 "memory_operand" "")]))
22384 (clobber (reg:CC 17))])]
22385 "! optimize_size && ! TARGET_READ_MODIFY"
22386 [(set (match_dup 2) (match_dup 1))
22387 (parallel [(set (match_dup 0)
22388 (match_op_dup 3 [(match_dup 0) (match_dup 2)]))
22389 (clobber (reg:CC 17))])]
22392 This pattern tries to split a load from its use in the hopes that we'll
22393 be able to schedule around the memory load latency. It allocates a
22394 single `SImode' register of class `GENERAL_REGS' (`"r"') that needs to
22395 be live only at the point just before the arithmetic.
22397 A real example requiring extended scratch lifetimes is harder to come
22398 by, so here's a silly made-up example:
22401 [(match_scratch:SI 4 "r")
22402 (set (match_operand:SI 0 "" "") (match_operand:SI 1 "" ""))
22403 (set (match_operand:SI 2 "" "") (match_dup 1))
22405 (set (match_operand:SI 3 "" "") (match_dup 1))]
22406 "/* determine 1 does not overlap 0 and 2 */"
22407 [(set (match_dup 4) (match_dup 1))
22408 (set (match_dup 0) (match_dup 4))
22409 (set (match_dup 2) (match_dup 4))]
22410 (set (match_dup 3) (match_dup 4))]
22413 If we had not added the `(match_dup 4)' in the middle of the input
22414 sequence, it might have been the case that the register we chose at the
22415 beginning of the sequence is killed by the first or second `set'.
22418 File: gccint.info, Node: Insn Attributes, Next: Conditional Execution, Prev: Peephole Definitions, Up: Machine Desc
22420 16.19 Instruction Attributes
22421 ============================
22423 In addition to describing the instruction supported by the target
22424 machine, the `md' file also defines a group of "attributes" and a set of
22425 values for each. Every generated insn is assigned a value for each
22426 attribute. One possible attribute would be the effect that the insn
22427 has on the machine's condition code. This attribute can then be used
22428 by `NOTICE_UPDATE_CC' to track the condition codes.
22432 * Defining Attributes:: Specifying attributes and their values.
22433 * Expressions:: Valid expressions for attribute values.
22434 * Tagging Insns:: Assigning attribute values to insns.
22435 * Attr Example:: An example of assigning attributes.
22436 * Insn Lengths:: Computing the length of insns.
22437 * Constant Attributes:: Defining attributes that are constant.
22438 * Delay Slots:: Defining delay slots required for a machine.
22439 * Processor pipeline description:: Specifying information for insn scheduling.
22442 File: gccint.info, Node: Defining Attributes, Next: Expressions, Up: Insn Attributes
22444 16.19.1 Defining Attributes and their Values
22445 --------------------------------------------
22447 The `define_attr' expression is used to define each attribute required
22448 by the target machine. It looks like:
22450 (define_attr NAME LIST-OF-VALUES DEFAULT)
22452 NAME is a string specifying the name of the attribute being defined.
22454 LIST-OF-VALUES is either a string that specifies a comma-separated
22455 list of values that can be assigned to the attribute, or a null string
22456 to indicate that the attribute takes numeric values.
22458 DEFAULT is an attribute expression that gives the value of this
22459 attribute for insns that match patterns whose definition does not
22460 include an explicit value for this attribute. *Note Attr Example::,
22461 for more information on the handling of defaults. *Note Constant
22462 Attributes::, for information on attributes that do not depend on any
22465 For each defined attribute, a number of definitions are written to the
22466 `insn-attr.h' file. For cases where an explicit set of values is
22467 specified for an attribute, the following are defined:
22469 * A `#define' is written for the symbol `HAVE_ATTR_NAME'.
22471 * An enumerated class is defined for `attr_NAME' with elements of
22472 the form `UPPER-NAME_UPPER-VALUE' where the attribute name and
22473 value are first converted to uppercase.
22475 * A function `get_attr_NAME' is defined that is passed an insn and
22476 returns the attribute value for that insn.
22478 For example, if the following is present in the `md' file:
22480 (define_attr "type" "branch,fp,load,store,arith" ...)
22482 the following lines will be written to the file `insn-attr.h'.
22484 #define HAVE_ATTR_type
22485 enum attr_type {TYPE_BRANCH, TYPE_FP, TYPE_LOAD,
22486 TYPE_STORE, TYPE_ARITH};
22487 extern enum attr_type get_attr_type ();
22489 If the attribute takes numeric values, no `enum' type will be defined
22490 and the function to obtain the attribute's value will return `int'.
22492 There are attributes which are tied to a specific meaning. These
22493 attributes are not free to use for other purposes:
22496 The `length' attribute is used to calculate the length of emitted
22497 code chunks. This is especially important when verifying branch
22498 distances. *Note Insn Lengths::.
22501 The `enabled' attribute can be defined to prevent certain
22502 alternatives of an insn definition from being used during code
22503 generation. *Note Disable Insn Alternatives::.
22507 File: gccint.info, Node: Expressions, Next: Tagging Insns, Prev: Defining Attributes, Up: Insn Attributes
22509 16.19.2 Attribute Expressions
22510 -----------------------------
22512 RTL expressions used to define attributes use the codes described above
22513 plus a few specific to attribute definitions, to be discussed below.
22514 Attribute value expressions must have one of the following forms:
22517 The integer I specifies the value of a numeric attribute. I must
22520 The value of a numeric attribute can be specified either with a
22521 `const_int', or as an integer represented as a string in
22522 `const_string', `eq_attr' (see below), `attr', `symbol_ref',
22523 simple arithmetic expressions, and `set_attr' overrides on
22524 specific instructions (*note Tagging Insns::).
22526 `(const_string VALUE)'
22527 The string VALUE specifies a constant attribute value. If VALUE
22528 is specified as `"*"', it means that the default value of the
22529 attribute is to be used for the insn containing this expression.
22530 `"*"' obviously cannot be used in the DEFAULT expression of a
22533 If the attribute whose value is being specified is numeric, VALUE
22534 must be a string containing a non-negative integer (normally
22535 `const_int' would be used in this case). Otherwise, it must
22536 contain one of the valid values for the attribute.
22538 `(if_then_else TEST TRUE-VALUE FALSE-VALUE)'
22539 TEST specifies an attribute test, whose format is defined below.
22540 The value of this expression is TRUE-VALUE if TEST is true,
22541 otherwise it is FALSE-VALUE.
22543 `(cond [TEST1 VALUE1 ...] DEFAULT)'
22544 The first operand of this expression is a vector containing an even
22545 number of expressions and consisting of pairs of TEST and VALUE
22546 expressions. The value of the `cond' expression is that of the
22547 VALUE corresponding to the first true TEST expression. If none of
22548 the TEST expressions are true, the value of the `cond' expression
22549 is that of the DEFAULT expression.
22551 TEST expressions can have one of the following forms:
22554 This test is true if I is nonzero and false otherwise.
22557 `(ior TEST1 TEST2)'
22558 `(and TEST1 TEST2)'
22559 These tests are true if the indicated logical function is true.
22561 `(match_operand:M N PRED CONSTRAINTS)'
22562 This test is true if operand N of the insn whose attribute value
22563 is being determined has mode M (this part of the test is ignored
22564 if M is `VOIDmode') and the function specified by the string PRED
22565 returns a nonzero value when passed operand N and mode M (this
22566 part of the test is ignored if PRED is the null string).
22568 The CONSTRAINTS operand is ignored and should be the null string.
22570 `(le ARITH1 ARITH2)'
22571 `(leu ARITH1 ARITH2)'
22572 `(lt ARITH1 ARITH2)'
22573 `(ltu ARITH1 ARITH2)'
22574 `(gt ARITH1 ARITH2)'
22575 `(gtu ARITH1 ARITH2)'
22576 `(ge ARITH1 ARITH2)'
22577 `(geu ARITH1 ARITH2)'
22578 `(ne ARITH1 ARITH2)'
22579 `(eq ARITH1 ARITH2)'
22580 These tests are true if the indicated comparison of the two
22581 arithmetic expressions is true. Arithmetic expressions are formed
22582 with `plus', `minus', `mult', `div', `mod', `abs', `neg', `and',
22583 `ior', `xor', `not', `ashift', `lshiftrt', and `ashiftrt'
22586 `const_int' and `symbol_ref' are always valid terms (*note Insn
22587 Lengths::,for additional forms). `symbol_ref' is a string
22588 denoting a C expression that yields an `int' when evaluated by the
22589 `get_attr_...' routine. It should normally be a global variable.
22591 `(eq_attr NAME VALUE)'
22592 NAME is a string specifying the name of an attribute.
22594 VALUE is a string that is either a valid value for attribute NAME,
22595 a comma-separated list of values, or `!' followed by a value or
22596 list. If VALUE does not begin with a `!', this test is true if
22597 the value of the NAME attribute of the current insn is in the list
22598 specified by VALUE. If VALUE begins with a `!', this test is true
22599 if the attribute's value is _not_ in the specified list.
22603 (eq_attr "type" "load,store")
22607 (ior (eq_attr "type" "load") (eq_attr "type" "store"))
22609 If NAME specifies an attribute of `alternative', it refers to the
22610 value of the compiler variable `which_alternative' (*note Output
22611 Statement::) and the values must be small integers. For example,
22613 (eq_attr "alternative" "2,3")
22617 (ior (eq (symbol_ref "which_alternative") (const_int 2))
22618 (eq (symbol_ref "which_alternative") (const_int 3)))
22620 Note that, for most attributes, an `eq_attr' test is simplified in
22621 cases where the value of the attribute being tested is known for
22622 all insns matching a particular pattern. This is by far the most
22626 The value of an `attr_flag' expression is true if the flag
22627 specified by NAME is true for the `insn' currently being scheduled.
22629 NAME is a string specifying one of a fixed set of flags to test.
22630 Test the flags `forward' and `backward' to determine the direction
22631 of a conditional branch. Test the flags `very_likely', `likely',
22632 `very_unlikely', and `unlikely' to determine if a conditional
22633 branch is expected to be taken.
22635 If the `very_likely' flag is true, then the `likely' flag is also
22636 true. Likewise for the `very_unlikely' and `unlikely' flags.
22638 This example describes a conditional branch delay slot which can
22639 be nullified for forward branches that are taken (annul-true) or
22640 for backward branches which are not taken (annul-false).
22642 (define_delay (eq_attr "type" "cbranch")
22643 [(eq_attr "in_branch_delay" "true")
22644 (and (eq_attr "in_branch_delay" "true")
22645 (attr_flag "forward"))
22646 (and (eq_attr "in_branch_delay" "true")
22647 (attr_flag "backward"))])
22649 The `forward' and `backward' flags are false if the current `insn'
22650 being scheduled is not a conditional branch.
22652 The `very_likely' and `likely' flags are true if the `insn' being
22653 scheduled is not a conditional branch. The `very_unlikely' and
22654 `unlikely' flags are false if the `insn' being scheduled is not a
22655 conditional branch.
22657 `attr_flag' is only used during delay slot scheduling and has no
22658 meaning to other passes of the compiler.
22661 The value of another attribute is returned. This is most useful
22662 for numeric attributes, as `eq_attr' and `attr_flag' produce more
22663 efficient code for non-numeric attributes.
22666 File: gccint.info, Node: Tagging Insns, Next: Attr Example, Prev: Expressions, Up: Insn Attributes
22668 16.19.3 Assigning Attribute Values to Insns
22669 -------------------------------------------
22671 The value assigned to an attribute of an insn is primarily determined by
22672 which pattern is matched by that insn (or which `define_peephole'
22673 generated it). Every `define_insn' and `define_peephole' can have an
22674 optional last argument to specify the values of attributes for matching
22675 insns. The value of any attribute not specified in a particular insn
22676 is set to the default value for that attribute, as specified in its
22677 `define_attr'. Extensive use of default values for attributes permits
22678 the specification of the values for only one or two attributes in the
22679 definition of most insn patterns, as seen in the example in the next
22682 The optional last argument of `define_insn' and `define_peephole' is a
22683 vector of expressions, each of which defines the value for a single
22684 attribute. The most general way of assigning an attribute's value is
22685 to use a `set' expression whose first operand is an `attr' expression
22686 giving the name of the attribute being set. The second operand of the
22687 `set' is an attribute expression (*note Expressions::) giving the value
22690 When the attribute value depends on the `alternative' attribute (i.e.,
22691 which is the applicable alternative in the constraint of the insn), the
22692 `set_attr_alternative' expression can be used. It allows the
22693 specification of a vector of attribute expressions, one for each
22696 When the generality of arbitrary attribute expressions is not required,
22697 the simpler `set_attr' expression can be used, which allows specifying
22698 a string giving either a single attribute value or a list of attribute
22699 values, one for each alternative.
22701 The form of each of the above specifications is shown below. In each
22702 case, NAME is a string specifying the attribute to be set.
22704 `(set_attr NAME VALUE-STRING)'
22705 VALUE-STRING is either a string giving the desired attribute value,
22706 or a string containing a comma-separated list giving the values for
22707 succeeding alternatives. The number of elements must match the
22708 number of alternatives in the constraint of the insn pattern.
22710 Note that it may be useful to specify `*' for some alternative, in
22711 which case the attribute will assume its default value for insns
22712 matching that alternative.
22714 `(set_attr_alternative NAME [VALUE1 VALUE2 ...])'
22715 Depending on the alternative of the insn, the value will be one of
22716 the specified values. This is a shorthand for using a `cond' with
22717 tests on the `alternative' attribute.
22719 `(set (attr NAME) VALUE)'
22720 The first operand of this `set' must be the special RTL expression
22721 `attr', whose sole operand is a string giving the name of the
22722 attribute being set. VALUE is the value of the attribute.
22724 The following shows three different ways of representing the same
22725 attribute value specification:
22727 (set_attr "type" "load,store,arith")
22729 (set_attr_alternative "type"
22730 [(const_string "load") (const_string "store")
22731 (const_string "arith")])
22734 (cond [(eq_attr "alternative" "1") (const_string "load")
22735 (eq_attr "alternative" "2") (const_string "store")]
22736 (const_string "arith")))
22738 The `define_asm_attributes' expression provides a mechanism to specify
22739 the attributes assigned to insns produced from an `asm' statement. It
22742 (define_asm_attributes [ATTR-SETS])
22744 where ATTR-SETS is specified the same as for both the `define_insn' and
22745 the `define_peephole' expressions.
22747 These values will typically be the "worst case" attribute values. For
22748 example, they might indicate that the condition code will be clobbered.
22750 A specification for a `length' attribute is handled specially. The
22751 way to compute the length of an `asm' insn is to multiply the length
22752 specified in the expression `define_asm_attributes' by the number of
22753 machine instructions specified in the `asm' statement, determined by
22754 counting the number of semicolons and newlines in the string.
22755 Therefore, the value of the `length' attribute specified in a
22756 `define_asm_attributes' should be the maximum possible length of a
22757 single machine instruction.
22760 File: gccint.info, Node: Attr Example, Next: Insn Lengths, Prev: Tagging Insns, Up: Insn Attributes
22762 16.19.4 Example of Attribute Specifications
22763 -------------------------------------------
22765 The judicious use of defaulting is important in the efficient use of
22766 insn attributes. Typically, insns are divided into "types" and an
22767 attribute, customarily called `type', is used to represent this value.
22768 This attribute is normally used only to define the default value for
22769 other attributes. An example will clarify this usage.
22771 Assume we have a RISC machine with a condition code and in which only
22772 full-word operations are performed in registers. Let us assume that we
22773 can divide all insns into loads, stores, (integer) arithmetic
22774 operations, floating point operations, and branches.
22776 Here we will concern ourselves with determining the effect of an insn
22777 on the condition code and will limit ourselves to the following possible
22778 effects: The condition code can be set unpredictably (clobbered), not
22779 be changed, be set to agree with the results of the operation, or only
22780 changed if the item previously set into the condition code has been
22783 Here is part of a sample `md' file for such a machine:
22785 (define_attr "type" "load,store,arith,fp,branch" (const_string "arith"))
22787 (define_attr "cc" "clobber,unchanged,set,change0"
22788 (cond [(eq_attr "type" "load")
22789 (const_string "change0")
22790 (eq_attr "type" "store,branch")
22791 (const_string "unchanged")
22792 (eq_attr "type" "arith")
22793 (if_then_else (match_operand:SI 0 "" "")
22794 (const_string "set")
22795 (const_string "clobber"))]
22796 (const_string "clobber")))
22799 [(set (match_operand:SI 0 "general_operand" "=r,r,m")
22800 (match_operand:SI 1 "general_operand" "r,m,r"))]
22806 [(set_attr "type" "arith,load,store")])
22808 Note that we assume in the above example that arithmetic operations
22809 performed on quantities smaller than a machine word clobber the
22810 condition code since they will set the condition code to a value
22811 corresponding to the full-word result.
22814 File: gccint.info, Node: Insn Lengths, Next: Constant Attributes, Prev: Attr Example, Up: Insn Attributes
22816 16.19.5 Computing the Length of an Insn
22817 ---------------------------------------
22819 For many machines, multiple types of branch instructions are provided,
22820 each for different length branch displacements. In most cases, the
22821 assembler will choose the correct instruction to use. However, when
22822 the assembler cannot do so, GCC can when a special attribute, the
22823 `length' attribute, is defined. This attribute must be defined to have
22824 numeric values by specifying a null string in its `define_attr'.
22826 In the case of the `length' attribute, two additional forms of
22827 arithmetic terms are allowed in test expressions:
22830 This refers to the address of operand N of the current insn, which
22831 must be a `label_ref'.
22834 This refers to the address of the _current_ insn. It might have
22835 been more consistent with other usage to make this the address of
22836 the _next_ insn but this would be confusing because the length of
22837 the current insn is to be computed.
22839 For normal insns, the length will be determined by value of the
22840 `length' attribute. In the case of `addr_vec' and `addr_diff_vec' insn
22841 patterns, the length is computed as the number of vectors multiplied by
22842 the size of each vector.
22844 Lengths are measured in addressable storage units (bytes).
22846 The following macros can be used to refine the length computation:
22848 `ADJUST_INSN_LENGTH (INSN, LENGTH)'
22849 If defined, modifies the length assigned to instruction INSN as a
22850 function of the context in which it is used. LENGTH is an lvalue
22851 that contains the initially computed length of the insn and should
22852 be updated with the correct length of the insn.
22854 This macro will normally not be required. A case in which it is
22855 required is the ROMP. On this machine, the size of an `addr_vec'
22856 insn must be increased by two to compensate for the fact that
22857 alignment may be required.
22859 The routine that returns `get_attr_length' (the value of the `length'
22860 attribute) can be used by the output routine to determine the form of
22861 the branch instruction to be written, as the example below illustrates.
22863 As an example of the specification of variable-length branches,
22864 consider the IBM 360. If we adopt the convention that a register will
22865 be set to the starting address of a function, we can jump to labels
22866 within 4k of the start using a four-byte instruction. Otherwise, we
22867 need a six-byte sequence to load the address from memory and then
22870 On such a machine, a pattern for a branch instruction might be
22871 specified as follows:
22873 (define_insn "jump"
22875 (label_ref (match_operand 0 "" "")))]
22878 return (get_attr_length (insn) == 4
22879 ? "b %l0" : "l r15,=a(%l0); br r15");
22881 [(set (attr "length")
22882 (if_then_else (lt (match_dup 0) (const_int 4096))
22887 File: gccint.info, Node: Constant Attributes, Next: Delay Slots, Prev: Insn Lengths, Up: Insn Attributes
22889 16.19.6 Constant Attributes
22890 ---------------------------
22892 A special form of `define_attr', where the expression for the default
22893 value is a `const' expression, indicates an attribute that is constant
22894 for a given run of the compiler. Constant attributes may be used to
22895 specify which variety of processor is used. For example,
22897 (define_attr "cpu" "m88100,m88110,m88000"
22899 (cond [(symbol_ref "TARGET_88100") (const_string "m88100")
22900 (symbol_ref "TARGET_88110") (const_string "m88110")]
22901 (const_string "m88000"))))
22903 (define_attr "memory" "fast,slow"
22905 (if_then_else (symbol_ref "TARGET_FAST_MEM")
22906 (const_string "fast")
22907 (const_string "slow"))))
22909 The routine generated for constant attributes has no parameters as it
22910 does not depend on any particular insn. RTL expressions used to define
22911 the value of a constant attribute may use the `symbol_ref' form, but
22912 may not use either the `match_operand' form or `eq_attr' forms
22913 involving insn attributes.
22916 File: gccint.info, Node: Delay Slots, Next: Processor pipeline description, Prev: Constant Attributes, Up: Insn Attributes
22918 16.19.7 Delay Slot Scheduling
22919 -----------------------------
22921 The insn attribute mechanism can be used to specify the requirements for
22922 delay slots, if any, on a target machine. An instruction is said to
22923 require a "delay slot" if some instructions that are physically after
22924 the instruction are executed as if they were located before it.
22925 Classic examples are branch and call instructions, which often execute
22926 the following instruction before the branch or call is performed.
22928 On some machines, conditional branch instructions can optionally
22929 "annul" instructions in the delay slot. This means that the
22930 instruction will not be executed for certain branch outcomes. Both
22931 instructions that annul if the branch is true and instructions that
22932 annul if the branch is false are supported.
22934 Delay slot scheduling differs from instruction scheduling in that
22935 determining whether an instruction needs a delay slot is dependent only
22936 on the type of instruction being generated, not on data flow between the
22937 instructions. See the next section for a discussion of data-dependent
22938 instruction scheduling.
22940 The requirement of an insn needing one or more delay slots is indicated
22941 via the `define_delay' expression. It has the following form:
22944 [DELAY-1 ANNUL-TRUE-1 ANNUL-FALSE-1
22945 DELAY-2 ANNUL-TRUE-2 ANNUL-FALSE-2
22948 TEST is an attribute test that indicates whether this `define_delay'
22949 applies to a particular insn. If so, the number of required delay
22950 slots is determined by the length of the vector specified as the second
22951 argument. An insn placed in delay slot N must satisfy attribute test
22952 DELAY-N. ANNUL-TRUE-N is an attribute test that specifies which insns
22953 may be annulled if the branch is true. Similarly, ANNUL-FALSE-N
22954 specifies which insns in the delay slot may be annulled if the branch
22955 is false. If annulling is not supported for that delay slot, `(nil)'
22958 For example, in the common case where branch and call insns require a
22959 single delay slot, which may contain any insn other than a branch or
22960 call, the following would be placed in the `md' file:
22962 (define_delay (eq_attr "type" "branch,call")
22963 [(eq_attr "type" "!branch,call") (nil) (nil)])
22965 Multiple `define_delay' expressions may be specified. In this case,
22966 each such expression specifies different delay slot requirements and
22967 there must be no insn for which tests in two `define_delay' expressions
22970 For example, if we have a machine that requires one delay slot for
22971 branches but two for calls, no delay slot can contain a branch or call
22972 insn, and any valid insn in the delay slot for the branch can be
22973 annulled if the branch is true, we might represent this as follows:
22975 (define_delay (eq_attr "type" "branch")
22976 [(eq_attr "type" "!branch,call")
22977 (eq_attr "type" "!branch,call")
22980 (define_delay (eq_attr "type" "call")
22981 [(eq_attr "type" "!branch,call") (nil) (nil)
22982 (eq_attr "type" "!branch,call") (nil) (nil)])
22985 File: gccint.info, Node: Processor pipeline description, Prev: Delay Slots, Up: Insn Attributes
22987 16.19.8 Specifying processor pipeline description
22988 -------------------------------------------------
22990 To achieve better performance, most modern processors (super-pipelined,
22991 superscalar RISC, and VLIW processors) have many "functional units" on
22992 which several instructions can be executed simultaneously. An
22993 instruction starts execution if its issue conditions are satisfied. If
22994 not, the instruction is stalled until its conditions are satisfied.
22995 Such "interlock (pipeline) delay" causes interruption of the fetching
22996 of successor instructions (or demands nop instructions, e.g. for some
22999 There are two major kinds of interlock delays in modern processors.
23000 The first one is a data dependence delay determining "instruction
23001 latency time". The instruction execution is not started until all
23002 source data have been evaluated by prior instructions (there are more
23003 complex cases when the instruction execution starts even when the data
23004 are not available but will be ready in given time after the instruction
23005 execution start). Taking the data dependence delays into account is
23006 simple. The data dependence (true, output, and anti-dependence) delay
23007 between two instructions is given by a constant. In most cases this
23008 approach is adequate. The second kind of interlock delays is a
23009 reservation delay. The reservation delay means that two instructions
23010 under execution will be in need of shared processors resources, i.e.
23011 buses, internal registers, and/or functional units, which are reserved
23012 for some time. Taking this kind of delay into account is complex
23013 especially for modern RISC processors.
23015 The task of exploiting more processor parallelism is solved by an
23016 instruction scheduler. For a better solution to this problem, the
23017 instruction scheduler has to have an adequate description of the
23018 processor parallelism (or "pipeline description"). GCC machine
23019 descriptions describe processor parallelism and functional unit
23020 reservations for groups of instructions with the aid of "regular
23023 The GCC instruction scheduler uses a "pipeline hazard recognizer" to
23024 figure out the possibility of the instruction issue by the processor on
23025 a given simulated processor cycle. The pipeline hazard recognizer is
23026 automatically generated from the processor pipeline description. The
23027 pipeline hazard recognizer generated from the machine description is
23028 based on a deterministic finite state automaton (DFA): the instruction
23029 issue is possible if there is a transition from one automaton state to
23030 another one. This algorithm is very fast, and furthermore, its speed
23031 is not dependent on processor complexity(1).
23033 The rest of this section describes the directives that constitute an
23034 automaton-based processor pipeline description. The order of these
23035 constructions within the machine description file is not important.
23037 The following optional construction describes names of automata
23038 generated and used for the pipeline hazards recognition. Sometimes the
23039 generated finite state automaton used by the pipeline hazard recognizer
23040 is large. If we use more than one automaton and bind functional units
23041 to the automata, the total size of the automata is usually less than
23042 the size of the single automaton. If there is no one such
23043 construction, only one finite state automaton is generated.
23045 (define_automaton AUTOMATA-NAMES)
23047 AUTOMATA-NAMES is a string giving names of the automata. The names
23048 are separated by commas. All the automata should have unique names.
23049 The automaton name is used in the constructions `define_cpu_unit' and
23050 `define_query_cpu_unit'.
23052 Each processor functional unit used in the description of instruction
23053 reservations should be described by the following construction.
23055 (define_cpu_unit UNIT-NAMES [AUTOMATON-NAME])
23057 UNIT-NAMES is a string giving the names of the functional units
23058 separated by commas. Don't use name `nothing', it is reserved for
23061 AUTOMATON-NAME is a string giving the name of the automaton with which
23062 the unit is bound. The automaton should be described in construction
23063 `define_automaton'. You should give "automaton-name", if there is a
23066 The assignment of units to automata are constrained by the uses of the
23067 units in insn reservations. The most important constraint is: if a
23068 unit reservation is present on a particular cycle of an alternative for
23069 an insn reservation, then some unit from the same automaton must be
23070 present on the same cycle for the other alternatives of the insn
23071 reservation. The rest of the constraints are mentioned in the
23072 description of the subsequent constructions.
23074 The following construction describes CPU functional units analogously
23075 to `define_cpu_unit'. The reservation of such units can be queried for
23076 an automaton state. The instruction scheduler never queries
23077 reservation of functional units for given automaton state. So as a
23078 rule, you don't need this construction. This construction could be
23079 used for future code generation goals (e.g. to generate VLIW insn
23082 (define_query_cpu_unit UNIT-NAMES [AUTOMATON-NAME])
23084 UNIT-NAMES is a string giving names of the functional units separated
23087 AUTOMATON-NAME is a string giving the name of the automaton with which
23090 The following construction is the major one to describe pipeline
23091 characteristics of an instruction.
23093 (define_insn_reservation INSN-NAME DEFAULT_LATENCY
23096 DEFAULT_LATENCY is a number giving latency time of the instruction.
23097 There is an important difference between the old description and the
23098 automaton based pipeline description. The latency time is used for all
23099 dependencies when we use the old description. In the automaton based
23100 pipeline description, the given latency time is only used for true
23101 dependencies. The cost of anti-dependencies is always zero and the
23102 cost of output dependencies is the difference between latency times of
23103 the producing and consuming insns (if the difference is negative, the
23104 cost is considered to be zero). You can always change the default
23105 costs for any description by using the target hook
23106 `TARGET_SCHED_ADJUST_COST' (*note Scheduling::).
23108 INSN-NAME is a string giving the internal name of the insn. The
23109 internal names are used in constructions `define_bypass' and in the
23110 automaton description file generated for debugging. The internal name
23111 has nothing in common with the names in `define_insn'. It is a good
23112 practice to use insn classes described in the processor manual.
23114 CONDITION defines what RTL insns are described by this construction.
23115 You should remember that you will be in trouble if CONDITION for two or
23116 more different `define_insn_reservation' constructions is TRUE for an
23117 insn. In this case what reservation will be used for the insn is not
23118 defined. Such cases are not checked during generation of the pipeline
23119 hazards recognizer because in general recognizing that two conditions
23120 may have the same value is quite difficult (especially if the conditions
23121 contain `symbol_ref'). It is also not checked during the pipeline
23122 hazard recognizer work because it would slow down the recognizer
23125 REGEXP is a string describing the reservation of the cpu's functional
23126 units by the instruction. The reservations are described by a regular
23127 expression according to the following syntax:
23129 regexp = regexp "," oneof
23132 oneof = oneof "|" allof
23135 allof = allof "+" repeat
23138 repeat = element "*" number
23141 element = cpu_function_unit_name
23147 * `,' is used for describing the start of the next cycle in the
23150 * `|' is used for describing a reservation described by the first
23151 regular expression *or* a reservation described by the second
23152 regular expression *or* etc.
23154 * `+' is used for describing a reservation described by the first
23155 regular expression *and* a reservation described by the second
23156 regular expression *and* etc.
23158 * `*' is used for convenience and simply means a sequence in which
23159 the regular expression are repeated NUMBER times with cycle
23160 advancing (see `,').
23162 * `cpu_function_unit_name' denotes reservation of the named
23165 * `reservation_name' -- see description of construction
23166 `define_reservation'.
23168 * `nothing' denotes no unit reservations.
23170 Sometimes unit reservations for different insns contain common parts.
23171 In such case, you can simplify the pipeline description by describing
23172 the common part by the following construction
23174 (define_reservation RESERVATION-NAME REGEXP)
23176 RESERVATION-NAME is a string giving name of REGEXP. Functional unit
23177 names and reservation names are in the same name space. So the
23178 reservation names should be different from the functional unit names
23179 and can not be the reserved name `nothing'.
23181 The following construction is used to describe exceptions in the
23182 latency time for given instruction pair. This is so called bypasses.
23184 (define_bypass NUMBER OUT_INSN_NAMES IN_INSN_NAMES
23187 NUMBER defines when the result generated by the instructions given in
23188 string OUT_INSN_NAMES will be ready for the instructions given in
23189 string IN_INSN_NAMES. The instructions in the string are separated by
23192 GUARD is an optional string giving the name of a C function which
23193 defines an additional guard for the bypass. The function will get the
23194 two insns as parameters. If the function returns zero the bypass will
23195 be ignored for this case. The additional guard is necessary to
23196 recognize complicated bypasses, e.g. when the consumer is only an
23197 address of insn `store' (not a stored value).
23199 The following five constructions are usually used to describe VLIW
23200 processors, or more precisely, to describe a placement of small
23201 instructions into VLIW instruction slots. They can be used for RISC
23204 (exclusion_set UNIT-NAMES UNIT-NAMES)
23205 (presence_set UNIT-NAMES PATTERNS)
23206 (final_presence_set UNIT-NAMES PATTERNS)
23207 (absence_set UNIT-NAMES PATTERNS)
23208 (final_absence_set UNIT-NAMES PATTERNS)
23210 UNIT-NAMES is a string giving names of functional units separated by
23213 PATTERNS is a string giving patterns of functional units separated by
23214 comma. Currently pattern is one unit or units separated by
23217 The first construction (`exclusion_set') means that each functional
23218 unit in the first string can not be reserved simultaneously with a unit
23219 whose name is in the second string and vice versa. For example, the
23220 construction is useful for describing processors (e.g. some SPARC
23221 processors) with a fully pipelined floating point functional unit which
23222 can execute simultaneously only single floating point insns or only
23223 double floating point insns.
23225 The second construction (`presence_set') means that each functional
23226 unit in the first string can not be reserved unless at least one of
23227 pattern of units whose names are in the second string is reserved.
23228 This is an asymmetric relation. For example, it is useful for
23229 description that VLIW `slot1' is reserved after `slot0' reservation.
23230 We could describe it by the following construction
23232 (presence_set "slot1" "slot0")
23234 Or `slot1' is reserved only after `slot0' and unit `b0' reservation.
23235 In this case we could write
23237 (presence_set "slot1" "slot0 b0")
23239 The third construction (`final_presence_set') is analogous to
23240 `presence_set'. The difference between them is when checking is done.
23241 When an instruction is issued in given automaton state reflecting all
23242 current and planned unit reservations, the automaton state is changed.
23243 The first state is a source state, the second one is a result state.
23244 Checking for `presence_set' is done on the source state reservation,
23245 checking for `final_presence_set' is done on the result reservation.
23246 This construction is useful to describe a reservation which is actually
23247 two subsequent reservations. For example, if we use
23249 (presence_set "slot1" "slot0")
23251 the following insn will be never issued (because `slot1' requires
23252 `slot0' which is absent in the source state).
23254 (define_reservation "insn_and_nop" "slot0 + slot1")
23256 but it can be issued if we use analogous `final_presence_set'.
23258 The forth construction (`absence_set') means that each functional unit
23259 in the first string can be reserved only if each pattern of units whose
23260 names are in the second string is not reserved. This is an asymmetric
23261 relation (actually `exclusion_set' is analogous to this one but it is
23262 symmetric). For example it might be useful in a VLIW description to
23263 say that `slot0' cannot be reserved after either `slot1' or `slot2'
23264 have been reserved. This can be described as:
23266 (absence_set "slot0" "slot1, slot2")
23268 Or `slot2' can not be reserved if `slot0' and unit `b0' are reserved
23269 or `slot1' and unit `b1' are reserved. In this case we could write
23271 (absence_set "slot2" "slot0 b0, slot1 b1")
23273 All functional units mentioned in a set should belong to the same
23276 The last construction (`final_absence_set') is analogous to
23277 `absence_set' but checking is done on the result (state) reservation.
23278 See comments for `final_presence_set'.
23280 You can control the generator of the pipeline hazard recognizer with
23281 the following construction.
23283 (automata_option OPTIONS)
23285 OPTIONS is a string giving options which affect the generated code.
23286 Currently there are the following options:
23288 * "no-minimization" makes no minimization of the automaton. This is
23289 only worth to do when we are debugging the description and need to
23290 look more accurately at reservations of states.
23292 * "time" means printing time statistics about the generation of
23295 * "stats" means printing statistics about the generated automata
23296 such as the number of DFA states, NDFA states and arcs.
23298 * "v" means a generation of the file describing the result automata.
23299 The file has suffix `.dfa' and can be used for the description
23300 verification and debugging.
23302 * "w" means a generation of warning instead of error for
23303 non-critical errors.
23305 * "ndfa" makes nondeterministic finite state automata. This affects
23306 the treatment of operator `|' in the regular expressions. The
23307 usual treatment of the operator is to try the first alternative
23308 and, if the reservation is not possible, the second alternative.
23309 The nondeterministic treatment means trying all alternatives, some
23310 of them may be rejected by reservations in the subsequent insns.
23312 * "progress" means output of a progress bar showing how many states
23313 were generated so far for automaton being processed. This is
23314 useful during debugging a DFA description. If you see too many
23315 generated states, you could interrupt the generator of the pipeline
23316 hazard recognizer and try to figure out a reason for generation of
23317 the huge automaton.
23319 As an example, consider a superscalar RISC machine which can issue
23320 three insns (two integer insns and one floating point insn) on the
23321 cycle but can finish only two insns. To describe this, we define the
23322 following functional units.
23324 (define_cpu_unit "i0_pipeline, i1_pipeline, f_pipeline")
23325 (define_cpu_unit "port0, port1")
23327 All simple integer insns can be executed in any integer pipeline and
23328 their result is ready in two cycles. The simple integer insns are
23329 issued into the first pipeline unless it is reserved, otherwise they
23330 are issued into the second pipeline. Integer division and
23331 multiplication insns can be executed only in the second integer
23332 pipeline and their results are ready correspondingly in 8 and 4 cycles.
23333 The integer division is not pipelined, i.e. the subsequent integer
23334 division insn can not be issued until the current division insn
23335 finished. Floating point insns are fully pipelined and their results
23336 are ready in 3 cycles. Where the result of a floating point insn is
23337 used by an integer insn, an additional delay of one cycle is incurred.
23338 To describe all of this we could specify
23340 (define_cpu_unit "div")
23342 (define_insn_reservation "simple" 2 (eq_attr "type" "int")
23343 "(i0_pipeline | i1_pipeline), (port0 | port1)")
23345 (define_insn_reservation "mult" 4 (eq_attr "type" "mult")
23346 "i1_pipeline, nothing*2, (port0 | port1)")
23348 (define_insn_reservation "div" 8 (eq_attr "type" "div")
23349 "i1_pipeline, div*7, div + (port0 | port1)")
23351 (define_insn_reservation "float" 3 (eq_attr "type" "float")
23352 "f_pipeline, nothing, (port0 | port1))
23354 (define_bypass 4 "float" "simple,mult,div")
23356 To simplify the description we could describe the following reservation
23358 (define_reservation "finish" "port0|port1")
23360 and use it in all `define_insn_reservation' as in the following
23363 (define_insn_reservation "simple" 2 (eq_attr "type" "int")
23364 "(i0_pipeline | i1_pipeline), finish")
23366 ---------- Footnotes ----------
23368 (1) However, the size of the automaton depends on processor
23369 complexity. To limit this effect, machine descriptions can split
23370 orthogonal parts of the machine description among several automata: but
23371 then, since each of these must be stepped independently, this does
23372 cause a small decrease in the algorithm's performance.
23375 File: gccint.info, Node: Conditional Execution, Next: Constant Definitions, Prev: Insn Attributes, Up: Machine Desc
23377 16.20 Conditional Execution
23378 ===========================
23380 A number of architectures provide for some form of conditional
23381 execution, or predication. The hallmark of this feature is the ability
23382 to nullify most of the instructions in the instruction set. When the
23383 instruction set is large and not entirely symmetric, it can be quite
23384 tedious to describe these forms directly in the `.md' file. An
23385 alternative is the `define_cond_exec' template.
23388 [PREDICATE-PATTERN]
23392 PREDICATE-PATTERN is the condition that must be true for the insn to
23393 be executed at runtime and should match a relational operator. One can
23394 use `match_operator' to match several relational operators at once.
23395 Any `match_operand' operands must have no more than one alternative.
23397 CONDITION is a C expression that must be true for the generated
23400 OUTPUT-TEMPLATE is a string similar to the `define_insn' output
23401 template (*note Output Template::), except that the `*' and `@' special
23402 cases do not apply. This is only useful if the assembly text for the
23403 predicate is a simple prefix to the main insn. In order to handle the
23404 general case, there is a global variable `current_insn_predicate' that
23405 will contain the entire predicate if the current insn is predicated,
23406 and will otherwise be `NULL'.
23408 When `define_cond_exec' is used, an implicit reference to the
23409 `predicable' instruction attribute is made. *Note Insn Attributes::.
23410 This attribute must be boolean (i.e. have exactly two elements in its
23411 LIST-OF-VALUES). Further, it must not be used with complex
23412 expressions. That is, the default and all uses in the insns must be a
23413 simple constant, not dependent on the alternative or anything else.
23415 For each `define_insn' for which the `predicable' attribute is true, a
23416 new `define_insn' pattern will be generated that matches a predicated
23417 version of the instruction. For example,
23419 (define_insn "addsi"
23420 [(set (match_operand:SI 0 "register_operand" "r")
23421 (plus:SI (match_operand:SI 1 "register_operand" "r")
23422 (match_operand:SI 2 "register_operand" "r")))]
23427 [(ne (match_operand:CC 0 "register_operand" "c")
23432 generates a new pattern
23436 (ne (match_operand:CC 3 "register_operand" "c") (const_int 0))
23437 (set (match_operand:SI 0 "register_operand" "r")
23438 (plus:SI (match_operand:SI 1 "register_operand" "r")
23439 (match_operand:SI 2 "register_operand" "r"))))]
23440 "(TEST2) && (TEST1)"
23441 "(%3) add %2,%1,%0")
23444 File: gccint.info, Node: Constant Definitions, Next: Iterators, Prev: Conditional Execution, Up: Machine Desc
23446 16.21 Constant Definitions
23447 ==========================
23449 Using literal constants inside instruction patterns reduces legibility
23450 and can be a maintenance problem.
23452 To overcome this problem, you may use the `define_constants'
23453 expression. It contains a vector of name-value pairs. From that point
23454 on, wherever any of the names appears in the MD file, it is as if the
23455 corresponding value had been written instead. You may use
23456 `define_constants' multiple times; each appearance adds more constants
23457 to the table. It is an error to redefine a constant with a different
23460 To come back to the a29k load multiple example, instead of
23463 [(match_parallel 0 "load_multiple_operation"
23464 [(set (match_operand:SI 1 "gpc_reg_operand" "=r")
23465 (match_operand:SI 2 "memory_operand" "m"))
23467 (clobber (reg:SI 179))])]
23473 (define_constants [
23481 [(match_parallel 0 "load_multiple_operation"
23482 [(set (match_operand:SI 1 "gpc_reg_operand" "=r")
23483 (match_operand:SI 2 "memory_operand" "m"))
23484 (use (reg:SI R_CR))
23485 (clobber (reg:SI R_CR))])]
23489 The constants that are defined with a define_constant are also output
23490 in the insn-codes.h header file as #defines.
23493 File: gccint.info, Node: Iterators, Prev: Constant Definitions, Up: Machine Desc
23498 Ports often need to define similar patterns for more than one machine
23499 mode or for more than one rtx code. GCC provides some simple iterator
23500 facilities to make this process easier.
23504 * Mode Iterators:: Generating variations of patterns for different modes.
23505 * Code Iterators:: Doing the same for codes.
23508 File: gccint.info, Node: Mode Iterators, Next: Code Iterators, Up: Iterators
23510 16.22.1 Mode Iterators
23511 ----------------------
23513 Ports often need to define similar patterns for two or more different
23514 modes. For example:
23516 * If a processor has hardware support for both single and double
23517 floating-point arithmetic, the `SFmode' patterns tend to be very
23518 similar to the `DFmode' ones.
23520 * If a port uses `SImode' pointers in one configuration and `DImode'
23521 pointers in another, it will usually have very similar `SImode'
23522 and `DImode' patterns for manipulating pointers.
23524 Mode iterators allow several patterns to be instantiated from one
23525 `.md' file template. They can be used with any type of rtx-based
23526 construct, such as a `define_insn', `define_split', or
23527 `define_peephole2'.
23531 * Defining Mode Iterators:: Defining a new mode iterator.
23532 * Substitutions:: Combining mode iterators with substitutions
23533 * Examples:: Examples
23536 File: gccint.info, Node: Defining Mode Iterators, Next: Substitutions, Up: Mode Iterators
23538 16.22.1.1 Defining Mode Iterators
23539 .................................
23541 The syntax for defining a mode iterator is:
23543 (define_mode_iterator NAME [(MODE1 "COND1") ... (MODEN "CONDN")])
23545 This allows subsequent `.md' file constructs to use the mode suffix
23546 `:NAME'. Every construct that does so will be expanded N times, once
23547 with every use of `:NAME' replaced by `:MODE1', once with every use
23548 replaced by `:MODE2', and so on. In the expansion for a particular
23549 MODEI, every C condition will also require that CONDI be true.
23553 (define_mode_iterator P [(SI "Pmode == SImode") (DI "Pmode == DImode")])
23555 defines a new mode suffix `:P'. Every construct that uses `:P' will
23556 be expanded twice, once with every `:P' replaced by `:SI' and once with
23557 every `:P' replaced by `:DI'. The `:SI' version will only apply if
23558 `Pmode == SImode' and the `:DI' version will only apply if `Pmode ==
23561 As with other `.md' conditions, an empty string is treated as "always
23562 true". `(MODE "")' can also be abbreviated to `MODE'. For example:
23564 (define_mode_iterator GPR [SI (DI "TARGET_64BIT")])
23566 means that the `:DI' expansion only applies if `TARGET_64BIT' but that
23567 the `:SI' expansion has no such constraint.
23569 Iterators are applied in the order they are defined. This can be
23570 significant if two iterators are used in a construct that requires
23571 substitutions. *Note Substitutions::.
23574 File: gccint.info, Node: Substitutions, Next: Examples, Prev: Defining Mode Iterators, Up: Mode Iterators
23576 16.22.1.2 Substitution in Mode Iterators
23577 ........................................
23579 If an `.md' file construct uses mode iterators, each version of the
23580 construct will often need slightly different strings or modes. For
23583 * When a `define_expand' defines several `addM3' patterns (*note
23584 Standard Names::), each expander will need to use the appropriate
23587 * When a `define_insn' defines several instruction patterns, each
23588 instruction will often use a different assembler mnemonic.
23590 * When a `define_insn' requires operands with different modes, using
23591 an iterator for one of the operand modes usually requires a
23592 specific mode for the other operand(s).
23594 GCC supports such variations through a system of "mode attributes".
23595 There are two standard attributes: `mode', which is the name of the
23596 mode in lower case, and `MODE', which is the same thing in upper case.
23597 You can define other attributes using:
23599 (define_mode_attr NAME [(MODE1 "VALUE1") ... (MODEN "VALUEN")])
23601 where NAME is the name of the attribute and VALUEI is the value
23602 associated with MODEI.
23604 When GCC replaces some :ITERATOR with :MODE, it will scan each string
23605 and mode in the pattern for sequences of the form `<ITERATOR:ATTR>',
23606 where ATTR is the name of a mode attribute. If the attribute is
23607 defined for MODE, the whole `<...>' sequence will be replaced by the
23608 appropriate attribute value.
23610 For example, suppose an `.md' file has:
23612 (define_mode_iterator P [(SI "Pmode == SImode") (DI "Pmode == DImode")])
23613 (define_mode_attr load [(SI "lw") (DI "ld")])
23615 If one of the patterns that uses `:P' contains the string
23616 `"<P:load>\t%0,%1"', the `SI' version of that pattern will use
23617 `"lw\t%0,%1"' and the `DI' version will use `"ld\t%0,%1"'.
23619 Here is an example of using an attribute for a mode:
23621 (define_mode_iterator LONG [SI DI])
23622 (define_mode_attr SHORT [(SI "HI") (DI "SI")])
23624 (sign_extend:LONG (match_operand:<LONG:SHORT> ...)) ...)
23626 The `ITERATOR:' prefix may be omitted, in which case the substitution
23627 will be attempted for every iterator expansion.
23630 File: gccint.info, Node: Examples, Prev: Substitutions, Up: Mode Iterators
23632 16.22.1.3 Mode Iterator Examples
23633 ................................
23635 Here is an example from the MIPS port. It defines the following modes
23636 and attributes (among others):
23638 (define_mode_iterator GPR [SI (DI "TARGET_64BIT")])
23639 (define_mode_attr d [(SI "") (DI "d")])
23641 and uses the following template to define both `subsi3' and `subdi3':
23643 (define_insn "sub<mode>3"
23644 [(set (match_operand:GPR 0 "register_operand" "=d")
23645 (minus:GPR (match_operand:GPR 1 "register_operand" "d")
23646 (match_operand:GPR 2 "register_operand" "d")))]
23648 "<d>subu\t%0,%1,%2"
23649 [(set_attr "type" "arith")
23650 (set_attr "mode" "<MODE>")])
23652 This is exactly equivalent to:
23654 (define_insn "subsi3"
23655 [(set (match_operand:SI 0 "register_operand" "=d")
23656 (minus:SI (match_operand:SI 1 "register_operand" "d")
23657 (match_operand:SI 2 "register_operand" "d")))]
23660 [(set_attr "type" "arith")
23661 (set_attr "mode" "SI")])
23663 (define_insn "subdi3"
23664 [(set (match_operand:DI 0 "register_operand" "=d")
23665 (minus:DI (match_operand:DI 1 "register_operand" "d")
23666 (match_operand:DI 2 "register_operand" "d")))]
23669 [(set_attr "type" "arith")
23670 (set_attr "mode" "DI")])
23673 File: gccint.info, Node: Code Iterators, Prev: Mode Iterators, Up: Iterators
23675 16.22.2 Code Iterators
23676 ----------------------
23678 Code iterators operate in a similar way to mode iterators. *Note Mode
23683 (define_code_iterator NAME [(CODE1 "COND1") ... (CODEN "CONDN")])
23685 defines a pseudo rtx code NAME that can be instantiated as CODEI if
23686 condition CONDI is true. Each CODEI must have the same rtx format.
23687 *Note RTL Classes::.
23689 As with mode iterators, each pattern that uses NAME will be expanded N
23690 times, once with all uses of NAME replaced by CODE1, once with all uses
23691 replaced by CODE2, and so on. *Note Defining Mode Iterators::.
23693 It is possible to define attributes for codes as well as for modes.
23694 There are two standard code attributes: `code', the name of the code in
23695 lower case, and `CODE', the name of the code in upper case. Other
23696 attributes are defined using:
23698 (define_code_attr NAME [(CODE1 "VALUE1") ... (CODEN "VALUEN")])
23700 Here's an example of code iterators in action, taken from the MIPS
23703 (define_code_iterator any_cond [unordered ordered unlt unge uneq ltgt unle ungt
23704 eq ne gt ge lt le gtu geu ltu leu])
23706 (define_expand "b<code>"
23708 (if_then_else (any_cond:CC (cc0)
23710 (label_ref (match_operand 0 ""))
23714 gen_conditional_branch (operands, <CODE>);
23718 This is equivalent to:
23720 (define_expand "bunordered"
23722 (if_then_else (unordered:CC (cc0)
23724 (label_ref (match_operand 0 ""))
23728 gen_conditional_branch (operands, UNORDERED);
23732 (define_expand "bordered"
23734 (if_then_else (ordered:CC (cc0)
23736 (label_ref (match_operand 0 ""))
23740 gen_conditional_branch (operands, ORDERED);
23747 File: gccint.info, Node: Target Macros, Next: Host Config, Prev: Machine Desc, Up: Top
23749 17 Target Description Macros and Functions
23750 ******************************************
23752 In addition to the file `MACHINE.md', a machine description includes a
23753 C header file conventionally given the name `MACHINE.h' and a C source
23754 file named `MACHINE.c'. The header file defines numerous macros that
23755 convey the information about the target machine that does not fit into
23756 the scheme of the `.md' file. The file `tm.h' should be a link to
23757 `MACHINE.h'. The header file `config.h' includes `tm.h' and most
23758 compiler source files include `config.h'. The source file defines a
23759 variable `targetm', which is a structure containing pointers to
23760 functions and data relating to the target machine. `MACHINE.c' should
23761 also contain their definitions, if they are not defined elsewhere in
23762 GCC, and other functions called through the macros defined in the `.h'
23767 * Target Structure:: The `targetm' variable.
23768 * Driver:: Controlling how the driver runs the compilation passes.
23769 * Run-time Target:: Defining `-m' options like `-m68000' and `-m68020'.
23770 * Per-Function Data:: Defining data structures for per-function information.
23771 * Storage Layout:: Defining sizes and alignments of data.
23772 * Type Layout:: Defining sizes and properties of basic user data types.
23773 * Registers:: Naming and describing the hardware registers.
23774 * Register Classes:: Defining the classes of hardware registers.
23775 * Old Constraints:: The old way to define machine-specific constraints.
23776 * Stack and Calling:: Defining which way the stack grows and by how much.
23777 * Varargs:: Defining the varargs macros.
23778 * Trampolines:: Code set up at run time to enter a nested function.
23779 * Library Calls:: Controlling how library routines are implicitly called.
23780 * Addressing Modes:: Defining addressing modes valid for memory operands.
23781 * Anchored Addresses:: Defining how `-fsection-anchors' should work.
23782 * Condition Code:: Defining how insns update the condition code.
23783 * Costs:: Defining relative costs of different operations.
23784 * Scheduling:: Adjusting the behavior of the instruction scheduler.
23785 * Sections:: Dividing storage into text, data, and other sections.
23786 * PIC:: Macros for position independent code.
23787 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
23788 * Debugging Info:: Defining the format of debugging output.
23789 * Floating Point:: Handling floating point for cross-compilers.
23790 * Mode Switching:: Insertion of mode-switching instructions.
23791 * Target Attributes:: Defining target-specific uses of `__attribute__'.
23792 * Emulated TLS:: Emulated TLS support.
23793 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
23794 * PCH Target:: Validity checking for precompiled headers.
23795 * C++ ABI:: Controlling C++ ABI changes.
23796 * Misc:: Everything else.
23799 File: gccint.info, Node: Target Structure, Next: Driver, Up: Target Macros
23801 17.1 The Global `targetm' Variable
23802 ==================================
23804 -- Variable: struct gcc_target targetm
23805 The target `.c' file must define the global `targetm' variable
23806 which contains pointers to functions and data relating to the
23807 target machine. The variable is declared in `target.h';
23808 `target-def.h' defines the macro `TARGET_INITIALIZER' which is
23809 used to initialize the variable, and macros for the default
23810 initializers for elements of the structure. The `.c' file should
23811 override those macros for which the default definition is
23812 inappropriate. For example:
23813 #include "target.h"
23814 #include "target-def.h"
23816 /* Initialize the GCC target structure. */
23818 #undef TARGET_COMP_TYPE_ATTRIBUTES
23819 #define TARGET_COMP_TYPE_ATTRIBUTES MACHINE_comp_type_attributes
23821 struct gcc_target targetm = TARGET_INITIALIZER;
23823 Where a macro should be defined in the `.c' file in this manner to form
23824 part of the `targetm' structure, it is documented below as a "Target
23825 Hook" with a prototype. Many macros will change in future from being
23826 defined in the `.h' file to being part of the `targetm' structure.
23829 File: gccint.info, Node: Driver, Next: Run-time Target, Prev: Target Structure, Up: Target Macros
23831 17.2 Controlling the Compilation Driver, `gcc'
23832 ==============================================
23834 You can control the compilation driver.
23836 -- Macro: SWITCH_TAKES_ARG (CHAR)
23837 A C expression which determines whether the option `-CHAR' takes
23838 arguments. The value should be the number of arguments that
23839 option takes-zero, for many options.
23841 By default, this macro is defined as `DEFAULT_SWITCH_TAKES_ARG',
23842 which handles the standard options properly. You need not define
23843 `SWITCH_TAKES_ARG' unless you wish to add additional options which
23844 take arguments. Any redefinition should call
23845 `DEFAULT_SWITCH_TAKES_ARG' and then check for additional options.
23847 -- Macro: WORD_SWITCH_TAKES_ARG (NAME)
23848 A C expression which determines whether the option `-NAME' takes
23849 arguments. The value should be the number of arguments that
23850 option takes-zero, for many options. This macro rather than
23851 `SWITCH_TAKES_ARG' is used for multi-character option names.
23853 By default, this macro is defined as
23854 `DEFAULT_WORD_SWITCH_TAKES_ARG', which handles the standard options
23855 properly. You need not define `WORD_SWITCH_TAKES_ARG' unless you
23856 wish to add additional options which take arguments. Any
23857 redefinition should call `DEFAULT_WORD_SWITCH_TAKES_ARG' and then
23858 check for additional options.
23860 -- Macro: SWITCH_CURTAILS_COMPILATION (CHAR)
23861 A C expression which determines whether the option `-CHAR' stops
23862 compilation before the generation of an executable. The value is
23863 boolean, nonzero if the option does stop an executable from being
23864 generated, zero otherwise.
23866 By default, this macro is defined as
23867 `DEFAULT_SWITCH_CURTAILS_COMPILATION', which handles the standard
23868 options properly. You need not define
23869 `SWITCH_CURTAILS_COMPILATION' unless you wish to add additional
23870 options which affect the generation of an executable. Any
23871 redefinition should call `DEFAULT_SWITCH_CURTAILS_COMPILATION' and
23872 then check for additional options.
23874 -- Macro: SWITCHES_NEED_SPACES
23875 A string-valued C expression which enumerates the options for which
23876 the linker needs a space between the option and its argument.
23878 If this macro is not defined, the default value is `""'.
23880 -- Macro: TARGET_OPTION_TRANSLATE_TABLE
23881 If defined, a list of pairs of strings, the first of which is a
23882 potential command line target to the `gcc' driver program, and the
23883 second of which is a space-separated (tabs and other whitespace
23884 are not supported) list of options with which to replace the first
23885 option. The target defining this list is responsible for assuring
23886 that the results are valid. Replacement options may not be the
23887 `--opt' style, they must be the `-opt' style. It is the intention
23888 of this macro to provide a mechanism for substitution that affects
23889 the multilibs chosen, such as one option that enables many
23890 options, some of which select multilibs. Example nonsensical
23891 definition, where `-malt-abi', `-EB', and `-mspoo' cause different
23892 multilibs to be chosen:
23894 #define TARGET_OPTION_TRANSLATE_TABLE \
23895 { "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" }, \
23896 { "-compat", "-EB -malign=4 -mspoo" }
23898 -- Macro: DRIVER_SELF_SPECS
23899 A list of specs for the driver itself. It should be a suitable
23900 initializer for an array of strings, with no surrounding braces.
23902 The driver applies these specs to its own command line between
23903 loading default `specs' files (but not command-line specified
23904 ones) and choosing the multilib directory or running any
23905 subcommands. It applies them in the order given, so each spec can
23906 depend on the options added by earlier ones. It is also possible
23907 to remove options using `%<OPTION' in the usual way.
23909 This macro can be useful when a port has several interdependent
23910 target options. It provides a way of standardizing the command
23911 line so that the other specs are easier to write.
23913 Do not define this macro if it does not need to do anything.
23915 -- Macro: OPTION_DEFAULT_SPECS
23916 A list of specs used to support configure-time default options
23917 (i.e. `--with' options) in the driver. It should be a suitable
23918 initializer for an array of structures, each containing two
23919 strings, without the outermost pair of surrounding braces.
23921 The first item in the pair is the name of the default. This must
23922 match the code in `config.gcc' for the target. The second item is
23923 a spec to apply if a default with this name was specified. The
23924 string `%(VALUE)' in the spec will be replaced by the value of the
23925 default everywhere it occurs.
23927 The driver will apply these specs to its own command line between
23928 loading default `specs' files and processing `DRIVER_SELF_SPECS',
23929 using the same mechanism as `DRIVER_SELF_SPECS'.
23931 Do not define this macro if it does not need to do anything.
23934 A C string constant that tells the GCC driver program options to
23935 pass to CPP. It can also specify how to translate options you
23936 give to GCC into options for GCC to pass to the CPP.
23938 Do not define this macro if it does not need to do anything.
23940 -- Macro: CPLUSPLUS_CPP_SPEC
23941 This macro is just like `CPP_SPEC', but is used for C++, rather
23942 than C. If you do not define this macro, then the value of
23943 `CPP_SPEC' (if any) will be used instead.
23946 A C string constant that tells the GCC driver program options to
23947 pass to `cc1', `cc1plus', `f771', and the other language front
23948 ends. It can also specify how to translate options you give to
23949 GCC into options for GCC to pass to front ends.
23951 Do not define this macro if it does not need to do anything.
23953 -- Macro: CC1PLUS_SPEC
23954 A C string constant that tells the GCC driver program options to
23955 pass to `cc1plus'. It can also specify how to translate options
23956 you give to GCC into options for GCC to pass to the `cc1plus'.
23958 Do not define this macro if it does not need to do anything. Note
23959 that everything defined in CC1_SPEC is already passed to `cc1plus'
23960 so there is no need to duplicate the contents of CC1_SPEC in
23964 A C string constant that tells the GCC driver program options to
23965 pass to the assembler. It can also specify how to translate
23966 options you give to GCC into options for GCC to pass to the
23967 assembler. See the file `sun3.h' for an example of this.
23969 Do not define this macro if it does not need to do anything.
23971 -- Macro: ASM_FINAL_SPEC
23972 A C string constant that tells the GCC driver program how to run
23973 any programs which cleanup after the normal assembler. Normally,
23974 this is not needed. See the file `mips.h' for an example of this.
23976 Do not define this macro if it does not need to do anything.
23978 -- Macro: AS_NEEDS_DASH_FOR_PIPED_INPUT
23979 Define this macro, with no value, if the driver should give the
23980 assembler an argument consisting of a single dash, `-', to
23981 instruct it to read from its standard input (which will be a pipe
23982 connected to the output of the compiler proper). This argument is
23983 given after any `-o' option specifying the name of the output file.
23985 If you do not define this macro, the assembler is assumed to read
23986 its standard input if given no non-option arguments. If your
23987 assembler cannot read standard input at all, use a `%{pipe:%e}'
23988 construct; see `mips.h' for instance.
23990 -- Macro: LINK_SPEC
23991 A C string constant that tells the GCC driver program options to
23992 pass to the linker. It can also specify how to translate options
23993 you give to GCC into options for GCC to pass to the linker.
23995 Do not define this macro if it does not need to do anything.
23998 Another C string constant used much like `LINK_SPEC'. The
23999 difference between the two is that `LIB_SPEC' is used at the end
24000 of the command given to the linker.
24002 If this macro is not defined, a default is provided that loads the
24003 standard C library from the usual place. See `gcc.c'.
24005 -- Macro: LIBGCC_SPEC
24006 Another C string constant that tells the GCC driver program how
24007 and when to place a reference to `libgcc.a' into the linker
24008 command line. This constant is placed both before and after the
24009 value of `LIB_SPEC'.
24011 If this macro is not defined, the GCC driver provides a default
24012 that passes the string `-lgcc' to the linker.
24014 -- Macro: REAL_LIBGCC_SPEC
24015 By default, if `ENABLE_SHARED_LIBGCC' is defined, the
24016 `LIBGCC_SPEC' is not directly used by the driver program but is
24017 instead modified to refer to different versions of `libgcc.a'
24018 depending on the values of the command line flags `-static',
24019 `-shared', `-static-libgcc', and `-shared-libgcc'. On targets
24020 where these modifications are inappropriate, define
24021 `REAL_LIBGCC_SPEC' instead. `REAL_LIBGCC_SPEC' tells the driver
24022 how to place a reference to `libgcc' on the link command line,
24023 but, unlike `LIBGCC_SPEC', it is used unmodified.
24025 -- Macro: USE_LD_AS_NEEDED
24026 A macro that controls the modifications to `LIBGCC_SPEC' mentioned
24027 in `REAL_LIBGCC_SPEC'. If nonzero, a spec will be generated that
24028 uses -as-needed and the shared libgcc in place of the static
24029 exception handler library, when linking without any of `-static',
24030 `-static-libgcc', or `-shared-libgcc'.
24032 -- Macro: LINK_EH_SPEC
24033 If defined, this C string constant is added to `LINK_SPEC'. When
24034 `USE_LD_AS_NEEDED' is zero or undefined, it also affects the
24035 modifications to `LIBGCC_SPEC' mentioned in `REAL_LIBGCC_SPEC'.
24037 -- Macro: STARTFILE_SPEC
24038 Another C string constant used much like `LINK_SPEC'. The
24039 difference between the two is that `STARTFILE_SPEC' is used at the
24040 very beginning of the command given to the linker.
24042 If this macro is not defined, a default is provided that loads the
24043 standard C startup file from the usual place. See `gcc.c'.
24045 -- Macro: ENDFILE_SPEC
24046 Another C string constant used much like `LINK_SPEC'. The
24047 difference between the two is that `ENDFILE_SPEC' is used at the
24048 very end of the command given to the linker.
24050 Do not define this macro if it does not need to do anything.
24052 -- Macro: THREAD_MODEL_SPEC
24053 GCC `-v' will print the thread model GCC was configured to use.
24054 However, this doesn't work on platforms that are multilibbed on
24055 thread models, such as AIX 4.3. On such platforms, define
24056 `THREAD_MODEL_SPEC' such that it evaluates to a string without
24057 blanks that names one of the recognized thread models. `%*', the
24058 default value of this macro, will expand to the value of
24059 `thread_file' set in `config.gcc'.
24061 -- Macro: SYSROOT_SUFFIX_SPEC
24062 Define this macro to add a suffix to the target sysroot when GCC is
24063 configured with a sysroot. This will cause GCC to search for
24064 usr/lib, et al, within sysroot+suffix.
24066 -- Macro: SYSROOT_HEADERS_SUFFIX_SPEC
24067 Define this macro to add a headers_suffix to the target sysroot
24068 when GCC is configured with a sysroot. This will cause GCC to
24069 pass the updated sysroot+headers_suffix to CPP, causing it to
24070 search for usr/include, et al, within sysroot+headers_suffix.
24072 -- Macro: EXTRA_SPECS
24073 Define this macro to provide additional specifications to put in
24074 the `specs' file that can be used in various specifications like
24077 The definition should be an initializer for an array of structures,
24078 containing a string constant, that defines the specification name,
24079 and a string constant that provides the specification.
24081 Do not define this macro if it does not need to do anything.
24083 `EXTRA_SPECS' is useful when an architecture contains several
24084 related targets, which have various `..._SPECS' which are similar
24085 to each other, and the maintainer would like one central place to
24086 keep these definitions.
24088 For example, the PowerPC System V.4 targets use `EXTRA_SPECS' to
24089 define either `_CALL_SYSV' when the System V calling sequence is
24090 used or `_CALL_AIX' when the older AIX-based calling sequence is
24093 The `config/rs6000/rs6000.h' target file defines:
24095 #define EXTRA_SPECS \
24096 { "cpp_sysv_default", CPP_SYSV_DEFAULT },
24098 #define CPP_SYS_DEFAULT ""
24100 The `config/rs6000/sysv.h' target file defines:
24103 "%{posix: -D_POSIX_SOURCE } \
24104 %{mcall-sysv: -D_CALL_SYSV } \
24105 %{!mcall-sysv: %(cpp_sysv_default) } \
24106 %{msoft-float: -D_SOFT_FLOAT} %{mcpu=403: -D_SOFT_FLOAT}"
24108 #undef CPP_SYSV_DEFAULT
24109 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
24111 while the `config/rs6000/eabiaix.h' target file defines
24112 `CPP_SYSV_DEFAULT' as:
24114 #undef CPP_SYSV_DEFAULT
24115 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
24117 -- Macro: LINK_LIBGCC_SPECIAL_1
24118 Define this macro if the driver program should find the library
24119 `libgcc.a'. If you do not define this macro, the driver program
24120 will pass the argument `-lgcc' to tell the linker to do the search.
24122 -- Macro: LINK_GCC_C_SEQUENCE_SPEC
24123 The sequence in which libgcc and libc are specified to the linker.
24124 By default this is `%G %L %G'.
24126 -- Macro: LINK_COMMAND_SPEC
24127 A C string constant giving the complete command line need to
24128 execute the linker. When you do this, you will need to update
24129 your port each time a change is made to the link command line
24130 within `gcc.c'. Therefore, define this macro only if you need to
24131 completely redefine the command line for invoking the linker and
24132 there is no other way to accomplish the effect you need.
24133 Overriding this macro may be avoidable by overriding
24134 `LINK_GCC_C_SEQUENCE_SPEC' instead.
24136 -- Macro: LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
24137 A nonzero value causes `collect2' to remove duplicate
24138 `-LDIRECTORY' search directories from linking commands. Do not
24139 give it a nonzero value if removing duplicate search directories
24140 changes the linker's semantics.
24142 -- Macro: MULTILIB_DEFAULTS
24143 Define this macro as a C expression for the initializer of an
24144 array of string to tell the driver program which options are
24145 defaults for this target and thus do not need to be handled
24146 specially when using `MULTILIB_OPTIONS'.
24148 Do not define this macro if `MULTILIB_OPTIONS' is not defined in
24149 the target makefile fragment or if none of the options listed in
24150 `MULTILIB_OPTIONS' are set by default. *Note Target Fragment::.
24152 -- Macro: RELATIVE_PREFIX_NOT_LINKDIR
24153 Define this macro to tell `gcc' that it should only translate a
24154 `-B' prefix into a `-L' linker option if the prefix indicates an
24155 absolute file name.
24157 -- Macro: MD_EXEC_PREFIX
24158 If defined, this macro is an additional prefix to try after
24159 `STANDARD_EXEC_PREFIX'. `MD_EXEC_PREFIX' is not searched when the
24160 `-b' option is used, or the compiler is built as a cross compiler.
24161 If you define `MD_EXEC_PREFIX', then be sure to add it to the
24162 list of directories used to find the assembler in `configure.in'.
24164 -- Macro: STANDARD_STARTFILE_PREFIX
24165 Define this macro as a C string constant if you wish to override
24166 the standard choice of `libdir' as the default prefix to try when
24167 searching for startup files such as `crt0.o'.
24168 `STANDARD_STARTFILE_PREFIX' is not searched when the compiler is
24169 built as a cross compiler.
24171 -- Macro: STANDARD_STARTFILE_PREFIX_1
24172 Define this macro as a C string constant if you wish to override
24173 the standard choice of `/lib' as a prefix to try after the default
24174 prefix when searching for startup files such as `crt0.o'.
24175 `STANDARD_STARTFILE_PREFIX_1' is not searched when the compiler is
24176 built as a cross compiler.
24178 -- Macro: STANDARD_STARTFILE_PREFIX_2
24179 Define this macro as a C string constant if you wish to override
24180 the standard choice of `/lib' as yet another prefix to try after
24181 the default prefix when searching for startup files such as
24182 `crt0.o'. `STANDARD_STARTFILE_PREFIX_2' is not searched when the
24183 compiler is built as a cross compiler.
24185 -- Macro: MD_STARTFILE_PREFIX
24186 If defined, this macro supplies an additional prefix to try after
24187 the standard prefixes. `MD_EXEC_PREFIX' is not searched when the
24188 `-b' option is used, or when the compiler is built as a cross
24191 -- Macro: MD_STARTFILE_PREFIX_1
24192 If defined, this macro supplies yet another prefix to try after the
24193 standard prefixes. It is not searched when the `-b' option is
24194 used, or when the compiler is built as a cross compiler.
24196 -- Macro: INIT_ENVIRONMENT
24197 Define this macro as a C string constant if you wish to set
24198 environment variables for programs called by the driver, such as
24199 the assembler and loader. The driver passes the value of this
24200 macro to `putenv' to initialize the necessary environment
24203 -- Macro: LOCAL_INCLUDE_DIR
24204 Define this macro as a C string constant if you wish to override
24205 the standard choice of `/usr/local/include' as the default prefix
24206 to try when searching for local header files. `LOCAL_INCLUDE_DIR'
24207 comes before `SYSTEM_INCLUDE_DIR' in the search order.
24209 Cross compilers do not search either `/usr/local/include' or its
24212 -- Macro: MODIFY_TARGET_NAME
24213 Define this macro if you wish to define command-line switches that
24214 modify the default target name.
24216 For each switch, you can include a string to be appended to the
24217 first part of the configuration name or a string to be deleted
24218 from the configuration name, if present. The definition should be
24219 an initializer for an array of structures. Each array element
24220 should have three elements: the switch name (a string constant,
24221 including the initial dash), one of the enumeration codes `ADD' or
24222 `DELETE' to indicate whether the string should be inserted or
24223 deleted, and the string to be inserted or deleted (a string
24226 For example, on a machine where `64' at the end of the
24227 configuration name denotes a 64-bit target and you want the `-32'
24228 and `-64' switches to select between 32- and 64-bit targets, you
24231 #define MODIFY_TARGET_NAME \
24232 { { "-32", DELETE, "64"}, \
24233 {"-64", ADD, "64"}}
24235 -- Macro: SYSTEM_INCLUDE_DIR
24236 Define this macro as a C string constant if you wish to specify a
24237 system-specific directory to search for header files before the
24238 standard directory. `SYSTEM_INCLUDE_DIR' comes before
24239 `STANDARD_INCLUDE_DIR' in the search order.
24241 Cross compilers do not use this macro and do not search the
24242 directory specified.
24244 -- Macro: STANDARD_INCLUDE_DIR
24245 Define this macro as a C string constant if you wish to override
24246 the standard choice of `/usr/include' as the default prefix to try
24247 when searching for header files.
24249 Cross compilers ignore this macro and do not search either
24250 `/usr/include' or its replacement.
24252 -- Macro: STANDARD_INCLUDE_COMPONENT
24253 The "component" corresponding to `STANDARD_INCLUDE_DIR'. See
24254 `INCLUDE_DEFAULTS', below, for the description of components. If
24255 you do not define this macro, no component is used.
24257 -- Macro: INCLUDE_DEFAULTS
24258 Define this macro if you wish to override the entire default
24259 search path for include files. For a native compiler, the default
24260 search path usually consists of `GCC_INCLUDE_DIR',
24261 `LOCAL_INCLUDE_DIR', `SYSTEM_INCLUDE_DIR',
24262 `GPLUSPLUS_INCLUDE_DIR', and `STANDARD_INCLUDE_DIR'. In addition,
24263 `GPLUSPLUS_INCLUDE_DIR' and `GCC_INCLUDE_DIR' are defined
24264 automatically by `Makefile', and specify private search areas for
24265 GCC. The directory `GPLUSPLUS_INCLUDE_DIR' is used only for C++
24268 The definition should be an initializer for an array of structures.
24269 Each array element should have four elements: the directory name (a
24270 string constant), the component name (also a string constant), a
24271 flag for C++-only directories, and a flag showing that the
24272 includes in the directory don't need to be wrapped in `extern `C''
24273 when compiling C++. Mark the end of the array with a null element.
24275 The component name denotes what GNU package the include file is
24276 part of, if any, in all uppercase letters. For example, it might
24277 be `GCC' or `BINUTILS'. If the package is part of a
24278 vendor-supplied operating system, code the component name as `0'.
24280 For example, here is the definition used for VAX/VMS:
24282 #define INCLUDE_DEFAULTS \
24284 { "GNU_GXX_INCLUDE:", "G++", 1, 1}, \
24285 { "GNU_CC_INCLUDE:", "GCC", 0, 0}, \
24286 { "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0}, \
24291 Here is the order of prefixes tried for exec files:
24293 1. Any prefixes specified by the user with `-B'.
24295 2. The environment variable `GCC_EXEC_PREFIX' or, if `GCC_EXEC_PREFIX'
24296 is not set and the compiler has not been installed in the
24297 configure-time PREFIX, the location in which the compiler has
24298 actually been installed.
24300 3. The directories specified by the environment variable
24303 4. The macro `STANDARD_EXEC_PREFIX', if the compiler has been
24304 installed in the configured-time PREFIX.
24306 5. The location `/usr/libexec/gcc/', but only if this is a native
24309 6. The location `/usr/lib/gcc/', but only if this is a native
24312 7. The macro `MD_EXEC_PREFIX', if defined, but only if this is a
24315 Here is the order of prefixes tried for startfiles:
24317 1. Any prefixes specified by the user with `-B'.
24319 2. The environment variable `GCC_EXEC_PREFIX' or its automatically
24320 determined value based on the installed toolchain location.
24322 3. The directories specified by the environment variable
24323 `LIBRARY_PATH' (or port-specific name; native only, cross
24324 compilers do not use this).
24326 4. The macro `STANDARD_EXEC_PREFIX', but only if the toolchain is
24327 installed in the configured PREFIX or this is a native compiler.
24329 5. The location `/usr/lib/gcc/', but only if this is a native
24332 6. The macro `MD_EXEC_PREFIX', if defined, but only if this is a
24335 7. The macro `MD_STARTFILE_PREFIX', if defined, but only if this is a
24336 native compiler, or we have a target system root.
24338 8. The macro `MD_STARTFILE_PREFIX_1', if defined, but only if this is
24339 a native compiler, or we have a target system root.
24341 9. The macro `STANDARD_STARTFILE_PREFIX', with any sysroot
24342 modifications. If this path is relative it will be prefixed by
24343 `GCC_EXEC_PREFIX' and the machine suffix or `STANDARD_EXEC_PREFIX'
24344 and the machine suffix.
24346 10. The macro `STANDARD_STARTFILE_PREFIX_1', but only if this is a
24347 native compiler, or we have a target system root. The default for
24348 this macro is `/lib/'.
24350 11. The macro `STANDARD_STARTFILE_PREFIX_2', but only if this is a
24351 native compiler, or we have a target system root. The default for
24352 this macro is `/usr/lib/'.
24355 File: gccint.info, Node: Run-time Target, Next: Per-Function Data, Prev: Driver, Up: Target Macros
24357 17.3 Run-time Target Specification
24358 ==================================
24360 Here are run-time target specifications.
24362 -- Macro: TARGET_CPU_CPP_BUILTINS ()
24363 This function-like macro expands to a block of code that defines
24364 built-in preprocessor macros and assertions for the target CPU,
24365 using the functions `builtin_define', `builtin_define_std' and
24366 `builtin_assert'. When the front end calls this macro it provides
24367 a trailing semicolon, and since it has finished command line
24368 option processing your code can use those results freely.
24370 `builtin_assert' takes a string in the form you pass to the
24371 command-line option `-A', such as `cpu=mips', and creates the
24372 assertion. `builtin_define' takes a string in the form accepted
24373 by option `-D' and unconditionally defines the macro.
24375 `builtin_define_std' takes a string representing the name of an
24376 object-like macro. If it doesn't lie in the user's namespace,
24377 `builtin_define_std' defines it unconditionally. Otherwise, it
24378 defines a version with two leading underscores, and another version
24379 with two leading and trailing underscores, and defines the original
24380 only if an ISO standard was not requested on the command line. For
24381 example, passing `unix' defines `__unix', `__unix__' and possibly
24382 `unix'; passing `_mips' defines `__mips', `__mips__' and possibly
24383 `_mips', and passing `_ABI64' defines only `_ABI64'.
24385 You can also test for the C dialect being compiled. The variable
24386 `c_language' is set to one of `clk_c', `clk_cplusplus' or
24387 `clk_objective_c'. Note that if we are preprocessing assembler,
24388 this variable will be `clk_c' but the function-like macro
24389 `preprocessing_asm_p()' will return true, so you might want to
24390 check for that first. If you need to check for strict ANSI, the
24391 variable `flag_iso' can be used. The function-like macro
24392 `preprocessing_trad_p()' can be used to check for traditional
24395 -- Macro: TARGET_OS_CPP_BUILTINS ()
24396 Similarly to `TARGET_CPU_CPP_BUILTINS' but this macro is optional
24397 and is used for the target operating system instead.
24399 -- Macro: TARGET_OBJFMT_CPP_BUILTINS ()
24400 Similarly to `TARGET_CPU_CPP_BUILTINS' but this macro is optional
24401 and is used for the target object format. `elfos.h' uses this
24402 macro to define `__ELF__', so you probably do not need to define
24405 -- Variable: extern int target_flags
24406 This variable is declared in `options.h', which is included before
24407 any target-specific headers.
24409 -- Variable: Target Hook int TARGET_DEFAULT_TARGET_FLAGS
24410 This variable specifies the initial value of `target_flags'. Its
24411 default setting is 0.
24413 -- Target Hook: bool TARGET_HANDLE_OPTION (size_t CODE, const char
24415 This hook is called whenever the user specifies one of the
24416 target-specific options described by the `.opt' definition files
24417 (*note Options::). It has the opportunity to do some
24418 option-specific processing and should return true if the option is
24419 valid. The default definition does nothing but return true.
24421 CODE specifies the `OPT_NAME' enumeration value associated with
24422 the selected option; NAME is just a rendering of the option name
24423 in which non-alphanumeric characters are replaced by underscores.
24424 ARG specifies the string argument and is null if no argument was
24425 given. If the option is flagged as a `UInteger' (*note Option
24426 properties::), VALUE is the numeric value of the argument.
24427 Otherwise VALUE is 1 if the positive form of the option was used
24428 and 0 if the "no-" form was.
24430 -- Target Hook: bool TARGET_HANDLE_C_OPTION (size_t CODE, const char
24432 This target hook is called whenever the user specifies one of the
24433 target-specific C language family options described by the `.opt'
24434 definition files(*note Options::). It has the opportunity to do
24435 some option-specific processing and should return true if the
24436 option is valid. The default definition does nothing but return
24439 In general, you should use `TARGET_HANDLE_OPTION' to handle
24440 options. However, if processing an option requires routines that
24441 are only available in the C (and related language) front ends,
24442 then you should use `TARGET_HANDLE_C_OPTION' instead.
24444 -- Macro: TARGET_VERSION
24445 This macro is a C statement to print on `stderr' a string
24446 describing the particular machine description choice. Every
24447 machine description should define `TARGET_VERSION'. For example:
24450 #define TARGET_VERSION \
24451 fprintf (stderr, " (68k, Motorola syntax)");
24453 #define TARGET_VERSION \
24454 fprintf (stderr, " (68k, MIT syntax)");
24457 -- Macro: OVERRIDE_OPTIONS
24458 Sometimes certain combinations of command options do not make
24459 sense on a particular target machine. You can define a macro
24460 `OVERRIDE_OPTIONS' to take account of this. This macro, if
24461 defined, is executed once just after all the command options have
24464 Don't use this macro to turn on various extra optimizations for
24465 `-O'. That is what `OPTIMIZATION_OPTIONS' is for.
24467 -- Macro: C_COMMON_OVERRIDE_OPTIONS
24468 This is similar to `OVERRIDE_OPTIONS' but is only used in the C
24469 language frontends (C, Objective-C, C++, Objective-C++) and so can
24470 be used to alter option flag variables which only exist in those
24473 -- Macro: OPTIMIZATION_OPTIONS (LEVEL, SIZE)
24474 Some machines may desire to change what optimizations are
24475 performed for various optimization levels. This macro, if
24476 defined, is executed once just after the optimization level is
24477 determined and before the remainder of the command options have
24478 been parsed. Values set in this macro are used as the default
24479 values for the other command line options.
24481 LEVEL is the optimization level specified; 2 if `-O2' is
24482 specified, 1 if `-O' is specified, and 0 if neither is specified.
24484 SIZE is nonzero if `-Os' is specified and zero otherwise.
24486 This macro is run once at program startup and when the optimization
24487 options are changed via `#pragma GCC optimize' or by using the
24488 `optimize' attribute.
24490 *Do not examine `write_symbols' in this macro!* The debugging
24491 options are not supposed to alter the generated code.
24493 -- Target Hook: bool TARGET_HELP (void)
24494 This hook is called in response to the user invoking
24495 `--target-help' on the command line. It gives the target a chance
24496 to display extra information on the target specific command line
24497 options found in its `.opt' file.
24499 -- Macro: CAN_DEBUG_WITHOUT_FP
24500 Define this macro if debugging can be performed even without a
24501 frame pointer. If this macro is defined, GCC will turn on the
24502 `-fomit-frame-pointer' option whenever `-O' is specified.
24505 File: gccint.info, Node: Per-Function Data, Next: Storage Layout, Prev: Run-time Target, Up: Target Macros
24507 17.4 Defining data structures for per-function information.
24508 ===========================================================
24510 If the target needs to store information on a per-function basis, GCC
24511 provides a macro and a couple of variables to allow this. Note, just
24512 using statics to store the information is a bad idea, since GCC supports
24513 nested functions, so you can be halfway through encoding one function
24514 when another one comes along.
24516 GCC defines a data structure called `struct function' which contains
24517 all of the data specific to an individual function. This structure
24518 contains a field called `machine' whose type is `struct
24519 machine_function *', which can be used by targets to point to their own
24522 If a target needs per-function specific data it should define the type
24523 `struct machine_function' and also the macro `INIT_EXPANDERS'. This
24524 macro should be used to initialize the function pointer
24525 `init_machine_status'. This pointer is explained below.
24527 One typical use of per-function, target specific data is to create an
24528 RTX to hold the register containing the function's return address. This
24529 RTX can then be used to implement the `__builtin_return_address'
24530 function, for level 0.
24532 Note--earlier implementations of GCC used a single data area to hold
24533 all of the per-function information. Thus when processing of a nested
24534 function began the old per-function data had to be pushed onto a stack,
24535 and when the processing was finished, it had to be popped off the
24536 stack. GCC used to provide function pointers called
24537 `save_machine_status' and `restore_machine_status' to handle the saving
24538 and restoring of the target specific information. Since the single
24539 data area approach is no longer used, these pointers are no longer
24542 -- Macro: INIT_EXPANDERS
24543 Macro called to initialize any target specific information. This
24544 macro is called once per function, before generation of any RTL
24545 has begun. The intention of this macro is to allow the
24546 initialization of the function pointer `init_machine_status'.
24548 -- Variable: void (*)(struct function *) init_machine_status
24549 If this function pointer is non-`NULL' it will be called once per
24550 function, before function compilation starts, in order to allow the
24551 target to perform any target specific initialization of the
24552 `struct function' structure. It is intended that this would be
24553 used to initialize the `machine' of that structure.
24555 `struct machine_function' structures are expected to be freed by
24556 GC. Generally, any memory that they reference must be allocated
24557 by using `ggc_alloc', including the structure itself.
24560 File: gccint.info, Node: Storage Layout, Next: Type Layout, Prev: Per-Function Data, Up: Target Macros
24562 17.5 Storage Layout
24563 ===================
24565 Note that the definitions of the macros in this table which are sizes or
24566 alignments measured in bits do not need to be constant. They can be C
24567 expressions that refer to static variables, such as the `target_flags'.
24568 *Note Run-time Target::.
24570 -- Macro: BITS_BIG_ENDIAN
24571 Define this macro to have the value 1 if the most significant bit
24572 in a byte has the lowest number; otherwise define it to have the
24573 value zero. This means that bit-field instructions count from the
24574 most significant bit. If the machine has no bit-field
24575 instructions, then this must still be defined, but it doesn't
24576 matter which value it is defined to. This macro need not be a
24579 This macro does not affect the way structure fields are packed into
24580 bytes or words; that is controlled by `BYTES_BIG_ENDIAN'.
24582 -- Macro: BYTES_BIG_ENDIAN
24583 Define this macro to have the value 1 if the most significant byte
24584 in a word has the lowest number. This macro need not be a
24587 -- Macro: WORDS_BIG_ENDIAN
24588 Define this macro to have the value 1 if, in a multiword object,
24589 the most significant word has the lowest number. This applies to
24590 both memory locations and registers; GCC fundamentally assumes
24591 that the order of words in memory is the same as the order in
24592 registers. This macro need not be a constant.
24594 -- Macro: LIBGCC2_WORDS_BIG_ENDIAN
24595 Define this macro if `WORDS_BIG_ENDIAN' is not constant. This
24596 must be a constant value with the same meaning as
24597 `WORDS_BIG_ENDIAN', which will be used only when compiling
24598 `libgcc2.c'. Typically the value will be set based on
24599 preprocessor defines.
24601 -- Macro: FLOAT_WORDS_BIG_ENDIAN
24602 Define this macro to have the value 1 if `DFmode', `XFmode' or
24603 `TFmode' floating point numbers are stored in memory with the word
24604 containing the sign bit at the lowest address; otherwise define it
24605 to have the value 0. This macro need not be a constant.
24607 You need not define this macro if the ordering is the same as for
24608 multi-word integers.
24610 -- Macro: BITS_PER_UNIT
24611 Define this macro to be the number of bits in an addressable
24612 storage unit (byte). If you do not define this macro the default
24615 -- Macro: BITS_PER_WORD
24616 Number of bits in a word. If you do not define this macro, the
24617 default is `BITS_PER_UNIT * UNITS_PER_WORD'.
24619 -- Macro: MAX_BITS_PER_WORD
24620 Maximum number of bits in a word. If this is undefined, the
24621 default is `BITS_PER_WORD'. Otherwise, it is the constant value
24622 that is the largest value that `BITS_PER_WORD' can have at
24625 -- Macro: UNITS_PER_WORD
24626 Number of storage units in a word; normally the size of a
24627 general-purpose register, a power of two from 1 or 8.
24629 -- Macro: MIN_UNITS_PER_WORD
24630 Minimum number of units in a word. If this is undefined, the
24631 default is `UNITS_PER_WORD'. Otherwise, it is the constant value
24632 that is the smallest value that `UNITS_PER_WORD' can have at
24635 -- Macro: UNITS_PER_SIMD_WORD (MODE)
24636 Number of units in the vectors that the vectorizer can produce for
24637 scalar mode MODE. The default is equal to `UNITS_PER_WORD',
24638 because the vectorizer can do some transformations even in absence
24639 of specialized SIMD hardware.
24641 -- Macro: POINTER_SIZE
24642 Width of a pointer, in bits. You must specify a value no wider
24643 than the width of `Pmode'. If it is not equal to the width of
24644 `Pmode', you must define `POINTERS_EXTEND_UNSIGNED'. If you do
24645 not specify a value the default is `BITS_PER_WORD'.
24647 -- Macro: POINTERS_EXTEND_UNSIGNED
24648 A C expression that determines how pointers should be extended from
24649 `ptr_mode' to either `Pmode' or `word_mode'. It is greater than
24650 zero if pointers should be zero-extended, zero if they should be
24651 sign-extended, and negative if some other sort of conversion is
24652 needed. In the last case, the extension is done by the target's
24653 `ptr_extend' instruction.
24655 You need not define this macro if the `ptr_mode', `Pmode' and
24656 `word_mode' are all the same width.
24658 -- Macro: PROMOTE_MODE (M, UNSIGNEDP, TYPE)
24659 A macro to update M and UNSIGNEDP when an object whose type is
24660 TYPE and which has the specified mode and signedness is to be
24661 stored in a register. This macro is only called when TYPE is a
24664 On most RISC machines, which only have operations that operate on
24665 a full register, define this macro to set M to `word_mode' if M is
24666 an integer mode narrower than `BITS_PER_WORD'. In most cases,
24667 only integer modes should be widened because wider-precision
24668 floating-point operations are usually more expensive than their
24669 narrower counterparts.
24671 For most machines, the macro definition does not change UNSIGNEDP.
24672 However, some machines, have instructions that preferentially
24673 handle either signed or unsigned quantities of certain modes. For
24674 example, on the DEC Alpha, 32-bit loads from memory and 32-bit add
24675 instructions sign-extend the result to 64 bits. On such machines,
24676 set UNSIGNEDP according to which kind of extension is more
24679 Do not define this macro if it would never modify M.
24681 -- Macro: PROMOTE_FUNCTION_MODE
24682 Like `PROMOTE_MODE', but is applied to outgoing function arguments
24683 or function return values, as specified by
24684 `TARGET_PROMOTE_FUNCTION_ARGS' and
24685 `TARGET_PROMOTE_FUNCTION_RETURN', respectively.
24687 The default is `PROMOTE_MODE'.
24689 -- Target Hook: bool TARGET_PROMOTE_FUNCTION_ARGS (tree FNTYPE)
24690 This target hook should return `true' if the promotion described by
24691 `PROMOTE_FUNCTION_MODE' should be done for outgoing function
24694 -- Target Hook: bool TARGET_PROMOTE_FUNCTION_RETURN (tree FNTYPE)
24695 This target hook should return `true' if the promotion described by
24696 `PROMOTE_FUNCTION_MODE' should be done for the return value of
24699 If this target hook returns `true', `TARGET_FUNCTION_VALUE' must
24700 perform the same promotions done by `PROMOTE_FUNCTION_MODE'.
24702 -- Macro: PARM_BOUNDARY
24703 Normal alignment required for function parameters on the stack, in
24704 bits. All stack parameters receive at least this much alignment
24705 regardless of data type. On most machines, this is the same as the
24706 size of an integer.
24708 -- Macro: STACK_BOUNDARY
24709 Define this macro to the minimum alignment enforced by hardware
24710 for the stack pointer on this machine. The definition is a C
24711 expression for the desired alignment (measured in bits). This
24712 value is used as a default if `PREFERRED_STACK_BOUNDARY' is not
24713 defined. On most machines, this should be the same as
24716 -- Macro: PREFERRED_STACK_BOUNDARY
24717 Define this macro if you wish to preserve a certain alignment for
24718 the stack pointer, greater than what the hardware enforces. The
24719 definition is a C expression for the desired alignment (measured
24720 in bits). This macro must evaluate to a value equal to or larger
24721 than `STACK_BOUNDARY'.
24723 -- Macro: INCOMING_STACK_BOUNDARY
24724 Define this macro if the incoming stack boundary may be different
24725 from `PREFERRED_STACK_BOUNDARY'. This macro must evaluate to a
24726 value equal to or larger than `STACK_BOUNDARY'.
24728 -- Macro: FUNCTION_BOUNDARY
24729 Alignment required for a function entry point, in bits.
24731 -- Macro: BIGGEST_ALIGNMENT
24732 Biggest alignment that any data type can require on this machine,
24733 in bits. Note that this is not the biggest alignment that is
24734 supported, just the biggest alignment that, when violated, may
24737 -- Macro: MALLOC_ABI_ALIGNMENT
24738 Alignment, in bits, a C conformant malloc implementation has to
24739 provide. If not defined, the default value is `BITS_PER_WORD'.
24741 -- Macro: ATTRIBUTE_ALIGNED_VALUE
24742 Alignment used by the `__attribute__ ((aligned))' construct. If
24743 not defined, the default value is `BIGGEST_ALIGNMENT'.
24745 -- Macro: MINIMUM_ATOMIC_ALIGNMENT
24746 If defined, the smallest alignment, in bits, that can be given to
24747 an object that can be referenced in one operation, without
24748 disturbing any nearby object. Normally, this is `BITS_PER_UNIT',
24749 but may be larger on machines that don't have byte or half-word
24752 -- Macro: BIGGEST_FIELD_ALIGNMENT
24753 Biggest alignment that any structure or union field can require on
24754 this machine, in bits. If defined, this overrides
24755 `BIGGEST_ALIGNMENT' for structure and union fields only, unless
24756 the field alignment has been set by the `__attribute__ ((aligned
24759 -- Macro: ADJUST_FIELD_ALIGN (FIELD, COMPUTED)
24760 An expression for the alignment of a structure field FIELD if the
24761 alignment computed in the usual way (including applying of
24762 `BIGGEST_ALIGNMENT' and `BIGGEST_FIELD_ALIGNMENT' to the
24763 alignment) is COMPUTED. It overrides alignment only if the field
24764 alignment has not been set by the `__attribute__ ((aligned (N)))'
24767 -- Macro: MAX_STACK_ALIGNMENT
24768 Biggest stack alignment guaranteed by the backend. Use this macro
24769 to specify the maximum alignment of a variable on stack.
24771 If not defined, the default value is `STACK_BOUNDARY'.
24774 -- Macro: MAX_OFILE_ALIGNMENT
24775 Biggest alignment supported by the object file format of this
24776 machine. Use this macro to limit the alignment which can be
24777 specified using the `__attribute__ ((aligned (N)))' construct. If
24778 not defined, the default value is `BIGGEST_ALIGNMENT'.
24780 On systems that use ELF, the default (in `config/elfos.h') is the
24781 largest supported 32-bit ELF section alignment representable on a
24782 32-bit host e.g. `(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)'. On
24783 32-bit ELF the largest supported section alignment in bits is
24784 `(0x80000000 * 8)', but this is not representable on 32-bit hosts.
24786 -- Macro: DATA_ALIGNMENT (TYPE, BASIC-ALIGN)
24787 If defined, a C expression to compute the alignment for a variable
24788 in the static store. TYPE is the data type, and BASIC-ALIGN is
24789 the alignment that the object would ordinarily have. The value of
24790 this macro is used instead of that alignment to align the object.
24792 If this macro is not defined, then BASIC-ALIGN is used.
24794 One use of this macro is to increase alignment of medium-size data
24795 to make it all fit in fewer cache lines. Another is to cause
24796 character arrays to be word-aligned so that `strcpy' calls that
24797 copy constants to character arrays can be done inline.
24799 -- Macro: CONSTANT_ALIGNMENT (CONSTANT, BASIC-ALIGN)
24800 If defined, a C expression to compute the alignment given to a
24801 constant that is being placed in memory. CONSTANT is the constant
24802 and BASIC-ALIGN is the alignment that the object would ordinarily
24803 have. The value of this macro is used instead of that alignment to
24806 If this macro is not defined, then BASIC-ALIGN is used.
24808 The typical use of this macro is to increase alignment for string
24809 constants to be word aligned so that `strcpy' calls that copy
24810 constants can be done inline.
24812 -- Macro: LOCAL_ALIGNMENT (TYPE, BASIC-ALIGN)
24813 If defined, a C expression to compute the alignment for a variable
24814 in the local store. TYPE is the data type, and BASIC-ALIGN is the
24815 alignment that the object would ordinarily have. The value of this
24816 macro is used instead of that alignment to align the object.
24818 If this macro is not defined, then BASIC-ALIGN is used.
24820 One use of this macro is to increase alignment of medium-size data
24821 to make it all fit in fewer cache lines.
24823 -- Macro: STACK_SLOT_ALIGNMENT (TYPE, MODE, BASIC-ALIGN)
24824 If defined, a C expression to compute the alignment for stack slot.
24825 TYPE is the data type, MODE is the widest mode available, and
24826 BASIC-ALIGN is the alignment that the slot would ordinarily have.
24827 The value of this macro is used instead of that alignment to align
24830 If this macro is not defined, then BASIC-ALIGN is used when TYPE
24831 is `NULL'. Otherwise, `LOCAL_ALIGNMENT' will be used.
24833 This macro is to set alignment of stack slot to the maximum
24834 alignment of all possible modes which the slot may have.
24836 -- Macro: LOCAL_DECL_ALIGNMENT (DECL)
24837 If defined, a C expression to compute the alignment for a local
24840 If this macro is not defined, then `LOCAL_ALIGNMENT (TREE_TYPE
24841 (DECL), DECL_ALIGN (DECL))' is used.
24843 One use of this macro is to increase alignment of medium-size data
24844 to make it all fit in fewer cache lines.
24846 -- Macro: MINIMUM_ALIGNMENT (EXP, MODE, ALIGN)
24847 If defined, a C expression to compute the minimum required
24848 alignment for dynamic stack realignment purposes for EXP (a type
24849 or decl), MODE, assuming normal alignment ALIGN.
24851 If this macro is not defined, then ALIGN will be used.
24853 -- Macro: EMPTY_FIELD_BOUNDARY
24854 Alignment in bits to be given to a structure bit-field that
24855 follows an empty field such as `int : 0;'.
24857 If `PCC_BITFIELD_TYPE_MATTERS' is true, it overrides this macro.
24859 -- Macro: STRUCTURE_SIZE_BOUNDARY
24860 Number of bits which any structure or union's size must be a
24861 multiple of. Each structure or union's size is rounded up to a
24864 If you do not define this macro, the default is the same as
24867 -- Macro: STRICT_ALIGNMENT
24868 Define this macro to be the value 1 if instructions will fail to
24869 work if given data not on the nominal alignment. If instructions
24870 will merely go slower in that case, define this macro as 0.
24872 -- Macro: PCC_BITFIELD_TYPE_MATTERS
24873 Define this if you wish to imitate the way many other C compilers
24874 handle alignment of bit-fields and the structures that contain
24877 The behavior is that the type written for a named bit-field (`int',
24878 `short', or other integer type) imposes an alignment for the entire
24879 structure, as if the structure really did contain an ordinary
24880 field of that type. In addition, the bit-field is placed within
24881 the structure so that it would fit within such a field, not
24882 crossing a boundary for it.
24884 Thus, on most machines, a named bit-field whose type is written as
24885 `int' would not cross a four-byte boundary, and would force
24886 four-byte alignment for the whole structure. (The alignment used
24887 may not be four bytes; it is controlled by the other alignment
24890 An unnamed bit-field will not affect the alignment of the
24891 containing structure.
24893 If the macro is defined, its definition should be a C expression;
24894 a nonzero value for the expression enables this behavior.
24896 Note that if this macro is not defined, or its value is zero, some
24897 bit-fields may cross more than one alignment boundary. The
24898 compiler can support such references if there are `insv', `extv',
24899 and `extzv' insns that can directly reference memory.
24901 The other known way of making bit-fields work is to define
24902 `STRUCTURE_SIZE_BOUNDARY' as large as `BIGGEST_ALIGNMENT'. Then
24903 every structure can be accessed with fullwords.
24905 Unless the machine has bit-field instructions or you define
24906 `STRUCTURE_SIZE_BOUNDARY' that way, you must define
24907 `PCC_BITFIELD_TYPE_MATTERS' to have a nonzero value.
24909 If your aim is to make GCC use the same conventions for laying out
24910 bit-fields as are used by another compiler, here is how to
24911 investigate what the other compiler does. Compile and run this
24930 printf ("Size of foo1 is %d\n",
24931 sizeof (struct foo1));
24932 printf ("Size of foo2 is %d\n",
24933 sizeof (struct foo2));
24937 If this prints 2 and 5, then the compiler's behavior is what you
24938 would get from `PCC_BITFIELD_TYPE_MATTERS'.
24940 -- Macro: BITFIELD_NBYTES_LIMITED
24941 Like `PCC_BITFIELD_TYPE_MATTERS' except that its effect is limited
24942 to aligning a bit-field within the structure.
24944 -- Target Hook: bool TARGET_ALIGN_ANON_BITFIELD (void)
24945 When `PCC_BITFIELD_TYPE_MATTERS' is true this hook will determine
24946 whether unnamed bitfields affect the alignment of the containing
24947 structure. The hook should return true if the structure should
24948 inherit the alignment requirements of an unnamed bitfield's type.
24950 -- Target Hook: bool TARGET_NARROW_VOLATILE_BITFIELD (void)
24951 This target hook should return `true' if accesses to volatile
24952 bitfields should use the narrowest mode possible. It should
24953 return `false' if these accesses should use the bitfield container
24956 The default is `!TARGET_STRICT_ALIGN'.
24958 -- Macro: MEMBER_TYPE_FORCES_BLK (FIELD, MODE)
24959 Return 1 if a structure or array containing FIELD should be
24960 accessed using `BLKMODE'.
24962 If FIELD is the only field in the structure, MODE is its mode,
24963 otherwise MODE is VOIDmode. MODE is provided in the case where
24964 structures of one field would require the structure's mode to
24965 retain the field's mode.
24967 Normally, this is not needed.
24969 -- Macro: ROUND_TYPE_ALIGN (TYPE, COMPUTED, SPECIFIED)
24970 Define this macro as an expression for the alignment of a type
24971 (given by TYPE as a tree node) if the alignment computed in the
24972 usual way is COMPUTED and the alignment explicitly specified was
24975 The default is to use SPECIFIED if it is larger; otherwise, use
24976 the smaller of COMPUTED and `BIGGEST_ALIGNMENT'
24978 -- Macro: MAX_FIXED_MODE_SIZE
24979 An integer expression for the size in bits of the largest integer
24980 machine mode that should actually be used. All integer machine
24981 modes of this size or smaller can be used for structures and
24982 unions with the appropriate sizes. If this macro is undefined,
24983 `GET_MODE_BITSIZE (DImode)' is assumed.
24985 -- Macro: STACK_SAVEAREA_MODE (SAVE_LEVEL)
24986 If defined, an expression of type `enum machine_mode' that
24987 specifies the mode of the save area operand of a
24988 `save_stack_LEVEL' named pattern (*note Standard Names::).
24989 SAVE_LEVEL is one of `SAVE_BLOCK', `SAVE_FUNCTION', or
24990 `SAVE_NONLOCAL' and selects which of the three named patterns is
24991 having its mode specified.
24993 You need not define this macro if it always returns `Pmode'. You
24994 would most commonly define this macro if the `save_stack_LEVEL'
24995 patterns need to support both a 32- and a 64-bit mode.
24997 -- Macro: STACK_SIZE_MODE
24998 If defined, an expression of type `enum machine_mode' that
24999 specifies the mode of the size increment operand of an
25000 `allocate_stack' named pattern (*note Standard Names::).
25002 You need not define this macro if it always returns `word_mode'.
25003 You would most commonly define this macro if the `allocate_stack'
25004 pattern needs to support both a 32- and a 64-bit mode.
25006 -- Target Hook: enum machine_mode TARGET_LIBGCC_CMP_RETURN_MODE ()
25007 This target hook should return the mode to be used for the return
25008 value of compare instructions expanded to libgcc calls. If not
25009 defined `word_mode' is returned which is the right choice for a
25010 majority of targets.
25012 -- Target Hook: enum machine_mode TARGET_LIBGCC_SHIFT_COUNT_MODE ()
25013 This target hook should return the mode to be used for the shift
25014 count operand of shift instructions expanded to libgcc calls. If
25015 not defined `word_mode' is returned which is the right choice for
25016 a majority of targets.
25018 -- Macro: ROUND_TOWARDS_ZERO
25019 If defined, this macro should be true if the prevailing rounding
25020 mode is towards zero.
25022 Defining this macro only affects the way `libgcc.a' emulates
25023 floating-point arithmetic.
25025 Not defining this macro is equivalent to returning zero.
25027 -- Macro: LARGEST_EXPONENT_IS_NORMAL (SIZE)
25028 This macro should return true if floats with SIZE bits do not have
25029 a NaN or infinity representation, but use the largest exponent for
25030 normal numbers instead.
25032 Defining this macro only affects the way `libgcc.a' emulates
25033 floating-point arithmetic.
25035 The default definition of this macro returns false for all sizes.
25037 -- Target Hook: bool TARGET_VECTOR_OPAQUE_P (tree TYPE)
25038 This target hook should return `true' a vector is opaque. That
25039 is, if no cast is needed when copying a vector value of type TYPE
25040 into another vector lvalue of the same size. Vector opaque types
25041 cannot be initialized. The default is that there are no such
25044 -- Target Hook: bool TARGET_MS_BITFIELD_LAYOUT_P (tree RECORD_TYPE)
25045 This target hook returns `true' if bit-fields in the given
25046 RECORD_TYPE are to be laid out following the rules of Microsoft
25047 Visual C/C++, namely: (i) a bit-field won't share the same storage
25048 unit with the previous bit-field if their underlying types have
25049 different sizes, and the bit-field will be aligned to the highest
25050 alignment of the underlying types of itself and of the previous
25051 bit-field; (ii) a zero-sized bit-field will affect the alignment of
25052 the whole enclosing structure, even if it is unnamed; except that
25053 (iii) a zero-sized bit-field will be disregarded unless it follows
25054 another bit-field of nonzero size. If this hook returns `true',
25055 other macros that control bit-field layout are ignored.
25057 When a bit-field is inserted into a packed record, the whole size
25058 of the underlying type is used by one or more same-size adjacent
25059 bit-fields (that is, if its long:3, 32 bits is used in the record,
25060 and any additional adjacent long bit-fields are packed into the
25061 same chunk of 32 bits. However, if the size changes, a new field
25062 of that size is allocated). In an unpacked record, this is the
25063 same as using alignment, but not equivalent when packing.
25065 If both MS bit-fields and `__attribute__((packed))' are used, the
25066 latter will take precedence. If `__attribute__((packed))' is used
25067 on a single field when MS bit-fields are in use, it will take
25068 precedence for that field, but the alignment of the rest of the
25069 structure may affect its placement.
25071 -- Target Hook: bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
25072 Returns true if the target supports decimal floating point.
25074 -- Target Hook: bool TARGET_FIXED_POINT_SUPPORTED_P (void)
25075 Returns true if the target supports fixed-point arithmetic.
25077 -- Target Hook: void TARGET_EXPAND_TO_RTL_HOOK (void)
25078 This hook is called just before expansion into rtl, allowing the
25079 target to perform additional initializations or analysis before
25080 the expansion. For example, the rs6000 port uses it to allocate a
25081 scratch stack slot for use in copying SDmode values between memory
25082 and floating point registers whenever the function being expanded
25083 has any SDmode usage.
25085 -- Target Hook: void TARGET_INSTANTIATE_DECLS (void)
25086 This hook allows the backend to perform additional instantiations
25087 on rtl that are not actually in any insns yet, but will be later.
25089 -- Target Hook: const char * TARGET_MANGLE_TYPE (tree TYPE)
25090 If your target defines any fundamental types, or any types your
25091 target uses should be mangled differently from the default, define
25092 this hook to return the appropriate encoding for these types as
25093 part of a C++ mangled name. The TYPE argument is the tree
25094 structure representing the type to be mangled. The hook may be
25095 applied to trees which are not target-specific fundamental types;
25096 it should return `NULL' for all such types, as well as arguments
25097 it does not recognize. If the return value is not `NULL', it must
25098 point to a statically-allocated string constant.
25100 Target-specific fundamental types might be new fundamental types or
25101 qualified versions of ordinary fundamental types. Encode new
25102 fundamental types as `u N NAME', where NAME is the name used for
25103 the type in source code, and N is the length of NAME in decimal.
25104 Encode qualified versions of ordinary types as `U N NAME CODE',
25105 where NAME is the name used for the type qualifier in source code,
25106 N is the length of NAME as above, and CODE is the code used to
25107 represent the unqualified version of this type. (See
25108 `write_builtin_type' in `cp/mangle.c' for the list of codes.) In
25109 both cases the spaces are for clarity; do not include any spaces
25112 This hook is applied to types prior to typedef resolution. If the
25113 mangled name for a particular type depends only on that type's
25114 main variant, you can perform typedef resolution yourself using
25115 `TYPE_MAIN_VARIANT' before mangling.
25117 The default version of this hook always returns `NULL', which is
25118 appropriate for a target that does not define any new fundamental
25122 File: gccint.info, Node: Type Layout, Next: Registers, Prev: Storage Layout, Up: Target Macros
25124 17.6 Layout of Source Language Data Types
25125 =========================================
25127 These macros define the sizes and other characteristics of the standard
25128 basic data types used in programs being compiled. Unlike the macros in
25129 the previous section, these apply to specific features of C and related
25130 languages, rather than to fundamental aspects of storage layout.
25132 -- Macro: INT_TYPE_SIZE
25133 A C expression for the size in bits of the type `int' on the
25134 target machine. If you don't define this, the default is one word.
25136 -- Macro: SHORT_TYPE_SIZE
25137 A C expression for the size in bits of the type `short' on the
25138 target machine. If you don't define this, the default is half a
25139 word. (If this would be less than one storage unit, it is rounded
25142 -- Macro: LONG_TYPE_SIZE
25143 A C expression for the size in bits of the type `long' on the
25144 target machine. If you don't define this, the default is one word.
25146 -- Macro: ADA_LONG_TYPE_SIZE
25147 On some machines, the size used for the Ada equivalent of the type
25148 `long' by a native Ada compiler differs from that used by C. In
25149 that situation, define this macro to be a C expression to be used
25150 for the size of that type. If you don't define this, the default
25151 is the value of `LONG_TYPE_SIZE'.
25153 -- Macro: LONG_LONG_TYPE_SIZE
25154 A C expression for the size in bits of the type `long long' on the
25155 target machine. If you don't define this, the default is two
25156 words. If you want to support GNU Ada on your machine, the value
25157 of this macro must be at least 64.
25159 -- Macro: CHAR_TYPE_SIZE
25160 A C expression for the size in bits of the type `char' on the
25161 target machine. If you don't define this, the default is
25164 -- Macro: BOOL_TYPE_SIZE
25165 A C expression for the size in bits of the C++ type `bool' and C99
25166 type `_Bool' on the target machine. If you don't define this, and
25167 you probably shouldn't, the default is `CHAR_TYPE_SIZE'.
25169 -- Macro: FLOAT_TYPE_SIZE
25170 A C expression for the size in bits of the type `float' on the
25171 target machine. If you don't define this, the default is one word.
25173 -- Macro: DOUBLE_TYPE_SIZE
25174 A C expression for the size in bits of the type `double' on the
25175 target machine. If you don't define this, the default is two
25178 -- Macro: LONG_DOUBLE_TYPE_SIZE
25179 A C expression for the size in bits of the type `long double' on
25180 the target machine. If you don't define this, the default is two
25183 -- Macro: SHORT_FRACT_TYPE_SIZE
25184 A C expression for the size in bits of the type `short _Fract' on
25185 the target machine. If you don't define this, the default is
25188 -- Macro: FRACT_TYPE_SIZE
25189 A C expression for the size in bits of the type `_Fract' on the
25190 target machine. If you don't define this, the default is
25191 `BITS_PER_UNIT * 2'.
25193 -- Macro: LONG_FRACT_TYPE_SIZE
25194 A C expression for the size in bits of the type `long _Fract' on
25195 the target machine. If you don't define this, the default is
25196 `BITS_PER_UNIT * 4'.
25198 -- Macro: LONG_LONG_FRACT_TYPE_SIZE
25199 A C expression for the size in bits of the type `long long _Fract'
25200 on the target machine. If you don't define this, the default is
25201 `BITS_PER_UNIT * 8'.
25203 -- Macro: SHORT_ACCUM_TYPE_SIZE
25204 A C expression for the size in bits of the type `short _Accum' on
25205 the target machine. If you don't define this, the default is
25206 `BITS_PER_UNIT * 2'.
25208 -- Macro: ACCUM_TYPE_SIZE
25209 A C expression for the size in bits of the type `_Accum' on the
25210 target machine. If you don't define this, the default is
25211 `BITS_PER_UNIT * 4'.
25213 -- Macro: LONG_ACCUM_TYPE_SIZE
25214 A C expression for the size in bits of the type `long _Accum' on
25215 the target machine. If you don't define this, the default is
25216 `BITS_PER_UNIT * 8'.
25218 -- Macro: LONG_LONG_ACCUM_TYPE_SIZE
25219 A C expression for the size in bits of the type `long long _Accum'
25220 on the target machine. If you don't define this, the default is
25221 `BITS_PER_UNIT * 16'.
25223 -- Macro: LIBGCC2_LONG_DOUBLE_TYPE_SIZE
25224 Define this macro if `LONG_DOUBLE_TYPE_SIZE' is not constant or if
25225 you want routines in `libgcc2.a' for a size other than
25226 `LONG_DOUBLE_TYPE_SIZE'. If you don't define this, the default is
25227 `LONG_DOUBLE_TYPE_SIZE'.
25229 -- Macro: LIBGCC2_HAS_DF_MODE
25230 Define this macro if neither `LIBGCC2_DOUBLE_TYPE_SIZE' nor
25231 `LIBGCC2_LONG_DOUBLE_TYPE_SIZE' is `DFmode' but you want `DFmode'
25232 routines in `libgcc2.a' anyway. If you don't define this and
25233 either `LIBGCC2_DOUBLE_TYPE_SIZE' or
25234 `LIBGCC2_LONG_DOUBLE_TYPE_SIZE' is 64 then the default is 1,
25237 -- Macro: LIBGCC2_HAS_XF_MODE
25238 Define this macro if `LIBGCC2_LONG_DOUBLE_TYPE_SIZE' is not
25239 `XFmode' but you want `XFmode' routines in `libgcc2.a' anyway. If
25240 you don't define this and `LIBGCC2_LONG_DOUBLE_TYPE_SIZE' is 80
25241 then the default is 1, otherwise it is 0.
25243 -- Macro: LIBGCC2_HAS_TF_MODE
25244 Define this macro if `LIBGCC2_LONG_DOUBLE_TYPE_SIZE' is not
25245 `TFmode' but you want `TFmode' routines in `libgcc2.a' anyway. If
25246 you don't define this and `LIBGCC2_LONG_DOUBLE_TYPE_SIZE' is 128
25247 then the default is 1, otherwise it is 0.
25253 Define these macros to be the size in bits of the mantissa of
25254 `SFmode', `DFmode', `XFmode' and `TFmode' values, if the defaults
25255 in `libgcc2.h' are inappropriate. By default, `FLT_MANT_DIG' is
25256 used for `SF_SIZE', `LDBL_MANT_DIG' for `XF_SIZE' and `TF_SIZE',
25257 and `DBL_MANT_DIG' or `LDBL_MANT_DIG' for `DF_SIZE' according to
25258 whether `LIBGCC2_DOUBLE_TYPE_SIZE' or
25259 `LIBGCC2_LONG_DOUBLE_TYPE_SIZE' is 64.
25261 -- Macro: TARGET_FLT_EVAL_METHOD
25262 A C expression for the value for `FLT_EVAL_METHOD' in `float.h',
25263 assuming, if applicable, that the floating-point control word is
25264 in its default state. If you do not define this macro the value of
25265 `FLT_EVAL_METHOD' will be zero.
25267 -- Macro: WIDEST_HARDWARE_FP_SIZE
25268 A C expression for the size in bits of the widest floating-point
25269 format supported by the hardware. If you define this macro, you
25270 must specify a value less than or equal to the value of
25271 `LONG_DOUBLE_TYPE_SIZE'. If you do not define this macro, the
25272 value of `LONG_DOUBLE_TYPE_SIZE' is the default.
25274 -- Macro: DEFAULT_SIGNED_CHAR
25275 An expression whose value is 1 or 0, according to whether the type
25276 `char' should be signed or unsigned by default. The user can
25277 always override this default with the options `-fsigned-char' and
25280 -- Target Hook: bool TARGET_DEFAULT_SHORT_ENUMS (void)
25281 This target hook should return true if the compiler should give an
25282 `enum' type only as many bytes as it takes to represent the range
25283 of possible values of that type. It should return false if all
25284 `enum' types should be allocated like `int'.
25286 The default is to return false.
25288 -- Macro: SIZE_TYPE
25289 A C expression for a string describing the name of the data type
25290 to use for size values. The typedef name `size_t' is defined
25291 using the contents of the string.
25293 The string can contain more than one keyword. If so, separate
25294 them with spaces, and write first any length keyword, then
25295 `unsigned' if appropriate, and finally `int'. The string must
25296 exactly match one of the data type names defined in the function
25297 `init_decl_processing' in the file `c-decl.c'. You may not omit
25298 `int' or change the order--that would cause the compiler to crash
25301 If you don't define this macro, the default is `"long unsigned
25304 -- Macro: PTRDIFF_TYPE
25305 A C expression for a string describing the name of the data type
25306 to use for the result of subtracting two pointers. The typedef
25307 name `ptrdiff_t' is defined using the contents of the string. See
25308 `SIZE_TYPE' above for more information.
25310 If you don't define this macro, the default is `"long int"'.
25312 -- Macro: WCHAR_TYPE
25313 A C expression for a string describing the name of the data type
25314 to use for wide characters. The typedef name `wchar_t' is defined
25315 using the contents of the string. See `SIZE_TYPE' above for more
25318 If you don't define this macro, the default is `"int"'.
25320 -- Macro: WCHAR_TYPE_SIZE
25321 A C expression for the size in bits of the data type for wide
25322 characters. This is used in `cpp', which cannot make use of
25325 -- Macro: WINT_TYPE
25326 A C expression for a string describing the name of the data type to
25327 use for wide characters passed to `printf' and returned from
25328 `getwc'. The typedef name `wint_t' is defined using the contents
25329 of the string. See `SIZE_TYPE' above for more information.
25331 If you don't define this macro, the default is `"unsigned int"'.
25333 -- Macro: INTMAX_TYPE
25334 A C expression for a string describing the name of the data type
25335 that can represent any value of any standard or extended signed
25336 integer type. The typedef name `intmax_t' is defined using the
25337 contents of the string. See `SIZE_TYPE' above for more
25340 If you don't define this macro, the default is the first of
25341 `"int"', `"long int"', or `"long long int"' that has as much
25342 precision as `long long int'.
25344 -- Macro: UINTMAX_TYPE
25345 A C expression for a string describing the name of the data type
25346 that can represent any value of any standard or extended unsigned
25347 integer type. The typedef name `uintmax_t' is defined using the
25348 contents of the string. See `SIZE_TYPE' above for more
25351 If you don't define this macro, the default is the first of
25352 `"unsigned int"', `"long unsigned int"', or `"long long unsigned
25353 int"' that has as much precision as `long long unsigned int'.
25355 -- Macro: TARGET_PTRMEMFUNC_VBIT_LOCATION
25356 The C++ compiler represents a pointer-to-member-function with a
25357 struct that looks like:
25362 ptrdiff_t vtable_index;
25367 The C++ compiler must use one bit to indicate whether the function
25368 that will be called through a pointer-to-member-function is
25369 virtual. Normally, we assume that the low-order bit of a function
25370 pointer must always be zero. Then, by ensuring that the
25371 vtable_index is odd, we can distinguish which variant of the union
25372 is in use. But, on some platforms function pointers can be odd,
25373 and so this doesn't work. In that case, we use the low-order bit
25374 of the `delta' field, and shift the remainder of the `delta' field
25377 GCC will automatically make the right selection about where to
25378 store this bit using the `FUNCTION_BOUNDARY' setting for your
25379 platform. However, some platforms such as ARM/Thumb have
25380 `FUNCTION_BOUNDARY' set such that functions always start at even
25381 addresses, but the lowest bit of pointers to functions indicate
25382 whether the function at that address is in ARM or Thumb mode. If
25383 this is the case of your architecture, you should define this
25384 macro to `ptrmemfunc_vbit_in_delta'.
25386 In general, you should not have to define this macro. On
25387 architectures in which function addresses are always even,
25388 according to `FUNCTION_BOUNDARY', GCC will automatically define
25389 this macro to `ptrmemfunc_vbit_in_pfn'.
25391 -- Macro: TARGET_VTABLE_USES_DESCRIPTORS
25392 Normally, the C++ compiler uses function pointers in vtables. This
25393 macro allows the target to change to use "function descriptors"
25394 instead. Function descriptors are found on targets for whom a
25395 function pointer is actually a small data structure. Normally the
25396 data structure consists of the actual code address plus a data
25397 pointer to which the function's data is relative.
25399 If vtables are used, the value of this macro should be the number
25400 of words that the function descriptor occupies.
25402 -- Macro: TARGET_VTABLE_ENTRY_ALIGN
25403 By default, the vtable entries are void pointers, the so the
25404 alignment is the same as pointer alignment. The value of this
25405 macro specifies the alignment of the vtable entry in bits. It
25406 should be defined only when special alignment is necessary. */
25408 -- Macro: TARGET_VTABLE_DATA_ENTRY_DISTANCE
25409 There are a few non-descriptor entries in the vtable at offsets
25410 below zero. If these entries must be padded (say, to preserve the
25411 alignment specified by `TARGET_VTABLE_ENTRY_ALIGN'), set this to
25412 the number of words in each data entry.
25415 File: gccint.info, Node: Registers, Next: Register Classes, Prev: Type Layout, Up: Target Macros
25417 17.7 Register Usage
25418 ===================
25420 This section explains how to describe what registers the target machine
25421 has, and how (in general) they can be used.
25423 The description of which registers a specific instruction can use is
25424 done with register classes; see *Note Register Classes::. For
25425 information on using registers to access a stack frame, see *Note Frame
25426 Registers::. For passing values in registers, see *Note Register
25427 Arguments::. For returning values in registers, see *Note Scalar
25432 * Register Basics:: Number and kinds of registers.
25433 * Allocation Order:: Order in which registers are allocated.
25434 * Values in Registers:: What kinds of values each reg can hold.
25435 * Leaf Functions:: Renumbering registers for leaf functions.
25436 * Stack Registers:: Handling a register stack such as 80387.
25439 File: gccint.info, Node: Register Basics, Next: Allocation Order, Up: Registers
25441 17.7.1 Basic Characteristics of Registers
25442 -----------------------------------------
25444 Registers have various characteristics.
25446 -- Macro: FIRST_PSEUDO_REGISTER
25447 Number of hardware registers known to the compiler. They receive
25448 numbers 0 through `FIRST_PSEUDO_REGISTER-1'; thus, the first
25449 pseudo register's number really is assigned the number
25450 `FIRST_PSEUDO_REGISTER'.
25452 -- Macro: FIXED_REGISTERS
25453 An initializer that says which registers are used for fixed
25454 purposes all throughout the compiled code and are therefore not
25455 available for general allocation. These would include the stack
25456 pointer, the frame pointer (except on machines where that can be
25457 used as a general register when no frame pointer is needed), the
25458 program counter on machines where that is considered one of the
25459 addressable registers, and any other numbered register with a
25462 This information is expressed as a sequence of numbers, separated
25463 by commas and surrounded by braces. The Nth number is 1 if
25464 register N is fixed, 0 otherwise.
25466 The table initialized from this macro, and the table initialized by
25467 the following one, may be overridden at run time either
25468 automatically, by the actions of the macro
25469 `CONDITIONAL_REGISTER_USAGE', or by the user with the command
25470 options `-ffixed-REG', `-fcall-used-REG' and `-fcall-saved-REG'.
25472 -- Macro: CALL_USED_REGISTERS
25473 Like `FIXED_REGISTERS' but has 1 for each register that is
25474 clobbered (in general) by function calls as well as for fixed
25475 registers. This macro therefore identifies the registers that are
25476 not available for general allocation of values that must live
25477 across function calls.
25479 If a register has 0 in `CALL_USED_REGISTERS', the compiler
25480 automatically saves it on function entry and restores it on
25481 function exit, if the register is used within the function.
25483 -- Macro: CALL_REALLY_USED_REGISTERS
25484 Like `CALL_USED_REGISTERS' except this macro doesn't require that
25485 the entire set of `FIXED_REGISTERS' be included.
25486 (`CALL_USED_REGISTERS' must be a superset of `FIXED_REGISTERS').
25487 This macro is optional. If not specified, it defaults to the value
25488 of `CALL_USED_REGISTERS'.
25490 -- Macro: HARD_REGNO_CALL_PART_CLOBBERED (REGNO, MODE)
25491 A C expression that is nonzero if it is not permissible to store a
25492 value of mode MODE in hard register number REGNO across a call
25493 without some part of it being clobbered. For most machines this
25494 macro need not be defined. It is only required for machines that
25495 do not preserve the entire contents of a register across a call.
25497 -- Macro: CONDITIONAL_REGISTER_USAGE
25498 Zero or more C statements that may conditionally modify five
25499 variables `fixed_regs', `call_used_regs', `global_regs',
25500 `reg_names', and `reg_class_contents', to take into account any
25501 dependence of these register sets on target flags. The first three
25502 of these are of type `char []' (interpreted as Boolean vectors).
25503 `global_regs' is a `const char *[]', and `reg_class_contents' is a
25504 `HARD_REG_SET'. Before the macro is called, `fixed_regs',
25505 `call_used_regs', `reg_class_contents', and `reg_names' have been
25506 initialized from `FIXED_REGISTERS', `CALL_USED_REGISTERS',
25507 `REG_CLASS_CONTENTS', and `REGISTER_NAMES', respectively.
25508 `global_regs' has been cleared, and any `-ffixed-REG',
25509 `-fcall-used-REG' and `-fcall-saved-REG' command options have been
25512 You need not define this macro if it has no work to do.
25514 If the usage of an entire class of registers depends on the target
25515 flags, you may indicate this to GCC by using this macro to modify
25516 `fixed_regs' and `call_used_regs' to 1 for each of the registers
25517 in the classes which should not be used by GCC. Also define the
25518 macro `REG_CLASS_FROM_LETTER' / `REG_CLASS_FROM_CONSTRAINT' to
25519 return `NO_REGS' if it is called with a letter for a class that
25522 (However, if this class is not included in `GENERAL_REGS' and all
25523 of the insn patterns whose constraints permit this class are
25524 controlled by target switches, then GCC will automatically avoid
25525 using these registers when the target switches are opposed to
25528 -- Macro: INCOMING_REGNO (OUT)
25529 Define this macro if the target machine has register windows.
25530 This C expression returns the register number as seen by the
25531 called function corresponding to the register number OUT as seen
25532 by the calling function. Return OUT if register number OUT is not
25533 an outbound register.
25535 -- Macro: OUTGOING_REGNO (IN)
25536 Define this macro if the target machine has register windows.
25537 This C expression returns the register number as seen by the
25538 calling function corresponding to the register number IN as seen
25539 by the called function. Return IN if register number IN is not an
25542 -- Macro: LOCAL_REGNO (REGNO)
25543 Define this macro if the target machine has register windows.
25544 This C expression returns true if the register is call-saved but
25545 is in the register window. Unlike most call-saved registers, such
25546 registers need not be explicitly restored on function exit or
25547 during non-local gotos.
25549 -- Macro: PC_REGNUM
25550 If the program counter has a register number, define this as that
25551 register number. Otherwise, do not define it.
25554 File: gccint.info, Node: Allocation Order, Next: Values in Registers, Prev: Register Basics, Up: Registers
25556 17.7.2 Order of Allocation of Registers
25557 ---------------------------------------
25559 Registers are allocated in order.
25561 -- Macro: REG_ALLOC_ORDER
25562 If defined, an initializer for a vector of integers, containing the
25563 numbers of hard registers in the order in which GCC should prefer
25564 to use them (from most preferred to least).
25566 If this macro is not defined, registers are used lowest numbered
25567 first (all else being equal).
25569 One use of this macro is on machines where the highest numbered
25570 registers must always be saved and the save-multiple-registers
25571 instruction supports only sequences of consecutive registers. On
25572 such machines, define `REG_ALLOC_ORDER' to be an initializer that
25573 lists the highest numbered allocable register first.
25575 -- Macro: ORDER_REGS_FOR_LOCAL_ALLOC
25576 A C statement (sans semicolon) to choose the order in which to
25577 allocate hard registers for pseudo-registers local to a basic
25580 Store the desired register order in the array `reg_alloc_order'.
25581 Element 0 should be the register to allocate first; element 1, the
25582 next register; and so on.
25584 The macro body should not assume anything about the contents of
25585 `reg_alloc_order' before execution of the macro.
25587 On most machines, it is not necessary to define this macro.
25589 -- Macro: IRA_HARD_REGNO_ADD_COST_MULTIPLIER (REGNO)
25590 In some case register allocation order is not enough for the
25591 Integrated Register Allocator (IRA) to generate a good code. If
25592 this macro is defined, it should return a floating point value
25593 based on REGNO. The cost of using REGNO for a pseudo will be
25594 increased by approximately the pseudo's usage frequency times the
25595 value returned by this macro. Not defining this macro is
25596 equivalent to having it always return `0.0'.
25598 On most machines, it is not necessary to define this macro.
25601 File: gccint.info, Node: Values in Registers, Next: Leaf Functions, Prev: Allocation Order, Up: Registers
25603 17.7.3 How Values Fit in Registers
25604 ----------------------------------
25606 This section discusses the macros that describe which kinds of values
25607 (specifically, which machine modes) each register can hold, and how many
25608 consecutive registers are needed for a given mode.
25610 -- Macro: HARD_REGNO_NREGS (REGNO, MODE)
25611 A C expression for the number of consecutive hard registers,
25612 starting at register number REGNO, required to hold a value of mode
25613 MODE. This macro must never return zero, even if a register
25614 cannot hold the requested mode - indicate that with
25615 HARD_REGNO_MODE_OK and/or CANNOT_CHANGE_MODE_CLASS instead.
25617 On a machine where all registers are exactly one word, a suitable
25618 definition of this macro is
25620 #define HARD_REGNO_NREGS(REGNO, MODE) \
25621 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
25624 -- Macro: HARD_REGNO_NREGS_HAS_PADDING (REGNO, MODE)
25625 A C expression that is nonzero if a value of mode MODE, stored in
25626 memory, ends with padding that causes it to take up more space than
25627 in registers starting at register number REGNO (as determined by
25628 multiplying GCC's notion of the size of the register when
25629 containing this mode by the number of registers returned by
25630 `HARD_REGNO_NREGS'). By default this is zero.
25632 For example, if a floating-point value is stored in three 32-bit
25633 registers but takes up 128 bits in memory, then this would be
25636 This macros only needs to be defined if there are cases where
25637 `subreg_get_info' would otherwise wrongly determine that a
25638 `subreg' can be represented by an offset to the register number,
25639 when in fact such a `subreg' would contain some of the padding not
25640 stored in registers and so not be representable.
25642 -- Macro: HARD_REGNO_NREGS_WITH_PADDING (REGNO, MODE)
25643 For values of REGNO and MODE for which
25644 `HARD_REGNO_NREGS_HAS_PADDING' returns nonzero, a C expression
25645 returning the greater number of registers required to hold the
25646 value including any padding. In the example above, the value
25649 -- Macro: REGMODE_NATURAL_SIZE (MODE)
25650 Define this macro if the natural size of registers that hold values
25651 of mode MODE is not the word size. It is a C expression that
25652 should give the natural size in bytes for the specified mode. It
25653 is used by the register allocator to try to optimize its results.
25654 This happens for example on SPARC 64-bit where the natural size of
25655 floating-point registers is still 32-bit.
25657 -- Macro: HARD_REGNO_MODE_OK (REGNO, MODE)
25658 A C expression that is nonzero if it is permissible to store a
25659 value of mode MODE in hard register number REGNO (or in several
25660 registers starting with that one). For a machine where all
25661 registers are equivalent, a suitable definition is
25663 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
25665 You need not include code to check for the numbers of fixed
25666 registers, because the allocation mechanism considers them to be
25669 On some machines, double-precision values must be kept in even/odd
25670 register pairs. You can implement that by defining this macro to
25671 reject odd register numbers for such modes.
25673 The minimum requirement for a mode to be OK in a register is that
25674 the `movMODE' instruction pattern support moves between the
25675 register and other hard register in the same class and that moving
25676 a value into the register and back out not alter it.
25678 Since the same instruction used to move `word_mode' will work for
25679 all narrower integer modes, it is not necessary on any machine for
25680 `HARD_REGNO_MODE_OK' to distinguish between these modes, provided
25681 you define patterns `movhi', etc., to take advantage of this. This
25682 is useful because of the interaction between `HARD_REGNO_MODE_OK'
25683 and `MODES_TIEABLE_P'; it is very desirable for all integer modes
25686 Many machines have special registers for floating point arithmetic.
25687 Often people assume that floating point machine modes are allowed
25688 only in floating point registers. This is not true. Any
25689 registers that can hold integers can safely _hold_ a floating
25690 point machine mode, whether or not floating arithmetic can be done
25691 on it in those registers. Integer move instructions can be used
25692 to move the values.
25694 On some machines, though, the converse is true: fixed-point machine
25695 modes may not go in floating registers. This is true if the
25696 floating registers normalize any value stored in them, because
25697 storing a non-floating value there would garble it. In this case,
25698 `HARD_REGNO_MODE_OK' should reject fixed-point machine modes in
25699 floating registers. But if the floating registers do not
25700 automatically normalize, if you can store any bit pattern in one
25701 and retrieve it unchanged without a trap, then any machine mode
25702 may go in a floating register, so you can define this macro to say
25705 The primary significance of special floating registers is rather
25706 that they are the registers acceptable in floating point arithmetic
25707 instructions. However, this is of no concern to
25708 `HARD_REGNO_MODE_OK'. You handle it by writing the proper
25709 constraints for those instructions.
25711 On some machines, the floating registers are especially slow to
25712 access, so that it is better to store a value in a stack frame
25713 than in such a register if floating point arithmetic is not being
25714 done. As long as the floating registers are not in class
25715 `GENERAL_REGS', they will not be used unless some pattern's
25716 constraint asks for one.
25718 -- Macro: HARD_REGNO_RENAME_OK (FROM, TO)
25719 A C expression that is nonzero if it is OK to rename a hard
25720 register FROM to another hard register TO.
25722 One common use of this macro is to prevent renaming of a register
25723 to another register that is not saved by a prologue in an interrupt
25726 The default is always nonzero.
25728 -- Macro: MODES_TIEABLE_P (MODE1, MODE2)
25729 A C expression that is nonzero if a value of mode MODE1 is
25730 accessible in mode MODE2 without copying.
25732 If `HARD_REGNO_MODE_OK (R, MODE1)' and `HARD_REGNO_MODE_OK (R,
25733 MODE2)' are always the same for any R, then `MODES_TIEABLE_P
25734 (MODE1, MODE2)' should be nonzero. If they differ for any R, you
25735 should define this macro to return zero unless some other
25736 mechanism ensures the accessibility of the value in a narrower
25739 You should define this macro to return nonzero in as many cases as
25740 possible since doing so will allow GCC to perform better register
25743 -- Target Hook: bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int REGNO)
25744 This target hook should return `true' if it is OK to use a hard
25745 register REGNO as scratch reg in peephole2.
25747 One common use of this macro is to prevent using of a register that
25748 is not saved by a prologue in an interrupt handler.
25750 The default version of this hook always returns `true'.
25752 -- Macro: AVOID_CCMODE_COPIES
25753 Define this macro if the compiler should avoid copies to/from
25754 `CCmode' registers. You should only define this macro if support
25755 for copying to/from `CCmode' is incomplete.
25758 File: gccint.info, Node: Leaf Functions, Next: Stack Registers, Prev: Values in Registers, Up: Registers
25760 17.7.4 Handling Leaf Functions
25761 ------------------------------
25763 On some machines, a leaf function (i.e., one which makes no calls) can
25764 run more efficiently if it does not make its own register window.
25765 Often this means it is required to receive its arguments in the
25766 registers where they are passed by the caller, instead of the registers
25767 where they would normally arrive.
25769 The special treatment for leaf functions generally applies only when
25770 other conditions are met; for example, often they may use only those
25771 registers for its own variables and temporaries. We use the term "leaf
25772 function" to mean a function that is suitable for this special
25773 handling, so that functions with no calls are not necessarily "leaf
25776 GCC assigns register numbers before it knows whether the function is
25777 suitable for leaf function treatment. So it needs to renumber the
25778 registers in order to output a leaf function. The following macros
25781 -- Macro: LEAF_REGISTERS
25782 Name of a char vector, indexed by hard register number, which
25783 contains 1 for a register that is allowable in a candidate for leaf
25784 function treatment.
25786 If leaf function treatment involves renumbering the registers,
25787 then the registers marked here should be the ones before
25788 renumbering--those that GCC would ordinarily allocate. The
25789 registers which will actually be used in the assembler code, after
25790 renumbering, should not be marked with 1 in this vector.
25792 Define this macro only if the target machine offers a way to
25793 optimize the treatment of leaf functions.
25795 -- Macro: LEAF_REG_REMAP (REGNO)
25796 A C expression whose value is the register number to which REGNO
25797 should be renumbered, when a function is treated as a leaf
25800 If REGNO is a register number which should not appear in a leaf
25801 function before renumbering, then the expression should yield -1,
25802 which will cause the compiler to abort.
25804 Define this macro only if the target machine offers a way to
25805 optimize the treatment of leaf functions, and registers need to be
25806 renumbered to do this.
25808 `TARGET_ASM_FUNCTION_PROLOGUE' and `TARGET_ASM_FUNCTION_EPILOGUE' must
25809 usually treat leaf functions specially. They can test the C variable
25810 `current_function_is_leaf' which is nonzero for leaf functions.
25811 `current_function_is_leaf' is set prior to local register allocation
25812 and is valid for the remaining compiler passes. They can also test the
25813 C variable `current_function_uses_only_leaf_regs' which is nonzero for
25814 leaf functions which only use leaf registers.
25815 `current_function_uses_only_leaf_regs' is valid after all passes that
25816 modify the instructions have been run and is only useful if
25817 `LEAF_REGISTERS' is defined.
25820 File: gccint.info, Node: Stack Registers, Prev: Leaf Functions, Up: Registers
25822 17.7.5 Registers That Form a Stack
25823 ----------------------------------
25825 There are special features to handle computers where some of the
25826 "registers" form a stack. Stack registers are normally written by
25827 pushing onto the stack, and are numbered relative to the top of the
25830 Currently, GCC can only handle one group of stack-like registers, and
25831 they must be consecutively numbered. Furthermore, the existing support
25832 for stack-like registers is specific to the 80387 floating point
25833 coprocessor. If you have a new architecture that uses stack-like
25834 registers, you will need to do substantial work on `reg-stack.c' and
25835 write your machine description to cooperate with it, as well as
25836 defining these macros.
25838 -- Macro: STACK_REGS
25839 Define this if the machine has any stack-like registers.
25841 -- Macro: FIRST_STACK_REG
25842 The number of the first stack-like register. This one is the top
25845 -- Macro: LAST_STACK_REG
25846 The number of the last stack-like register. This one is the
25847 bottom of the stack.
25850 File: gccint.info, Node: Register Classes, Next: Old Constraints, Prev: Registers, Up: Target Macros
25852 17.8 Register Classes
25853 =====================
25855 On many machines, the numbered registers are not all equivalent. For
25856 example, certain registers may not be allowed for indexed addressing;
25857 certain registers may not be allowed in some instructions. These
25858 machine restrictions are described to the compiler using "register
25861 You define a number of register classes, giving each one a name and
25862 saying which of the registers belong to it. Then you can specify
25863 register classes that are allowed as operands to particular instruction
25866 In general, each register will belong to several classes. In fact, one
25867 class must be named `ALL_REGS' and contain all the registers. Another
25868 class must be named `NO_REGS' and contain no registers. Often the
25869 union of two classes will be another class; however, this is not
25872 One of the classes must be named `GENERAL_REGS'. There is nothing
25873 terribly special about the name, but the operand constraint letters `r'
25874 and `g' specify this class. If `GENERAL_REGS' is the same as
25875 `ALL_REGS', just define it as a macro which expands to `ALL_REGS'.
25877 Order the classes so that if class X is contained in class Y then X
25878 has a lower class number than Y.
25880 The way classes other than `GENERAL_REGS' are specified in operand
25881 constraints is through machine-dependent operand constraint letters.
25882 You can define such letters to correspond to various classes, then use
25883 them in operand constraints.
25885 You should define a class for the union of two classes whenever some
25886 instruction allows both classes. For example, if an instruction allows
25887 either a floating point (coprocessor) register or a general register
25888 for a certain operand, you should define a class `FLOAT_OR_GENERAL_REGS'
25889 which includes both of them. Otherwise you will get suboptimal code.
25891 You must also specify certain redundant information about the register
25892 classes: for each class, which classes contain it and which ones are
25893 contained in it; for each pair of classes, the largest class contained
25896 When a value occupying several consecutive registers is expected in a
25897 certain class, all the registers used must belong to that class.
25898 Therefore, register classes cannot be used to enforce a requirement for
25899 a register pair to start with an even-numbered register. The way to
25900 specify this requirement is with `HARD_REGNO_MODE_OK'.
25902 Register classes used for input-operands of bitwise-and or shift
25903 instructions have a special requirement: each such class must have, for
25904 each fixed-point machine mode, a subclass whose registers can transfer
25905 that mode to or from memory. For example, on some machines, the
25906 operations for single-byte values (`QImode') are limited to certain
25907 registers. When this is so, each register class that is used in a
25908 bitwise-and or shift instruction must have a subclass consisting of
25909 registers from which single-byte values can be loaded or stored. This
25910 is so that `PREFERRED_RELOAD_CLASS' can always have a possible value to
25913 -- Data type: enum reg_class
25914 An enumerated type that must be defined with all the register
25915 class names as enumerated values. `NO_REGS' must be first.
25916 `ALL_REGS' must be the last register class, followed by one more
25917 enumerated value, `LIM_REG_CLASSES', which is not a register class
25918 but rather tells how many classes there are.
25920 Each register class has a number, which is the value of casting
25921 the class name to type `int'. The number serves as an index in
25922 many of the tables described below.
25924 -- Macro: N_REG_CLASSES
25925 The number of distinct register classes, defined as follows:
25927 #define N_REG_CLASSES (int) LIM_REG_CLASSES
25929 -- Macro: REG_CLASS_NAMES
25930 An initializer containing the names of the register classes as C
25931 string constants. These names are used in writing some of the
25934 -- Macro: REG_CLASS_CONTENTS
25935 An initializer containing the contents of the register classes, as
25936 integers which are bit masks. The Nth integer specifies the
25937 contents of class N. The way the integer MASK is interpreted is
25938 that register R is in the class if `MASK & (1 << R)' is 1.
25940 When the machine has more than 32 registers, an integer does not
25941 suffice. Then the integers are replaced by sub-initializers,
25942 braced groupings containing several integers. Each
25943 sub-initializer must be suitable as an initializer for the type
25944 `HARD_REG_SET' which is defined in `hard-reg-set.h'. In this
25945 situation, the first integer in each sub-initializer corresponds to
25946 registers 0 through 31, the second integer to registers 32 through
25949 -- Macro: REGNO_REG_CLASS (REGNO)
25950 A C expression whose value is a register class containing hard
25951 register REGNO. In general there is more than one such class;
25952 choose a class which is "minimal", meaning that no smaller class
25953 also contains the register.
25955 -- Macro: BASE_REG_CLASS
25956 A macro whose definition is the name of the class to which a valid
25957 base register must belong. A base register is one used in an
25958 address which is the register value plus a displacement.
25960 -- Macro: MODE_BASE_REG_CLASS (MODE)
25961 This is a variation of the `BASE_REG_CLASS' macro which allows the
25962 selection of a base register in a mode dependent manner. If MODE
25963 is VOIDmode then it should return the same value as
25966 -- Macro: MODE_BASE_REG_REG_CLASS (MODE)
25967 A C expression whose value is the register class to which a valid
25968 base register must belong in order to be used in a base plus index
25969 register address. You should define this macro if base plus index
25970 addresses have different requirements than other base register
25973 -- Macro: MODE_CODE_BASE_REG_CLASS (MODE, OUTER_CODE, INDEX_CODE)
25974 A C expression whose value is the register class to which a valid
25975 base register must belong. OUTER_CODE and INDEX_CODE define the
25976 context in which the base register occurs. OUTER_CODE is the code
25977 of the immediately enclosing expression (`MEM' for the top level
25978 of an address, `ADDRESS' for something that occurs in an
25979 `address_operand'). INDEX_CODE is the code of the corresponding
25980 index expression if OUTER_CODE is `PLUS'; `SCRATCH' otherwise.
25982 -- Macro: INDEX_REG_CLASS
25983 A macro whose definition is the name of the class to which a valid
25984 index register must belong. An index register is one used in an
25985 address where its value is either multiplied by a scale factor or
25986 added to another register (as well as added to a displacement).
25988 -- Macro: REGNO_OK_FOR_BASE_P (NUM)
25989 A C expression which is nonzero if register number NUM is suitable
25990 for use as a base register in operand addresses. It may be either
25991 a suitable hard register or a pseudo register that has been
25992 allocated such a hard register.
25994 -- Macro: REGNO_MODE_OK_FOR_BASE_P (NUM, MODE)
25995 A C expression that is just like `REGNO_OK_FOR_BASE_P', except that
25996 that expression may examine the mode of the memory reference in
25997 MODE. You should define this macro if the mode of the memory
25998 reference affects whether a register may be used as a base
25999 register. If you define this macro, the compiler will use it
26000 instead of `REGNO_OK_FOR_BASE_P'. The mode may be `VOIDmode' for
26001 addresses that appear outside a `MEM', i.e., as an
26005 -- Macro: REGNO_MODE_OK_FOR_REG_BASE_P (NUM, MODE)
26006 A C expression which is nonzero if register number NUM is suitable
26007 for use as a base register in base plus index operand addresses,
26008 accessing memory in mode MODE. It may be either a suitable hard
26009 register or a pseudo register that has been allocated such a hard
26010 register. You should define this macro if base plus index
26011 addresses have different requirements than other base register
26014 Use of this macro is deprecated; please use the more general
26015 `REGNO_MODE_CODE_OK_FOR_BASE_P'.
26017 -- Macro: REGNO_MODE_CODE_OK_FOR_BASE_P (NUM, MODE, OUTER_CODE,
26019 A C expression that is just like `REGNO_MODE_OK_FOR_BASE_P', except
26020 that that expression may examine the context in which the register
26021 appears in the memory reference. OUTER_CODE is the code of the
26022 immediately enclosing expression (`MEM' if at the top level of the
26023 address, `ADDRESS' for something that occurs in an
26024 `address_operand'). INDEX_CODE is the code of the corresponding
26025 index expression if OUTER_CODE is `PLUS'; `SCRATCH' otherwise.
26026 The mode may be `VOIDmode' for addresses that appear outside a
26027 `MEM', i.e., as an `address_operand'.
26029 -- Macro: REGNO_OK_FOR_INDEX_P (NUM)
26030 A C expression which is nonzero if register number NUM is suitable
26031 for use as an index register in operand addresses. It may be
26032 either a suitable hard register or a pseudo register that has been
26033 allocated such a hard register.
26035 The difference between an index register and a base register is
26036 that the index register may be scaled. If an address involves the
26037 sum of two registers, neither one of them scaled, then either one
26038 may be labeled the "base" and the other the "index"; but whichever
26039 labeling is used must fit the machine's constraints of which
26040 registers may serve in each capacity. The compiler will try both
26041 labelings, looking for one that is valid, and will reload one or
26042 both registers only if neither labeling works.
26044 -- Macro: PREFERRED_RELOAD_CLASS (X, CLASS)
26045 A C expression that places additional restrictions on the register
26046 class to use when it is necessary to copy value X into a register
26047 in class CLASS. The value is a register class; perhaps CLASS, or
26048 perhaps another, smaller class. On many machines, the following
26049 definition is safe:
26051 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
26053 Sometimes returning a more restrictive class makes better code.
26054 For example, on the 68000, when X is an integer constant that is
26055 in range for a `moveq' instruction, the value of this macro is
26056 always `DATA_REGS' as long as CLASS includes the data registers.
26057 Requiring a data register guarantees that a `moveq' will be used.
26059 One case where `PREFERRED_RELOAD_CLASS' must not return CLASS is
26060 if X is a legitimate constant which cannot be loaded into some
26061 register class. By returning `NO_REGS' you can force X into a
26062 memory location. For example, rs6000 can load immediate values
26063 into general-purpose registers, but does not have an instruction
26064 for loading an immediate value into a floating-point register, so
26065 `PREFERRED_RELOAD_CLASS' returns `NO_REGS' when X is a
26066 floating-point constant. If the constant can't be loaded into any
26067 kind of register, code generation will be better if
26068 `LEGITIMATE_CONSTANT_P' makes the constant illegitimate instead of
26069 using `PREFERRED_RELOAD_CLASS'.
26071 If an insn has pseudos in it after register allocation, reload
26072 will go through the alternatives and call repeatedly
26073 `PREFERRED_RELOAD_CLASS' to find the best one. Returning
26074 `NO_REGS', in this case, makes reload add a `!' in front of the
26075 constraint: the x86 back-end uses this feature to discourage usage
26076 of 387 registers when math is done in the SSE registers (and vice
26079 -- Macro: PREFERRED_OUTPUT_RELOAD_CLASS (X, CLASS)
26080 Like `PREFERRED_RELOAD_CLASS', but for output reloads instead of
26081 input reloads. If you don't define this macro, the default is to
26082 use CLASS, unchanged.
26084 You can also use `PREFERRED_OUTPUT_RELOAD_CLASS' to discourage
26085 reload from using some alternatives, like `PREFERRED_RELOAD_CLASS'.
26087 -- Macro: LIMIT_RELOAD_CLASS (MODE, CLASS)
26088 A C expression that places additional restrictions on the register
26089 class to use when it is necessary to be able to hold a value of
26090 mode MODE in a reload register for which class CLASS would
26091 ordinarily be used.
26093 Unlike `PREFERRED_RELOAD_CLASS', this macro should be used when
26094 there are certain modes that simply can't go in certain reload
26097 The value is a register class; perhaps CLASS, or perhaps another,
26100 Don't define this macro unless the target machine has limitations
26101 which require the macro to do something nontrivial.
26103 -- Target Hook: enum reg_class TARGET_SECONDARY_RELOAD (bool IN_P, rtx
26104 X, enum reg_class RELOAD_CLASS, enum machine_mode
26105 RELOAD_MODE, secondary_reload_info *SRI)
26106 Many machines have some registers that cannot be copied directly
26107 to or from memory or even from other types of registers. An
26108 example is the `MQ' register, which on most machines, can only be
26109 copied to or from general registers, but not memory. Below, we
26110 shall be using the term 'intermediate register' when a move
26111 operation cannot be performed directly, but has to be done by
26112 copying the source into the intermediate register first, and then
26113 copying the intermediate register to the destination. An
26114 intermediate register always has the same mode as source and
26115 destination. Since it holds the actual value being copied, reload
26116 might apply optimizations to re-use an intermediate register and
26117 eliding the copy from the source when it can determine that the
26118 intermediate register still holds the required value.
26120 Another kind of secondary reload is required on some machines which
26121 allow copying all registers to and from memory, but require a
26122 scratch register for stores to some memory locations (e.g., those
26123 with symbolic address on the RT, and those with certain symbolic
26124 address on the SPARC when compiling PIC). Scratch registers need
26125 not have the same mode as the value being copied, and usually hold
26126 a different value that that being copied. Special patterns in the
26127 md file are needed to describe how the copy is performed with the
26128 help of the scratch register; these patterns also describe the
26129 number, register class(es) and mode(s) of the scratch register(s).
26131 In some cases, both an intermediate and a scratch register are
26134 For input reloads, this target hook is called with nonzero IN_P,
26135 and X is an rtx that needs to be copied to a register of class
26136 RELOAD_CLASS in RELOAD_MODE. For output reloads, this target hook
26137 is called with zero IN_P, and a register of class RELOAD_CLASS
26138 needs to be copied to rtx X in RELOAD_MODE.
26140 If copying a register of RELOAD_CLASS from/to X requires an
26141 intermediate register, the hook `secondary_reload' should return
26142 the register class required for this intermediate register. If no
26143 intermediate register is required, it should return NO_REGS. If
26144 more than one intermediate register is required, describe the one
26145 that is closest in the copy chain to the reload register.
26147 If scratch registers are needed, you also have to describe how to
26148 perform the copy from/to the reload register to/from this closest
26149 intermediate register. Or if no intermediate register is
26150 required, but still a scratch register is needed, describe the
26151 copy from/to the reload register to/from the reload operand X.
26153 You do this by setting `sri->icode' to the instruction code of a
26154 pattern in the md file which performs the move. Operands 0 and 1
26155 are the output and input of this copy, respectively. Operands
26156 from operand 2 onward are for scratch operands. These scratch
26157 operands must have a mode, and a single-register-class output
26160 When an intermediate register is used, the `secondary_reload' hook
26161 will be called again to determine how to copy the intermediate
26162 register to/from the reload operand X, so your hook must also have
26163 code to handle the register class of the intermediate operand.
26165 X might be a pseudo-register or a `subreg' of a pseudo-register,
26166 which could either be in a hard register or in memory. Use
26167 `true_regnum' to find out; it will return -1 if the pseudo is in
26168 memory and the hard register number if it is in a register.
26170 Scratch operands in memory (constraint `"=m"' / `"=&m"') are
26171 currently not supported. For the time being, you will have to
26172 continue to use `SECONDARY_MEMORY_NEEDED' for that purpose.
26174 `copy_cost' also uses this target hook to find out how values are
26175 copied. If you want it to include some extra cost for the need to
26176 allocate (a) scratch register(s), set `sri->extra_cost' to the
26177 additional cost. Or if two dependent moves are supposed to have a
26178 lower cost than the sum of the individual moves due to expected
26179 fortuitous scheduling and/or special forwarding logic, you can set
26180 `sri->extra_cost' to a negative amount.
26182 -- Macro: SECONDARY_RELOAD_CLASS (CLASS, MODE, X)
26183 -- Macro: SECONDARY_INPUT_RELOAD_CLASS (CLASS, MODE, X)
26184 -- Macro: SECONDARY_OUTPUT_RELOAD_CLASS (CLASS, MODE, X)
26185 These macros are obsolete, new ports should use the target hook
26186 `TARGET_SECONDARY_RELOAD' instead.
26188 These are obsolete macros, replaced by the
26189 `TARGET_SECONDARY_RELOAD' target hook. Older ports still define
26190 these macros to indicate to the reload phase that it may need to
26191 allocate at least one register for a reload in addition to the
26192 register to contain the data. Specifically, if copying X to a
26193 register CLASS in MODE requires an intermediate register, you were
26194 supposed to define `SECONDARY_INPUT_RELOAD_CLASS' to return the
26195 largest register class all of whose registers can be used as
26196 intermediate registers or scratch registers.
26198 If copying a register CLASS in MODE to X requires an intermediate
26199 or scratch register, `SECONDARY_OUTPUT_RELOAD_CLASS' was supposed
26200 to be defined be defined to return the largest register class
26201 required. If the requirements for input and output reloads were
26202 the same, the macro `SECONDARY_RELOAD_CLASS' should have been used
26203 instead of defining both macros identically.
26205 The values returned by these macros are often `GENERAL_REGS'.
26206 Return `NO_REGS' if no spare register is needed; i.e., if X can be
26207 directly copied to or from a register of CLASS in MODE without
26208 requiring a scratch register. Do not define this macro if it
26209 would always return `NO_REGS'.
26211 If a scratch register is required (either with or without an
26212 intermediate register), you were supposed to define patterns for
26213 `reload_inM' or `reload_outM', as required (*note Standard
26214 Names::. These patterns, which were normally implemented with a
26215 `define_expand', should be similar to the `movM' patterns, except
26216 that operand 2 is the scratch register.
26218 These patterns need constraints for the reload register and scratch
26219 register that contain a single register class. If the original
26220 reload register (whose class is CLASS) can meet the constraint
26221 given in the pattern, the value returned by these macros is used
26222 for the class of the scratch register. Otherwise, two additional
26223 reload registers are required. Their classes are obtained from
26224 the constraints in the insn pattern.
26226 X might be a pseudo-register or a `subreg' of a pseudo-register,
26227 which could either be in a hard register or in memory. Use
26228 `true_regnum' to find out; it will return -1 if the pseudo is in
26229 memory and the hard register number if it is in a register.
26231 These macros should not be used in the case where a particular
26232 class of registers can only be copied to memory and not to another
26233 class of registers. In that case, secondary reload registers are
26234 not needed and would not be helpful. Instead, a stack location
26235 must be used to perform the copy and the `movM' pattern should use
26236 memory as an intermediate storage. This case often occurs between
26237 floating-point and general registers.
26239 -- Macro: SECONDARY_MEMORY_NEEDED (CLASS1, CLASS2, M)
26240 Certain machines have the property that some registers cannot be
26241 copied to some other registers without using memory. Define this
26242 macro on those machines to be a C expression that is nonzero if
26243 objects of mode M in registers of CLASS1 can only be copied to
26244 registers of class CLASS2 by storing a register of CLASS1 into
26245 memory and loading that memory location into a register of CLASS2.
26247 Do not define this macro if its value would always be zero.
26249 -- Macro: SECONDARY_MEMORY_NEEDED_RTX (MODE)
26250 Normally when `SECONDARY_MEMORY_NEEDED' is defined, the compiler
26251 allocates a stack slot for a memory location needed for register
26252 copies. If this macro is defined, the compiler instead uses the
26253 memory location defined by this macro.
26255 Do not define this macro if you do not define
26256 `SECONDARY_MEMORY_NEEDED'.
26258 -- Macro: SECONDARY_MEMORY_NEEDED_MODE (MODE)
26259 When the compiler needs a secondary memory location to copy
26260 between two registers of mode MODE, it normally allocates
26261 sufficient memory to hold a quantity of `BITS_PER_WORD' bits and
26262 performs the store and load operations in a mode that many bits
26263 wide and whose class is the same as that of MODE.
26265 This is right thing to do on most machines because it ensures that
26266 all bits of the register are copied and prevents accesses to the
26267 registers in a narrower mode, which some machines prohibit for
26268 floating-point registers.
26270 However, this default behavior is not correct on some machines,
26271 such as the DEC Alpha, that store short integers in floating-point
26272 registers differently than in integer registers. On those
26273 machines, the default widening will not work correctly and you
26274 must define this macro to suppress that widening in some cases.
26275 See the file `alpha.h' for details.
26277 Do not define this macro if you do not define
26278 `SECONDARY_MEMORY_NEEDED' or if widening MODE to a mode that is
26279 `BITS_PER_WORD' bits wide is correct for your machine.
26281 -- Macro: SMALL_REGISTER_CLASSES
26282 On some machines, it is risky to let hard registers live across
26283 arbitrary insns. Typically, these machines have instructions that
26284 require values to be in specific registers (like an accumulator),
26285 and reload will fail if the required hard register is used for
26286 another purpose across such an insn.
26288 Define `SMALL_REGISTER_CLASSES' to be an expression with a nonzero
26289 value on these machines. When this macro has a nonzero value, the
26290 compiler will try to minimize the lifetime of hard registers.
26292 It is always safe to define this macro with a nonzero value, but
26293 if you unnecessarily define it, you will reduce the amount of
26294 optimizations that can be performed in some cases. If you do not
26295 define this macro with a nonzero value when it is required, the
26296 compiler will run out of spill registers and print a fatal error
26297 message. For most machines, you should not define this macro at
26300 -- Macro: CLASS_LIKELY_SPILLED_P (CLASS)
26301 A C expression whose value is nonzero if pseudos that have been
26302 assigned to registers of class CLASS would likely be spilled
26303 because registers of CLASS are needed for spill registers.
26305 The default value of this macro returns 1 if CLASS has exactly one
26306 register and zero otherwise. On most machines, this default
26307 should be used. Only define this macro to some other expression
26308 if pseudos allocated by `local-alloc.c' end up in memory because
26309 their hard registers were needed for spill registers. If this
26310 macro returns nonzero for those classes, those pseudos will only
26311 be allocated by `global.c', which knows how to reallocate the
26312 pseudo to another register. If there would not be another
26313 register available for reallocation, you should not change the
26314 definition of this macro since the only effect of such a
26315 definition would be to slow down register allocation.
26317 -- Macro: CLASS_MAX_NREGS (CLASS, MODE)
26318 A C expression for the maximum number of consecutive registers of
26319 class CLASS needed to hold a value of mode MODE.
26321 This is closely related to the macro `HARD_REGNO_NREGS'. In fact,
26322 the value of the macro `CLASS_MAX_NREGS (CLASS, MODE)' should be
26323 the maximum value of `HARD_REGNO_NREGS (REGNO, MODE)' for all
26324 REGNO values in the class CLASS.
26326 This macro helps control the handling of multiple-word values in
26329 -- Macro: CANNOT_CHANGE_MODE_CLASS (FROM, TO, CLASS)
26330 If defined, a C expression that returns nonzero for a CLASS for
26331 which a change from mode FROM to mode TO is invalid.
26333 For the example, loading 32-bit integer or floating-point objects
26334 into floating-point registers on the Alpha extends them to 64 bits.
26335 Therefore loading a 64-bit object and then storing it as a 32-bit
26336 object does not store the low-order 32 bits, as would be the case
26337 for a normal register. Therefore, `alpha.h' defines
26338 `CANNOT_CHANGE_MODE_CLASS' as below:
26340 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
26341 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
26342 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
26344 -- Target Hook: const enum reg_class * TARGET_IRA_COVER_CLASSES ()
26345 Return an array of cover classes for the Integrated Register
26346 Allocator (IRA). Cover classes are a set of non-intersecting
26347 register classes covering all hard registers used for register
26348 allocation purposes. If a move between two registers in the same
26349 cover class is possible, it should be cheaper than a load or store
26350 of the registers. The array is terminated by a `LIM_REG_CLASSES'
26353 This hook is called once at compiler startup, after the
26354 command-line options have been processed. It is then re-examined
26355 by every call to `target_reinit'.
26357 The default implementation returns `IRA_COVER_CLASSES', if defined,
26358 otherwise there is no default implementation. You must define
26359 either this macro or `IRA_COVER_CLASSES' in order to use the
26360 integrated register allocator with Chaitin-Briggs coloring. If the
26361 macro is not defined, the only available coloring algorithm is
26362 Chow's priority coloring.
26364 -- Macro: IRA_COVER_CLASSES
26365 See the documentation for `TARGET_IRA_COVER_CLASSES'.
26368 File: gccint.info, Node: Old Constraints, Next: Stack and Calling, Prev: Register Classes, Up: Target Macros
26370 17.9 Obsolete Macros for Defining Constraints
26371 =============================================
26373 Machine-specific constraints can be defined with these macros instead
26374 of the machine description constructs described in *Note Define
26375 Constraints::. This mechanism is obsolete. New ports should not use
26376 it; old ports should convert to the new mechanism.
26378 -- Macro: CONSTRAINT_LEN (CHAR, STR)
26379 For the constraint at the start of STR, which starts with the
26380 letter C, return the length. This allows you to have register
26381 class / constant / extra constraints that are longer than a single
26382 letter; you don't need to define this macro if you can do with
26383 single-letter constraints only. The definition of this macro
26384 should use DEFAULT_CONSTRAINT_LEN for all the characters that you
26385 don't want to handle specially. There are some sanity checks in
26386 genoutput.c that check the constraint lengths for the md file, so
26387 you can also use this macro to help you while you are
26388 transitioning from a byzantine single-letter-constraint scheme:
26389 when you return a negative length for a constraint you want to
26390 re-use, genoutput will complain about every instance where it is
26391 used in the md file.
26393 -- Macro: REG_CLASS_FROM_LETTER (CHAR)
26394 A C expression which defines the machine-dependent operand
26395 constraint letters for register classes. If CHAR is such a
26396 letter, the value should be the register class corresponding to
26397 it. Otherwise, the value should be `NO_REGS'. The register
26398 letter `r', corresponding to class `GENERAL_REGS', will not be
26399 passed to this macro; you do not need to handle it.
26401 -- Macro: REG_CLASS_FROM_CONSTRAINT (CHAR, STR)
26402 Like `REG_CLASS_FROM_LETTER', but you also get the constraint
26403 string passed in STR, so that you can use suffixes to distinguish
26404 between different variants.
26406 -- Macro: CONST_OK_FOR_LETTER_P (VALUE, C)
26407 A C expression that defines the machine-dependent operand
26408 constraint letters (`I', `J', `K', ... `P') that specify
26409 particular ranges of integer values. If C is one of those
26410 letters, the expression should check that VALUE, an integer, is in
26411 the appropriate range and return 1 if so, 0 otherwise. If C is
26412 not one of those letters, the value should be 0 regardless of
26415 -- Macro: CONST_OK_FOR_CONSTRAINT_P (VALUE, C, STR)
26416 Like `CONST_OK_FOR_LETTER_P', but you also get the constraint
26417 string passed in STR, so that you can use suffixes to distinguish
26418 between different variants.
26420 -- Macro: CONST_DOUBLE_OK_FOR_LETTER_P (VALUE, C)
26421 A C expression that defines the machine-dependent operand
26422 constraint letters that specify particular ranges of
26423 `const_double' values (`G' or `H').
26425 If C is one of those letters, the expression should check that
26426 VALUE, an RTX of code `const_double', is in the appropriate range
26427 and return 1 if so, 0 otherwise. If C is not one of those
26428 letters, the value should be 0 regardless of VALUE.
26430 `const_double' is used for all floating-point constants and for
26431 `DImode' fixed-point constants. A given letter can accept either
26432 or both kinds of values. It can use `GET_MODE' to distinguish
26433 between these kinds.
26435 -- Macro: CONST_DOUBLE_OK_FOR_CONSTRAINT_P (VALUE, C, STR)
26436 Like `CONST_DOUBLE_OK_FOR_LETTER_P', but you also get the
26437 constraint string passed in STR, so that you can use suffixes to
26438 distinguish between different variants.
26440 -- Macro: EXTRA_CONSTRAINT (VALUE, C)
26441 A C expression that defines the optional machine-dependent
26442 constraint letters that can be used to segregate specific types of
26443 operands, usually memory references, for the target machine. Any
26444 letter that is not elsewhere defined and not matched by
26445 `REG_CLASS_FROM_LETTER' / `REG_CLASS_FROM_CONSTRAINT' may be used.
26446 Normally this macro will not be defined.
26448 If it is required for a particular target machine, it should
26449 return 1 if VALUE corresponds to the operand type represented by
26450 the constraint letter C. If C is not defined as an extra
26451 constraint, the value returned should be 0 regardless of VALUE.
26453 For example, on the ROMP, load instructions cannot have their
26454 output in r0 if the memory reference contains a symbolic address.
26455 Constraint letter `Q' is defined as representing a memory address
26456 that does _not_ contain a symbolic address. An alternative is
26457 specified with a `Q' constraint on the input and `r' on the
26458 output. The next alternative specifies `m' on the input and a
26459 register class that does not include r0 on the output.
26461 -- Macro: EXTRA_CONSTRAINT_STR (VALUE, C, STR)
26462 Like `EXTRA_CONSTRAINT', but you also get the constraint string
26463 passed in STR, so that you can use suffixes to distinguish between
26464 different variants.
26466 -- Macro: EXTRA_MEMORY_CONSTRAINT (C, STR)
26467 A C expression that defines the optional machine-dependent
26468 constraint letters, amongst those accepted by `EXTRA_CONSTRAINT',
26469 that should be treated like memory constraints by the reload pass.
26471 It should return 1 if the operand type represented by the
26472 constraint at the start of STR, the first letter of which is the
26473 letter C, comprises a subset of all memory references including
26474 all those whose address is simply a base register. This allows
26475 the reload pass to reload an operand, if it does not directly
26476 correspond to the operand type of C, by copying its address into a
26479 For example, on the S/390, some instructions do not accept
26480 arbitrary memory references, but only those that do not make use
26481 of an index register. The constraint letter `Q' is defined via
26482 `EXTRA_CONSTRAINT' as representing a memory address of this type.
26483 If the letter `Q' is marked as `EXTRA_MEMORY_CONSTRAINT', a `Q'
26484 constraint can handle any memory operand, because the reload pass
26485 knows it can be reloaded by copying the memory address into a base
26486 register if required. This is analogous to the way a `o'
26487 constraint can handle any memory operand.
26489 -- Macro: EXTRA_ADDRESS_CONSTRAINT (C, STR)
26490 A C expression that defines the optional machine-dependent
26491 constraint letters, amongst those accepted by `EXTRA_CONSTRAINT' /
26492 `EXTRA_CONSTRAINT_STR', that should be treated like address
26493 constraints by the reload pass.
26495 It should return 1 if the operand type represented by the
26496 constraint at the start of STR, which starts with the letter C,
26497 comprises a subset of all memory addresses including all those
26498 that consist of just a base register. This allows the reload pass
26499 to reload an operand, if it does not directly correspond to the
26500 operand type of STR, by copying it into a base register.
26502 Any constraint marked as `EXTRA_ADDRESS_CONSTRAINT' can only be
26503 used with the `address_operand' predicate. It is treated
26504 analogously to the `p' constraint.
26507 File: gccint.info, Node: Stack and Calling, Next: Varargs, Prev: Old Constraints, Up: Target Macros
26509 17.10 Stack Layout and Calling Conventions
26510 ==========================================
26512 This describes the stack layout and calling conventions.
26517 * Exception Handling::
26519 * Frame Registers::
26521 * Stack Arguments::
26522 * Register Arguments::
26524 * Aggregate Return::
26529 * Stack Smashing Protection::
26532 File: gccint.info, Node: Frame Layout, Next: Exception Handling, Up: Stack and Calling
26534 17.10.1 Basic Stack Layout
26535 --------------------------
26537 Here is the basic stack layout.
26539 -- Macro: STACK_GROWS_DOWNWARD
26540 Define this macro if pushing a word onto the stack moves the stack
26541 pointer to a smaller address.
26543 When we say, "define this macro if ...", it means that the
26544 compiler checks this macro only with `#ifdef' so the precise
26545 definition used does not matter.
26547 -- Macro: STACK_PUSH_CODE
26548 This macro defines the operation used when something is pushed on
26549 the stack. In RTL, a push operation will be `(set (mem
26550 (STACK_PUSH_CODE (reg sp))) ...)'
26552 The choices are `PRE_DEC', `POST_DEC', `PRE_INC', and `POST_INC'.
26553 Which of these is correct depends on the stack direction and on
26554 whether the stack pointer points to the last item on the stack or
26555 whether it points to the space for the next item on the stack.
26557 The default is `PRE_DEC' when `STACK_GROWS_DOWNWARD' is defined,
26558 which is almost always right, and `PRE_INC' otherwise, which is
26561 -- Macro: FRAME_GROWS_DOWNWARD
26562 Define this macro to nonzero value if the addresses of local
26563 variable slots are at negative offsets from the frame pointer.
26565 -- Macro: ARGS_GROW_DOWNWARD
26566 Define this macro if successive arguments to a function occupy
26567 decreasing addresses on the stack.
26569 -- Macro: STARTING_FRAME_OFFSET
26570 Offset from the frame pointer to the first local variable slot to
26573 If `FRAME_GROWS_DOWNWARD', find the next slot's offset by
26574 subtracting the first slot's length from `STARTING_FRAME_OFFSET'.
26575 Otherwise, it is found by adding the length of the first slot to
26576 the value `STARTING_FRAME_OFFSET'.
26578 -- Macro: STACK_ALIGNMENT_NEEDED
26579 Define to zero to disable final alignment of the stack during
26580 reload. The nonzero default for this macro is suitable for most
26583 On ports where `STARTING_FRAME_OFFSET' is nonzero or where there
26584 is a register save block following the local block that doesn't
26585 require alignment to `STACK_BOUNDARY', it may be beneficial to
26586 disable stack alignment and do it in the backend.
26588 -- Macro: STACK_POINTER_OFFSET
26589 Offset from the stack pointer register to the first location at
26590 which outgoing arguments are placed. If not specified, the
26591 default value of zero is used. This is the proper value for most
26594 If `ARGS_GROW_DOWNWARD', this is the offset to the location above
26595 the first location at which outgoing arguments are placed.
26597 -- Macro: FIRST_PARM_OFFSET (FUNDECL)
26598 Offset from the argument pointer register to the first argument's
26599 address. On some machines it may depend on the data type of the
26602 If `ARGS_GROW_DOWNWARD', this is the offset to the location above
26603 the first argument's address.
26605 -- Macro: STACK_DYNAMIC_OFFSET (FUNDECL)
26606 Offset from the stack pointer register to an item dynamically
26607 allocated on the stack, e.g., by `alloca'.
26609 The default value for this macro is `STACK_POINTER_OFFSET' plus the
26610 length of the outgoing arguments. The default is correct for most
26611 machines. See `function.c' for details.
26613 -- Macro: INITIAL_FRAME_ADDRESS_RTX
26614 A C expression whose value is RTL representing the address of the
26615 initial stack frame. This address is passed to `RETURN_ADDR_RTX'
26616 and `DYNAMIC_CHAIN_ADDRESS'. If you don't define this macro, a
26617 reasonable default value will be used. Define this macro in order
26618 to make frame pointer elimination work in the presence of
26619 `__builtin_frame_address (count)' and `__builtin_return_address
26620 (count)' for `count' not equal to zero.
26622 -- Macro: DYNAMIC_CHAIN_ADDRESS (FRAMEADDR)
26623 A C expression whose value is RTL representing the address in a
26624 stack frame where the pointer to the caller's frame is stored.
26625 Assume that FRAMEADDR is an RTL expression for the address of the
26626 stack frame itself.
26628 If you don't define this macro, the default is to return the value
26629 of FRAMEADDR--that is, the stack frame address is also the address
26630 of the stack word that points to the previous frame.
26632 -- Macro: SETUP_FRAME_ADDRESSES
26633 If defined, a C expression that produces the machine-specific code
26634 to setup the stack so that arbitrary frames can be accessed. For
26635 example, on the SPARC, we must flush all of the register windows
26636 to the stack before we can access arbitrary stack frames. You
26637 will seldom need to define this macro.
26639 -- Target Hook: bool TARGET_BUILTIN_SETJMP_FRAME_VALUE ()
26640 This target hook should return an rtx that is used to store the
26641 address of the current frame into the built in `setjmp' buffer.
26642 The default value, `virtual_stack_vars_rtx', is correct for most
26643 machines. One reason you may need to define this target hook is if
26644 `hard_frame_pointer_rtx' is the appropriate value on your machine.
26646 -- Macro: FRAME_ADDR_RTX (FRAMEADDR)
26647 A C expression whose value is RTL representing the value of the
26648 frame address for the current frame. FRAMEADDR is the frame
26649 pointer of the current frame. This is used for
26650 __builtin_frame_address. You need only define this macro if the
26651 frame address is not the same as the frame pointer. Most machines
26652 do not need to define it.
26654 -- Macro: RETURN_ADDR_RTX (COUNT, FRAMEADDR)
26655 A C expression whose value is RTL representing the value of the
26656 return address for the frame COUNT steps up from the current
26657 frame, after the prologue. FRAMEADDR is the frame pointer of the
26658 COUNT frame, or the frame pointer of the COUNT - 1 frame if
26659 `RETURN_ADDR_IN_PREVIOUS_FRAME' is defined.
26661 The value of the expression must always be the correct address when
26662 COUNT is zero, but may be `NULL_RTX' if there is no way to
26663 determine the return address of other frames.
26665 -- Macro: RETURN_ADDR_IN_PREVIOUS_FRAME
26666 Define this if the return address of a particular stack frame is
26667 accessed from the frame pointer of the previous stack frame.
26669 -- Macro: INCOMING_RETURN_ADDR_RTX
26670 A C expression whose value is RTL representing the location of the
26671 incoming return address at the beginning of any function, before
26672 the prologue. This RTL is either a `REG', indicating that the
26673 return value is saved in `REG', or a `MEM' representing a location
26676 You only need to define this macro if you want to support call
26677 frame debugging information like that provided by DWARF 2.
26679 If this RTL is a `REG', you should also define
26680 `DWARF_FRAME_RETURN_COLUMN' to `DWARF_FRAME_REGNUM (REGNO)'.
26682 -- Macro: DWARF_ALT_FRAME_RETURN_COLUMN
26683 A C expression whose value is an integer giving a DWARF 2 column
26684 number that may be used as an alternative return column. The
26685 column must not correspond to any gcc hard register (that is, it
26686 must not be in the range of `DWARF_FRAME_REGNUM').
26688 This macro can be useful if `DWARF_FRAME_RETURN_COLUMN' is set to a
26689 general register, but an alternative column needs to be used for
26690 signal frames. Some targets have also used different frame return
26693 -- Macro: DWARF_ZERO_REG
26694 A C expression whose value is an integer giving a DWARF 2 register
26695 number that is considered to always have the value zero. This
26696 should only be defined if the target has an architected zero
26697 register, and someone decided it was a good idea to use that
26698 register number to terminate the stack backtrace. New ports
26701 -- Target Hook: void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char
26702 *LABEL, rtx PATTERN, int INDEX)
26703 This target hook allows the backend to emit frame-related insns
26704 that contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame
26705 debugging info engine will invoke it on insns of the form
26706 (set (reg) (unspec [...] UNSPEC_INDEX))
26708 (set (reg) (unspec_volatile [...] UNSPECV_INDEX)).
26709 to let the backend emit the call frame instructions. LABEL is the
26710 CFI label attached to the insn, PATTERN is the pattern of the insn
26711 and INDEX is `UNSPEC_INDEX' or `UNSPECV_INDEX'.
26713 -- Macro: INCOMING_FRAME_SP_OFFSET
26714 A C expression whose value is an integer giving the offset, in
26715 bytes, from the value of the stack pointer register to the top of
26716 the stack frame at the beginning of any function, before the
26717 prologue. The top of the frame is defined to be the value of the
26718 stack pointer in the previous frame, just before the call
26721 You only need to define this macro if you want to support call
26722 frame debugging information like that provided by DWARF 2.
26724 -- Macro: ARG_POINTER_CFA_OFFSET (FUNDECL)
26725 A C expression whose value is an integer giving the offset, in
26726 bytes, from the argument pointer to the canonical frame address
26727 (cfa). The final value should coincide with that calculated by
26728 `INCOMING_FRAME_SP_OFFSET'. Which is unfortunately not usable
26729 during virtual register instantiation.
26731 The default value for this macro is `FIRST_PARM_OFFSET (fundecl)',
26732 which is correct for most machines; in general, the arguments are
26733 found immediately before the stack frame. Note that this is not
26734 the case on some targets that save registers into the caller's
26735 frame, such as SPARC and rs6000, and so such targets need to
26738 You only need to define this macro if the default is incorrect,
26739 and you want to support call frame debugging information like that
26740 provided by DWARF 2.
26742 -- Macro: FRAME_POINTER_CFA_OFFSET (FUNDECL)
26743 If defined, a C expression whose value is an integer giving the
26744 offset in bytes from the frame pointer to the canonical frame
26745 address (cfa). The final value should coincide with that
26746 calculated by `INCOMING_FRAME_SP_OFFSET'.
26748 Normally the CFA is calculated as an offset from the argument
26749 pointer, via `ARG_POINTER_CFA_OFFSET', but if the argument pointer
26750 is variable due to the ABI, this may not be possible. If this
26751 macro is defined, it implies that the virtual register
26752 instantiation should be based on the frame pointer instead of the
26753 argument pointer. Only one of `FRAME_POINTER_CFA_OFFSET' and
26754 `ARG_POINTER_CFA_OFFSET' should be defined.
26756 -- Macro: CFA_FRAME_BASE_OFFSET (FUNDECL)
26757 If defined, a C expression whose value is an integer giving the
26758 offset in bytes from the canonical frame address (cfa) to the
26759 frame base used in DWARF 2 debug information. The default is
26760 zero. A different value may reduce the size of debug information
26764 File: gccint.info, Node: Exception Handling, Next: Stack Checking, Prev: Frame Layout, Up: Stack and Calling
26766 17.10.2 Exception Handling Support
26767 ----------------------------------
26769 -- Macro: EH_RETURN_DATA_REGNO (N)
26770 A C expression whose value is the Nth register number used for
26771 data by exception handlers, or `INVALID_REGNUM' if fewer than N
26772 registers are usable.
26774 The exception handling library routines communicate with the
26775 exception handlers via a set of agreed upon registers. Ideally
26776 these registers should be call-clobbered; it is possible to use
26777 call-saved registers, but may negatively impact code size. The
26778 target must support at least 2 data registers, but should define 4
26779 if there are enough free registers.
26781 You must define this macro if you want to support call frame
26782 exception handling like that provided by DWARF 2.
26784 -- Macro: EH_RETURN_STACKADJ_RTX
26785 A C expression whose value is RTL representing a location in which
26786 to store a stack adjustment to be applied before function return.
26787 This is used to unwind the stack to an exception handler's call
26788 frame. It will be assigned zero on code paths that return
26791 Typically this is a call-clobbered hard register that is otherwise
26792 untouched by the epilogue, but could also be a stack slot.
26794 Do not define this macro if the stack pointer is saved and restored
26795 by the regular prolog and epilog code in the call frame itself; in
26796 this case, the exception handling library routines will update the
26797 stack location to be restored in place. Otherwise, you must define
26798 this macro if you want to support call frame exception handling
26799 like that provided by DWARF 2.
26801 -- Macro: EH_RETURN_HANDLER_RTX
26802 A C expression whose value is RTL representing a location in which
26803 to store the address of an exception handler to which we should
26804 return. It will not be assigned on code paths that return
26807 Typically this is the location in the call frame at which the
26808 normal return address is stored. For targets that return by
26809 popping an address off the stack, this might be a memory address
26810 just below the _target_ call frame rather than inside the current
26811 call frame. If defined, `EH_RETURN_STACKADJ_RTX' will have already
26812 been assigned, so it may be used to calculate the location of the
26815 Some targets have more complex requirements than storing to an
26816 address calculable during initial code generation. In that case
26817 the `eh_return' instruction pattern should be used instead.
26819 If you want to support call frame exception handling, you must
26820 define either this macro or the `eh_return' instruction pattern.
26822 -- Macro: RETURN_ADDR_OFFSET
26823 If defined, an integer-valued C expression for which rtl will be
26824 generated to add it to the exception handler address before it is
26825 searched in the exception handling tables, and to subtract it
26826 again from the address before using it to return to the exception
26829 -- Macro: ASM_PREFERRED_EH_DATA_FORMAT (CODE, GLOBAL)
26830 This macro chooses the encoding of pointers embedded in the
26831 exception handling sections. If at all possible, this should be
26832 defined such that the exception handling section will not require
26833 dynamic relocations, and so may be read-only.
26835 CODE is 0 for data, 1 for code labels, 2 for function pointers.
26836 GLOBAL is true if the symbol may be affected by dynamic
26837 relocations. The macro should return a combination of the
26838 `DW_EH_PE_*' defines as found in `dwarf2.h'.
26840 If this macro is not defined, pointers will not be encoded but
26841 represented directly.
26843 -- Macro: ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (FILE, ENCODING, SIZE,
26845 This macro allows the target to emit whatever special magic is
26846 required to represent the encoding chosen by
26847 `ASM_PREFERRED_EH_DATA_FORMAT'. Generic code takes care of
26848 pc-relative and indirect encodings; this must be defined if the
26849 target uses text-relative or data-relative encodings.
26851 This is a C statement that branches to DONE if the format was
26852 handled. ENCODING is the format chosen, SIZE is the number of
26853 bytes that the format occupies, ADDR is the `SYMBOL_REF' to be
26856 -- Macro: MD_UNWIND_SUPPORT
26857 A string specifying a file to be #include'd in unwind-dw2.c. The
26858 file so included typically defines `MD_FALLBACK_FRAME_STATE_FOR'.
26860 -- Macro: MD_FALLBACK_FRAME_STATE_FOR (CONTEXT, FS)
26861 This macro allows the target to add CPU and operating system
26862 specific code to the call-frame unwinder for use when there is no
26863 unwind data available. The most common reason to implement this
26864 macro is to unwind through signal frames.
26866 This macro is called from `uw_frame_state_for' in `unwind-dw2.c',
26867 `unwind-dw2-xtensa.c' and `unwind-ia64.c'. CONTEXT is an
26868 `_Unwind_Context'; FS is an `_Unwind_FrameState'. Examine
26869 `context->ra' for the address of the code being executed and
26870 `context->cfa' for the stack pointer value. If the frame can be
26871 decoded, the register save addresses should be updated in FS and
26872 the macro should evaluate to `_URC_NO_REASON'. If the frame
26873 cannot be decoded, the macro should evaluate to
26874 `_URC_END_OF_STACK'.
26876 For proper signal handling in Java this macro is accompanied by
26877 `MAKE_THROW_FRAME', defined in `libjava/include/*-signal.h'
26880 -- Macro: MD_HANDLE_UNWABI (CONTEXT, FS)
26881 This macro allows the target to add operating system specific code
26882 to the call-frame unwinder to handle the IA-64 `.unwabi' unwinding
26883 directive, usually used for signal or interrupt frames.
26885 This macro is called from `uw_update_context' in `unwind-ia64.c'.
26886 CONTEXT is an `_Unwind_Context'; FS is an `_Unwind_FrameState'.
26887 Examine `fs->unwabi' for the abi and context in the `.unwabi'
26888 directive. If the `.unwabi' directive can be handled, the
26889 register save addresses should be updated in FS.
26891 -- Macro: TARGET_USES_WEAK_UNWIND_INFO
26892 A C expression that evaluates to true if the target requires unwind
26893 info to be given comdat linkage. Define it to be `1' if comdat
26894 linkage is necessary. The default is `0'.
26897 File: gccint.info, Node: Stack Checking, Next: Frame Registers, Prev: Exception Handling, Up: Stack and Calling
26899 17.10.3 Specifying How Stack Checking is Done
26900 ---------------------------------------------
26902 GCC will check that stack references are within the boundaries of the
26903 stack, if the option `-fstack-check' is specified, in one of three ways:
26905 1. If the value of the `STACK_CHECK_BUILTIN' macro is nonzero, GCC
26906 will assume that you have arranged for full stack checking to be
26907 done at appropriate places in the configuration files. GCC will
26908 not do other special processing.
26910 2. If `STACK_CHECK_BUILTIN' is zero and the value of the
26911 `STACK_CHECK_STATIC_BUILTIN' macro is nonzero, GCC will assume
26912 that you have arranged for static stack checking (checking of the
26913 static stack frame of functions) to be done at appropriate places
26914 in the configuration files. GCC will only emit code to do dynamic
26915 stack checking (checking on dynamic stack allocations) using the
26916 third approach below.
26918 3. If neither of the above are true, GCC will generate code to
26919 periodically "probe" the stack pointer using the values of the
26920 macros defined below.
26922 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is
26923 defined, GCC will change its allocation strategy for large objects if
26924 the option `-fstack-check' is specified: they will always be allocated
26925 dynamically if their size exceeds `STACK_CHECK_MAX_VAR_SIZE' bytes.
26927 -- Macro: STACK_CHECK_BUILTIN
26928 A nonzero value if stack checking is done by the configuration
26929 files in a machine-dependent manner. You should define this macro
26930 if stack checking is require by the ABI of your machine or if you
26931 would like to do stack checking in some more efficient way than
26932 the generic approach. The default value of this macro is zero.
26934 -- Macro: STACK_CHECK_STATIC_BUILTIN
26935 A nonzero value if static stack checking is done by the
26936 configuration files in a machine-dependent manner. You should
26937 define this macro if you would like to do static stack checking in
26938 some more efficient way than the generic approach. The default
26939 value of this macro is zero.
26941 -- Macro: STACK_CHECK_PROBE_INTERVAL
26942 An integer representing the interval at which GCC must generate
26943 stack probe instructions. You will normally define this macro to
26944 be no larger than the size of the "guard pages" at the end of a
26945 stack area. The default value of 4096 is suitable for most
26948 -- Macro: STACK_CHECK_PROBE_LOAD
26949 An integer which is nonzero if GCC should perform the stack probe
26950 as a load instruction and zero if GCC should use a store
26951 instruction. The default is zero, which is the most efficient
26952 choice on most systems.
26954 -- Macro: STACK_CHECK_PROTECT
26955 The number of bytes of stack needed to recover from a stack
26956 overflow, for languages where such a recovery is supported. The
26957 default value of 75 words should be adequate for most machines.
26959 The following macros are relevant only if neither STACK_CHECK_BUILTIN
26960 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
26961 in the opposite case.
26963 -- Macro: STACK_CHECK_MAX_FRAME_SIZE
26964 The maximum size of a stack frame, in bytes. GCC will generate
26965 probe instructions in non-leaf functions to ensure at least this
26966 many bytes of stack are available. If a stack frame is larger
26967 than this size, stack checking will not be reliable and GCC will
26968 issue a warning. The default is chosen so that GCC only generates
26969 one instruction on most systems. You should normally not change
26970 the default value of this macro.
26972 -- Macro: STACK_CHECK_FIXED_FRAME_SIZE
26973 GCC uses this value to generate the above warning message. It
26974 represents the amount of fixed frame used by a function, not
26975 including space for any callee-saved registers, temporaries and
26976 user variables. You need only specify an upper bound for this
26977 amount and will normally use the default of four words.
26979 -- Macro: STACK_CHECK_MAX_VAR_SIZE
26980 The maximum size, in bytes, of an object that GCC will place in the
26981 fixed area of the stack frame when the user specifies
26982 `-fstack-check'. GCC computed the default from the values of the
26983 above macros and you will normally not need to override that
26987 File: gccint.info, Node: Frame Registers, Next: Elimination, Prev: Stack Checking, Up: Stack and Calling
26989 17.10.4 Registers That Address the Stack Frame
26990 ----------------------------------------------
26992 This discusses registers that address the stack frame.
26994 -- Macro: STACK_POINTER_REGNUM
26995 The register number of the stack pointer register, which must also
26996 be a fixed register according to `FIXED_REGISTERS'. On most
26997 machines, the hardware determines which register this is.
26999 -- Macro: FRAME_POINTER_REGNUM
27000 The register number of the frame pointer register, which is used to
27001 access automatic variables in the stack frame. On some machines,
27002 the hardware determines which register this is. On other
27003 machines, you can choose any register you wish for this purpose.
27005 -- Macro: HARD_FRAME_POINTER_REGNUM
27006 On some machines the offset between the frame pointer and starting
27007 offset of the automatic variables is not known until after register
27008 allocation has been done (for example, because the saved registers
27009 are between these two locations). On those machines, define
27010 `FRAME_POINTER_REGNUM' the number of a special, fixed register to
27011 be used internally until the offset is known, and define
27012 `HARD_FRAME_POINTER_REGNUM' to be the actual hard register number
27013 used for the frame pointer.
27015 You should define this macro only in the very rare circumstances
27016 when it is not possible to calculate the offset between the frame
27017 pointer and the automatic variables until after register
27018 allocation has been completed. When this macro is defined, you
27019 must also indicate in your definition of `ELIMINABLE_REGS' how to
27020 eliminate `FRAME_POINTER_REGNUM' into either
27021 `HARD_FRAME_POINTER_REGNUM' or `STACK_POINTER_REGNUM'.
27023 Do not define this macro if it would be the same as
27024 `FRAME_POINTER_REGNUM'.
27026 -- Macro: ARG_POINTER_REGNUM
27027 The register number of the arg pointer register, which is used to
27028 access the function's argument list. On some machines, this is
27029 the same as the frame pointer register. On some machines, the
27030 hardware determines which register this is. On other machines,
27031 you can choose any register you wish for this purpose. If this is
27032 not the same register as the frame pointer register, then you must
27033 mark it as a fixed register according to `FIXED_REGISTERS', or
27034 arrange to be able to eliminate it (*note Elimination::).
27036 -- Macro: RETURN_ADDRESS_POINTER_REGNUM
27037 The register number of the return address pointer register, which
27038 is used to access the current function's return address from the
27039 stack. On some machines, the return address is not at a fixed
27040 offset from the frame pointer or stack pointer or argument
27041 pointer. This register can be defined to point to the return
27042 address on the stack, and then be converted by `ELIMINABLE_REGS'
27043 into either the frame pointer or stack pointer.
27045 Do not define this macro unless there is no other way to get the
27046 return address from the stack.
27048 -- Macro: STATIC_CHAIN_REGNUM
27049 -- Macro: STATIC_CHAIN_INCOMING_REGNUM
27050 Register numbers used for passing a function's static chain
27051 pointer. If register windows are used, the register number as
27052 seen by the called function is `STATIC_CHAIN_INCOMING_REGNUM',
27053 while the register number as seen by the calling function is
27054 `STATIC_CHAIN_REGNUM'. If these registers are the same,
27055 `STATIC_CHAIN_INCOMING_REGNUM' need not be defined.
27057 The static chain register need not be a fixed register.
27059 If the static chain is passed in memory, these macros should not be
27060 defined; instead, the next two macros should be defined.
27062 -- Macro: STATIC_CHAIN
27063 -- Macro: STATIC_CHAIN_INCOMING
27064 If the static chain is passed in memory, these macros provide rtx
27065 giving `mem' expressions that denote where they are stored.
27066 `STATIC_CHAIN' and `STATIC_CHAIN_INCOMING' give the locations as
27067 seen by the calling and called functions, respectively. Often the
27068 former will be at an offset from the stack pointer and the latter
27069 at an offset from the frame pointer.
27071 The variables `stack_pointer_rtx', `frame_pointer_rtx', and
27072 `arg_pointer_rtx' will have been initialized prior to the use of
27073 these macros and should be used to refer to those items.
27075 If the static chain is passed in a register, the two previous
27076 macros should be defined instead.
27078 -- Macro: DWARF_FRAME_REGISTERS
27079 This macro specifies the maximum number of hard registers that can
27080 be saved in a call frame. This is used to size data structures
27081 used in DWARF2 exception handling.
27083 Prior to GCC 3.0, this macro was needed in order to establish a
27084 stable exception handling ABI in the face of adding new hard
27085 registers for ISA extensions. In GCC 3.0 and later, the EH ABI is
27086 insulated from changes in the number of hard registers.
27087 Nevertheless, this macro can still be used to reduce the runtime
27088 memory requirements of the exception handling routines, which can
27089 be substantial if the ISA contains a lot of registers that are not
27092 If this macro is not defined, it defaults to
27093 `FIRST_PSEUDO_REGISTER'.
27095 -- Macro: PRE_GCC3_DWARF_FRAME_REGISTERS
27096 This macro is similar to `DWARF_FRAME_REGISTERS', but is provided
27097 for backward compatibility in pre GCC 3.0 compiled code.
27099 If this macro is not defined, it defaults to
27100 `DWARF_FRAME_REGISTERS'.
27102 -- Macro: DWARF_REG_TO_UNWIND_COLUMN (REGNO)
27103 Define this macro if the target's representation for dwarf
27104 registers is different than the internal representation for unwind
27105 column. Given a dwarf register, this macro should return the
27106 internal unwind column number to use instead.
27108 See the PowerPC's SPE target for an example.
27110 -- Macro: DWARF_FRAME_REGNUM (REGNO)
27111 Define this macro if the target's representation for dwarf
27112 registers used in .eh_frame or .debug_frame is different from that
27113 used in other debug info sections. Given a GCC hard register
27114 number, this macro should return the .eh_frame register number.
27115 The default is `DBX_REGISTER_NUMBER (REGNO)'.
27118 -- Macro: DWARF2_FRAME_REG_OUT (REGNO, FOR_EH)
27119 Define this macro to map register numbers held in the call frame
27120 info that GCC has collected using `DWARF_FRAME_REGNUM' to those
27121 that should be output in .debug_frame (`FOR_EH' is zero) and
27122 .eh_frame (`FOR_EH' is nonzero). The default is to return `REGNO'.
27126 File: gccint.info, Node: Elimination, Next: Stack Arguments, Prev: Frame Registers, Up: Stack and Calling
27128 17.10.5 Eliminating Frame Pointer and Arg Pointer
27129 -------------------------------------------------
27131 This is about eliminating the frame pointer and arg pointer.
27133 -- Macro: FRAME_POINTER_REQUIRED
27134 A C expression which is nonzero if a function must have and use a
27135 frame pointer. This expression is evaluated in the reload pass.
27136 If its value is nonzero the function will have a frame pointer.
27138 The expression can in principle examine the current function and
27139 decide according to the facts, but on most machines the constant 0
27140 or the constant 1 suffices. Use 0 when the machine allows code to
27141 be generated with no frame pointer, and doing so saves some time
27142 or space. Use 1 when there is no possible advantage to avoiding a
27145 In certain cases, the compiler does not know how to produce valid
27146 code without a frame pointer. The compiler recognizes those cases
27147 and automatically gives the function a frame pointer regardless of
27148 what `FRAME_POINTER_REQUIRED' says. You don't need to worry about
27151 In a function that does not require a frame pointer, the frame
27152 pointer register can be allocated for ordinary usage, unless you
27153 mark it as a fixed register. See `FIXED_REGISTERS' for more
27156 -- Macro: INITIAL_FRAME_POINTER_OFFSET (DEPTH-VAR)
27157 A C statement to store in the variable DEPTH-VAR the difference
27158 between the frame pointer and the stack pointer values immediately
27159 after the function prologue. The value would be computed from
27160 information such as the result of `get_frame_size ()' and the
27161 tables of registers `regs_ever_live' and `call_used_regs'.
27163 If `ELIMINABLE_REGS' is defined, this macro will be not be used and
27164 need not be defined. Otherwise, it must be defined even if
27165 `FRAME_POINTER_REQUIRED' is defined to always be true; in that
27166 case, you may set DEPTH-VAR to anything.
27168 -- Macro: ELIMINABLE_REGS
27169 If defined, this macro specifies a table of register pairs used to
27170 eliminate unneeded registers that point into the stack frame. If
27171 it is not defined, the only elimination attempted by the compiler
27172 is to replace references to the frame pointer with references to
27175 The definition of this macro is a list of structure
27176 initializations, each of which specifies an original and
27177 replacement register.
27179 On some machines, the position of the argument pointer is not
27180 known until the compilation is completed. In such a case, a
27181 separate hard register must be used for the argument pointer.
27182 This register can be eliminated by replacing it with either the
27183 frame pointer or the argument pointer, depending on whether or not
27184 the frame pointer has been eliminated.
27186 In this case, you might specify:
27187 #define ELIMINABLE_REGS \
27188 {{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM}, \
27189 {ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM}, \
27190 {FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM}}
27192 Note that the elimination of the argument pointer with the stack
27193 pointer is specified first since that is the preferred elimination.
27195 -- Macro: CAN_ELIMINATE (FROM-REG, TO-REG)
27196 A C expression that returns nonzero if the compiler is allowed to
27197 try to replace register number FROM-REG with register number
27198 TO-REG. This macro need only be defined if `ELIMINABLE_REGS' is
27199 defined, and will usually be the constant 1, since most of the
27200 cases preventing register elimination are things that the compiler
27201 already knows about.
27203 -- Macro: INITIAL_ELIMINATION_OFFSET (FROM-REG, TO-REG, OFFSET-VAR)
27204 This macro is similar to `INITIAL_FRAME_POINTER_OFFSET'. It
27205 specifies the initial difference between the specified pair of
27206 registers. This macro must be defined if `ELIMINABLE_REGS' is
27210 File: gccint.info, Node: Stack Arguments, Next: Register Arguments, Prev: Elimination, Up: Stack and Calling
27212 17.10.6 Passing Function Arguments on the Stack
27213 -----------------------------------------------
27215 The macros in this section control how arguments are passed on the
27216 stack. See the following section for other macros that control passing
27217 certain arguments in registers.
27219 -- Target Hook: bool TARGET_PROMOTE_PROTOTYPES (tree FNTYPE)
27220 This target hook returns `true' if an argument declared in a
27221 prototype as an integral type smaller than `int' should actually be
27222 passed as an `int'. In addition to avoiding errors in certain
27223 cases of mismatch, it also makes for better code on certain
27224 machines. The default is to not promote prototypes.
27226 -- Macro: PUSH_ARGS
27227 A C expression. If nonzero, push insns will be used to pass
27228 outgoing arguments. If the target machine does not have a push
27229 instruction, set it to zero. That directs GCC to use an alternate
27230 strategy: to allocate the entire argument block and then store the
27231 arguments into it. When `PUSH_ARGS' is nonzero, `PUSH_ROUNDING'
27232 must be defined too.
27234 -- Macro: PUSH_ARGS_REVERSED
27235 A C expression. If nonzero, function arguments will be evaluated
27236 from last to first, rather than from first to last. If this macro
27237 is not defined, it defaults to `PUSH_ARGS' on targets where the
27238 stack and args grow in opposite directions, and 0 otherwise.
27240 -- Macro: PUSH_ROUNDING (NPUSHED)
27241 A C expression that is the number of bytes actually pushed onto the
27242 stack when an instruction attempts to push NPUSHED bytes.
27244 On some machines, the definition
27246 #define PUSH_ROUNDING(BYTES) (BYTES)
27248 will suffice. But on other machines, instructions that appear to
27249 push one byte actually push two bytes in an attempt to maintain
27250 alignment. Then the definition should be
27252 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
27254 -- Macro: ACCUMULATE_OUTGOING_ARGS
27255 A C expression. If nonzero, the maximum amount of space required
27256 for outgoing arguments will be computed and placed into the
27257 variable `current_function_outgoing_args_size'. No space will be
27258 pushed onto the stack for each call; instead, the function
27259 prologue should increase the stack frame size by this amount.
27261 Setting both `PUSH_ARGS' and `ACCUMULATE_OUTGOING_ARGS' is not
27264 -- Macro: REG_PARM_STACK_SPACE (FNDECL)
27265 Define this macro if functions should assume that stack space has
27266 been allocated for arguments even when their values are passed in
27269 The value of this macro is the size, in bytes, of the area
27270 reserved for arguments passed in registers for the function
27271 represented by FNDECL, which can be zero if GCC is calling a
27272 library function. The argument FNDECL can be the FUNCTION_DECL,
27273 or the type itself of the function.
27275 This space can be allocated by the caller, or be a part of the
27276 machine-dependent stack frame: `OUTGOING_REG_PARM_STACK_SPACE' says
27279 -- Macro: OUTGOING_REG_PARM_STACK_SPACE (FNTYPE)
27280 Define this to a nonzero value if it is the responsibility of the
27281 caller to allocate the area reserved for arguments passed in
27282 registers when calling a function of FNTYPE. FNTYPE may be NULL
27283 if the function called is a library function.
27285 If `ACCUMULATE_OUTGOING_ARGS' is defined, this macro controls
27286 whether the space for these arguments counts in the value of
27287 `current_function_outgoing_args_size'.
27289 -- Macro: STACK_PARMS_IN_REG_PARM_AREA
27290 Define this macro if `REG_PARM_STACK_SPACE' is defined, but the
27291 stack parameters don't skip the area specified by it.
27293 Normally, when a parameter is not passed in registers, it is
27294 placed on the stack beyond the `REG_PARM_STACK_SPACE' area.
27295 Defining this macro suppresses this behavior and causes the
27296 parameter to be passed on the stack in its natural location.
27298 -- Macro: RETURN_POPS_ARGS (FUNDECL, FUNTYPE, STACK-SIZE)
27299 A C expression that should indicate the number of bytes of its own
27300 arguments that a function pops on returning, or 0 if the function
27301 pops no arguments and the caller must therefore pop them all after
27302 the function returns.
27304 FUNDECL is a C variable whose value is a tree node that describes
27305 the function in question. Normally it is a node of type
27306 `FUNCTION_DECL' that describes the declaration of the function.
27307 From this you can obtain the `DECL_ATTRIBUTES' of the function.
27309 FUNTYPE is a C variable whose value is a tree node that describes
27310 the function in question. Normally it is a node of type
27311 `FUNCTION_TYPE' that describes the data type of the function.
27312 From this it is possible to obtain the data types of the value and
27313 arguments (if known).
27315 When a call to a library function is being considered, FUNDECL
27316 will contain an identifier node for the library function. Thus, if
27317 you need to distinguish among various library functions, you can
27318 do so by their names. Note that "library function" in this
27319 context means a function used to perform arithmetic, whose name is
27320 known specially in the compiler and was not mentioned in the C
27321 code being compiled.
27323 STACK-SIZE is the number of bytes of arguments passed on the
27324 stack. If a variable number of bytes is passed, it is zero, and
27325 argument popping will always be the responsibility of the calling
27328 On the VAX, all functions always pop their arguments, so the
27329 definition of this macro is STACK-SIZE. On the 68000, using the
27330 standard calling convention, no functions pop their arguments, so
27331 the value of the macro is always 0 in this case. But an
27332 alternative calling convention is available in which functions
27333 that take a fixed number of arguments pop them but other functions
27334 (such as `printf') pop nothing (the caller pops all). When this
27335 convention is in use, FUNTYPE is examined to determine whether a
27336 function takes a fixed number of arguments.
27338 -- Macro: CALL_POPS_ARGS (CUM)
27339 A C expression that should indicate the number of bytes a call
27340 sequence pops off the stack. It is added to the value of
27341 `RETURN_POPS_ARGS' when compiling a function call.
27343 CUM is the variable in which all arguments to the called function
27344 have been accumulated.
27346 On certain architectures, such as the SH5, a call trampoline is
27347 used that pops certain registers off the stack, depending on the
27348 arguments that have been passed to the function. Since this is a
27349 property of the call site, not of the called function,
27350 `RETURN_POPS_ARGS' is not appropriate.
27353 File: gccint.info, Node: Register Arguments, Next: Scalar Return, Prev: Stack Arguments, Up: Stack and Calling
27355 17.10.7 Passing Arguments in Registers
27356 --------------------------------------
27358 This section describes the macros which let you control how various
27359 types of arguments are passed in registers or how they are arranged in
27362 -- Macro: FUNCTION_ARG (CUM, MODE, TYPE, NAMED)
27363 A C expression that controls whether a function argument is passed
27364 in a register, and which register.
27366 The arguments are CUM, which summarizes all the previous
27367 arguments; MODE, the machine mode of the argument; TYPE, the data
27368 type of the argument as a tree node or 0 if that is not known
27369 (which happens for C support library functions); and NAMED, which
27370 is 1 for an ordinary argument and 0 for nameless arguments that
27371 correspond to `...' in the called function's prototype. TYPE can
27372 be an incomplete type if a syntax error has previously occurred.
27374 The value of the expression is usually either a `reg' RTX for the
27375 hard register in which to pass the argument, or zero to pass the
27376 argument on the stack.
27378 For machines like the VAX and 68000, where normally all arguments
27379 are pushed, zero suffices as a definition.
27381 The value of the expression can also be a `parallel' RTX. This is
27382 used when an argument is passed in multiple locations. The mode
27383 of the `parallel' should be the mode of the entire argument. The
27384 `parallel' holds any number of `expr_list' pairs; each one
27385 describes where part of the argument is passed. In each
27386 `expr_list' the first operand must be a `reg' RTX for the hard
27387 register in which to pass this part of the argument, and the mode
27388 of the register RTX indicates how large this part of the argument
27389 is. The second operand of the `expr_list' is a `const_int' which
27390 gives the offset in bytes into the entire argument of where this
27391 part starts. As a special exception the first `expr_list' in the
27392 `parallel' RTX may have a first operand of zero. This indicates
27393 that the entire argument is also stored on the stack.
27395 The last time this macro is called, it is called with `MODE ==
27396 VOIDmode', and its result is passed to the `call' or `call_value'
27397 pattern as operands 2 and 3 respectively.
27399 The usual way to make the ISO library `stdarg.h' work on a machine
27400 where some arguments are usually passed in registers, is to cause
27401 nameless arguments to be passed on the stack instead. This is done
27402 by making `FUNCTION_ARG' return 0 whenever NAMED is 0.
27404 You may use the hook `targetm.calls.must_pass_in_stack' in the
27405 definition of this macro to determine if this argument is of a
27406 type that must be passed in the stack. If `REG_PARM_STACK_SPACE'
27407 is not defined and `FUNCTION_ARG' returns nonzero for such an
27408 argument, the compiler will abort. If `REG_PARM_STACK_SPACE' is
27409 defined, the argument will be computed in the stack and then
27410 loaded into a register.
27412 -- Target Hook: bool TARGET_MUST_PASS_IN_STACK (enum machine_mode
27414 This target hook should return `true' if we should not pass TYPE
27415 solely in registers. The file `expr.h' defines a definition that
27416 is usually appropriate, refer to `expr.h' for additional
27419 -- Macro: FUNCTION_INCOMING_ARG (CUM, MODE, TYPE, NAMED)
27420 Define this macro if the target machine has "register windows", so
27421 that the register in which a function sees an arguments is not
27422 necessarily the same as the one in which the caller passed the
27425 For such machines, `FUNCTION_ARG' computes the register in which
27426 the caller passes the value, and `FUNCTION_INCOMING_ARG' should be
27427 defined in a similar fashion to tell the function being called
27428 where the arguments will arrive.
27430 If `FUNCTION_INCOMING_ARG' is not defined, `FUNCTION_ARG' serves
27433 -- Target Hook: int TARGET_ARG_PARTIAL_BYTES (CUMULATIVE_ARGS *CUM,
27434 enum machine_mode MODE, tree TYPE, bool NAMED)
27435 This target hook returns the number of bytes at the beginning of an
27436 argument that must be put in registers. The value must be zero for
27437 arguments that are passed entirely in registers or that are
27438 entirely pushed on the stack.
27440 On some machines, certain arguments must be passed partially in
27441 registers and partially in memory. On these machines, typically
27442 the first few words of arguments are passed in registers, and the
27443 rest on the stack. If a multi-word argument (a `double' or a
27444 structure) crosses that boundary, its first few words must be
27445 passed in registers and the rest must be pushed. This macro tells
27446 the compiler when this occurs, and how many bytes should go in
27449 `FUNCTION_ARG' for these arguments should return the first
27450 register to be used by the caller for this argument; likewise
27451 `FUNCTION_INCOMING_ARG', for the called function.
27453 -- Target Hook: bool TARGET_PASS_BY_REFERENCE (CUMULATIVE_ARGS *CUM,
27454 enum machine_mode MODE, tree TYPE, bool NAMED)
27455 This target hook should return `true' if an argument at the
27456 position indicated by CUM should be passed by reference. This
27457 predicate is queried after target independent reasons for being
27458 passed by reference, such as `TREE_ADDRESSABLE (type)'.
27460 If the hook returns true, a copy of that argument is made in
27461 memory and a pointer to the argument is passed instead of the
27462 argument itself. The pointer is passed in whatever way is
27463 appropriate for passing a pointer to that type.
27465 -- Target Hook: bool TARGET_CALLEE_COPIES (CUMULATIVE_ARGS *CUM, enum
27466 machine_mode MODE, tree TYPE, bool NAMED)
27467 The function argument described by the parameters to this hook is
27468 known to be passed by reference. The hook should return true if
27469 the function argument should be copied by the callee instead of
27470 copied by the caller.
27472 For any argument for which the hook returns true, if it can be
27473 determined that the argument is not modified, then a copy need not
27476 The default version of this hook always returns false.
27478 -- Macro: CUMULATIVE_ARGS
27479 A C type for declaring a variable that is used as the first
27480 argument of `FUNCTION_ARG' and other related values. For some
27481 target machines, the type `int' suffices and can hold the number
27482 of bytes of argument so far.
27484 There is no need to record in `CUMULATIVE_ARGS' anything about the
27485 arguments that have been passed on the stack. The compiler has
27486 other variables to keep track of that. For target machines on
27487 which all arguments are passed on the stack, there is no need to
27488 store anything in `CUMULATIVE_ARGS'; however, the data structure
27489 must exist and should not be empty, so use `int'.
27491 -- Macro: OVERRIDE_ABI_FORMAT (FNDECL)
27492 If defined, this macro is called before generating any code for a
27493 function, but after the CFUN descriptor for the function has been
27494 created. The back end may use this macro to update CFUN to
27495 reflect an ABI other than that which would normally be used by
27496 default. If the compiler is generating code for a
27497 compiler-generated function, FNDECL may be `NULL'.
27499 -- Macro: INIT_CUMULATIVE_ARGS (CUM, FNTYPE, LIBNAME, FNDECL,
27501 A C statement (sans semicolon) for initializing the variable CUM
27502 for the state at the beginning of the argument list. The variable
27503 has type `CUMULATIVE_ARGS'. The value of FNTYPE is the tree node
27504 for the data type of the function which will receive the args, or
27505 0 if the args are to a compiler support library function. For
27506 direct calls that are not libcalls, FNDECL contain the declaration
27507 node of the function. FNDECL is also set when
27508 `INIT_CUMULATIVE_ARGS' is used to find arguments for the function
27509 being compiled. N_NAMED_ARGS is set to the number of named
27510 arguments, including a structure return address if it is passed as
27511 a parameter, when making a call. When processing incoming
27512 arguments, N_NAMED_ARGS is set to -1.
27514 When processing a call to a compiler support library function,
27515 LIBNAME identifies which one. It is a `symbol_ref' rtx which
27516 contains the name of the function, as a string. LIBNAME is 0 when
27517 an ordinary C function call is being processed. Thus, each time
27518 this macro is called, either LIBNAME or FNTYPE is nonzero, but
27519 never both of them at once.
27521 -- Macro: INIT_CUMULATIVE_LIBCALL_ARGS (CUM, MODE, LIBNAME)
27522 Like `INIT_CUMULATIVE_ARGS' but only used for outgoing libcalls,
27523 it gets a `MODE' argument instead of FNTYPE, that would be `NULL'.
27524 INDIRECT would always be zero, too. If this macro is not
27525 defined, `INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 0)' is
27528 -- Macro: INIT_CUMULATIVE_INCOMING_ARGS (CUM, FNTYPE, LIBNAME)
27529 Like `INIT_CUMULATIVE_ARGS' but overrides it for the purposes of
27530 finding the arguments for the function being compiled. If this
27531 macro is undefined, `INIT_CUMULATIVE_ARGS' is used instead.
27533 The value passed for LIBNAME is always 0, since library routines
27534 with special calling conventions are never compiled with GCC. The
27535 argument LIBNAME exists for symmetry with `INIT_CUMULATIVE_ARGS'.
27537 -- Macro: FUNCTION_ARG_ADVANCE (CUM, MODE, TYPE, NAMED)
27538 A C statement (sans semicolon) to update the summarizer variable
27539 CUM to advance past an argument in the argument list. The values
27540 MODE, TYPE and NAMED describe that argument. Once this is done,
27541 the variable CUM is suitable for analyzing the _following_
27542 argument with `FUNCTION_ARG', etc.
27544 This macro need not do anything if the argument in question was
27545 passed on the stack. The compiler knows how to track the amount
27546 of stack space used for arguments without any special help.
27548 -- Macro: FUNCTION_ARG_OFFSET (MODE, TYPE)
27549 If defined, a C expression that is the number of bytes to add to
27550 the offset of the argument passed in memory. This is needed for
27551 the SPU, which passes `char' and `short' arguments in the preferred
27552 slot that is in the middle of the quad word instead of starting at
27555 -- Macro: FUNCTION_ARG_PADDING (MODE, TYPE)
27556 If defined, a C expression which determines whether, and in which
27557 direction, to pad out an argument with extra space. The value
27558 should be of type `enum direction': either `upward' to pad above
27559 the argument, `downward' to pad below, or `none' to inhibit
27562 The _amount_ of padding is always just enough to reach the next
27563 multiple of `FUNCTION_ARG_BOUNDARY'; this macro does not control
27566 This macro has a default definition which is right for most
27567 systems. For little-endian machines, the default is to pad
27568 upward. For big-endian machines, the default is to pad downward
27569 for an argument of constant size shorter than an `int', and upward
27572 -- Macro: PAD_VARARGS_DOWN
27573 If defined, a C expression which determines whether the default
27574 implementation of va_arg will attempt to pad down before reading
27575 the next argument, if that argument is smaller than its aligned
27576 space as controlled by `PARM_BOUNDARY'. If this macro is not
27577 defined, all such arguments are padded down if `BYTES_BIG_ENDIAN'
27580 -- Macro: BLOCK_REG_PADDING (MODE, TYPE, FIRST)
27581 Specify padding for the last element of a block move between
27582 registers and memory. FIRST is nonzero if this is the only
27583 element. Defining this macro allows better control of register
27584 function parameters on big-endian machines, without using
27585 `PARALLEL' rtl. In particular, `MUST_PASS_IN_STACK' need not test
27586 padding and mode of types in registers, as there is no longer a
27587 "wrong" part of a register; For example, a three byte aggregate
27588 may be passed in the high part of a register if so required.
27590 -- Macro: FUNCTION_ARG_BOUNDARY (MODE, TYPE)
27591 If defined, a C expression that gives the alignment boundary, in
27592 bits, of an argument with the specified mode and type. If it is
27593 not defined, `PARM_BOUNDARY' is used for all arguments.
27595 -- Macro: FUNCTION_ARG_REGNO_P (REGNO)
27596 A C expression that is nonzero if REGNO is the number of a hard
27597 register in which function arguments are sometimes passed. This
27598 does _not_ include implicit arguments such as the static chain and
27599 the structure-value address. On many machines, no registers can be
27600 used for this purpose since all function arguments are pushed on
27603 -- Target Hook: bool TARGET_SPLIT_COMPLEX_ARG (tree TYPE)
27604 This hook should return true if parameter of type TYPE are passed
27605 as two scalar parameters. By default, GCC will attempt to pack
27606 complex arguments into the target's word size. Some ABIs require
27607 complex arguments to be split and treated as their individual
27608 components. For example, on AIX64, complex floats should be
27609 passed in a pair of floating point registers, even though a
27610 complex float would fit in one 64-bit floating point register.
27612 The default value of this hook is `NULL', which is treated as
27615 -- Target Hook: tree TARGET_BUILD_BUILTIN_VA_LIST (void)
27616 This hook returns a type node for `va_list' for the target. The
27617 default version of the hook returns `void*'.
27619 -- Target Hook: tree TARGET_FN_ABI_VA_LIST (tree FNDECL)
27620 This hook returns the va_list type of the calling convention
27621 specified by FNDECL. The default version of this hook returns
27622 `va_list_type_node'.
27624 -- Target Hook: tree TARGET_CANONICAL_VA_LIST_TYPE (tree TYPE)
27625 This hook returns the va_list type of the calling convention
27626 specified by the type of TYPE. If TYPE is not a valid va_list
27627 type, it returns `NULL_TREE'.
27629 -- Target Hook: tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree VALIST, tree
27630 TYPE, tree *PRE_P, tree *POST_P)
27631 This hook performs target-specific gimplification of
27632 `VA_ARG_EXPR'. The first two parameters correspond to the
27633 arguments to `va_arg'; the latter two are as in
27634 `gimplify.c:gimplify_expr'.
27636 -- Target Hook: bool TARGET_VALID_POINTER_MODE (enum machine_mode MODE)
27637 Define this to return nonzero if the port can handle pointers with
27638 machine mode MODE. The default version of this hook returns true
27639 for both `ptr_mode' and `Pmode'.
27641 -- Target Hook: bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode
27643 Define this to return nonzero if the port is prepared to handle
27644 insns involving scalar mode MODE. For a scalar mode to be
27645 considered supported, all the basic arithmetic and comparisons
27648 The default version of this hook returns true for any mode
27649 required to handle the basic C types (as defined by the port).
27650 Included here are the double-word arithmetic supported by the code
27653 -- Target Hook: bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode
27655 Define this to return nonzero if the port is prepared to handle
27656 insns involving vector mode MODE. At the very least, it must have
27657 move patterns for this mode.
27660 File: gccint.info, Node: Scalar Return, Next: Aggregate Return, Prev: Register Arguments, Up: Stack and Calling
27662 17.10.8 How Scalar Function Values Are Returned
27663 -----------------------------------------------
27665 This section discusses the macros that control returning scalars as
27666 values--values that can fit in registers.
27668 -- Target Hook: rtx TARGET_FUNCTION_VALUE (tree RET_TYPE, tree
27669 FN_DECL_OR_TYPE, bool OUTGOING)
27670 Define this to return an RTX representing the place where a
27671 function returns or receives a value of data type RET_TYPE, a tree
27672 node node representing a data type. FN_DECL_OR_TYPE is a tree node
27673 representing `FUNCTION_DECL' or `FUNCTION_TYPE' of a function
27674 being called. If OUTGOING is false, the hook should compute the
27675 register in which the caller will see the return value.
27676 Otherwise, the hook should return an RTX representing the place
27677 where a function returns a value.
27679 On many machines, only `TYPE_MODE (RET_TYPE)' is relevant.
27680 (Actually, on most machines, scalar values are returned in the same
27681 place regardless of mode.) The value of the expression is usually
27682 a `reg' RTX for the hard register where the return value is stored.
27683 The value can also be a `parallel' RTX, if the return value is in
27684 multiple places. See `FUNCTION_ARG' for an explanation of the
27685 `parallel' form. Note that the callee will populate every
27686 location specified in the `parallel', but if the first element of
27687 the `parallel' contains the whole return value, callers will use
27688 that element as the canonical location and ignore the others. The
27689 m68k port uses this type of `parallel' to return pointers in both
27690 `%a0' (the canonical location) and `%d0'.
27692 If `TARGET_PROMOTE_FUNCTION_RETURN' returns true, you must apply
27693 the same promotion rules specified in `PROMOTE_MODE' if VALTYPE is
27696 If the precise function being called is known, FUNC is a tree node
27697 (`FUNCTION_DECL') for it; otherwise, FUNC is a null pointer. This
27698 makes it possible to use a different value-returning convention
27699 for specific functions when all their calls are known.
27701 Some target machines have "register windows" so that the register
27702 in which a function returns its value is not the same as the one
27703 in which the caller sees the value. For such machines, you should
27704 return different RTX depending on OUTGOING.
27706 `TARGET_FUNCTION_VALUE' is not used for return values with
27707 aggregate data types, because these are returned in another way.
27708 See `TARGET_STRUCT_VALUE_RTX' and related macros, below.
27710 -- Macro: FUNCTION_VALUE (VALTYPE, FUNC)
27711 This macro has been deprecated. Use `TARGET_FUNCTION_VALUE' for a
27712 new target instead.
27714 -- Macro: FUNCTION_OUTGOING_VALUE (VALTYPE, FUNC)
27715 This macro has been deprecated. Use `TARGET_FUNCTION_VALUE' for a
27716 new target instead.
27718 -- Macro: LIBCALL_VALUE (MODE)
27719 A C expression to create an RTX representing the place where a
27720 library function returns a value of mode MODE.
27722 Note that "library function" in this context means a compiler
27723 support routine, used to perform arithmetic, whose name is known
27724 specially by the compiler and was not mentioned in the C code being
27727 -- Macro: FUNCTION_VALUE_REGNO_P (REGNO)
27728 A C expression that is nonzero if REGNO is the number of a hard
27729 register in which the values of called function may come back.
27731 A register whose use for returning values is limited to serving as
27732 the second of a pair (for a value of type `double', say) need not
27733 be recognized by this macro. So for most machines, this definition
27736 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
27738 If the machine has register windows, so that the caller and the
27739 called function use different registers for the return value, this
27740 macro should recognize only the caller's register numbers.
27742 -- Macro: TARGET_ENUM_VA_LIST (IDX, PNAME, PTYPE)
27743 This target macro is used in function `c_common_nodes_and_builtins'
27744 to iterate through the target specific builtin types for va_list.
27745 The variable IDX is used as iterator. PNAME has to be a pointer to
27746 a `const char *' and PTYPE a pointer to a `tree' typed variable.
27747 The arguments PNAME and PTYPE are used to store the result of this
27748 macro and are set to the name of the va_list builtin type and its
27749 internal type. If the return value of this macro is zero, then
27750 there is no more element. Otherwise the IDX should be increased
27751 for the next call of this macro to iterate through all types.
27753 -- Macro: APPLY_RESULT_SIZE
27754 Define this macro if `untyped_call' and `untyped_return' need more
27755 space than is implied by `FUNCTION_VALUE_REGNO_P' for saving and
27756 restoring an arbitrary return value.
27758 -- Target Hook: bool TARGET_RETURN_IN_MSB (tree TYPE)
27759 This hook should return true if values of type TYPE are returned
27760 at the most significant end of a register (in other words, if they
27761 are padded at the least significant end). You can assume that TYPE
27762 is returned in a register; the caller is required to check this.
27764 Note that the register provided by `TARGET_FUNCTION_VALUE' must be
27765 able to hold the complete return value. For example, if a 1-, 2-
27766 or 3-byte structure is returned at the most significant end of a
27767 4-byte register, `TARGET_FUNCTION_VALUE' should provide an
27771 File: gccint.info, Node: Aggregate Return, Next: Caller Saves, Prev: Scalar Return, Up: Stack and Calling
27773 17.10.9 How Large Values Are Returned
27774 -------------------------------------
27776 When a function value's mode is `BLKmode' (and in some other cases),
27777 the value is not returned according to `TARGET_FUNCTION_VALUE' (*note
27778 Scalar Return::). Instead, the caller passes the address of a block of
27779 memory in which the value should be stored. This address is called the
27780 "structure value address".
27782 This section describes how to control returning structure values in
27785 -- Target Hook: bool TARGET_RETURN_IN_MEMORY (tree TYPE, tree FNTYPE)
27786 This target hook should return a nonzero value to say to return the
27787 function value in memory, just as large structures are always
27788 returned. Here TYPE will be the data type of the value, and FNTYPE
27789 will be the type of the function doing the returning, or `NULL' for
27792 Note that values of mode `BLKmode' must be explicitly handled by
27793 this function. Also, the option `-fpcc-struct-return' takes
27794 effect regardless of this macro. On most systems, it is possible
27795 to leave the hook undefined; this causes a default definition to
27796 be used, whose value is the constant 1 for `BLKmode' values, and 0
27799 Do not use this hook to indicate that structures and unions should
27800 always be returned in memory. You should instead use
27801 `DEFAULT_PCC_STRUCT_RETURN' to indicate this.
27803 -- Macro: DEFAULT_PCC_STRUCT_RETURN
27804 Define this macro to be 1 if all structure and union return values
27805 must be in memory. Since this results in slower code, this should
27806 be defined only if needed for compatibility with other compilers
27807 or with an ABI. If you define this macro to be 0, then the
27808 conventions used for structure and union return values are decided
27809 by the `TARGET_RETURN_IN_MEMORY' target hook.
27811 If not defined, this defaults to the value 1.
27813 -- Target Hook: rtx TARGET_STRUCT_VALUE_RTX (tree FNDECL, int INCOMING)
27814 This target hook should return the location of the structure value
27815 address (normally a `mem' or `reg'), or 0 if the address is passed
27816 as an "invisible" first argument. Note that FNDECL may be `NULL',
27817 for libcalls. You do not need to define this target hook if the
27818 address is always passed as an "invisible" first argument.
27820 On some architectures the place where the structure value address
27821 is found by the called function is not the same place that the
27822 caller put it. This can be due to register windows, or it could
27823 be because the function prologue moves it to a different place.
27824 INCOMING is `1' or `2' when the location is needed in the context
27825 of the called function, and `0' in the context of the caller.
27827 If INCOMING is nonzero and the address is to be found on the
27828 stack, return a `mem' which refers to the frame pointer. If
27829 INCOMING is `2', the result is being used to fetch the structure
27830 value address at the beginning of a function. If you need to emit
27831 adjusting code, you should do it at this point.
27833 -- Macro: PCC_STATIC_STRUCT_RETURN
27834 Define this macro if the usual system convention on the target
27835 machine for returning structures and unions is for the called
27836 function to return the address of a static variable containing the
27839 Do not define this if the usual system convention is for the
27840 caller to pass an address to the subroutine.
27842 This macro has effect in `-fpcc-struct-return' mode, but it does
27843 nothing when you use `-freg-struct-return' mode.
27846 File: gccint.info, Node: Caller Saves, Next: Function Entry, Prev: Aggregate Return, Up: Stack and Calling
27848 17.10.10 Caller-Saves Register Allocation
27849 -----------------------------------------
27851 If you enable it, GCC can save registers around function calls. This
27852 makes it possible to use call-clobbered registers to hold variables that
27853 must live across calls.
27855 -- Macro: CALLER_SAVE_PROFITABLE (REFS, CALLS)
27856 A C expression to determine whether it is worthwhile to consider
27857 placing a pseudo-register in a call-clobbered hard register and
27858 saving and restoring it around each function call. The expression
27859 should be 1 when this is worth doing, and 0 otherwise.
27861 If you don't define this macro, a default is used which is good on
27862 most machines: `4 * CALLS < REFS'.
27864 -- Macro: HARD_REGNO_CALLER_SAVE_MODE (REGNO, NREGS)
27865 A C expression specifying which mode is required for saving NREGS
27866 of a pseudo-register in call-clobbered hard register REGNO. If
27867 REGNO is unsuitable for caller save, `VOIDmode' should be
27868 returned. For most machines this macro need not be defined since
27869 GCC will select the smallest suitable mode.
27872 File: gccint.info, Node: Function Entry, Next: Profiling, Prev: Caller Saves, Up: Stack and Calling
27874 17.10.11 Function Entry and Exit
27875 --------------------------------
27877 This section describes the macros that output function entry
27878 ("prologue") and exit ("epilogue") code.
27880 -- Target Hook: void TARGET_ASM_FUNCTION_PROLOGUE (FILE *FILE,
27881 HOST_WIDE_INT SIZE)
27882 If defined, a function that outputs the assembler code for entry
27883 to a function. The prologue is responsible for setting up the
27884 stack frame, initializing the frame pointer register, saving
27885 registers that must be saved, and allocating SIZE additional bytes
27886 of storage for the local variables. SIZE is an integer. FILE is
27887 a stdio stream to which the assembler code should be output.
27889 The label for the beginning of the function need not be output by
27890 this macro. That has already been done when the macro is run.
27892 To determine which registers to save, the macro can refer to the
27893 array `regs_ever_live': element R is nonzero if hard register R is
27894 used anywhere within the function. This implies the function
27895 prologue should save register R, provided it is not one of the
27896 call-used registers. (`TARGET_ASM_FUNCTION_EPILOGUE' must
27897 likewise use `regs_ever_live'.)
27899 On machines that have "register windows", the function entry code
27900 does not save on the stack the registers that are in the windows,
27901 even if they are supposed to be preserved by function calls;
27902 instead it takes appropriate steps to "push" the register stack,
27903 if any non-call-used registers are used in the function.
27905 On machines where functions may or may not have frame-pointers, the
27906 function entry code must vary accordingly; it must set up the frame
27907 pointer if one is wanted, and not otherwise. To determine whether
27908 a frame pointer is in wanted, the macro can refer to the variable
27909 `frame_pointer_needed'. The variable's value will be 1 at run
27910 time in a function that needs a frame pointer. *Note
27913 The function entry code is responsible for allocating any stack
27914 space required for the function. This stack space consists of the
27915 regions listed below. In most cases, these regions are allocated
27916 in the order listed, with the last listed region closest to the
27917 top of the stack (the lowest address if `STACK_GROWS_DOWNWARD' is
27918 defined, and the highest address if it is not defined). You can
27919 use a different order for a machine if doing so is more convenient
27920 or required for compatibility reasons. Except in cases where
27921 required by standard or by a debugger, there is no reason why the
27922 stack layout used by GCC need agree with that used by other
27923 compilers for a machine.
27925 -- Target Hook: void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *FILE)
27926 If defined, a function that outputs assembler code at the end of a
27927 prologue. This should be used when the function prologue is being
27928 emitted as RTL, and you have some extra assembler that needs to be
27929 emitted. *Note prologue instruction pattern::.
27931 -- Target Hook: void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *FILE)
27932 If defined, a function that outputs assembler code at the start of
27933 an epilogue. This should be used when the function epilogue is
27934 being emitted as RTL, and you have some extra assembler that needs
27935 to be emitted. *Note epilogue instruction pattern::.
27937 -- Target Hook: void TARGET_ASM_FUNCTION_EPILOGUE (FILE *FILE,
27938 HOST_WIDE_INT SIZE)
27939 If defined, a function that outputs the assembler code for exit
27940 from a function. The epilogue is responsible for restoring the
27941 saved registers and stack pointer to their values when the
27942 function was called, and returning control to the caller. This
27943 macro takes the same arguments as the macro
27944 `TARGET_ASM_FUNCTION_PROLOGUE', and the registers to restore are
27945 determined from `regs_ever_live' and `CALL_USED_REGISTERS' in the
27948 On some machines, there is a single instruction that does all the
27949 work of returning from the function. On these machines, give that
27950 instruction the name `return' and do not define the macro
27951 `TARGET_ASM_FUNCTION_EPILOGUE' at all.
27953 Do not define a pattern named `return' if you want the
27954 `TARGET_ASM_FUNCTION_EPILOGUE' to be used. If you want the target
27955 switches to control whether return instructions or epilogues are
27956 used, define a `return' pattern with a validity condition that
27957 tests the target switches appropriately. If the `return'
27958 pattern's validity condition is false, epilogues will be used.
27960 On machines where functions may or may not have frame-pointers, the
27961 function exit code must vary accordingly. Sometimes the code for
27962 these two cases is completely different. To determine whether a
27963 frame pointer is wanted, the macro can refer to the variable
27964 `frame_pointer_needed'. The variable's value will be 1 when
27965 compiling a function that needs a frame pointer.
27967 Normally, `TARGET_ASM_FUNCTION_PROLOGUE' and
27968 `TARGET_ASM_FUNCTION_EPILOGUE' must treat leaf functions specially.
27969 The C variable `current_function_is_leaf' is nonzero for such a
27970 function. *Note Leaf Functions::.
27972 On some machines, some functions pop their arguments on exit while
27973 others leave that for the caller to do. For example, the 68020
27974 when given `-mrtd' pops arguments in functions that take a fixed
27975 number of arguments.
27977 Your definition of the macro `RETURN_POPS_ARGS' decides which
27978 functions pop their own arguments. `TARGET_ASM_FUNCTION_EPILOGUE'
27979 needs to know what was decided. The variable that is called
27980 `current_function_pops_args' is the number of bytes of its
27981 arguments that a function should pop. *Note Scalar Return::.
27983 * A region of `current_function_pretend_args_size' bytes of
27984 uninitialized space just underneath the first argument arriving on
27985 the stack. (This may not be at the very start of the allocated
27986 stack region if the calling sequence has pushed anything else
27987 since pushing the stack arguments. But usually, on such machines,
27988 nothing else has been pushed yet, because the function prologue
27989 itself does all the pushing.) This region is used on machines
27990 where an argument may be passed partly in registers and partly in
27991 memory, and, in some cases to support the features in `<stdarg.h>'.
27993 * An area of memory used to save certain registers used by the
27994 function. The size of this area, which may also include space for
27995 such things as the return address and pointers to previous stack
27996 frames, is machine-specific and usually depends on which registers
27997 have been used in the function. Machines with register windows
27998 often do not require a save area.
28000 * A region of at least SIZE bytes, possibly rounded up to an
28001 allocation boundary, to contain the local variables of the
28002 function. On some machines, this region and the save area may
28003 occur in the opposite order, with the save area closer to the top
28006 * Optionally, when `ACCUMULATE_OUTGOING_ARGS' is defined, a region of
28007 `current_function_outgoing_args_size' bytes to be used for outgoing
28008 argument lists of the function. *Note Stack Arguments::.
28010 -- Macro: EXIT_IGNORE_STACK
28011 Define this macro as a C expression that is nonzero if the return
28012 instruction or the function epilogue ignores the value of the stack
28013 pointer; in other words, if it is safe to delete an instruction to
28014 adjust the stack pointer before a return from the function. The
28017 Note that this macro's value is relevant only for functions for
28018 which frame pointers are maintained. It is never safe to delete a
28019 final stack adjustment in a function that has no frame pointer,
28020 and the compiler knows this regardless of `EXIT_IGNORE_STACK'.
28022 -- Macro: EPILOGUE_USES (REGNO)
28023 Define this macro as a C expression that is nonzero for registers
28024 that are used by the epilogue or the `return' pattern. The stack
28025 and frame pointer registers are already assumed to be used as
28028 -- Macro: EH_USES (REGNO)
28029 Define this macro as a C expression that is nonzero for registers
28030 that are used by the exception handling mechanism, and so should
28031 be considered live on entry to an exception edge.
28033 -- Macro: DELAY_SLOTS_FOR_EPILOGUE
28034 Define this macro if the function epilogue contains delay slots to
28035 which instructions from the rest of the function can be "moved".
28036 The definition should be a C expression whose value is an integer
28037 representing the number of delay slots there.
28039 -- Macro: ELIGIBLE_FOR_EPILOGUE_DELAY (INSN, N)
28040 A C expression that returns 1 if INSN can be placed in delay slot
28041 number N of the epilogue.
28043 The argument N is an integer which identifies the delay slot now
28044 being considered (since different slots may have different rules of
28045 eligibility). It is never negative and is always less than the
28046 number of epilogue delay slots (what `DELAY_SLOTS_FOR_EPILOGUE'
28047 returns). If you reject a particular insn for a given delay slot,
28048 in principle, it may be reconsidered for a subsequent delay slot.
28049 Also, other insns may (at least in principle) be considered for
28050 the so far unfilled delay slot.
28052 The insns accepted to fill the epilogue delay slots are put in an
28053 RTL list made with `insn_list' objects, stored in the variable
28054 `current_function_epilogue_delay_list'. The insn for the first
28055 delay slot comes first in the list. Your definition of the macro
28056 `TARGET_ASM_FUNCTION_EPILOGUE' should fill the delay slots by
28057 outputting the insns in this list, usually by calling
28060 You need not define this macro if you did not define
28061 `DELAY_SLOTS_FOR_EPILOGUE'.
28063 -- Target Hook: void TARGET_ASM_OUTPUT_MI_THUNK (FILE *FILE, tree
28064 THUNK_FNDECL, HOST_WIDE_INT DELTA, HOST_WIDE_INT
28065 VCALL_OFFSET, tree FUNCTION)
28066 A function that outputs the assembler code for a thunk function,
28067 used to implement C++ virtual function calls with multiple
28068 inheritance. The thunk acts as a wrapper around a virtual
28069 function, adjusting the implicit object parameter before handing
28070 control off to the real function.
28072 First, emit code to add the integer DELTA to the location that
28073 contains the incoming first argument. Assume that this argument
28074 contains a pointer, and is the one used to pass the `this' pointer
28075 in C++. This is the incoming argument _before_ the function
28076 prologue, e.g. `%o0' on a sparc. The addition must preserve the
28077 values of all other incoming arguments.
28079 Then, if VCALL_OFFSET is nonzero, an additional adjustment should
28080 be made after adding `delta'. In particular, if P is the adjusted
28081 pointer, the following adjustment should be made:
28083 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
28085 After the additions, emit code to jump to FUNCTION, which is a
28086 `FUNCTION_DECL'. This is a direct pure jump, not a call, and does
28087 not touch the return address. Hence returning from FUNCTION will
28088 return to whoever called the current `thunk'.
28090 The effect must be as if FUNCTION had been called directly with
28091 the adjusted first argument. This macro is responsible for
28092 emitting all of the code for a thunk function;
28093 `TARGET_ASM_FUNCTION_PROLOGUE' and `TARGET_ASM_FUNCTION_EPILOGUE'
28096 The THUNK_FNDECL is redundant. (DELTA and FUNCTION have already
28097 been extracted from it.) It might possibly be useful on some
28098 targets, but probably not.
28100 If you do not define this macro, the target-independent code in
28101 the C++ front end will generate a less efficient heavyweight thunk
28102 that calls FUNCTION instead of jumping to it. The generic
28103 approach does not support varargs.
28105 -- Target Hook: bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (tree
28106 THUNK_FNDECL, HOST_WIDE_INT DELTA, HOST_WIDE_INT
28107 VCALL_OFFSET, tree FUNCTION)
28108 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would
28109 be able to output the assembler code for the thunk function
28110 specified by the arguments it is passed, and false otherwise. In
28111 the latter case, the generic approach will be used by the C++
28112 front end, with the limitations previously exposed.
28115 File: gccint.info, Node: Profiling, Next: Tail Calls, Prev: Function Entry, Up: Stack and Calling
28117 17.10.12 Generating Code for Profiling
28118 --------------------------------------
28120 These macros will help you generate code for profiling.
28122 -- Macro: FUNCTION_PROFILER (FILE, LABELNO)
28123 A C statement or compound statement to output to FILE some
28124 assembler code to call the profiling subroutine `mcount'.
28126 The details of how `mcount' expects to be called are determined by
28127 your operating system environment, not by GCC. To figure them out,
28128 compile a small program for profiling using the system's installed
28129 C compiler and look at the assembler code that results.
28131 Older implementations of `mcount' expect the address of a counter
28132 variable to be loaded into some register. The name of this
28133 variable is `LP' followed by the number LABELNO, so you would
28134 generate the name using `LP%d' in a `fprintf'.
28136 -- Macro: PROFILE_HOOK
28137 A C statement or compound statement to output to FILE some assembly
28138 code to call the profiling subroutine `mcount' even the target does
28139 not support profiling.
28141 -- Macro: NO_PROFILE_COUNTERS
28142 Define this macro to be an expression with a nonzero value if the
28143 `mcount' subroutine on your system does not need a counter variable
28144 allocated for each function. This is true for almost all modern
28145 implementations. If you define this macro, you must not use the
28146 LABELNO argument to `FUNCTION_PROFILER'.
28148 -- Macro: PROFILE_BEFORE_PROLOGUE
28149 Define this macro if the code for function profiling should come
28150 before the function prologue. Normally, the profiling code comes
28154 File: gccint.info, Node: Tail Calls, Next: Stack Smashing Protection, Prev: Profiling, Up: Stack and Calling
28156 17.10.13 Permitting tail calls
28157 ------------------------------
28159 -- Target Hook: bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree DECL, tree
28161 True if it is ok to do sibling call optimization for the specified
28162 call expression EXP. DECL will be the called function, or `NULL'
28163 if this is an indirect call.
28165 It is not uncommon for limitations of calling conventions to
28166 prevent tail calls to functions outside the current unit of
28167 translation, or during PIC compilation. The hook is used to
28168 enforce these restrictions, as the `sibcall' md pattern can not
28169 fail, or fall over to a "normal" call. The criteria for
28170 successful sibling call optimization may vary greatly between
28171 different architectures.
28173 -- Target Hook: void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap *REGS)
28174 Add any hard registers to REGS that are live on entry to the
28175 function. This hook only needs to be defined to provide registers
28176 that cannot be found by examination of FUNCTION_ARG_REGNO_P, the
28177 callee saved registers, STATIC_CHAIN_INCOMING_REGNUM,
28178 STATIC_CHAIN_REGNUM, TARGET_STRUCT_VALUE_RTX,
28179 FRAME_POINTER_REGNUM, EH_USES, FRAME_POINTER_REGNUM,
28180 ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
28183 File: gccint.info, Node: Stack Smashing Protection, Prev: Tail Calls, Up: Stack and Calling
28185 17.10.14 Stack smashing protection
28186 ----------------------------------
28188 -- Target Hook: tree TARGET_STACK_PROTECT_GUARD (void)
28189 This hook returns a `DECL' node for the external variable to use
28190 for the stack protection guard. This variable is initialized by
28191 the runtime to some random value and is used to initialize the
28192 guard value that is placed at the top of the local stack frame.
28193 The type of this variable must be `ptr_type_node'.
28195 The default version of this hook creates a variable called
28196 `__stack_chk_guard', which is normally defined in `libgcc2.c'.
28198 -- Target Hook: tree TARGET_STACK_PROTECT_FAIL (void)
28199 This hook returns a tree expression that alerts the runtime that
28200 the stack protect guard variable has been modified. This
28201 expression should involve a call to a `noreturn' function.
28203 The default version of this hook invokes a function called
28204 `__stack_chk_fail', taking no arguments. This function is
28205 normally defined in `libgcc2.c'.
28208 File: gccint.info, Node: Varargs, Next: Trampolines, Prev: Stack and Calling, Up: Target Macros
28210 17.11 Implementing the Varargs Macros
28211 =====================================
28213 GCC comes with an implementation of `<varargs.h>' and `<stdarg.h>' that
28214 work without change on machines that pass arguments on the stack.
28215 Other machines require their own implementations of varargs, and the
28216 two machine independent header files must have conditionals to include
28219 ISO `<stdarg.h>' differs from traditional `<varargs.h>' mainly in the
28220 calling convention for `va_start'. The traditional implementation
28221 takes just one argument, which is the variable in which to store the
28222 argument pointer. The ISO implementation of `va_start' takes an
28223 additional second argument. The user is supposed to write the last
28224 named argument of the function here.
28226 However, `va_start' should not use this argument. The way to find the
28227 end of the named arguments is with the built-in functions described
28230 -- Macro: __builtin_saveregs ()
28231 Use this built-in function to save the argument registers in
28232 memory so that the varargs mechanism can access them. Both ISO
28233 and traditional versions of `va_start' must use
28234 `__builtin_saveregs', unless you use
28235 `TARGET_SETUP_INCOMING_VARARGS' (see below) instead.
28237 On some machines, `__builtin_saveregs' is open-coded under the
28238 control of the target hook `TARGET_EXPAND_BUILTIN_SAVEREGS'. On
28239 other machines, it calls a routine written in assembler language,
28240 found in `libgcc2.c'.
28242 Code generated for the call to `__builtin_saveregs' appears at the
28243 beginning of the function, as opposed to where the call to
28244 `__builtin_saveregs' is written, regardless of what the code is.
28245 This is because the registers must be saved before the function
28246 starts to use them for its own purposes.
28248 -- Macro: __builtin_args_info (CATEGORY)
28249 Use this built-in function to find the first anonymous arguments in
28252 In general, a machine may have several categories of registers
28253 used for arguments, each for a particular category of data types.
28254 (For example, on some machines, floating-point registers are used
28255 for floating-point arguments while other arguments are passed in
28256 the general registers.) To make non-varargs functions use the
28257 proper calling convention, you have defined the `CUMULATIVE_ARGS'
28258 data type to record how many registers in each category have been
28261 `__builtin_args_info' accesses the same data structure of type
28262 `CUMULATIVE_ARGS' after the ordinary argument layout is finished
28263 with it, with CATEGORY specifying which word to access. Thus, the
28264 value indicates the first unused register in a given category.
28266 Normally, you would use `__builtin_args_info' in the implementation
28267 of `va_start', accessing each category just once and storing the
28268 value in the `va_list' object. This is because `va_list' will
28269 have to update the values, and there is no way to alter the values
28270 accessed by `__builtin_args_info'.
28272 -- Macro: __builtin_next_arg (LASTARG)
28273 This is the equivalent of `__builtin_args_info', for stack
28274 arguments. It returns the address of the first anonymous stack
28275 argument, as type `void *'. If `ARGS_GROW_DOWNWARD', it returns
28276 the address of the location above the first anonymous stack
28277 argument. Use it in `va_start' to initialize the pointer for
28278 fetching arguments from the stack. Also use it in `va_start' to
28279 verify that the second parameter LASTARG is the last named argument
28280 of the current function.
28282 -- Macro: __builtin_classify_type (OBJECT)
28283 Since each machine has its own conventions for which data types are
28284 passed in which kind of register, your implementation of `va_arg'
28285 has to embody these conventions. The easiest way to categorize the
28286 specified data type is to use `__builtin_classify_type' together
28287 with `sizeof' and `__alignof__'.
28289 `__builtin_classify_type' ignores the value of OBJECT, considering
28290 only its data type. It returns an integer describing what kind of
28291 type that is--integer, floating, pointer, structure, and so on.
28293 The file `typeclass.h' defines an enumeration that you can use to
28294 interpret the values of `__builtin_classify_type'.
28296 These machine description macros help implement varargs:
28298 -- Target Hook: rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
28299 If defined, this hook produces the machine-specific code for a
28300 call to `__builtin_saveregs'. This code will be moved to the very
28301 beginning of the function, before any parameter access are made.
28302 The return value of this function should be an RTX that contains
28303 the value to use as the return of `__builtin_saveregs'.
28305 -- Target Hook: void TARGET_SETUP_INCOMING_VARARGS (CUMULATIVE_ARGS
28306 *ARGS_SO_FAR, enum machine_mode MODE, tree TYPE, int
28307 *PRETEND_ARGS_SIZE, int SECOND_TIME)
28308 This target hook offers an alternative to using
28309 `__builtin_saveregs' and defining the hook
28310 `TARGET_EXPAND_BUILTIN_SAVEREGS'. Use it to store the anonymous
28311 register arguments into the stack so that all the arguments appear
28312 to have been passed consecutively on the stack. Once this is
28313 done, you can use the standard implementation of varargs that
28314 works for machines that pass all their arguments on the stack.
28316 The argument ARGS_SO_FAR points to the `CUMULATIVE_ARGS' data
28317 structure, containing the values that are obtained after
28318 processing the named arguments. The arguments MODE and TYPE
28319 describe the last named argument--its machine mode and its data
28320 type as a tree node.
28322 The target hook should do two things: first, push onto the stack
28323 all the argument registers _not_ used for the named arguments, and
28324 second, store the size of the data thus pushed into the
28325 `int'-valued variable pointed to by PRETEND_ARGS_SIZE. The value
28326 that you store here will serve as additional offset for setting up
28329 Because you must generate code to push the anonymous arguments at
28330 compile time without knowing their data types,
28331 `TARGET_SETUP_INCOMING_VARARGS' is only useful on machines that
28332 have just a single category of argument register and use it
28333 uniformly for all data types.
28335 If the argument SECOND_TIME is nonzero, it means that the
28336 arguments of the function are being analyzed for the second time.
28337 This happens for an inline function, which is not actually
28338 compiled until the end of the source file. The hook
28339 `TARGET_SETUP_INCOMING_VARARGS' should not generate any
28340 instructions in this case.
28342 -- Target Hook: bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS
28344 Define this hook to return `true' if the location where a function
28345 argument is passed depends on whether or not it is a named
28348 This hook controls how the NAMED argument to `FUNCTION_ARG' is set
28349 for varargs and stdarg functions. If this hook returns `true',
28350 the NAMED argument is always true for named arguments, and false
28351 for unnamed arguments. If it returns `false', but
28352 `TARGET_PRETEND_OUTGOING_VARARGS_NAMED' returns `true', then all
28353 arguments are treated as named. Otherwise, all named arguments
28354 except the last are treated as named.
28356 You need not define this hook if it always returns zero.
28358 -- Target Hook: bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED
28359 If you need to conditionally change ABIs so that one works with
28360 `TARGET_SETUP_INCOMING_VARARGS', but the other works like neither
28361 `TARGET_SETUP_INCOMING_VARARGS' nor
28362 `TARGET_STRICT_ARGUMENT_NAMING' was defined, then define this hook
28363 to return `true' if `TARGET_SETUP_INCOMING_VARARGS' is used,
28364 `false' otherwise. Otherwise, you should not define this hook.
28367 File: gccint.info, Node: Trampolines, Next: Library Calls, Prev: Varargs, Up: Target Macros
28369 17.12 Trampolines for Nested Functions
28370 ======================================
28372 A "trampoline" is a small piece of code that is created at run time
28373 when the address of a nested function is taken. It normally resides on
28374 the stack, in the stack frame of the containing function. These macros
28375 tell GCC how to generate code to allocate and initialize a trampoline.
28377 The instructions in the trampoline must do two things: load a constant
28378 address into the static chain register, and jump to the real address of
28379 the nested function. On CISC machines such as the m68k, this requires
28380 two instructions, a move immediate and a jump. Then the two addresses
28381 exist in the trampoline as word-long immediate operands. On RISC
28382 machines, it is often necessary to load each address into a register in
28383 two parts. Then pieces of each address form separate immediate
28386 The code generated to initialize the trampoline must store the variable
28387 parts--the static chain value and the function address--into the
28388 immediate operands of the instructions. On a CISC machine, this is
28389 simply a matter of copying each address to a memory reference at the
28390 proper offset from the start of the trampoline. On a RISC machine, it
28391 may be necessary to take out pieces of the address and store them
28394 -- Macro: TRAMPOLINE_TEMPLATE (FILE)
28395 A C statement to output, on the stream FILE, assembler code for a
28396 block of data that contains the constant parts of a trampoline.
28397 This code should not include a label--the label is taken care of
28400 If you do not define this macro, it means no template is needed
28401 for the target. Do not define this macro on systems where the
28402 block move code to copy the trampoline into place would be larger
28403 than the code to generate it on the spot.
28405 -- Macro: TRAMPOLINE_SECTION
28406 Return the section into which the trampoline template is to be
28407 placed (*note Sections::). The default value is
28408 `readonly_data_section'.
28410 -- Macro: TRAMPOLINE_SIZE
28411 A C expression for the size in bytes of the trampoline, as an
28414 -- Macro: TRAMPOLINE_ALIGNMENT
28415 Alignment required for trampolines, in bits.
28417 If you don't define this macro, the value of `BIGGEST_ALIGNMENT'
28418 is used for aligning trampolines.
28420 -- Macro: INITIALIZE_TRAMPOLINE (ADDR, FNADDR, STATIC_CHAIN)
28421 A C statement to initialize the variable parts of a trampoline.
28422 ADDR is an RTX for the address of the trampoline; FNADDR is an RTX
28423 for the address of the nested function; STATIC_CHAIN is an RTX for
28424 the static chain value that should be passed to the function when
28427 -- Macro: TRAMPOLINE_ADJUST_ADDRESS (ADDR)
28428 A C statement that should perform any machine-specific adjustment
28429 in the address of the trampoline. Its argument contains the
28430 address that was passed to `INITIALIZE_TRAMPOLINE'. In case the
28431 address to be used for a function call should be different from
28432 the address in which the template was stored, the different
28433 address should be assigned to ADDR. If this macro is not defined,
28434 ADDR will be used for function calls.
28436 If this macro is not defined, by default the trampoline is
28437 allocated as a stack slot. This default is right for most
28438 machines. The exceptions are machines where it is impossible to
28439 execute instructions in the stack area. On such machines, you may
28440 have to implement a separate stack, using this macro in
28441 conjunction with `TARGET_ASM_FUNCTION_PROLOGUE' and
28442 `TARGET_ASM_FUNCTION_EPILOGUE'.
28444 FP points to a data structure, a `struct function', which
28445 describes the compilation status of the immediate containing
28446 function of the function which the trampoline is for. The stack
28447 slot for the trampoline is in the stack frame of this containing
28448 function. Other allocation strategies probably must do something
28449 analogous with this information.
28451 Implementing trampolines is difficult on many machines because they
28452 have separate instruction and data caches. Writing into a stack
28453 location fails to clear the memory in the instruction cache, so when
28454 the program jumps to that location, it executes the old contents.
28456 Here are two possible solutions. One is to clear the relevant parts of
28457 the instruction cache whenever a trampoline is set up. The other is to
28458 make all trampolines identical, by having them jump to a standard
28459 subroutine. The former technique makes trampoline execution faster; the
28460 latter makes initialization faster.
28462 To clear the instruction cache when a trampoline is initialized, define
28463 the following macro.
28465 -- Macro: CLEAR_INSN_CACHE (BEG, END)
28466 If defined, expands to a C expression clearing the _instruction
28467 cache_ in the specified interval. The definition of this macro
28468 would typically be a series of `asm' statements. Both BEG and END
28469 are both pointer expressions.
28471 The operating system may also require the stack to be made executable
28472 before calling the trampoline. To implement this requirement, define
28473 the following macro.
28475 -- Macro: ENABLE_EXECUTE_STACK
28476 Define this macro if certain operations must be performed before
28477 executing code located on the stack. The macro should expand to a
28478 series of C file-scope constructs (e.g. functions) and provide a
28479 unique entry point named `__enable_execute_stack'. The target is
28480 responsible for emitting calls to the entry point in the code, for
28481 example from the `INITIALIZE_TRAMPOLINE' macro.
28483 To use a standard subroutine, define the following macro. In addition,
28484 you must make sure that the instructions in a trampoline fill an entire
28485 cache line with identical instructions, or else ensure that the
28486 beginning of the trampoline code is always aligned at the same point in
28487 its cache line. Look in `m68k.h' as a guide.
28489 -- Macro: TRANSFER_FROM_TRAMPOLINE
28490 Define this macro if trampolines need a special subroutine to do
28491 their work. The macro should expand to a series of `asm'
28492 statements which will be compiled with GCC. They go in a library
28493 function named `__transfer_from_trampoline'.
28495 If you need to avoid executing the ordinary prologue code of a
28496 compiled C function when you jump to the subroutine, you can do so
28497 by placing a special label of your own in the assembler code. Use
28498 one `asm' statement to generate an assembler label, and another to
28499 make the label global. Then trampolines can use that label to
28500 jump directly to your special assembler code.
28503 File: gccint.info, Node: Library Calls, Next: Addressing Modes, Prev: Trampolines, Up: Target Macros
28505 17.13 Implicit Calls to Library Routines
28506 ========================================
28508 Here is an explanation of implicit calls to library routines.
28510 -- Macro: DECLARE_LIBRARY_RENAMES
28511 This macro, if defined, should expand to a piece of C code that
28512 will get expanded when compiling functions for libgcc.a. It can
28513 be used to provide alternate names for GCC's internal library
28514 functions if there are ABI-mandated names that the compiler should
28517 -- Target Hook: void TARGET_INIT_LIBFUNCS (void)
28518 This hook should declare additional library routines or rename
28519 existing ones, using the functions `set_optab_libfunc' and
28520 `init_one_libfunc' defined in `optabs.c'. `init_optabs' calls
28521 this macro after initializing all the normal library routines.
28523 The default is to do nothing. Most ports don't need to define
28526 -- Macro: FLOAT_LIB_COMPARE_RETURNS_BOOL (MODE, COMPARISON)
28527 This macro should return `true' if the library routine that
28528 implements the floating point comparison operator COMPARISON in
28529 mode MODE will return a boolean, and FALSE if it will return a
28532 GCC's own floating point libraries return tristates from the
28533 comparison operators, so the default returns false always. Most
28534 ports don't need to define this macro.
28536 -- Macro: TARGET_LIB_INT_CMP_BIASED
28537 This macro should evaluate to `true' if the integer comparison
28538 functions (like `__cmpdi2') return 0 to indicate that the first
28539 operand is smaller than the second, 1 to indicate that they are
28540 equal, and 2 to indicate that the first operand is greater than
28541 the second. If this macro evaluates to `false' the comparison
28542 functions return -1, 0, and 1 instead of 0, 1, and 2. If the
28543 target uses the routines in `libgcc.a', you do not need to define
28546 -- Macro: US_SOFTWARE_GOFAST
28547 Define this macro if your system C library uses the US Software
28548 GOFAST library to provide floating point emulation.
28550 In addition to defining this macro, your architecture must set
28551 `TARGET_INIT_LIBFUNCS' to `gofast_maybe_init_libfuncs', or else
28552 call that function from its version of that hook. It is defined
28553 in `config/gofast.h', which must be included by your
28554 architecture's `CPU.c' file. See `sparc/sparc.c' for an example.
28556 If this macro is defined, the
28557 `TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL' target hook must return
28558 false for `SFmode' and `DFmode' comparisons.
28560 -- Macro: TARGET_EDOM
28561 The value of `EDOM' on the target machine, as a C integer constant
28562 expression. If you don't define this macro, GCC does not attempt
28563 to deposit the value of `EDOM' into `errno' directly. Look in
28564 `/usr/include/errno.h' to find the value of `EDOM' on your system.
28566 If you do not define `TARGET_EDOM', then compiled code reports
28567 domain errors by calling the library function and letting it
28568 report the error. If mathematical functions on your system use
28569 `matherr' when there is an error, then you should leave
28570 `TARGET_EDOM' undefined so that `matherr' is used normally.
28572 -- Macro: GEN_ERRNO_RTX
28573 Define this macro as a C expression to create an rtl expression
28574 that refers to the global "variable" `errno'. (On certain systems,
28575 `errno' may not actually be a variable.) If you don't define this
28576 macro, a reasonable default is used.
28578 -- Macro: TARGET_C99_FUNCTIONS
28579 When this macro is nonzero, GCC will implicitly optimize `sin'
28580 calls into `sinf' and similarly for other functions defined by C99
28581 standard. The default is zero because a number of existing
28582 systems lack support for these functions in their runtime so this
28583 macro needs to be redefined to one on systems that do support the
28586 -- Macro: TARGET_HAS_SINCOS
28587 When this macro is nonzero, GCC will implicitly optimize calls to
28588 `sin' and `cos' with the same argument to a call to `sincos'. The
28589 default is zero. The target has to provide the following
28591 void sincos(double x, double *sin, double *cos);
28592 void sincosf(float x, float *sin, float *cos);
28593 void sincosl(long double x, long double *sin, long double *cos);
28595 -- Macro: NEXT_OBJC_RUNTIME
28596 Define this macro to generate code for Objective-C message sending
28597 using the calling convention of the NeXT system. This calling
28598 convention involves passing the object, the selector and the
28599 method arguments all at once to the method-lookup library function.
28601 The default calling convention passes just the object and the
28602 selector to the lookup function, which returns a pointer to the
28606 File: gccint.info, Node: Addressing Modes, Next: Anchored Addresses, Prev: Library Calls, Up: Target Macros
28608 17.14 Addressing Modes
28609 ======================
28611 This is about addressing modes.
28613 -- Macro: HAVE_PRE_INCREMENT
28614 -- Macro: HAVE_PRE_DECREMENT
28615 -- Macro: HAVE_POST_INCREMENT
28616 -- Macro: HAVE_POST_DECREMENT
28617 A C expression that is nonzero if the machine supports
28618 pre-increment, pre-decrement, post-increment, or post-decrement
28619 addressing respectively.
28621 -- Macro: HAVE_PRE_MODIFY_DISP
28622 -- Macro: HAVE_POST_MODIFY_DISP
28623 A C expression that is nonzero if the machine supports pre- or
28624 post-address side-effect generation involving constants other than
28625 the size of the memory operand.
28627 -- Macro: HAVE_PRE_MODIFY_REG
28628 -- Macro: HAVE_POST_MODIFY_REG
28629 A C expression that is nonzero if the machine supports pre- or
28630 post-address side-effect generation involving a register
28633 -- Macro: CONSTANT_ADDRESS_P (X)
28634 A C expression that is 1 if the RTX X is a constant which is a
28635 valid address. On most machines, this can be defined as
28636 `CONSTANT_P (X)', but a few machines are more restrictive in which
28637 constant addresses are supported.
28639 -- Macro: CONSTANT_P (X)
28640 `CONSTANT_P', which is defined by target-independent code, accepts
28641 integer-values expressions whose values are not explicitly known,
28642 such as `symbol_ref', `label_ref', and `high' expressions and
28643 `const' arithmetic expressions, in addition to `const_int' and
28644 `const_double' expressions.
28646 -- Macro: MAX_REGS_PER_ADDRESS
28647 A number, the maximum number of registers that can appear in a
28648 valid memory address. Note that it is up to you to specify a
28649 value equal to the maximum number that `GO_IF_LEGITIMATE_ADDRESS'
28652 -- Macro: GO_IF_LEGITIMATE_ADDRESS (MODE, X, LABEL)
28653 A C compound statement with a conditional `goto LABEL;' executed
28654 if X (an RTX) is a legitimate memory address on the target machine
28655 for a memory operand of mode MODE.
28657 It usually pays to define several simpler macros to serve as
28658 subroutines for this one. Otherwise it may be too complicated to
28661 This macro must exist in two variants: a strict variant and a
28662 non-strict one. The strict variant is used in the reload pass. It
28663 must be defined so that any pseudo-register that has not been
28664 allocated a hard register is considered a memory reference. In
28665 contexts where some kind of register is required, a pseudo-register
28666 with no hard register must be rejected.
28668 The non-strict variant is used in other passes. It must be
28669 defined to accept all pseudo-registers in every context where some
28670 kind of register is required.
28672 Compiler source files that want to use the strict variant of this
28673 macro define the macro `REG_OK_STRICT'. You should use an `#ifdef
28674 REG_OK_STRICT' conditional to define the strict variant in that
28675 case and the non-strict variant otherwise.
28677 Subroutines to check for acceptable registers for various purposes
28678 (one for base registers, one for index registers, and so on) are
28679 typically among the subroutines used to define
28680 `GO_IF_LEGITIMATE_ADDRESS'. Then only these subroutine macros
28681 need have two variants; the higher levels of macros may be the
28682 same whether strict or not.
28684 Normally, constant addresses which are the sum of a `symbol_ref'
28685 and an integer are stored inside a `const' RTX to mark them as
28686 constant. Therefore, there is no need to recognize such sums
28687 specifically as legitimate addresses. Normally you would simply
28688 recognize any `const' as legitimate.
28690 Usually `PRINT_OPERAND_ADDRESS' is not prepared to handle constant
28691 sums that are not marked with `const'. It assumes that a naked
28692 `plus' indicates indexing. If so, then you _must_ reject such
28693 naked constant sums as illegitimate addresses, so that none of
28694 them will be given to `PRINT_OPERAND_ADDRESS'.
28696 On some machines, whether a symbolic address is legitimate depends
28697 on the section that the address refers to. On these machines,
28698 define the target hook `TARGET_ENCODE_SECTION_INFO' to store the
28699 information into the `symbol_ref', and then check for it here.
28700 When you see a `const', you will have to look inside it to find the
28701 `symbol_ref' in order to determine the section. *Note Assembler
28704 -- Macro: TARGET_MEM_CONSTRAINT
28705 A single character to be used instead of the default `'m''
28706 character for general memory addresses. This defines the
28707 constraint letter which matches the memory addresses accepted by
28708 `GO_IF_LEGITIMATE_ADDRESS_P'. Define this macro if you want to
28709 support new address formats in your back end without changing the
28710 semantics of the `'m'' constraint. This is necessary in order to
28711 preserve functionality of inline assembly constructs using the
28714 -- Macro: FIND_BASE_TERM (X)
28715 A C expression to determine the base term of address X, or to
28716 provide a simplified version of X from which `alias.c' can easily
28717 find the base term. This macro is used in only two places:
28718 `find_base_value' and `find_base_term' in `alias.c'.
28720 It is always safe for this macro to not be defined. It exists so
28721 that alias analysis can understand machine-dependent addresses.
28723 The typical use of this macro is to handle addresses containing a
28724 label_ref or symbol_ref within an UNSPEC.
28726 -- Macro: LEGITIMIZE_ADDRESS (X, OLDX, MODE, WIN)
28727 A C compound statement that attempts to replace X with a valid
28728 memory address for an operand of mode MODE. WIN will be a C
28729 statement label elsewhere in the code; the macro definition may use
28731 GO_IF_LEGITIMATE_ADDRESS (MODE, X, WIN);
28733 to avoid further processing if the address has become legitimate.
28735 X will always be the result of a call to `break_out_memory_refs',
28736 and OLDX will be the operand that was given to that function to
28739 The code generated by this macro should not alter the substructure
28740 of X. If it transforms X into a more legitimate form, it should
28741 assign X (which will always be a C variable) a new value.
28743 It is not necessary for this macro to come up with a legitimate
28744 address. The compiler has standard ways of doing so in all cases.
28745 In fact, it is safe to omit this macro. But often a
28746 machine-dependent strategy can generate better code.
28748 -- Macro: LEGITIMIZE_RELOAD_ADDRESS (X, MODE, OPNUM, TYPE, IND_LEVELS,
28750 A C compound statement that attempts to replace X, which is an
28751 address that needs reloading, with a valid memory address for an
28752 operand of mode MODE. WIN will be a C statement label elsewhere
28753 in the code. It is not necessary to define this macro, but it
28754 might be useful for performance reasons.
28756 For example, on the i386, it is sometimes possible to use a single
28757 reload register instead of two by reloading a sum of two pseudo
28758 registers into a register. On the other hand, for number of RISC
28759 processors offsets are limited so that often an intermediate
28760 address needs to be generated in order to address a stack slot.
28761 By defining `LEGITIMIZE_RELOAD_ADDRESS' appropriately, the
28762 intermediate addresses generated for adjacent some stack slots can
28763 be made identical, and thus be shared.
28765 _Note_: This macro should be used with caution. It is necessary
28766 to know something of how reload works in order to effectively use
28767 this, and it is quite easy to produce macros that build in too
28768 much knowledge of reload internals.
28770 _Note_: This macro must be able to reload an address created by a
28771 previous invocation of this macro. If it fails to handle such
28772 addresses then the compiler may generate incorrect code or abort.
28774 The macro definition should use `push_reload' to indicate parts
28775 that need reloading; OPNUM, TYPE and IND_LEVELS are usually
28776 suitable to be passed unaltered to `push_reload'.
28778 The code generated by this macro must not alter the substructure of
28779 X. If it transforms X into a more legitimate form, it should
28780 assign X (which will always be a C variable) a new value. This
28781 also applies to parts that you change indirectly by calling
28784 The macro definition may use `strict_memory_address_p' to test if
28785 the address has become legitimate.
28787 If you want to change only a part of X, one standard way of doing
28788 this is to use `copy_rtx'. Note, however, that it unshares only a
28789 single level of rtl. Thus, if the part to be changed is not at the
28790 top level, you'll need to replace first the top level. It is not
28791 necessary for this macro to come up with a legitimate address;
28792 but often a machine-dependent strategy can generate better code.
28794 -- Macro: GO_IF_MODE_DEPENDENT_ADDRESS (ADDR, LABEL)
28795 A C statement or compound statement with a conditional `goto
28796 LABEL;' executed if memory address X (an RTX) can have different
28797 meanings depending on the machine mode of the memory reference it
28798 is used for or if the address is valid for some modes but not
28801 Autoincrement and autodecrement addresses typically have
28802 mode-dependent effects because the amount of the increment or
28803 decrement is the size of the operand being addressed. Some
28804 machines have other mode-dependent addresses. Many RISC machines
28805 have no mode-dependent addresses.
28807 You may assume that ADDR is a valid address for the machine.
28809 -- Macro: LEGITIMATE_CONSTANT_P (X)
28810 A C expression that is nonzero if X is a legitimate constant for
28811 an immediate operand on the target machine. You can assume that X
28812 satisfies `CONSTANT_P', so you need not check this. In fact, `1'
28813 is a suitable definition for this macro on machines where anything
28814 `CONSTANT_P' is valid.
28816 -- Target Hook: rtx TARGET_DELEGITIMIZE_ADDRESS (rtx X)
28817 This hook is used to undo the possibly obfuscating effects of the
28818 `LEGITIMIZE_ADDRESS' and `LEGITIMIZE_RELOAD_ADDRESS' target
28819 macros. Some backend implementations of these macros wrap symbol
28820 references inside an `UNSPEC' rtx to represent PIC or similar
28821 addressing modes. This target hook allows GCC's optimizers to
28822 understand the semantics of these opaque `UNSPEC's by converting
28823 them back into their original form.
28825 -- Target Hook: bool TARGET_CANNOT_FORCE_CONST_MEM (rtx X)
28826 This hook should return true if X is of a form that cannot (or
28827 should not) be spilled to the constant pool. The default version
28828 of this hook returns false.
28830 The primary reason to define this hook is to prevent reload from
28831 deciding that a non-legitimate constant would be better reloaded
28832 from the constant pool instead of spilling and reloading a register
28833 holding the constant. This restriction is often true of addresses
28834 of TLS symbols for various targets.
28836 -- Target Hook: bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum
28837 machine_mode MODE, rtx X)
28838 This hook should return true if pool entries for constant X can be
28839 placed in an `object_block' structure. MODE is the mode of X.
28841 The default version returns false for all constants.
28843 -- Target Hook: tree TARGET_BUILTIN_RECIPROCAL (enum tree_code FN,
28844 bool TM_FN, bool SQRT)
28845 This hook should return the DECL of a function that implements
28846 reciprocal of the builtin function with builtin function code FN,
28847 or `NULL_TREE' if such a function is not available. TM_FN is true
28848 when FN is a code of a machine-dependent builtin function. When
28849 SQRT is true, additional optimizations that apply only to the
28850 reciprocal of a square root function are performed, and only
28851 reciprocals of `sqrt' function are valid.
28853 -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
28854 This hook should return the DECL of a function F that given an
28855 address ADDR as an argument returns a mask M that can be used to
28856 extract from two vectors the relevant data that resides in ADDR in
28857 case ADDR is not properly aligned.
28859 The autovectorizer, when vectorizing a load operation from an
28860 address ADDR that may be unaligned, will generate two vector loads
28861 from the two aligned addresses around ADDR. It then generates a
28862 `REALIGN_LOAD' operation to extract the relevant data from the two
28863 loaded vectors. The first two arguments to `REALIGN_LOAD', V1 and
28864 V2, are the two vectors, each of size VS, and the third argument,
28865 OFF, defines how the data will be extracted from these two
28866 vectors: if OFF is 0, then the returned vector is V2; otherwise,
28867 the returned vector is composed from the last VS-OFF elements of
28868 V1 concatenated to the first OFF elements of V2.
28870 If this hook is defined, the autovectorizer will generate a call
28871 to F (using the DECL tree that this hook returns) and will use the
28872 return value of F as the argument OFF to `REALIGN_LOAD'.
28873 Therefore, the mask M returned by F should comply with the
28874 semantics expected by `REALIGN_LOAD' described above. If this
28875 hook is not defined, then ADDR will be used as the argument OFF to
28876 `REALIGN_LOAD', in which case the low log2(VS)-1 bits of ADDR will
28879 -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN (tree X)
28880 This hook should return the DECL of a function F that implements
28881 widening multiplication of the even elements of two input vectors
28884 If this hook is defined, the autovectorizer will use it along with
28885 the `TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD' target hook when
28886 vectorizing widening multiplication in cases that the order of the
28887 results does not have to be preserved (e.g. used only by a
28888 reduction computation). Otherwise, the `widen_mult_hi/lo' idioms
28891 -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD (tree X)
28892 This hook should return the DECL of a function F that implements
28893 widening multiplication of the odd elements of two input vectors
28896 If this hook is defined, the autovectorizer will use it along with
28897 the `TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN' target hook when
28898 vectorizing widening multiplication in cases that the order of the
28899 results does not have to be preserved (e.g. used only by a
28900 reduction computation). Otherwise, the `widen_mult_hi/lo' idioms
28903 -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_CONVERSION (enum
28904 tree_code CODE, tree TYPE)
28905 This hook should return the DECL of a function that implements
28906 conversion of the input vector of type TYPE. If TYPE is an
28907 integral type, the result of the conversion is a vector of
28908 floating-point type of the same size. If TYPE is a floating-point
28909 type, the result of the conversion is a vector of integral type of
28910 the same size. CODE specifies how the conversion is to be applied
28911 (truncation, rounding, etc.).
28913 If this hook is defined, the autovectorizer will use the
28914 `TARGET_VECTORIZE_BUILTIN_CONVERSION' target hook when vectorizing
28915 conversion. Otherwise, it will return `NULL_TREE'.
28917 -- Target Hook: tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION
28918 (enum built_in_function CODE, tree VEC_TYPE_OUT, tree
28920 This hook should return the decl of a function that implements the
28921 vectorized variant of the builtin function with builtin function
28922 code CODE or `NULL_TREE' if such a function is not available. The
28923 return type of the vectorized function shall be of vector type
28924 VEC_TYPE_OUT and the argument types should be VEC_TYPE_IN.
28927 File: gccint.info, Node: Anchored Addresses, Next: Condition Code, Prev: Addressing Modes, Up: Target Macros
28929 17.15 Anchored Addresses
28930 ========================
28932 GCC usually addresses every static object as a separate entity. For
28933 example, if we have:
28935 static int a, b, c;
28936 int foo (void) { return a + b + c; }
28938 the code for `foo' will usually calculate three separate symbolic
28939 addresses: those of `a', `b' and `c'. On some targets, it would be
28940 better to calculate just one symbolic address and access the three
28941 variables relative to it. The equivalent pseudocode would be something
28946 register int *xr = &x;
28947 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
28950 (which isn't valid C). We refer to shared addresses like `x' as
28951 "section anchors". Their use is controlled by `-fsection-anchors'.
28953 The hooks below describe the target properties that GCC needs to know
28954 in order to make effective use of section anchors. It won't use
28955 section anchors at all unless either `TARGET_MIN_ANCHOR_OFFSET' or
28956 `TARGET_MAX_ANCHOR_OFFSET' is set to a nonzero value.
28958 -- Variable: Target Hook HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
28959 The minimum offset that should be applied to a section anchor. On
28960 most targets, it should be the smallest offset that can be applied
28961 to a base register while still giving a legitimate address for
28962 every mode. The default value is 0.
28964 -- Variable: Target Hook HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
28965 Like `TARGET_MIN_ANCHOR_OFFSET', but the maximum (inclusive)
28966 offset that should be applied to section anchors. The default
28969 -- Target Hook: void TARGET_ASM_OUTPUT_ANCHOR (rtx X)
28970 Write the assembly code to define section anchor X, which is a
28971 `SYMBOL_REF' for which `SYMBOL_REF_ANCHOR_P (X)' is true. The
28972 hook is called with the assembly output position set to the
28973 beginning of `SYMBOL_REF_BLOCK (X)'.
28975 If `ASM_OUTPUT_DEF' is available, the hook's default definition
28976 uses it to define the symbol as `. + SYMBOL_REF_BLOCK_OFFSET (X)'.
28977 If `ASM_OUTPUT_DEF' is not available, the hook's default definition
28978 is `NULL', which disables the use of section anchors altogether.
28980 -- Target Hook: bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (rtx X)
28981 Return true if GCC should attempt to use anchors to access
28982 `SYMBOL_REF' X. You can assume `SYMBOL_REF_HAS_BLOCK_INFO_P (X)'
28983 and `!SYMBOL_REF_ANCHOR_P (X)'.
28985 The default version is correct for most targets, but you might
28986 need to intercept this hook to handle things like target-specific
28987 attributes or target-specific sections.
28990 File: gccint.info, Node: Condition Code, Next: Costs, Prev: Anchored Addresses, Up: Target Macros
28992 17.16 Condition Code Status
28993 ===========================
28995 This describes the condition code status.
28997 The file `conditions.h' defines a variable `cc_status' to describe how
28998 the condition code was computed (in case the interpretation of the
28999 condition code depends on the instruction that it was set by). This
29000 variable contains the RTL expressions on which the condition code is
29001 currently based, and several standard flags.
29003 Sometimes additional machine-specific flags must be defined in the
29004 machine description header file. It can also add additional
29005 machine-specific information by defining `CC_STATUS_MDEP'.
29007 -- Macro: CC_STATUS_MDEP
29008 C code for a data type which is used for declaring the `mdep'
29009 component of `cc_status'. It defaults to `int'.
29011 This macro is not used on machines that do not use `cc0'.
29013 -- Macro: CC_STATUS_MDEP_INIT
29014 A C expression to initialize the `mdep' field to "empty". The
29015 default definition does nothing, since most machines don't use the
29016 field anyway. If you want to use the field, you should probably
29017 define this macro to initialize it.
29019 This macro is not used on machines that do not use `cc0'.
29021 -- Macro: NOTICE_UPDATE_CC (EXP, INSN)
29022 A C compound statement to set the components of `cc_status'
29023 appropriately for an insn INSN whose body is EXP. It is this
29024 macro's responsibility to recognize insns that set the condition
29025 code as a byproduct of other activity as well as those that
29026 explicitly set `(cc0)'.
29028 This macro is not used on machines that do not use `cc0'.
29030 If there are insns that do not set the condition code but do alter
29031 other machine registers, this macro must check to see whether they
29032 invalidate the expressions that the condition code is recorded as
29033 reflecting. For example, on the 68000, insns that store in address
29034 registers do not set the condition code, which means that usually
29035 `NOTICE_UPDATE_CC' can leave `cc_status' unaltered for such insns.
29036 But suppose that the previous insn set the condition code based
29037 on location `a4@(102)' and the current insn stores a new value in
29038 `a4'. Although the condition code is not changed by this, it will
29039 no longer be true that it reflects the contents of `a4@(102)'.
29040 Therefore, `NOTICE_UPDATE_CC' must alter `cc_status' in this case
29041 to say that nothing is known about the condition code value.
29043 The definition of `NOTICE_UPDATE_CC' must be prepared to deal with
29044 the results of peephole optimization: insns whose patterns are
29045 `parallel' RTXs containing various `reg', `mem' or constants which
29046 are just the operands. The RTL structure of these insns is not
29047 sufficient to indicate what the insns actually do. What
29048 `NOTICE_UPDATE_CC' should do when it sees one is just to run
29051 A possible definition of `NOTICE_UPDATE_CC' is to call a function
29052 that looks at an attribute (*note Insn Attributes::) named, for
29053 example, `cc'. This avoids having detailed information about
29054 patterns in two places, the `md' file and in `NOTICE_UPDATE_CC'.
29056 -- Macro: SELECT_CC_MODE (OP, X, Y)
29057 Returns a mode from class `MODE_CC' to be used when comparison
29058 operation code OP is applied to rtx X and Y. For example, on the
29059 SPARC, `SELECT_CC_MODE' is defined as (see *note Jump Patterns::
29060 for a description of the reason for this definition)
29062 #define SELECT_CC_MODE(OP,X,Y) \
29063 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
29064 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
29065 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
29066 || GET_CODE (X) == NEG) \
29067 ? CC_NOOVmode : CCmode))
29069 You should define this macro if and only if you define extra CC
29070 modes in `MACHINE-modes.def'.
29072 -- Macro: CANONICALIZE_COMPARISON (CODE, OP0, OP1)
29073 On some machines not all possible comparisons are defined, but you
29074 can convert an invalid comparison into a valid one. For example,
29075 the Alpha does not have a `GT' comparison, but you can use an `LT'
29076 comparison instead and swap the order of the operands.
29078 On such machines, define this macro to be a C statement to do any
29079 required conversions. CODE is the initial comparison code and OP0
29080 and OP1 are the left and right operands of the comparison,
29081 respectively. You should modify CODE, OP0, and OP1 as required.
29083 GCC will not assume that the comparison resulting from this macro
29084 is valid but will see if the resulting insn matches a pattern in
29087 You need not define this macro if it would never change the
29088 comparison code or operands.
29090 -- Macro: REVERSIBLE_CC_MODE (MODE)
29091 A C expression whose value is one if it is always safe to reverse a
29092 comparison whose mode is MODE. If `SELECT_CC_MODE' can ever
29093 return MODE for a floating-point inequality comparison, then
29094 `REVERSIBLE_CC_MODE (MODE)' must be zero.
29096 You need not define this macro if it would always returns zero or
29097 if the floating-point format is anything other than
29098 `IEEE_FLOAT_FORMAT'. For example, here is the definition used on
29099 the SPARC, where floating-point inequality comparisons are always
29102 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
29104 -- Macro: REVERSE_CONDITION (CODE, MODE)
29105 A C expression whose value is reversed condition code of the CODE
29106 for comparison done in CC_MODE MODE. The macro is used only in
29107 case `REVERSIBLE_CC_MODE (MODE)' is nonzero. Define this macro in
29108 case machine has some non-standard way how to reverse certain
29109 conditionals. For instance in case all floating point conditions
29110 are non-trapping, compiler may freely convert unordered compares
29111 to ordered one. Then definition may look like:
29113 #define REVERSE_CONDITION(CODE, MODE) \
29114 ((MODE) != CCFPmode ? reverse_condition (CODE) \
29115 : reverse_condition_maybe_unordered (CODE))
29117 -- Macro: REVERSE_CONDEXEC_PREDICATES_P (OP1, OP2)
29118 A C expression that returns true if the conditional execution
29119 predicate OP1, a comparison operation, is the inverse of OP2 and
29120 vice versa. Define this to return 0 if the target has conditional
29121 execution predicates that cannot be reversed safely. There is no
29122 need to validate that the arguments of op1 and op2 are the same,
29123 this is done separately. If no expansion is specified, this macro
29124 is defined as follows:
29126 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
29127 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
29129 -- Target Hook: bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *,
29131 On targets which do not use `(cc0)', and which use a hard register
29132 rather than a pseudo-register to hold condition codes, the regular
29133 CSE passes are often not able to identify cases in which the hard
29134 register is set to a common value. Use this hook to enable a
29135 small pass which optimizes such cases. This hook should return
29136 true to enable this pass, and it should set the integers to which
29137 its arguments point to the hard register numbers used for
29138 condition codes. When there is only one such register, as is true
29139 on most systems, the integer pointed to by the second argument
29140 should be set to `INVALID_REGNUM'.
29142 The default version of this hook returns false.
29144 -- Target Hook: enum machine_mode TARGET_CC_MODES_COMPATIBLE (enum
29145 machine_mode, enum machine_mode)
29146 On targets which use multiple condition code modes in class
29147 `MODE_CC', it is sometimes the case that a comparison can be
29148 validly done in more than one mode. On such a system, define this
29149 target hook to take two mode arguments and to return a mode in
29150 which both comparisons may be validly done. If there is no such
29151 mode, return `VOIDmode'.
29153 The default version of this hook checks whether the modes are the
29154 same. If they are, it returns that mode. If they are different,
29155 it returns `VOIDmode'.
29158 File: gccint.info, Node: Costs, Next: Scheduling, Prev: Condition Code, Up: Target Macros
29160 17.17 Describing Relative Costs of Operations
29161 =============================================
29163 These macros let you describe the relative speed of various operations
29164 on the target machine.
29166 -- Macro: REGISTER_MOVE_COST (MODE, FROM, TO)
29167 A C expression for the cost of moving data of mode MODE from a
29168 register in class FROM to one in class TO. The classes are
29169 expressed using the enumeration values such as `GENERAL_REGS'. A
29170 value of 2 is the default; other values are interpreted relative to
29173 It is not required that the cost always equal 2 when FROM is the
29174 same as TO; on some machines it is expensive to move between
29175 registers if they are not general registers.
29177 If reload sees an insn consisting of a single `set' between two
29178 hard registers, and if `REGISTER_MOVE_COST' applied to their
29179 classes returns a value of 2, reload does not check to ensure that
29180 the constraints of the insn are met. Setting a cost of other than
29181 2 will allow reload to verify that the constraints are met. You
29182 should do this if the `movM' pattern's constraints do not allow
29185 -- Macro: MEMORY_MOVE_COST (MODE, CLASS, IN)
29186 A C expression for the cost of moving data of mode MODE between a
29187 register of class CLASS and memory; IN is zero if the value is to
29188 be written to memory, nonzero if it is to be read in. This cost
29189 is relative to those in `REGISTER_MOVE_COST'. If moving between
29190 registers and memory is more expensive than between two registers,
29191 you should define this macro to express the relative cost.
29193 If you do not define this macro, GCC uses a default cost of 4 plus
29194 the cost of copying via a secondary reload register, if one is
29195 needed. If your machine requires a secondary reload register to
29196 copy between memory and a register of CLASS but the reload
29197 mechanism is more complex than copying via an intermediate, define
29198 this macro to reflect the actual cost of the move.
29200 GCC defines the function `memory_move_secondary_cost' if secondary
29201 reloads are needed. It computes the costs due to copying via a
29202 secondary register. If your machine copies from memory using a
29203 secondary register in the conventional way but the default base
29204 value of 4 is not correct for your machine, define this macro to
29205 add some other value to the result of that function. The
29206 arguments to that function are the same as to this macro.
29208 -- Macro: BRANCH_COST (SPEED_P, PREDICTABLE_P)
29209 A C expression for the cost of a branch instruction. A value of 1
29210 is the default; other values are interpreted relative to that.
29211 Parameter SPEED_P is true when the branch in question should be
29212 optimized for speed. When it is false, `BRANCH_COST' should be
29213 returning value optimal for code size rather then performance
29214 considerations. PREDICTABLE_P is true for well predictable
29215 branches. On many architectures the `BRANCH_COST' can be reduced
29218 Here are additional macros which do not specify precise relative costs,
29219 but only that certain actions are more expensive than GCC would
29222 -- Macro: SLOW_BYTE_ACCESS
29223 Define this macro as a C expression which is nonzero if accessing
29224 less than a word of memory (i.e. a `char' or a `short') is no
29225 faster than accessing a word of memory, i.e., if such access
29226 require more than one instruction or if there is no difference in
29227 cost between byte and (aligned) word loads.
29229 When this macro is not defined, the compiler will access a field by
29230 finding the smallest containing object; when it is defined, a
29231 fullword load will be used if alignment permits. Unless bytes
29232 accesses are faster than word accesses, using word accesses is
29233 preferable since it may eliminate subsequent memory access if
29234 subsequent accesses occur to other fields in the same word of the
29235 structure, but to different bytes.
29237 -- Macro: SLOW_UNALIGNED_ACCESS (MODE, ALIGNMENT)
29238 Define this macro to be the value 1 if memory accesses described
29239 by the MODE and ALIGNMENT parameters have a cost many times greater
29240 than aligned accesses, for example if they are emulated in a trap
29243 When this macro is nonzero, the compiler will act as if
29244 `STRICT_ALIGNMENT' were nonzero when generating code for block
29245 moves. This can cause significantly more instructions to be
29246 produced. Therefore, do not set this macro nonzero if unaligned
29247 accesses only add a cycle or two to the time for a memory access.
29249 If the value of this macro is always zero, it need not be defined.
29250 If this macro is defined, it should produce a nonzero value when
29251 `STRICT_ALIGNMENT' is nonzero.
29253 -- Macro: MOVE_RATIO
29254 The threshold of number of scalar memory-to-memory move insns,
29255 _below_ which a sequence of insns should be generated instead of a
29256 string move insn or a library call. Increasing the value will
29257 always make code faster, but eventually incurs high cost in
29258 increased code size.
29260 Note that on machines where the corresponding move insn is a
29261 `define_expand' that emits a sequence of insns, this macro counts
29262 the number of such sequences.
29264 If you don't define this, a reasonable default is used.
29266 -- Macro: MOVE_BY_PIECES_P (SIZE, ALIGNMENT)
29267 A C expression used to determine whether `move_by_pieces' will be
29268 used to copy a chunk of memory, or whether some other block move
29269 mechanism will be used. Defaults to 1 if `move_by_pieces_ninsns'
29270 returns less than `MOVE_RATIO'.
29272 -- Macro: MOVE_MAX_PIECES
29273 A C expression used by `move_by_pieces' to determine the largest
29274 unit a load or store used to copy memory is. Defaults to
29277 -- Macro: CLEAR_RATIO
29278 The threshold of number of scalar move insns, _below_ which a
29279 sequence of insns should be generated to clear memory instead of a
29280 string clear insn or a library call. Increasing the value will
29281 always make code faster, but eventually incurs high cost in
29282 increased code size.
29284 If you don't define this, a reasonable default is used.
29286 -- Macro: CLEAR_BY_PIECES_P (SIZE, ALIGNMENT)
29287 A C expression used to determine whether `clear_by_pieces' will be
29288 used to clear a chunk of memory, or whether some other block clear
29289 mechanism will be used. Defaults to 1 if `move_by_pieces_ninsns'
29290 returns less than `CLEAR_RATIO'.
29292 -- Macro: SET_RATIO
29293 The threshold of number of scalar move insns, _below_ which a
29294 sequence of insns should be generated to set memory to a constant
29295 value, instead of a block set insn or a library call. Increasing
29296 the value will always make code faster, but eventually incurs high
29297 cost in increased code size.
29299 If you don't define this, it defaults to the value of `MOVE_RATIO'.
29301 -- Macro: SET_BY_PIECES_P (SIZE, ALIGNMENT)
29302 A C expression used to determine whether `store_by_pieces' will be
29303 used to set a chunk of memory to a constant value, or whether some
29304 other mechanism will be used. Used by `__builtin_memset' when
29305 storing values other than constant zero. Defaults to 1 if
29306 `move_by_pieces_ninsns' returns less than `SET_RATIO'.
29308 -- Macro: STORE_BY_PIECES_P (SIZE, ALIGNMENT)
29309 A C expression used to determine whether `store_by_pieces' will be
29310 used to set a chunk of memory to a constant string value, or
29311 whether some other mechanism will be used. Used by
29312 `__builtin_strcpy' when called with a constant source string.
29313 Defaults to 1 if `move_by_pieces_ninsns' returns less than
29316 -- Macro: USE_LOAD_POST_INCREMENT (MODE)
29317 A C expression used to determine whether a load postincrement is a
29318 good thing to use for a given mode. Defaults to the value of
29319 `HAVE_POST_INCREMENT'.
29321 -- Macro: USE_LOAD_POST_DECREMENT (MODE)
29322 A C expression used to determine whether a load postdecrement is a
29323 good thing to use for a given mode. Defaults to the value of
29324 `HAVE_POST_DECREMENT'.
29326 -- Macro: USE_LOAD_PRE_INCREMENT (MODE)
29327 A C expression used to determine whether a load preincrement is a
29328 good thing to use for a given mode. Defaults to the value of
29329 `HAVE_PRE_INCREMENT'.
29331 -- Macro: USE_LOAD_PRE_DECREMENT (MODE)
29332 A C expression used to determine whether a load predecrement is a
29333 good thing to use for a given mode. Defaults to the value of
29334 `HAVE_PRE_DECREMENT'.
29336 -- Macro: USE_STORE_POST_INCREMENT (MODE)
29337 A C expression used to determine whether a store postincrement is
29338 a good thing to use for a given mode. Defaults to the value of
29339 `HAVE_POST_INCREMENT'.
29341 -- Macro: USE_STORE_POST_DECREMENT (MODE)
29342 A C expression used to determine whether a store postdecrement is
29343 a good thing to use for a given mode. Defaults to the value of
29344 `HAVE_POST_DECREMENT'.
29346 -- Macro: USE_STORE_PRE_INCREMENT (MODE)
29347 This macro is used to determine whether a store preincrement is a
29348 good thing to use for a given mode. Defaults to the value of
29349 `HAVE_PRE_INCREMENT'.
29351 -- Macro: USE_STORE_PRE_DECREMENT (MODE)
29352 This macro is used to determine whether a store predecrement is a
29353 good thing to use for a given mode. Defaults to the value of
29354 `HAVE_PRE_DECREMENT'.
29356 -- Macro: NO_FUNCTION_CSE
29357 Define this macro if it is as good or better to call a constant
29358 function address than to call an address kept in a register.
29360 -- Macro: RANGE_TEST_NON_SHORT_CIRCUIT
29361 Define this macro if a non-short-circuit operation produced by
29362 `fold_range_test ()' is optimal. This macro defaults to true if
29363 `BRANCH_COST' is greater than or equal to the value 2.
29365 -- Target Hook: bool TARGET_RTX_COSTS (rtx X, int CODE, int
29366 OUTER_CODE, int *TOTAL)
29367 This target hook describes the relative costs of RTL expressions.
29369 The cost may depend on the precise form of the expression, which is
29370 available for examination in X, and the rtx code of the expression
29371 in which it is contained, found in OUTER_CODE. CODE is the
29372 expression code--redundant, since it can be obtained with
29375 In implementing this hook, you can use the construct
29376 `COSTS_N_INSNS (N)' to specify a cost equal to N fast instructions.
29378 On entry to the hook, `*TOTAL' contains a default estimate for the
29379 cost of the expression. The hook should modify this value as
29380 necessary. Traditionally, the default costs are `COSTS_N_INSNS
29381 (5)' for multiplications, `COSTS_N_INSNS (7)' for division and
29382 modulus operations, and `COSTS_N_INSNS (1)' for all other
29385 When optimizing for code size, i.e. when `optimize_size' is
29386 nonzero, this target hook should be used to estimate the relative
29387 size cost of an expression, again relative to `COSTS_N_INSNS'.
29389 The hook returns true when all subexpressions of X have been
29390 processed, and false when `rtx_cost' should recurse.
29392 -- Target Hook: int TARGET_ADDRESS_COST (rtx ADDRESS)
29393 This hook computes the cost of an addressing mode that contains
29394 ADDRESS. If not defined, the cost is computed from the ADDRESS
29395 expression and the `TARGET_RTX_COST' hook.
29397 For most CISC machines, the default cost is a good approximation
29398 of the true cost of the addressing mode. However, on RISC
29399 machines, all instructions normally have the same length and
29400 execution time. Hence all addresses will have equal costs.
29402 In cases where more than one form of an address is known, the form
29403 with the lowest cost will be used. If multiple forms have the
29404 same, lowest, cost, the one that is the most complex will be used.
29406 For example, suppose an address that is equal to the sum of a
29407 register and a constant is used twice in the same basic block.
29408 When this macro is not defined, the address will be computed in a
29409 register and memory references will be indirect through that
29410 register. On machines where the cost of the addressing mode
29411 containing the sum is no higher than that of a simple indirect
29412 reference, this will produce an additional instruction and
29413 possibly require an additional register. Proper specification of
29414 this macro eliminates this overhead for such machines.
29416 This hook is never called with an invalid address.
29418 On machines where an address involving more than one register is as
29419 cheap as an address computation involving only one register,
29420 defining `TARGET_ADDRESS_COST' to reflect this can cause two
29421 registers to be live over a region of code where only one would
29422 have been if `TARGET_ADDRESS_COST' were not defined in that
29423 manner. This effect should be considered in the definition of
29424 this macro. Equivalent costs should probably only be given to
29425 addresses with different numbers of registers on machines with
29429 File: gccint.info, Node: Scheduling, Next: Sections, Prev: Costs, Up: Target Macros
29431 17.18 Adjusting the Instruction Scheduler
29432 =========================================
29434 The instruction scheduler may need a fair amount of machine-specific
29435 adjustment in order to produce good code. GCC provides several target
29436 hooks for this purpose. It is usually enough to define just a few of
29437 them: try the first ones in this list first.
29439 -- Target Hook: int TARGET_SCHED_ISSUE_RATE (void)
29440 This hook returns the maximum number of instructions that can ever
29441 issue at the same time on the target machine. The default is one.
29442 Although the insn scheduler can define itself the possibility of
29443 issue an insn on the same cycle, the value can serve as an
29444 additional constraint to issue insns on the same simulated
29445 processor cycle (see hooks `TARGET_SCHED_REORDER' and
29446 `TARGET_SCHED_REORDER2'). This value must be constant over the
29447 entire compilation. If you need it to vary depending on what the
29448 instructions are, you must use `TARGET_SCHED_VARIABLE_ISSUE'.
29450 -- Target Hook: int TARGET_SCHED_VARIABLE_ISSUE (FILE *FILE, int
29451 VERBOSE, rtx INSN, int MORE)
29452 This hook is executed by the scheduler after it has scheduled an
29453 insn from the ready list. It should return the number of insns
29454 which can still be issued in the current cycle. The default is
29455 `MORE - 1' for insns other than `CLOBBER' and `USE', which
29456 normally are not counted against the issue rate. You should
29457 define this hook if some insns take more machine resources than
29458 others, so that fewer insns can follow them in the same cycle.
29459 FILE is either a null pointer, or a stdio stream to write any
29460 debug output to. VERBOSE is the verbose level provided by
29461 `-fsched-verbose-N'. INSN is the instruction that was scheduled.
29463 -- Target Hook: int TARGET_SCHED_ADJUST_COST (rtx INSN, rtx LINK, rtx
29464 DEP_INSN, int COST)
29465 This function corrects the value of COST based on the relationship
29466 between INSN and DEP_INSN through the dependence LINK. It should
29467 return the new value. The default is to make no adjustment to
29468 COST. This can be used for example to specify to the scheduler
29469 using the traditional pipeline description that an output- or
29470 anti-dependence does not incur the same cost as a data-dependence.
29471 If the scheduler using the automaton based pipeline description,
29472 the cost of anti-dependence is zero and the cost of
29473 output-dependence is maximum of one and the difference of latency
29474 times of the first and the second insns. If these values are not
29475 acceptable, you could use the hook to modify them too. See also
29476 *note Processor pipeline description::.
29478 -- Target Hook: int TARGET_SCHED_ADJUST_PRIORITY (rtx INSN, int
29480 This hook adjusts the integer scheduling priority PRIORITY of
29481 INSN. It should return the new priority. Increase the priority to
29482 execute INSN earlier, reduce the priority to execute INSN later.
29483 Do not define this hook if you do not need to adjust the
29484 scheduling priorities of insns.
29486 -- Target Hook: int TARGET_SCHED_REORDER (FILE *FILE, int VERBOSE, rtx
29487 *READY, int *N_READYP, int CLOCK)
29488 This hook is executed by the scheduler after it has scheduled the
29489 ready list, to allow the machine description to reorder it (for
29490 example to combine two small instructions together on `VLIW'
29491 machines). FILE is either a null pointer, or a stdio stream to
29492 write any debug output to. VERBOSE is the verbose level provided
29493 by `-fsched-verbose-N'. READY is a pointer to the ready list of
29494 instructions that are ready to be scheduled. N_READYP is a
29495 pointer to the number of elements in the ready list. The scheduler
29496 reads the ready list in reverse order, starting with
29497 READY[*N_READYP-1] and going to READY[0]. CLOCK is the timer tick
29498 of the scheduler. You may modify the ready list and the number of
29499 ready insns. The return value is the number of insns that can
29500 issue this cycle; normally this is just `issue_rate'. See also
29501 `TARGET_SCHED_REORDER2'.
29503 -- Target Hook: int TARGET_SCHED_REORDER2 (FILE *FILE, int VERBOSE,
29504 rtx *READY, int *N_READY, CLOCK)
29505 Like `TARGET_SCHED_REORDER', but called at a different time. That
29506 function is called whenever the scheduler starts a new cycle.
29507 This one is called once per iteration over a cycle, immediately
29508 after `TARGET_SCHED_VARIABLE_ISSUE'; it can reorder the ready list
29509 and return the number of insns to be scheduled in the same cycle.
29510 Defining this hook can be useful if there are frequent situations
29511 where scheduling one insn causes other insns to become ready in
29512 the same cycle. These other insns can then be taken into account
29515 -- Target Hook: void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx
29517 This hook is called after evaluation forward dependencies of insns
29518 in chain given by two parameter values (HEAD and TAIL
29519 correspondingly) but before insns scheduling of the insn chain.
29520 For example, it can be used for better insn classification if it
29521 requires analysis of dependencies. This hook can use backward and
29522 forward dependencies of the insn scheduler because they are already
29525 -- Target Hook: void TARGET_SCHED_INIT (FILE *FILE, int VERBOSE, int
29527 This hook is executed by the scheduler at the beginning of each
29528 block of instructions that are to be scheduled. FILE is either a
29529 null pointer, or a stdio stream to write any debug output to.
29530 VERBOSE is the verbose level provided by `-fsched-verbose-N'.
29531 MAX_READY is the maximum number of insns in the current scheduling
29532 region that can be live at the same time. This can be used to
29533 allocate scratch space if it is needed, e.g. by
29534 `TARGET_SCHED_REORDER'.
29536 -- Target Hook: void TARGET_SCHED_FINISH (FILE *FILE, int VERBOSE)
29537 This hook is executed by the scheduler at the end of each block of
29538 instructions that are to be scheduled. It can be used to perform
29539 cleanup of any actions done by the other scheduling hooks. FILE
29540 is either a null pointer, or a stdio stream to write any debug
29541 output to. VERBOSE is the verbose level provided by
29542 `-fsched-verbose-N'.
29544 -- Target Hook: void TARGET_SCHED_INIT_GLOBAL (FILE *FILE, int
29545 VERBOSE, int OLD_MAX_UID)
29546 This hook is executed by the scheduler after function level
29547 initializations. FILE is either a null pointer, or a stdio stream
29548 to write any debug output to. VERBOSE is the verbose level
29549 provided by `-fsched-verbose-N'. OLD_MAX_UID is the maximum insn
29550 uid when scheduling begins.
29552 -- Target Hook: void TARGET_SCHED_FINISH_GLOBAL (FILE *FILE, int
29554 This is the cleanup hook corresponding to
29555 `TARGET_SCHED_INIT_GLOBAL'. FILE is either a null pointer, or a
29556 stdio stream to write any debug output to. VERBOSE is the verbose
29557 level provided by `-fsched-verbose-N'.
29559 -- Target Hook: int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
29560 The hook returns an RTL insn. The automaton state used in the
29561 pipeline hazard recognizer is changed as if the insn were scheduled
29562 when the new simulated processor cycle starts. Usage of the hook
29563 may simplify the automaton pipeline description for some VLIW
29564 processors. If the hook is defined, it is used only for the
29565 automaton based pipeline description. The default is not to
29566 change the state when the new simulated processor cycle starts.
29568 -- Target Hook: void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
29569 The hook can be used to initialize data used by the previous hook.
29571 -- Target Hook: int TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
29572 The hook is analogous to `TARGET_SCHED_DFA_PRE_CYCLE_INSN' but used
29573 to changed the state as if the insn were scheduled when the new
29574 simulated processor cycle finishes.
29576 -- Target Hook: void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
29577 The hook is analogous to `TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN' but
29578 used to initialize data used by the previous hook.
29580 -- Target Hook: void TARGET_SCHED_DFA_PRE_CYCLE_ADVANCE (void)
29581 The hook to notify target that the current simulated cycle is
29582 about to finish. The hook is analogous to
29583 `TARGET_SCHED_DFA_PRE_CYCLE_INSN' but used to change the state in
29584 more complicated situations - e.g., when advancing state on a
29585 single insn is not enough.
29587 -- Target Hook: void TARGET_SCHED_DFA_POST_CYCLE_ADVANCE (void)
29588 The hook to notify target that new simulated cycle has just
29589 started. The hook is analogous to
29590 `TARGET_SCHED_DFA_POST_CYCLE_INSN' but used to change the state in
29591 more complicated situations - e.g., when advancing state on a
29592 single insn is not enough.
29594 -- Target Hook: int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD
29596 This hook controls better choosing an insn from the ready insn
29597 queue for the DFA-based insn scheduler. Usually the scheduler
29598 chooses the first insn from the queue. If the hook returns a
29599 positive value, an additional scheduler code tries all
29600 permutations of `TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD
29601 ()' subsequent ready insns to choose an insn whose issue will
29602 result in maximal number of issued insns on the same cycle. For
29603 the VLIW processor, the code could actually solve the problem of
29604 packing simple insns into the VLIW insn. Of course, if the rules
29605 of VLIW packing are described in the automaton.
29607 This code also could be used for superscalar RISC processors. Let
29608 us consider a superscalar RISC processor with 3 pipelines. Some
29609 insns can be executed in pipelines A or B, some insns can be
29610 executed only in pipelines B or C, and one insn can be executed in
29611 pipeline B. The processor may issue the 1st insn into A and the
29612 2nd one into B. In this case, the 3rd insn will wait for freeing B
29613 until the next cycle. If the scheduler issues the 3rd insn the
29614 first, the processor could issue all 3 insns per cycle.
29616 Actually this code demonstrates advantages of the automaton based
29617 pipeline hazard recognizer. We try quickly and easy many insn
29618 schedules to choose the best one.
29620 The default is no multipass scheduling.
29622 -- Target Hook: int
29623 TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx)
29624 This hook controls what insns from the ready insn queue will be
29625 considered for the multipass insn scheduling. If the hook returns
29626 zero for insn passed as the parameter, the insn will be not chosen
29629 The default is that any ready insns can be chosen to be issued.
29631 -- Target Hook: int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int,
29633 This hook is called by the insn scheduler before issuing insn
29634 passed as the third parameter on given cycle. If the hook returns
29635 nonzero, the insn is not issued on given processors cycle.
29636 Instead of that, the processor cycle is advanced. If the value
29637 passed through the last parameter is zero, the insn ready queue is
29638 not sorted on the new cycle start as usually. The first parameter
29639 passes file for debugging output. The second one passes the
29640 scheduler verbose level of the debugging output. The forth and
29641 the fifth parameter values are correspondingly processor cycle on
29642 which the previous insn has been issued and the current processor
29645 -- Target Hook: bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct dep_def
29646 *_DEP, int COST, int DISTANCE)
29647 This hook is used to define which dependences are considered
29648 costly by the target, so costly that it is not advisable to
29649 schedule the insns that are involved in the dependence too close
29650 to one another. The parameters to this hook are as follows: The
29651 first parameter _DEP is the dependence being evaluated. The
29652 second parameter COST is the cost of the dependence, and the third
29653 parameter DISTANCE is the distance in cycles between the two insns.
29654 The hook returns `true' if considering the distance between the two
29655 insns the dependence between them is considered costly by the
29656 target, and `false' otherwise.
29658 Defining this hook can be useful in multiple-issue out-of-order
29659 machines, where (a) it's practically hopeless to predict the
29660 actual data/resource delays, however: (b) there's a better chance
29661 to predict the actual grouping that will be formed, and (c)
29662 correctly emulating the grouping can be very important. In such
29663 targets one may want to allow issuing dependent insns closer to
29664 one another--i.e., closer than the dependence distance; however,
29665 not in cases of "costly dependences", which this hooks allows to
29668 -- Target Hook: void TARGET_SCHED_H_I_D_EXTENDED (void)
29669 This hook is called by the insn scheduler after emitting a new
29670 instruction to the instruction stream. The hook notifies a target
29671 backend to extend its per instruction data structures.
29673 -- Target Hook: void * TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)
29674 Return a pointer to a store large enough to hold target scheduling
29677 -- Target Hook: void TARGET_SCHED_INIT_SCHED_CONTEXT (void *TC, bool
29679 Initialize store pointed to by TC to hold target scheduling
29680 context. It CLEAN_P is true then initialize TC as if scheduler is
29681 at the beginning of the block. Otherwise, make a copy of the
29682 current context in TC.
29684 -- Target Hook: void TARGET_SCHED_SET_SCHED_CONTEXT (void *TC)
29685 Copy target scheduling context pointer to by TC to the current
29688 -- Target Hook: void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *TC)
29689 Deallocate internal data in target scheduling context pointed to
29692 -- Target Hook: void TARGET_SCHED_FREE_SCHED_CONTEXT (void *TC)
29693 Deallocate a store for target scheduling context pointed to by TC.
29695 -- Target Hook: void * TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)
29696 Return a pointer to a store large enough to hold target scheduling
29699 -- Target Hook: void TARGET_SCHED_INIT_SCHED_CONTEXT (void *TC, bool
29701 Initialize store pointed to by TC to hold target scheduling
29702 context. It CLEAN_P is true then initialize TC as if scheduler is
29703 at the beginning of the block. Otherwise, make a copy of the
29704 current context in TC.
29706 -- Target Hook: void TARGET_SCHED_SET_SCHED_CONTEXT (void *TC)
29707 Copy target scheduling context pointer to by TC to the current
29710 -- Target Hook: void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *TC)
29711 Deallocate internal data in target scheduling context pointed to
29714 -- Target Hook: void TARGET_SCHED_FREE_SCHED_CONTEXT (void *TC)
29715 Deallocate a store for target scheduling context pointed to by TC.
29717 -- Target Hook: int TARGET_SCHED_SPECULATE_INSN (rtx INSN, int
29718 REQUEST, rtx *NEW_PAT)
29719 This hook is called by the insn scheduler when INSN has only
29720 speculative dependencies and therefore can be scheduled
29721 speculatively. The hook is used to check if the pattern of INSN
29722 has a speculative version and, in case of successful check, to
29723 generate that speculative pattern. The hook should return 1, if
29724 the instruction has a speculative form, or -1, if it doesn't.
29725 REQUEST describes the type of requested speculation. If the
29726 return value equals 1 then NEW_PAT is assigned the generated
29727 speculative pattern.
29729 -- Target Hook: int TARGET_SCHED_NEEDS_BLOCK_P (rtx INSN)
29730 This hook is called by the insn scheduler during generation of
29731 recovery code for INSN. It should return nonzero, if the
29732 corresponding check instruction should branch to recovery code, or
29735 -- Target Hook: rtx TARGET_SCHED_GEN_CHECK (rtx INSN, rtx LABEL, int
29737 This hook is called by the insn scheduler to generate a pattern
29738 for recovery check instruction. If MUTATE_P is zero, then INSN is
29739 a speculative instruction for which the check should be generated.
29740 LABEL is either a label of a basic block, where recovery code
29741 should be emitted, or a null pointer, when requested check doesn't
29742 branch to recovery code (a simple check). If MUTATE_P is nonzero,
29743 then a pattern for a branchy check corresponding to a simple check
29744 denoted by INSN should be generated. In this case LABEL can't be
29747 -- Target Hook: int
29748 TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC (rtx INSN)
29749 This hook is used as a workaround for
29750 `TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD' not being
29751 called on the first instruction of the ready list. The hook is
29752 used to discard speculative instruction that stand first in the
29753 ready list from being scheduled on the current cycle. For
29754 non-speculative instructions, the hook should always return
29755 nonzero. For example, in the ia64 backend the hook is used to
29756 cancel data speculative insns when the ALAT table is nearly full.
29758 -- Target Hook: void TARGET_SCHED_SET_SCHED_FLAGS (unsigned int
29759 *FLAGS, spec_info_t SPEC_INFO)
29760 This hook is used by the insn scheduler to find out what features
29761 should be enabled/used. FLAGS initially may have either the
29762 SCHED_RGN or SCHED_EBB bit set. This denotes the scheduler pass
29763 for which the data should be provided. The target backend should
29764 modify FLAGS by modifying the bits corresponding to the following
29765 features: USE_DEPS_LIST, USE_GLAT, DETACH_LIFE_INFO, and
29766 DO_SPECULATION. For the DO_SPECULATION feature an additional
29767 structure SPEC_INFO should be filled by the target. The structure
29768 describes speculation types that can be used in the scheduler.
29770 -- Target Hook: int TARGET_SCHED_SMS_RES_MII (struct ddg *G)
29771 This hook is called by the swing modulo scheduler to calculate a
29772 resource-based lower bound which is based on the resources
29773 available in the machine and the resources required by each
29774 instruction. The target backend can use G to calculate such
29775 bound. A very simple lower bound will be used in case this hook
29776 is not implemented: the total number of instructions divided by
29780 File: gccint.info, Node: Sections, Next: PIC, Prev: Scheduling, Up: Target Macros
29782 17.19 Dividing the Output into Sections (Texts, Data, ...)
29783 ==========================================================
29785 An object file is divided into sections containing different types of
29786 data. In the most common case, there are three sections: the "text
29787 section", which holds instructions and read-only data; the "data
29788 section", which holds initialized writable data; and the "bss section",
29789 which holds uninitialized data. Some systems have other kinds of
29792 `varasm.c' provides several well-known sections, such as
29793 `text_section', `data_section' and `bss_section'. The normal way of
29794 controlling a `FOO_section' variable is to define the associated
29795 `FOO_SECTION_ASM_OP' macro, as described below. The macros are only
29796 read once, when `varasm.c' initializes itself, so their values must be
29797 run-time constants. They may however depend on command-line flags.
29799 _Note:_ Some run-time files, such `crtstuff.c', also make use of the
29800 `FOO_SECTION_ASM_OP' macros, and expect them to be string literals.
29802 Some assemblers require a different string to be written every time a
29803 section is selected. If your assembler falls into this category, you
29804 should define the `TARGET_ASM_INIT_SECTIONS' hook and use
29805 `get_unnamed_section' to set up the sections.
29807 You must always create a `text_section', either by defining
29808 `TEXT_SECTION_ASM_OP' or by initializing `text_section' in
29809 `TARGET_ASM_INIT_SECTIONS'. The same is true of `data_section' and
29810 `DATA_SECTION_ASM_OP'. If you do not create a distinct
29811 `readonly_data_section', the default is to reuse `text_section'.
29813 All the other `varasm.c' sections are optional, and are null if the
29814 target does not provide them.
29816 -- Macro: TEXT_SECTION_ASM_OP
29817 A C expression whose value is a string, including spacing,
29818 containing the assembler operation that should precede
29819 instructions and read-only data. Normally `"\t.text"' is right.
29821 -- Macro: HOT_TEXT_SECTION_NAME
29822 If defined, a C string constant for the name of the section
29823 containing most frequently executed functions of the program. If
29824 not defined, GCC will provide a default definition if the target
29825 supports named sections.
29827 -- Macro: UNLIKELY_EXECUTED_TEXT_SECTION_NAME
29828 If defined, a C string constant for the name of the section
29829 containing unlikely executed functions in the program.
29831 -- Macro: DATA_SECTION_ASM_OP
29832 A C expression whose value is a string, including spacing,
29833 containing the assembler operation to identify the following data
29834 as writable initialized data. Normally `"\t.data"' is right.
29836 -- Macro: SDATA_SECTION_ASM_OP
29837 If defined, a C expression whose value is a string, including
29838 spacing, containing the assembler operation to identify the
29839 following data as initialized, writable small data.
29841 -- Macro: READONLY_DATA_SECTION_ASM_OP
29842 A C expression whose value is a string, including spacing,
29843 containing the assembler operation to identify the following data
29844 as read-only initialized data.
29846 -- Macro: BSS_SECTION_ASM_OP
29847 If defined, a C expression whose value is a string, including
29848 spacing, containing the assembler operation to identify the
29849 following data as uninitialized global data. If not defined, and
29850 neither `ASM_OUTPUT_BSS' nor `ASM_OUTPUT_ALIGNED_BSS' are defined,
29851 uninitialized global data will be output in the data section if
29852 `-fno-common' is passed, otherwise `ASM_OUTPUT_COMMON' will be
29855 -- Macro: SBSS_SECTION_ASM_OP
29856 If defined, a C expression whose value is a string, including
29857 spacing, containing the assembler operation to identify the
29858 following data as uninitialized, writable small data.
29860 -- Macro: INIT_SECTION_ASM_OP
29861 If defined, a C expression whose value is a string, including
29862 spacing, containing the assembler operation to identify the
29863 following data as initialization code. If not defined, GCC will
29864 assume such a section does not exist. This section has no
29865 corresponding `init_section' variable; it is used entirely in
29868 -- Macro: FINI_SECTION_ASM_OP
29869 If defined, a C expression whose value is a string, including
29870 spacing, containing the assembler operation to identify the
29871 following data as finalization code. If not defined, GCC will
29872 assume such a section does not exist. This section has no
29873 corresponding `fini_section' variable; it is used entirely in
29876 -- Macro: INIT_ARRAY_SECTION_ASM_OP
29877 If defined, a C expression whose value is a string, including
29878 spacing, containing the assembler operation to identify the
29879 following data as part of the `.init_array' (or equivalent)
29880 section. If not defined, GCC will assume such a section does not
29881 exist. Do not define both this macro and `INIT_SECTION_ASM_OP'.
29883 -- Macro: FINI_ARRAY_SECTION_ASM_OP
29884 If defined, a C expression whose value is a string, including
29885 spacing, containing the assembler operation to identify the
29886 following data as part of the `.fini_array' (or equivalent)
29887 section. If not defined, GCC will assume such a section does not
29888 exist. Do not define both this macro and `FINI_SECTION_ASM_OP'.
29890 -- Macro: CRT_CALL_STATIC_FUNCTION (SECTION_OP, FUNCTION)
29891 If defined, an ASM statement that switches to a different section
29892 via SECTION_OP, calls FUNCTION, and switches back to the text
29893 section. This is used in `crtstuff.c' if `INIT_SECTION_ASM_OP' or
29894 `FINI_SECTION_ASM_OP' to calls to initialization and finalization
29895 functions from the init and fini sections. By default, this macro
29896 uses a simple function call. Some ports need hand-crafted
29897 assembly code to avoid dependencies on registers initialized in
29898 the function prologue or to ensure that constant pools don't end
29899 up too far way in the text section.
29901 -- Macro: TARGET_LIBGCC_SDATA_SECTION
29902 If defined, a string which names the section into which small
29903 variables defined in crtstuff and libgcc should go. This is useful
29904 when the target has options for optimizing access to small data,
29905 and you want the crtstuff and libgcc routines to be conservative
29906 in what they expect of your application yet liberal in what your
29907 application expects. For example, for targets with a `.sdata'
29908 section (like MIPS), you could compile crtstuff with `-G 0' so
29909 that it doesn't require small data support from your application,
29910 but use this macro to put small data into `.sdata' so that your
29911 application can access these variables whether it uses small data
29914 -- Macro: FORCE_CODE_SECTION_ALIGN
29915 If defined, an ASM statement that aligns a code section to some
29916 arbitrary boundary. This is used to force all fragments of the
29917 `.init' and `.fini' sections to have to same alignment and thus
29918 prevent the linker from having to add any padding.
29920 -- Macro: JUMP_TABLES_IN_TEXT_SECTION
29921 Define this macro to be an expression with a nonzero value if jump
29922 tables (for `tablejump' insns) should be output in the text
29923 section, along with the assembler instructions. Otherwise, the
29924 readonly data section is used.
29926 This macro is irrelevant if there is no separate readonly data
29929 -- Target Hook: void TARGET_ASM_INIT_SECTIONS (void)
29930 Define this hook if you need to do something special to set up the
29931 `varasm.c' sections, or if your target has some special sections
29932 of its own that you need to create.
29934 GCC calls this hook after processing the command line, but before
29935 writing any assembly code, and before calling any of the
29936 section-returning hooks described below.
29938 -- Target Hook: TARGET_ASM_RELOC_RW_MASK (void)
29939 Return a mask describing how relocations should be treated when
29940 selecting sections. Bit 1 should be set if global relocations
29941 should be placed in a read-write section; bit 0 should be set if
29942 local relocations should be placed in a read-write section.
29944 The default version of this function returns 3 when `-fpic' is in
29945 effect, and 0 otherwise. The hook is typically redefined when the
29946 target cannot support (some kinds of) dynamic relocations in
29947 read-only sections even in executables.
29949 -- Target Hook: section * TARGET_ASM_SELECT_SECTION (tree EXP, int
29950 RELOC, unsigned HOST_WIDE_INT ALIGN)
29951 Return the section into which EXP should be placed. You can
29952 assume that EXP is either a `VAR_DECL' node or a constant of some
29953 sort. RELOC indicates whether the initial value of EXP requires
29954 link-time relocations. Bit 0 is set when variable contains local
29955 relocations only, while bit 1 is set for global relocations.
29956 ALIGN is the constant alignment in bits.
29958 The default version of this function takes care of putting
29959 read-only variables in `readonly_data_section'.
29961 See also USE_SELECT_SECTION_FOR_FUNCTIONS.
29963 -- Macro: USE_SELECT_SECTION_FOR_FUNCTIONS
29964 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be
29965 called for `FUNCTION_DECL's as well as for variables and constants.
29967 In the case of a `FUNCTION_DECL', RELOC will be zero if the
29968 function has been determined to be likely to be called, and
29969 nonzero if it is unlikely to be called.
29971 -- Target Hook: void TARGET_ASM_UNIQUE_SECTION (tree DECL, int RELOC)
29972 Build up a unique section name, expressed as a `STRING_CST' node,
29973 and assign it to `DECL_SECTION_NAME (DECL)'. As with
29974 `TARGET_ASM_SELECT_SECTION', RELOC indicates whether the initial
29975 value of EXP requires link-time relocations.
29977 The default version of this function appends the symbol name to the
29978 ELF section name that would normally be used for the symbol. For
29979 example, the function `foo' would be placed in `.text.foo'.
29980 Whatever the actual target object format, this is often good
29983 -- Target Hook: section * TARGET_ASM_FUNCTION_RODATA_SECTION (tree
29985 Return the readonly data section associated with
29986 `DECL_SECTION_NAME (DECL)'. The default version of this function
29987 selects `.gnu.linkonce.r.name' if the function's section is
29988 `.gnu.linkonce.t.name', `.rodata.name' if function is in
29989 `.text.name', and the normal readonly-data section otherwise.
29991 -- Target Hook: section * TARGET_ASM_SELECT_RTX_SECTION (enum
29992 machine_mode MODE, rtx X, unsigned HOST_WIDE_INT ALIGN)
29993 Return the section into which a constant X, of mode MODE, should
29994 be placed. You can assume that X is some kind of constant in RTL.
29995 The argument MODE is redundant except in the case of a
29996 `const_int' rtx. ALIGN is the constant alignment in bits.
29998 The default version of this function takes care of putting symbolic
29999 constants in `flag_pic' mode in `data_section' and everything else
30000 in `readonly_data_section'.
30002 -- Target Hook: void TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree DECL,
30004 Define this hook if you need to postprocess the assembler name
30005 generated by target-independent code. The ID provided to this
30006 hook will be the computed name (e.g., the macro `DECL_NAME' of the
30007 DECL in C, or the mangled name of the DECL in C++). The return
30008 value of the hook is an `IDENTIFIER_NODE' for the appropriate
30009 mangled name on your target system. The default implementation of
30010 this hook just returns the ID provided.
30012 -- Target Hook: void TARGET_ENCODE_SECTION_INFO (tree DECL, rtx RTL,
30014 Define this hook if references to a symbol or a constant must be
30015 treated differently depending on something about the variable or
30016 function named by the symbol (such as what section it is in).
30018 The hook is executed immediately after rtl has been created for
30019 DECL, which may be a variable or function declaration or an entry
30020 in the constant pool. In either case, RTL is the rtl in question.
30021 Do _not_ use `DECL_RTL (DECL)' in this hook; that field may not
30022 have been initialized yet.
30024 In the case of a constant, it is safe to assume that the rtl is a
30025 `mem' whose address is a `symbol_ref'. Most decls will also have
30026 this form, but that is not guaranteed. Global register variables,
30027 for instance, will have a `reg' for their rtl. (Normally the
30028 right thing to do with such unusual rtl is leave it alone.)
30030 The NEW_DECL_P argument will be true if this is the first time
30031 that `TARGET_ENCODE_SECTION_INFO' has been invoked on this decl.
30032 It will be false for subsequent invocations, which will happen for
30033 duplicate declarations. Whether or not anything must be done for
30034 the duplicate declaration depends on whether the hook examines
30035 `DECL_ATTRIBUTES'. NEW_DECL_P is always true when the hook is
30036 called for a constant.
30038 The usual thing for this hook to do is to record flags in the
30039 `symbol_ref', using `SYMBOL_REF_FLAG' or `SYMBOL_REF_FLAGS'.
30040 Historically, the name string was modified if it was necessary to
30041 encode more than one bit of information, but this practice is now
30042 discouraged; use `SYMBOL_REF_FLAGS'.
30044 The default definition of this hook, `default_encode_section_info'
30045 in `varasm.c', sets a number of commonly-useful bits in
30046 `SYMBOL_REF_FLAGS'. Check whether the default does what you need
30047 before overriding it.
30049 -- Target Hook: const char *TARGET_STRIP_NAME_ENCODING (const char
30051 Decode NAME and return the real name part, sans the characters
30052 that `TARGET_ENCODE_SECTION_INFO' may have added.
30054 -- Target Hook: bool TARGET_IN_SMALL_DATA_P (tree EXP)
30055 Returns true if EXP should be placed into a "small data" section.
30056 The default version of this hook always returns false.
30058 -- Variable: Target Hook bool TARGET_HAVE_SRODATA_SECTION
30059 Contains the value true if the target places read-only "small
30060 data" into a separate section. The default value is false.
30062 -- Target Hook: bool TARGET_BINDS_LOCAL_P (tree EXP)
30063 Returns true if EXP names an object for which name resolution
30064 rules must resolve to the current "module" (dynamic shared library
30065 or executable image).
30067 The default version of this hook implements the name resolution
30068 rules for ELF, which has a looser model of global name binding
30069 than other currently supported object file formats.
30071 -- Variable: Target Hook bool TARGET_HAVE_TLS
30072 Contains the value true if the target supports thread-local
30073 storage. The default value is false.
30076 File: gccint.info, Node: PIC, Next: Assembler Format, Prev: Sections, Up: Target Macros
30078 17.20 Position Independent Code
30079 ===============================
30081 This section describes macros that help implement generation of position
30082 independent code. Simply defining these macros is not enough to
30083 generate valid PIC; you must also add support to the macros
30084 `GO_IF_LEGITIMATE_ADDRESS' and `PRINT_OPERAND_ADDRESS', as well as
30085 `LEGITIMIZE_ADDRESS'. You must modify the definition of `movsi' to do
30086 something appropriate when the source operand contains a symbolic
30087 address. You may also need to alter the handling of switch statements
30088 so that they use relative addresses.
30090 -- Macro: PIC_OFFSET_TABLE_REGNUM
30091 The register number of the register used to address a table of
30092 static data addresses in memory. In some cases this register is
30093 defined by a processor's "application binary interface" (ABI).
30094 When this macro is defined, RTL is generated for this register
30095 once, as with the stack pointer and frame pointer registers. If
30096 this macro is not defined, it is up to the machine-dependent files
30097 to allocate such a register (if necessary). Note that this
30098 register must be fixed when in use (e.g. when `flag_pic' is true).
30100 -- Macro: PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
30101 Define this macro if the register defined by
30102 `PIC_OFFSET_TABLE_REGNUM' is clobbered by calls. Do not define
30103 this macro if `PIC_OFFSET_TABLE_REGNUM' is not defined.
30105 -- Macro: LEGITIMATE_PIC_OPERAND_P (X)
30106 A C expression that is nonzero if X is a legitimate immediate
30107 operand on the target machine when generating position independent
30108 code. You can assume that X satisfies `CONSTANT_P', so you need
30109 not check this. You can also assume FLAG_PIC is true, so you need
30110 not check it either. You need not define this macro if all
30111 constants (including `SYMBOL_REF') can be immediate operands when
30112 generating position independent code.
30115 File: gccint.info, Node: Assembler Format, Next: Debugging Info, Prev: PIC, Up: Target Macros
30117 17.21 Defining the Output Assembler Language
30118 ============================================
30120 This section describes macros whose principal purpose is to describe how
30121 to write instructions in assembler language--rather than what the
30126 * File Framework:: Structural information for the assembler file.
30127 * Data Output:: Output of constants (numbers, strings, addresses).
30128 * Uninitialized Data:: Output of uninitialized variables.
30129 * Label Output:: Output and generation of labels.
30130 * Initialization:: General principles of initialization
30131 and termination routines.
30132 * Macros for Initialization::
30133 Specific macros that control the handling of
30134 initialization and termination routines.
30135 * Instruction Output:: Output of actual instructions.
30136 * Dispatch Tables:: Output of jump tables.
30137 * Exception Region Output:: Output of exception region code.
30138 * Alignment Output:: Pseudo ops for alignment and skipping data.
30141 File: gccint.info, Node: File Framework, Next: Data Output, Up: Assembler Format
30143 17.21.1 The Overall Framework of an Assembler File
30144 --------------------------------------------------
30146 This describes the overall framework of an assembly file.
30148 -- Target Hook: void TARGET_ASM_FILE_START ()
30149 Output to `asm_out_file' any text which the assembler expects to
30150 find at the beginning of a file. The default behavior is
30151 controlled by two flags, documented below. Unless your target's
30152 assembler is quite unusual, if you override the default, you
30153 should call `default_file_start' at some point in your target
30154 hook. This lets other target files rely on these variables.
30156 -- Target Hook: bool TARGET_ASM_FILE_START_APP_OFF
30157 If this flag is true, the text of the macro `ASM_APP_OFF' will be
30158 printed as the very first line in the assembly file, unless
30159 `-fverbose-asm' is in effect. (If that macro has been defined to
30160 the empty string, this variable has no effect.) With the normal
30161 definition of `ASM_APP_OFF', the effect is to notify the GNU
30162 assembler that it need not bother stripping comments or extra
30163 whitespace from its input. This allows it to work a bit faster.
30165 The default is false. You should not set it to true unless you
30166 have verified that your port does not generate any extra
30167 whitespace or comments that will cause GAS to issue errors in
30170 -- Target Hook: bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
30171 If this flag is true, `output_file_directive' will be called for
30172 the primary source file, immediately after printing `ASM_APP_OFF'
30173 (if that is enabled). Most ELF assemblers expect this to be done.
30174 The default is false.
30176 -- Target Hook: void TARGET_ASM_FILE_END ()
30177 Output to `asm_out_file' any text which the assembler expects to
30178 find at the end of a file. The default is to output nothing.
30180 -- Function: void file_end_indicate_exec_stack ()
30181 Some systems use a common convention, the `.note.GNU-stack'
30182 special section, to indicate whether or not an object file relies
30183 on the stack being executable. If your system uses this
30184 convention, you should define `TARGET_ASM_FILE_END' to this
30185 function. If you need to do other things in that hook, have your
30186 hook function call this function.
30188 -- Macro: ASM_COMMENT_START
30189 A C string constant describing how to begin a comment in the target
30190 assembler language. The compiler assumes that the comment will
30191 end at the end of the line.
30193 -- Macro: ASM_APP_ON
30194 A C string constant for text to be output before each `asm'
30195 statement or group of consecutive ones. Normally this is
30196 `"#APP"', which is a comment that has no effect on most assemblers
30197 but tells the GNU assembler that it must check the lines that
30198 follow for all valid assembler constructs.
30200 -- Macro: ASM_APP_OFF
30201 A C string constant for text to be output after each `asm'
30202 statement or group of consecutive ones. Normally this is
30203 `"#NO_APP"', which tells the GNU assembler to resume making the
30204 time-saving assumptions that are valid for ordinary compiler
30207 -- Macro: ASM_OUTPUT_SOURCE_FILENAME (STREAM, NAME)
30208 A C statement to output COFF information or DWARF debugging
30209 information which indicates that filename NAME is the current
30210 source file to the stdio stream STREAM.
30212 This macro need not be defined if the standard form of output for
30213 the file format in use is appropriate.
30215 -- Macro: OUTPUT_QUOTED_STRING (STREAM, STRING)
30216 A C statement to output the string STRING to the stdio stream
30217 STREAM. If you do not call the function `output_quoted_string' in
30218 your config files, GCC will only call it to output filenames to
30219 the assembler source. So you can use it to canonicalize the format
30220 of the filename using this macro.
30222 -- Macro: ASM_OUTPUT_IDENT (STREAM, STRING)
30223 A C statement to output something to the assembler file to handle a
30224 `#ident' directive containing the text STRING. If this macro is
30225 not defined, nothing is output for a `#ident' directive.
30227 -- Target Hook: void TARGET_ASM_NAMED_SECTION (const char *NAME,
30228 unsigned int FLAGS, unsigned int ALIGN)
30229 Output assembly directives to switch to section NAME. The section
30230 should have attributes as specified by FLAGS, which is a bit mask
30231 of the `SECTION_*' flags defined in `output.h'. If ALIGN is
30232 nonzero, it contains an alignment in bytes to be used for the
30233 section, otherwise some target default should be used. Only
30234 targets that must specify an alignment within the section
30235 directive need pay attention to ALIGN - we will still use
30236 `ASM_OUTPUT_ALIGN'.
30238 -- Target Hook: bool TARGET_HAVE_NAMED_SECTIONS
30239 This flag is true if the target supports
30240 `TARGET_ASM_NAMED_SECTION'.
30242 -- Target Hook: bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
30243 This flag is true if we can create zeroed data by switching to a
30244 BSS section and then using `ASM_OUTPUT_SKIP' to allocate the space.
30245 This is true on most ELF targets.
30247 -- Target Hook: unsigned int TARGET_SECTION_TYPE_FLAGS (tree DECL,
30248 const char *NAME, int RELOC)
30249 Choose a set of section attributes for use by
30250 `TARGET_ASM_NAMED_SECTION' based on a variable or function decl, a
30251 section name, and whether or not the declaration's initializer may
30252 contain runtime relocations. DECL may be null, in which case
30253 read-write data should be assumed.
30255 The default version of this function handles choosing code vs data,
30256 read-only vs read-write data, and `flag_pic'. You should only
30257 need to override this if your target has special flags that might
30258 be set via `__attribute__'.
30260 -- Target Hook: int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type
30261 TYPE, const char * TEXT)
30262 Provides the target with the ability to record the gcc command line
30263 switches that have been passed to the compiler, and options that
30264 are enabled. The TYPE argument specifies what is being recorded.
30265 It can take the following values:
30267 `SWITCH_TYPE_PASSED'
30268 TEXT is a command line switch that has been set by the user.
30270 `SWITCH_TYPE_ENABLED'
30271 TEXT is an option which has been enabled. This might be as a
30272 direct result of a command line switch, or because it is
30273 enabled by default or because it has been enabled as a side
30274 effect of a different command line switch. For example, the
30275 `-O2' switch enables various different individual
30276 optimization passes.
30278 `SWITCH_TYPE_DESCRIPTIVE'
30279 TEXT is either NULL or some descriptive text which should be
30280 ignored. If TEXT is NULL then it is being used to warn the
30281 target hook that either recording is starting or ending. The
30282 first time TYPE is SWITCH_TYPE_DESCRIPTIVE and TEXT is NULL,
30283 the warning is for start up and the second time the warning
30284 is for wind down. This feature is to allow the target hook
30285 to make any necessary preparations before it starts to record
30286 switches and to perform any necessary tidying up after it has
30287 finished recording switches.
30289 `SWITCH_TYPE_LINE_START'
30290 This option can be ignored by this target hook.
30292 `SWITCH_TYPE_LINE_END'
30293 This option can be ignored by this target hook.
30295 The hook's return value must be zero. Other return values may be
30296 supported in the future.
30298 By default this hook is set to NULL, but an example implementation
30299 is provided for ELF based targets. Called ELF_RECORD_GCC_SWITCHES,
30300 it records the switches as ASCII text inside a new, string
30301 mergeable section in the assembler output file. The name of the
30302 new section is provided by the
30303 `TARGET_ASM_RECORD_GCC_SWITCHES_SECTION' target hook.
30305 -- Target Hook: const char * TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
30306 This is the name of the section that will be created by the example
30307 ELF implementation of the `TARGET_ASM_RECORD_GCC_SWITCHES' target
30311 File: gccint.info, Node: Data Output, Next: Uninitialized Data, Prev: File Framework, Up: Assembler Format
30313 17.21.2 Output of Data
30314 ----------------------
30316 -- Target Hook: const char * TARGET_ASM_BYTE_OP
30317 -- Target Hook: const char * TARGET_ASM_ALIGNED_HI_OP
30318 -- Target Hook: const char * TARGET_ASM_ALIGNED_SI_OP
30319 -- Target Hook: const char * TARGET_ASM_ALIGNED_DI_OP
30320 -- Target Hook: const char * TARGET_ASM_ALIGNED_TI_OP
30321 -- Target Hook: const char * TARGET_ASM_UNALIGNED_HI_OP
30322 -- Target Hook: const char * TARGET_ASM_UNALIGNED_SI_OP
30323 -- Target Hook: const char * TARGET_ASM_UNALIGNED_DI_OP
30324 -- Target Hook: const char * TARGET_ASM_UNALIGNED_TI_OP
30325 These hooks specify assembly directives for creating certain kinds
30326 of integer object. The `TARGET_ASM_BYTE_OP' directive creates a
30327 byte-sized object, the `TARGET_ASM_ALIGNED_HI_OP' one creates an
30328 aligned two-byte object, and so on. Any of the hooks may be
30329 `NULL', indicating that no suitable directive is available.
30331 The compiler will print these strings at the start of a new line,
30332 followed immediately by the object's initial value. In most cases,
30333 the string should contain a tab, a pseudo-op, and then another tab.
30335 -- Target Hook: bool TARGET_ASM_INTEGER (rtx X, unsigned int SIZE, int
30337 The `assemble_integer' function uses this hook to output an
30338 integer object. X is the object's value, SIZE is its size in
30339 bytes and ALIGNED_P indicates whether it is aligned. The function
30340 should return `true' if it was able to output the object. If it
30341 returns false, `assemble_integer' will try to split the object
30342 into smaller parts.
30344 The default implementation of this hook will use the
30345 `TARGET_ASM_BYTE_OP' family of strings, returning `false' when the
30346 relevant string is `NULL'.
30348 -- Macro: OUTPUT_ADDR_CONST_EXTRA (STREAM, X, FAIL)
30349 A C statement to recognize RTX patterns that `output_addr_const'
30350 can't deal with, and output assembly code to STREAM corresponding
30351 to the pattern X. This may be used to allow machine-dependent
30352 `UNSPEC's to appear within constants.
30354 If `OUTPUT_ADDR_CONST_EXTRA' fails to recognize a pattern, it must
30355 `goto fail', so that a standard error message is printed. If it
30356 prints an error message itself, by calling, for example,
30357 `output_operand_lossage', it may just complete normally.
30359 -- Macro: ASM_OUTPUT_ASCII (STREAM, PTR, LEN)
30360 A C statement to output to the stdio stream STREAM an assembler
30361 instruction to assemble a string constant containing the LEN bytes
30362 at PTR. PTR will be a C expression of type `char *' and LEN a C
30363 expression of type `int'.
30365 If the assembler has a `.ascii' pseudo-op as found in the Berkeley
30366 Unix assembler, do not define the macro `ASM_OUTPUT_ASCII'.
30368 -- Macro: ASM_OUTPUT_FDESC (STREAM, DECL, N)
30369 A C statement to output word N of a function descriptor for DECL.
30370 This must be defined if `TARGET_VTABLE_USES_DESCRIPTORS' is
30371 defined, and is otherwise unused.
30373 -- Macro: CONSTANT_POOL_BEFORE_FUNCTION
30374 You may define this macro as a C expression. You should define the
30375 expression to have a nonzero value if GCC should output the
30376 constant pool for a function before the code for the function, or
30377 a zero value if GCC should output the constant pool after the
30378 function. If you do not define this macro, the usual case, GCC
30379 will output the constant pool before the function.
30381 -- Macro: ASM_OUTPUT_POOL_PROLOGUE (FILE, FUNNAME, FUNDECL, SIZE)
30382 A C statement to output assembler commands to define the start of
30383 the constant pool for a function. FUNNAME is a string giving the
30384 name of the function. Should the return type of the function be
30385 required, it can be obtained via FUNDECL. SIZE is the size, in
30386 bytes, of the constant pool that will be written immediately after
30389 If no constant-pool prefix is required, the usual case, this macro
30390 need not be defined.
30392 -- Macro: ASM_OUTPUT_SPECIAL_POOL_ENTRY (FILE, X, MODE, ALIGN,
30394 A C statement (with or without semicolon) to output a constant in
30395 the constant pool, if it needs special treatment. (This macro
30396 need not do anything for RTL expressions that can be output
30399 The argument FILE is the standard I/O stream to output the
30400 assembler code on. X is the RTL expression for the constant to
30401 output, and MODE is the machine mode (in case X is a `const_int').
30402 ALIGN is the required alignment for the value X; you should
30403 output an assembler directive to force this much alignment.
30405 The argument LABELNO is a number to use in an internal label for
30406 the address of this pool entry. The definition of this macro is
30407 responsible for outputting the label definition at the proper
30408 place. Here is how to do this:
30410 `(*targetm.asm_out.internal_label)' (FILE, "LC", LABELNO);
30412 When you output a pool entry specially, you should end with a
30413 `goto' to the label JUMPTO. This will prevent the same pool entry
30414 from being output a second time in the usual manner.
30416 You need not define this macro if it would do nothing.
30418 -- Macro: ASM_OUTPUT_POOL_EPILOGUE (FILE FUNNAME FUNDECL SIZE)
30419 A C statement to output assembler commands to at the end of the
30420 constant pool for a function. FUNNAME is a string giving the name
30421 of the function. Should the return type of the function be
30422 required, you can obtain it via FUNDECL. SIZE is the size, in
30423 bytes, of the constant pool that GCC wrote immediately before this
30426 If no constant-pool epilogue is required, the usual case, you need
30427 not define this macro.
30429 -- Macro: IS_ASM_LOGICAL_LINE_SEPARATOR (C, STR)
30430 Define this macro as a C expression which is nonzero if C is used
30431 as a logical line separator by the assembler. STR points to the
30432 position in the string where C was found; this can be used if a
30433 line separator uses multiple characters.
30435 If you do not define this macro, the default is that only the
30436 character `;' is treated as a logical line separator.
30438 -- Target Hook: const char * TARGET_ASM_OPEN_PAREN
30439 -- Target Hook: const char * TARGET_ASM_CLOSE_PAREN
30440 These target hooks are C string constants, describing the syntax
30441 in the assembler for grouping arithmetic expressions. If not
30442 overridden, they default to normal parentheses, which is correct
30443 for most assemblers.
30445 These macros are provided by `real.h' for writing the definitions of
30446 `ASM_OUTPUT_DOUBLE' and the like:
30448 -- Macro: REAL_VALUE_TO_TARGET_SINGLE (X, L)
30449 -- Macro: REAL_VALUE_TO_TARGET_DOUBLE (X, L)
30450 -- Macro: REAL_VALUE_TO_TARGET_LONG_DOUBLE (X, L)
30451 -- Macro: REAL_VALUE_TO_TARGET_DECIMAL32 (X, L)
30452 -- Macro: REAL_VALUE_TO_TARGET_DECIMAL64 (X, L)
30453 -- Macro: REAL_VALUE_TO_TARGET_DECIMAL128 (X, L)
30454 These translate X, of type `REAL_VALUE_TYPE', to the target's
30455 floating point representation, and store its bit pattern in the
30456 variable L. For `REAL_VALUE_TO_TARGET_SINGLE' and
30457 `REAL_VALUE_TO_TARGET_DECIMAL32', this variable should be a simple
30458 `long int'. For the others, it should be an array of `long int'.
30459 The number of elements in this array is determined by the size of
30460 the desired target floating point data type: 32 bits of it go in
30461 each `long int' array element. Each array element holds 32 bits
30462 of the result, even if `long int' is wider than 32 bits on the
30465 The array element values are designed so that you can print them
30466 out using `fprintf' in the order they should appear in the target
30470 File: gccint.info, Node: Uninitialized Data, Next: Label Output, Prev: Data Output, Up: Assembler Format
30472 17.21.3 Output of Uninitialized Variables
30473 -----------------------------------------
30475 Each of the macros in this section is used to do the whole job of
30476 outputting a single uninitialized variable.
30478 -- Macro: ASM_OUTPUT_COMMON (STREAM, NAME, SIZE, ROUNDED)
30479 A C statement (sans semicolon) to output to the stdio stream
30480 STREAM the assembler definition of a common-label named NAME whose
30481 size is SIZE bytes. The variable ROUNDED is the size rounded up
30482 to whatever alignment the caller wants.
30484 Use the expression `assemble_name (STREAM, NAME)' to output the
30485 name itself; before and after that, output the additional
30486 assembler syntax for defining the name, and a newline.
30488 This macro controls how the assembler definitions of uninitialized
30489 common global variables are output.
30491 -- Macro: ASM_OUTPUT_ALIGNED_COMMON (STREAM, NAME, SIZE, ALIGNMENT)
30492 Like `ASM_OUTPUT_COMMON' except takes the required alignment as a
30493 separate, explicit argument. If you define this macro, it is used
30494 in place of `ASM_OUTPUT_COMMON', and gives you more flexibility in
30495 handling the required alignment of the variable. The alignment is
30496 specified as the number of bits.
30498 -- Macro: ASM_OUTPUT_ALIGNED_DECL_COMMON (STREAM, DECL, NAME, SIZE,
30500 Like `ASM_OUTPUT_ALIGNED_COMMON' except that DECL of the variable
30501 to be output, if there is one, or `NULL_TREE' if there is no
30502 corresponding variable. If you define this macro, GCC will use it
30503 in place of both `ASM_OUTPUT_COMMON' and
30504 `ASM_OUTPUT_ALIGNED_COMMON'. Define this macro when you need to
30505 see the variable's decl in order to chose what to output.
30507 -- Macro: ASM_OUTPUT_BSS (STREAM, DECL, NAME, SIZE, ROUNDED)
30508 A C statement (sans semicolon) to output to the stdio stream
30509 STREAM the assembler definition of uninitialized global DECL named
30510 NAME whose size is SIZE bytes. The variable ROUNDED is the size
30511 rounded up to whatever alignment the caller wants.
30513 Try to use function `asm_output_bss' defined in `varasm.c' when
30514 defining this macro. If unable, use the expression `assemble_name
30515 (STREAM, NAME)' to output the name itself; before and after that,
30516 output the additional assembler syntax for defining the name, and
30519 There are two ways of handling global BSS. One is to define either
30520 this macro or its aligned counterpart, `ASM_OUTPUT_ALIGNED_BSS'.
30521 The other is to have `TARGET_ASM_SELECT_SECTION' return a
30522 switchable BSS section (*note
30523 TARGET_HAVE_SWITCHABLE_BSS_SECTIONS::). You do not need to do
30526 Some languages do not have `common' data, and require a non-common
30527 form of global BSS in order to handle uninitialized globals
30528 efficiently. C++ is one example of this. However, if the target
30529 does not support global BSS, the front end may choose to make
30530 globals common in order to save space in the object file.
30532 -- Macro: ASM_OUTPUT_ALIGNED_BSS (STREAM, DECL, NAME, SIZE, ALIGNMENT)
30533 Like `ASM_OUTPUT_BSS' except takes the required alignment as a
30534 separate, explicit argument. If you define this macro, it is used
30535 in place of `ASM_OUTPUT_BSS', and gives you more flexibility in
30536 handling the required alignment of the variable. The alignment is
30537 specified as the number of bits.
30539 Try to use function `asm_output_aligned_bss' defined in file
30540 `varasm.c' when defining this macro.
30542 -- Macro: ASM_OUTPUT_LOCAL (STREAM, NAME, SIZE, ROUNDED)
30543 A C statement (sans semicolon) to output to the stdio stream
30544 STREAM the assembler definition of a local-common-label named NAME
30545 whose size is SIZE bytes. The variable ROUNDED is the size
30546 rounded up to whatever alignment the caller wants.
30548 Use the expression `assemble_name (STREAM, NAME)' to output the
30549 name itself; before and after that, output the additional
30550 assembler syntax for defining the name, and a newline.
30552 This macro controls how the assembler definitions of uninitialized
30553 static variables are output.
30555 -- Macro: ASM_OUTPUT_ALIGNED_LOCAL (STREAM, NAME, SIZE, ALIGNMENT)
30556 Like `ASM_OUTPUT_LOCAL' except takes the required alignment as a
30557 separate, explicit argument. If you define this macro, it is used
30558 in place of `ASM_OUTPUT_LOCAL', and gives you more flexibility in
30559 handling the required alignment of the variable. The alignment is
30560 specified as the number of bits.
30562 -- Macro: ASM_OUTPUT_ALIGNED_DECL_LOCAL (STREAM, DECL, NAME, SIZE,
30564 Like `ASM_OUTPUT_ALIGNED_DECL' except that DECL of the variable to
30565 be output, if there is one, or `NULL_TREE' if there is no
30566 corresponding variable. If you define this macro, GCC will use it
30567 in place of both `ASM_OUTPUT_DECL' and `ASM_OUTPUT_ALIGNED_DECL'.
30568 Define this macro when you need to see the variable's decl in
30569 order to chose what to output.
30572 File: gccint.info, Node: Label Output, Next: Initialization, Prev: Uninitialized Data, Up: Assembler Format
30574 17.21.4 Output and Generation of Labels
30575 ---------------------------------------
30577 This is about outputting labels.
30579 -- Macro: ASM_OUTPUT_LABEL (STREAM, NAME)
30580 A C statement (sans semicolon) to output to the stdio stream
30581 STREAM the assembler definition of a label named NAME. Use the
30582 expression `assemble_name (STREAM, NAME)' to output the name
30583 itself; before and after that, output the additional assembler
30584 syntax for defining the name, and a newline. A default definition
30585 of this macro is provided which is correct for most systems.
30587 -- Macro: ASM_OUTPUT_INTERNAL_LABEL (STREAM, NAME)
30588 Identical to `ASM_OUTPUT_LABEL', except that NAME is known to
30589 refer to a compiler-generated label. The default definition uses
30590 `assemble_name_raw', which is like `assemble_name' except that it
30593 -- Macro: SIZE_ASM_OP
30594 A C string containing the appropriate assembler directive to
30595 specify the size of a symbol, without any arguments. On systems
30596 that use ELF, the default (in `config/elfos.h') is `"\t.size\t"';
30597 on other systems, the default is not to define this macro.
30599 Define this macro only if it is correct to use the default
30600 definitions of `ASM_OUTPUT_SIZE_DIRECTIVE' and
30601 `ASM_OUTPUT_MEASURED_SIZE' for your system. If you need your own
30602 custom definitions of those macros, or if you do not need explicit
30603 symbol sizes at all, do not define this macro.
30605 -- Macro: ASM_OUTPUT_SIZE_DIRECTIVE (STREAM, NAME, SIZE)
30606 A C statement (sans semicolon) to output to the stdio stream
30607 STREAM a directive telling the assembler that the size of the
30608 symbol NAME is SIZE. SIZE is a `HOST_WIDE_INT'. If you define
30609 `SIZE_ASM_OP', a default definition of this macro is provided.
30611 -- Macro: ASM_OUTPUT_MEASURED_SIZE (STREAM, NAME)
30612 A C statement (sans semicolon) to output to the stdio stream
30613 STREAM a directive telling the assembler to calculate the size of
30614 the symbol NAME by subtracting its address from the current
30617 If you define `SIZE_ASM_OP', a default definition of this macro is
30618 provided. The default assumes that the assembler recognizes a
30619 special `.' symbol as referring to the current address, and can
30620 calculate the difference between this and another symbol. If your
30621 assembler does not recognize `.' or cannot do calculations with
30622 it, you will need to redefine `ASM_OUTPUT_MEASURED_SIZE' to use
30623 some other technique.
30625 -- Macro: TYPE_ASM_OP
30626 A C string containing the appropriate assembler directive to
30627 specify the type of a symbol, without any arguments. On systems
30628 that use ELF, the default (in `config/elfos.h') is `"\t.type\t"';
30629 on other systems, the default is not to define this macro.
30631 Define this macro only if it is correct to use the default
30632 definition of `ASM_OUTPUT_TYPE_DIRECTIVE' for your system. If you
30633 need your own custom definition of this macro, or if you do not
30634 need explicit symbol types at all, do not define this macro.
30636 -- Macro: TYPE_OPERAND_FMT
30637 A C string which specifies (using `printf' syntax) the format of
30638 the second operand to `TYPE_ASM_OP'. On systems that use ELF, the
30639 default (in `config/elfos.h') is `"@%s"'; on other systems, the
30640 default is not to define this macro.
30642 Define this macro only if it is correct to use the default
30643 definition of `ASM_OUTPUT_TYPE_DIRECTIVE' for your system. If you
30644 need your own custom definition of this macro, or if you do not
30645 need explicit symbol types at all, do not define this macro.
30647 -- Macro: ASM_OUTPUT_TYPE_DIRECTIVE (STREAM, TYPE)
30648 A C statement (sans semicolon) to output to the stdio stream
30649 STREAM a directive telling the assembler that the type of the
30650 symbol NAME is TYPE. TYPE is a C string; currently, that string
30651 is always either `"function"' or `"object"', but you should not
30654 If you define `TYPE_ASM_OP' and `TYPE_OPERAND_FMT', a default
30655 definition of this macro is provided.
30657 -- Macro: ASM_DECLARE_FUNCTION_NAME (STREAM, NAME, DECL)
30658 A C statement (sans semicolon) to output to the stdio stream
30659 STREAM any text necessary for declaring the name NAME of a
30660 function which is being defined. This macro is responsible for
30661 outputting the label definition (perhaps using
30662 `ASM_OUTPUT_LABEL'). The argument DECL is the `FUNCTION_DECL'
30663 tree node representing the function.
30665 If this macro is not defined, then the function name is defined in
30666 the usual manner as a label (by means of `ASM_OUTPUT_LABEL').
30668 You may wish to use `ASM_OUTPUT_TYPE_DIRECTIVE' in the definition
30671 -- Macro: ASM_DECLARE_FUNCTION_SIZE (STREAM, NAME, DECL)
30672 A C statement (sans semicolon) to output to the stdio stream
30673 STREAM any text necessary for declaring the size of a function
30674 which is being defined. The argument NAME is the name of the
30675 function. The argument DECL is the `FUNCTION_DECL' tree node
30676 representing the function.
30678 If this macro is not defined, then the function size is not
30681 You may wish to use `ASM_OUTPUT_MEASURED_SIZE' in the definition
30684 -- Macro: ASM_DECLARE_OBJECT_NAME (STREAM, NAME, DECL)
30685 A C statement (sans semicolon) to output to the stdio stream
30686 STREAM any text necessary for declaring the name NAME of an
30687 initialized variable which is being defined. This macro must
30688 output the label definition (perhaps using `ASM_OUTPUT_LABEL').
30689 The argument DECL is the `VAR_DECL' tree node representing the
30692 If this macro is not defined, then the variable name is defined in
30693 the usual manner as a label (by means of `ASM_OUTPUT_LABEL').
30695 You may wish to use `ASM_OUTPUT_TYPE_DIRECTIVE' and/or
30696 `ASM_OUTPUT_SIZE_DIRECTIVE' in the definition of this macro.
30698 -- Macro: ASM_DECLARE_CONSTANT_NAME (STREAM, NAME, EXP, SIZE)
30699 A C statement (sans semicolon) to output to the stdio stream
30700 STREAM any text necessary for declaring the name NAME of a
30701 constant which is being defined. This macro is responsible for
30702 outputting the label definition (perhaps using
30703 `ASM_OUTPUT_LABEL'). The argument EXP is the value of the
30704 constant, and SIZE is the size of the constant in bytes. NAME
30705 will be an internal label.
30707 If this macro is not defined, then the NAME is defined in the
30708 usual manner as a label (by means of `ASM_OUTPUT_LABEL').
30710 You may wish to use `ASM_OUTPUT_TYPE_DIRECTIVE' in the definition
30713 -- Macro: ASM_DECLARE_REGISTER_GLOBAL (STREAM, DECL, REGNO, NAME)
30714 A C statement (sans semicolon) to output to the stdio stream
30715 STREAM any text necessary for claiming a register REGNO for a
30716 global variable DECL with name NAME.
30718 If you don't define this macro, that is equivalent to defining it
30721 -- Macro: ASM_FINISH_DECLARE_OBJECT (STREAM, DECL, TOPLEVEL, ATEND)
30722 A C statement (sans semicolon) to finish up declaring a variable
30723 name once the compiler has processed its initializer fully and
30724 thus has had a chance to determine the size of an array when
30725 controlled by an initializer. This is used on systems where it's
30726 necessary to declare something about the size of the object.
30728 If you don't define this macro, that is equivalent to defining it
30731 You may wish to use `ASM_OUTPUT_SIZE_DIRECTIVE' and/or
30732 `ASM_OUTPUT_MEASURED_SIZE' in the definition of this macro.
30734 -- Target Hook: void TARGET_ASM_GLOBALIZE_LABEL (FILE *STREAM, const
30736 This target hook is a function to output to the stdio stream
30737 STREAM some commands that will make the label NAME global; that
30738 is, available for reference from other files.
30740 The default implementation relies on a proper definition of
30743 -- Target Hook: void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *STREAM,
30745 This target hook is a function to output to the stdio stream
30746 STREAM some commands that will make the name associated with DECL
30747 global; that is, available for reference from other files.
30749 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL
30752 -- Macro: ASM_WEAKEN_LABEL (STREAM, NAME)
30753 A C statement (sans semicolon) to output to the stdio stream
30754 STREAM some commands that will make the label NAME weak; that is,
30755 available for reference from other files but only used if no other
30756 definition is available. Use the expression `assemble_name
30757 (STREAM, NAME)' to output the name itself; before and after that,
30758 output the additional assembler syntax for making that name weak,
30761 If you don't define this macro or `ASM_WEAKEN_DECL', GCC will not
30762 support weak symbols and you should not define the `SUPPORTS_WEAK'
30765 -- Macro: ASM_WEAKEN_DECL (STREAM, DECL, NAME, VALUE)
30766 Combines (and replaces) the function of `ASM_WEAKEN_LABEL' and
30767 `ASM_OUTPUT_WEAK_ALIAS', allowing access to the associated function
30768 or variable decl. If VALUE is not `NULL', this C statement should
30769 output to the stdio stream STREAM assembler code which defines
30770 (equates) the weak symbol NAME to have the value VALUE. If VALUE
30771 is `NULL', it should output commands to make NAME weak.
30773 -- Macro: ASM_OUTPUT_WEAKREF (STREAM, DECL, NAME, VALUE)
30774 Outputs a directive that enables NAME to be used to refer to
30775 symbol VALUE with weak-symbol semantics. `decl' is the
30776 declaration of `name'.
30778 -- Macro: SUPPORTS_WEAK
30779 A C expression which evaluates to true if the target supports weak
30782 If you don't define this macro, `defaults.h' provides a default
30783 definition. If either `ASM_WEAKEN_LABEL' or `ASM_WEAKEN_DECL' is
30784 defined, the default definition is `1'; otherwise, it is `0'.
30785 Define this macro if you want to control weak symbol support with
30786 a compiler flag such as `-melf'.
30788 -- Macro: MAKE_DECL_ONE_ONLY (DECL)
30789 A C statement (sans semicolon) to mark DECL to be emitted as a
30790 public symbol such that extra copies in multiple translation units
30791 will be discarded by the linker. Define this macro if your object
30792 file format provides support for this concept, such as the `COMDAT'
30793 section flags in the Microsoft Windows PE/COFF format, and this
30794 support requires changes to DECL, such as putting it in a separate
30797 -- Macro: SUPPORTS_ONE_ONLY
30798 A C expression which evaluates to true if the target supports
30799 one-only semantics.
30801 If you don't define this macro, `varasm.c' provides a default
30802 definition. If `MAKE_DECL_ONE_ONLY' is defined, the default
30803 definition is `1'; otherwise, it is `0'. Define this macro if you
30804 want to control one-only symbol support with a compiler flag, or if
30805 setting the `DECL_ONE_ONLY' flag is enough to mark a declaration to
30806 be emitted as one-only.
30808 -- Target Hook: void TARGET_ASM_ASSEMBLE_VISIBILITY (tree DECL, const
30810 This target hook is a function to output to ASM_OUT_FILE some
30811 commands that will make the symbol(s) associated with DECL have
30812 hidden, protected or internal visibility as specified by
30815 -- Macro: TARGET_WEAK_NOT_IN_ARCHIVE_TOC
30816 A C expression that evaluates to true if the target's linker
30817 expects that weak symbols do not appear in a static archive's
30818 table of contents. The default is `0'.
30820 Leaving weak symbols out of an archive's table of contents means
30821 that, if a symbol will only have a definition in one translation
30822 unit and will have undefined references from other translation
30823 units, that symbol should not be weak. Defining this macro to be
30824 nonzero will thus have the effect that certain symbols that would
30825 normally be weak (explicit template instantiations, and vtables
30826 for polymorphic classes with noninline key methods) will instead
30829 The C++ ABI requires this macro to be zero. Define this macro for
30830 targets where full C++ ABI compliance is impossible and where
30831 linker restrictions require weak symbols to be left out of a
30832 static archive's table of contents.
30834 -- Macro: ASM_OUTPUT_EXTERNAL (STREAM, DECL, NAME)
30835 A C statement (sans semicolon) to output to the stdio stream
30836 STREAM any text necessary for declaring the name of an external
30837 symbol named NAME which is referenced in this compilation but not
30838 defined. The value of DECL is the tree node for the declaration.
30840 This macro need not be defined if it does not need to output
30841 anything. The GNU assembler and most Unix assemblers don't
30844 -- Target Hook: void TARGET_ASM_EXTERNAL_LIBCALL (rtx SYMREF)
30845 This target hook is a function to output to ASM_OUT_FILE an
30846 assembler pseudo-op to declare a library function name external.
30847 The name of the library function is given by SYMREF, which is a
30850 -- Target Hook: void TARGET_ASM_MARK_DECL_PRESERVED (tree DECL)
30851 This target hook is a function to output to ASM_OUT_FILE an
30852 assembler directive to annotate used symbol. Darwin target use
30853 .no_dead_code_strip directive.
30855 -- Macro: ASM_OUTPUT_LABELREF (STREAM, NAME)
30856 A C statement (sans semicolon) to output to the stdio stream
30857 STREAM a reference in assembler syntax to a label named NAME.
30858 This should add `_' to the front of the name, if that is customary
30859 on your operating system, as it is in most Berkeley Unix systems.
30860 This macro is used in `assemble_name'.
30862 -- Macro: ASM_OUTPUT_SYMBOL_REF (STREAM, SYM)
30863 A C statement (sans semicolon) to output a reference to
30864 `SYMBOL_REF' SYM. If not defined, `assemble_name' will be used to
30865 output the name of the symbol. This macro may be used to modify
30866 the way a symbol is referenced depending on information encoded by
30867 `TARGET_ENCODE_SECTION_INFO'.
30869 -- Macro: ASM_OUTPUT_LABEL_REF (STREAM, BUF)
30870 A C statement (sans semicolon) to output a reference to BUF, the
30871 result of `ASM_GENERATE_INTERNAL_LABEL'. If not defined,
30872 `assemble_name' will be used to output the name of the symbol.
30873 This macro is not used by `output_asm_label', or the `%l'
30874 specifier that calls it; the intention is that this macro should
30875 be set when it is necessary to output a label differently when its
30876 address is being taken.
30878 -- Target Hook: void TARGET_ASM_INTERNAL_LABEL (FILE *STREAM, const
30879 char *PREFIX, unsigned long LABELNO)
30880 A function to output to the stdio stream STREAM a label whose name
30881 is made from the string PREFIX and the number LABELNO.
30883 It is absolutely essential that these labels be distinct from the
30884 labels used for user-level functions and variables. Otherwise,
30885 certain programs will have name conflicts with internal labels.
30887 It is desirable to exclude internal labels from the symbol table
30888 of the object file. Most assemblers have a naming convention for
30889 labels that should be excluded; on many systems, the letter `L' at
30890 the beginning of a label has this effect. You should find out what
30891 convention your system uses, and follow it.
30893 The default version of this function utilizes
30894 `ASM_GENERATE_INTERNAL_LABEL'.
30896 -- Macro: ASM_OUTPUT_DEBUG_LABEL (STREAM, PREFIX, NUM)
30897 A C statement to output to the stdio stream STREAM a debug info
30898 label whose name is made from the string PREFIX and the number
30899 NUM. This is useful for VLIW targets, where debug info labels may
30900 need to be treated differently than branch target labels. On some
30901 systems, branch target labels must be at the beginning of
30902 instruction bundles, but debug info labels can occur in the middle
30903 of instruction bundles.
30905 If this macro is not defined, then
30906 `(*targetm.asm_out.internal_label)' will be used.
30908 -- Macro: ASM_GENERATE_INTERNAL_LABEL (STRING, PREFIX, NUM)
30909 A C statement to store into the string STRING a label whose name
30910 is made from the string PREFIX and the number NUM.
30912 This string, when output subsequently by `assemble_name', should
30913 produce the output that `(*targetm.asm_out.internal_label)' would
30914 produce with the same PREFIX and NUM.
30916 If the string begins with `*', then `assemble_name' will output
30917 the rest of the string unchanged. It is often convenient for
30918 `ASM_GENERATE_INTERNAL_LABEL' to use `*' in this way. If the
30919 string doesn't start with `*', then `ASM_OUTPUT_LABELREF' gets to
30920 output the string, and may change it. (Of course,
30921 `ASM_OUTPUT_LABELREF' is also part of your machine description, so
30922 you should know what it does on your machine.)
30924 -- Macro: ASM_FORMAT_PRIVATE_NAME (OUTVAR, NAME, NUMBER)
30925 A C expression to assign to OUTVAR (which is a variable of type
30926 `char *') a newly allocated string made from the string NAME and
30927 the number NUMBER, with some suitable punctuation added. Use
30928 `alloca' to get space for the string.
30930 The string will be used as an argument to `ASM_OUTPUT_LABELREF' to
30931 produce an assembler label for an internal static variable whose
30932 name is NAME. Therefore, the string must be such as to result in
30933 valid assembler code. The argument NUMBER is different each time
30934 this macro is executed; it prevents conflicts between
30935 similarly-named internal static variables in different scopes.
30937 Ideally this string should not be a valid C identifier, to prevent
30938 any conflict with the user's own symbols. Most assemblers allow
30939 periods or percent signs in assembler symbols; putting at least
30940 one of these between the name and the number will suffice.
30942 If this macro is not defined, a default definition will be provided
30943 which is correct for most systems.
30945 -- Macro: ASM_OUTPUT_DEF (STREAM, NAME, VALUE)
30946 A C statement to output to the stdio stream STREAM assembler code
30947 which defines (equates) the symbol NAME to have the value VALUE.
30949 If `SET_ASM_OP' is defined, a default definition is provided which
30950 is correct for most systems.
30952 -- Macro: ASM_OUTPUT_DEF_FROM_DECLS (STREAM, DECL_OF_NAME,
30954 A C statement to output to the stdio stream STREAM assembler code
30955 which defines (equates) the symbol whose tree node is DECL_OF_NAME
30956 to have the value of the tree node DECL_OF_VALUE. This macro will
30957 be used in preference to `ASM_OUTPUT_DEF' if it is defined and if
30958 the tree nodes are available.
30960 If `SET_ASM_OP' is defined, a default definition is provided which
30961 is correct for most systems.
30963 -- Macro: TARGET_DEFERRED_OUTPUT_DEFS (DECL_OF_NAME, DECL_OF_VALUE)
30964 A C statement that evaluates to true if the assembler code which
30965 defines (equates) the symbol whose tree node is DECL_OF_NAME to
30966 have the value of the tree node DECL_OF_VALUE should be emitted
30967 near the end of the current compilation unit. The default is to
30968 not defer output of defines. This macro affects defines output by
30969 `ASM_OUTPUT_DEF' and `ASM_OUTPUT_DEF_FROM_DECLS'.
30971 -- Macro: ASM_OUTPUT_WEAK_ALIAS (STREAM, NAME, VALUE)
30972 A C statement to output to the stdio stream STREAM assembler code
30973 which defines (equates) the weak symbol NAME to have the value
30974 VALUE. If VALUE is `NULL', it defines NAME as an undefined weak
30977 Define this macro if the target only supports weak aliases; define
30978 `ASM_OUTPUT_DEF' instead if possible.
30980 -- Macro: OBJC_GEN_METHOD_LABEL (BUF, IS_INST, CLASS_NAME, CAT_NAME,
30982 Define this macro to override the default assembler names used for
30983 Objective-C methods.
30985 The default name is a unique method number followed by the name of
30986 the class (e.g. `_1_Foo'). For methods in categories, the name of
30987 the category is also included in the assembler name (e.g.
30990 These names are safe on most systems, but make debugging difficult
30991 since the method's selector is not present in the name.
30992 Therefore, particular systems define other ways of computing names.
30994 BUF is an expression of type `char *' which gives you a buffer in
30995 which to store the name; its length is as long as CLASS_NAME,
30996 CAT_NAME and SEL_NAME put together, plus 50 characters extra.
30998 The argument IS_INST specifies whether the method is an instance
30999 method or a class method; CLASS_NAME is the name of the class;
31000 CAT_NAME is the name of the category (or `NULL' if the method is
31001 not in a category); and SEL_NAME is the name of the selector.
31003 On systems where the assembler can handle quoted names, you can
31004 use this macro to provide more human-readable names.
31006 -- Macro: ASM_DECLARE_CLASS_REFERENCE (STREAM, NAME)
31007 A C statement (sans semicolon) to output to the stdio stream
31008 STREAM commands to declare that the label NAME is an Objective-C
31009 class reference. This is only needed for targets whose linkers
31010 have special support for NeXT-style runtimes.
31012 -- Macro: ASM_DECLARE_UNRESOLVED_REFERENCE (STREAM, NAME)
31013 A C statement (sans semicolon) to output to the stdio stream
31014 STREAM commands to declare that the label NAME is an unresolved
31015 Objective-C class reference. This is only needed for targets
31016 whose linkers have special support for NeXT-style runtimes.
31019 File: gccint.info, Node: Initialization, Next: Macros for Initialization, Prev: Label Output, Up: Assembler Format
31021 17.21.5 How Initialization Functions Are Handled
31022 ------------------------------------------------
31024 The compiled code for certain languages includes "constructors" (also
31025 called "initialization routines")--functions to initialize data in the
31026 program when the program is started. These functions need to be called
31027 before the program is "started"--that is to say, before `main' is
31030 Compiling some languages generates "destructors" (also called
31031 "termination routines") that should be called when the program
31034 To make the initialization and termination functions work, the compiler
31035 must output something in the assembler code to cause those functions to
31036 be called at the appropriate time. When you port the compiler to a new
31037 system, you need to specify how to do this.
31039 There are two major ways that GCC currently supports the execution of
31040 initialization and termination functions. Each way has two variants.
31041 Much of the structure is common to all four variations.
31043 The linker must build two lists of these functions--a list of
31044 initialization functions, called `__CTOR_LIST__', and a list of
31045 termination functions, called `__DTOR_LIST__'.
31047 Each list always begins with an ignored function pointer (which may
31048 hold 0, -1, or a count of the function pointers after it, depending on
31049 the environment). This is followed by a series of zero or more function
31050 pointers to constructors (or destructors), followed by a function
31051 pointer containing zero.
31053 Depending on the operating system and its executable file format,
31054 either `crtstuff.c' or `libgcc2.c' traverses these lists at startup
31055 time and exit time. Constructors are called in reverse order of the
31056 list; destructors in forward order.
31058 The best way to handle static constructors works only for object file
31059 formats which provide arbitrarily-named sections. A section is set
31060 aside for a list of constructors, and another for a list of destructors.
31061 Traditionally these are called `.ctors' and `.dtors'. Each object file
31062 that defines an initialization function also puts a word in the
31063 constructor section to point to that function. The linker accumulates
31064 all these words into one contiguous `.ctors' section. Termination
31065 functions are handled similarly.
31067 This method will be chosen as the default by `target-def.h' if
31068 `TARGET_ASM_NAMED_SECTION' is defined. A target that does not support
31069 arbitrary sections, but does support special designated constructor and
31070 destructor sections may define `CTORS_SECTION_ASM_OP' and
31071 `DTORS_SECTION_ASM_OP' to achieve the same effect.
31073 When arbitrary sections are available, there are two variants,
31074 depending upon how the code in `crtstuff.c' is called. On systems that
31075 support a ".init" section which is executed at program startup, parts
31076 of `crtstuff.c' are compiled into that section. The program is linked
31077 by the `gcc' driver like this:
31079 ld -o OUTPUT_FILE crti.o crtbegin.o ... -lgcc crtend.o crtn.o
31081 The prologue of a function (`__init') appears in the `.init' section
31082 of `crti.o'; the epilogue appears in `crtn.o'. Likewise for the
31083 function `__fini' in the ".fini" section. Normally these files are
31084 provided by the operating system or by the GNU C library, but are
31085 provided by GCC for a few targets.
31087 The objects `crtbegin.o' and `crtend.o' are (for most targets)
31088 compiled from `crtstuff.c'. They contain, among other things, code
31089 fragments within the `.init' and `.fini' sections that branch to
31090 routines in the `.text' section. The linker will pull all parts of a
31091 section together, which results in a complete `__init' function that
31092 invokes the routines we need at startup.
31094 To use this variant, you must define the `INIT_SECTION_ASM_OP' macro
31097 If no init section is available, when GCC compiles any function called
31098 `main' (or more accurately, any function designated as a program entry
31099 point by the language front end calling `expand_main_function'), it
31100 inserts a procedure call to `__main' as the first executable code after
31101 the function prologue. The `__main' function is defined in `libgcc2.c'
31102 and runs the global constructors.
31104 In file formats that don't support arbitrary sections, there are again
31105 two variants. In the simplest variant, the GNU linker (GNU `ld') and
31106 an `a.out' format must be used. In this case, `TARGET_ASM_CONSTRUCTOR'
31107 is defined to produce a `.stabs' entry of type `N_SETT', referencing
31108 the name `__CTOR_LIST__', and with the address of the void function
31109 containing the initialization code as its value. The GNU linker
31110 recognizes this as a request to add the value to a "set"; the values
31111 are accumulated, and are eventually placed in the executable as a
31112 vector in the format described above, with a leading (ignored) count
31113 and a trailing zero element. `TARGET_ASM_DESTRUCTOR' is handled
31114 similarly. Since no init section is available, the absence of
31115 `INIT_SECTION_ASM_OP' causes the compilation of `main' to call `__main'
31116 as above, starting the initialization process.
31118 The last variant uses neither arbitrary sections nor the GNU linker.
31119 This is preferable when you want to do dynamic linking and when using
31120 file formats which the GNU linker does not support, such as `ECOFF'. In
31121 this case, `TARGET_HAVE_CTORS_DTORS' is false, initialization and
31122 termination functions are recognized simply by their names. This
31123 requires an extra program in the linkage step, called `collect2'. This
31124 program pretends to be the linker, for use with GCC; it does its job by
31125 running the ordinary linker, but also arranges to include the vectors of
31126 initialization and termination functions. These functions are called
31127 via `__main' as described above. In order to use this method,
31128 `use_collect2' must be defined in the target in `config.gcc'.
31130 The following section describes the specific macros that control and
31131 customize the handling of initialization and termination functions.
31134 File: gccint.info, Node: Macros for Initialization, Next: Instruction Output, Prev: Initialization, Up: Assembler Format
31136 17.21.6 Macros Controlling Initialization Routines
31137 --------------------------------------------------
31139 Here are the macros that control how the compiler handles initialization
31140 and termination functions:
31142 -- Macro: INIT_SECTION_ASM_OP
31143 If defined, a C string constant, including spacing, for the
31144 assembler operation to identify the following data as
31145 initialization code. If not defined, GCC will assume such a
31146 section does not exist. When you are using special sections for
31147 initialization and termination functions, this macro also controls
31148 how `crtstuff.c' and `libgcc2.c' arrange to run the initialization
31151 -- Macro: HAS_INIT_SECTION
31152 If defined, `main' will not call `__main' as described above.
31153 This macro should be defined for systems that control start-up code
31154 on a symbol-by-symbol basis, such as OSF/1, and should not be
31155 defined explicitly for systems that support `INIT_SECTION_ASM_OP'.
31157 -- Macro: LD_INIT_SWITCH
31158 If defined, a C string constant for a switch that tells the linker
31159 that the following symbol is an initialization routine.
31161 -- Macro: LD_FINI_SWITCH
31162 If defined, a C string constant for a switch that tells the linker
31163 that the following symbol is a finalization routine.
31165 -- Macro: COLLECT_SHARED_INIT_FUNC (STREAM, FUNC)
31166 If defined, a C statement that will write a function that can be
31167 automatically called when a shared library is loaded. The function
31168 should call FUNC, which takes no arguments. If not defined, and
31169 the object format requires an explicit initialization function,
31170 then a function called `_GLOBAL__DI' will be generated.
31172 This function and the following one are used by collect2 when
31173 linking a shared library that needs constructors or destructors,
31174 or has DWARF2 exception tables embedded in the code.
31176 -- Macro: COLLECT_SHARED_FINI_FUNC (STREAM, FUNC)
31177 If defined, a C statement that will write a function that can be
31178 automatically called when a shared library is unloaded. The
31179 function should call FUNC, which takes no arguments. If not
31180 defined, and the object format requires an explicit finalization
31181 function, then a function called `_GLOBAL__DD' will be generated.
31183 -- Macro: INVOKE__main
31184 If defined, `main' will call `__main' despite the presence of
31185 `INIT_SECTION_ASM_OP'. This macro should be defined for systems
31186 where the init section is not actually run automatically, but is
31187 still useful for collecting the lists of constructors and
31190 -- Macro: SUPPORTS_INIT_PRIORITY
31191 If nonzero, the C++ `init_priority' attribute is supported and the
31192 compiler should emit instructions to control the order of
31193 initialization of objects. If zero, the compiler will issue an
31194 error message upon encountering an `init_priority' attribute.
31196 -- Target Hook: bool TARGET_HAVE_CTORS_DTORS
31197 This value is true if the target supports some "native" method of
31198 collecting constructors and destructors to be run at startup and
31199 exit. It is false if we must use `collect2'.
31201 -- Target Hook: void TARGET_ASM_CONSTRUCTOR (rtx SYMBOL, int PRIORITY)
31202 If defined, a function that outputs assembler code to arrange to
31203 call the function referenced by SYMBOL at initialization time.
31205 Assume that SYMBOL is a `SYMBOL_REF' for a function taking no
31206 arguments and with no return value. If the target supports
31207 initialization priorities, PRIORITY is a value between 0 and
31208 `MAX_INIT_PRIORITY'; otherwise it must be `DEFAULT_INIT_PRIORITY'.
31210 If this macro is not defined by the target, a suitable default will
31211 be chosen if (1) the target supports arbitrary section names, (2)
31212 the target defines `CTORS_SECTION_ASM_OP', or (3) `USE_COLLECT2'
31215 -- Target Hook: void TARGET_ASM_DESTRUCTOR (rtx SYMBOL, int PRIORITY)
31216 This is like `TARGET_ASM_CONSTRUCTOR' but used for termination
31217 functions rather than initialization functions.
31219 If `TARGET_HAVE_CTORS_DTORS' is true, the initialization routine
31220 generated for the generated object file will have static linkage.
31222 If your system uses `collect2' as the means of processing
31223 constructors, then that program normally uses `nm' to scan an object
31224 file for constructor functions to be called.
31226 On certain kinds of systems, you can define this macro to make
31227 `collect2' work faster (and, in some cases, make it work at all):
31229 -- Macro: OBJECT_FORMAT_COFF
31230 Define this macro if the system uses COFF (Common Object File
31231 Format) object files, so that `collect2' can assume this format
31232 and scan object files directly for dynamic constructor/destructor
31235 This macro is effective only in a native compiler; `collect2' as
31236 part of a cross compiler always uses `nm' for the target machine.
31238 -- Macro: REAL_NM_FILE_NAME
31239 Define this macro as a C string constant containing the file name
31240 to use to execute `nm'. The default is to search the path
31243 If your system supports shared libraries and has a program to list
31244 the dynamic dependencies of a given library or executable, you can
31245 define these macros to enable support for running initialization
31246 and termination functions in shared libraries:
31248 -- Macro: LDD_SUFFIX
31249 Define this macro to a C string constant containing the name of
31250 the program which lists dynamic dependencies, like `"ldd"' under
31253 -- Macro: PARSE_LDD_OUTPUT (PTR)
31254 Define this macro to be C code that extracts filenames from the
31255 output of the program denoted by `LDD_SUFFIX'. PTR is a variable
31256 of type `char *' that points to the beginning of a line of output
31257 from `LDD_SUFFIX'. If the line lists a dynamic dependency, the
31258 code must advance PTR to the beginning of the filename on that
31259 line. Otherwise, it must set PTR to `NULL'.
31261 -- Macro: SHLIB_SUFFIX
31262 Define this macro to a C string constant containing the default
31263 shared library extension of the target (e.g., `".so"'). `collect2'
31264 strips version information after this suffix when generating global
31265 constructor and destructor names. This define is only needed on
31266 targets that use `collect2' to process constructors and
31270 File: gccint.info, Node: Instruction Output, Next: Dispatch Tables, Prev: Macros for Initialization, Up: Assembler Format
31272 17.21.7 Output of Assembler Instructions
31273 ----------------------------------------
31275 This describes assembler instruction output.
31277 -- Macro: REGISTER_NAMES
31278 A C initializer containing the assembler's names for the machine
31279 registers, each one as a C string constant. This is what
31280 translates register numbers in the compiler into assembler
31283 -- Macro: ADDITIONAL_REGISTER_NAMES
31284 If defined, a C initializer for an array of structures containing
31285 a name and a register number. This macro defines additional names
31286 for hard registers, thus allowing the `asm' option in declarations
31287 to refer to registers using alternate names.
31289 -- Macro: ASM_OUTPUT_OPCODE (STREAM, PTR)
31290 Define this macro if you are using an unusual assembler that
31291 requires different names for the machine instructions.
31293 The definition is a C statement or statements which output an
31294 assembler instruction opcode to the stdio stream STREAM. The
31295 macro-operand PTR is a variable of type `char *' which points to
31296 the opcode name in its "internal" form--the form that is written
31297 in the machine description. The definition should output the
31298 opcode name to STREAM, performing any translation you desire, and
31299 increment the variable PTR to point at the end of the opcode so
31300 that it will not be output twice.
31302 In fact, your macro definition may process less than the entire
31303 opcode name, or more than the opcode name; but if you want to
31304 process text that includes `%'-sequences to substitute operands,
31305 you must take care of the substitution yourself. Just be sure to
31306 increment PTR over whatever text should not be output normally.
31308 If you need to look at the operand values, they can be found as the
31309 elements of `recog_data.operand'.
31311 If the macro definition does nothing, the instruction is output in
31314 -- Macro: FINAL_PRESCAN_INSN (INSN, OPVEC, NOPERANDS)
31315 If defined, a C statement to be executed just prior to the output
31316 of assembler code for INSN, to modify the extracted operands so
31317 they will be output differently.
31319 Here the argument OPVEC is the vector containing the operands
31320 extracted from INSN, and NOPERANDS is the number of elements of
31321 the vector which contain meaningful data for this insn. The
31322 contents of this vector are what will be used to convert the insn
31323 template into assembler code, so you can change the assembler
31324 output by changing the contents of the vector.
31326 This macro is useful when various assembler syntaxes share a single
31327 file of instruction patterns; by defining this macro differently,
31328 you can cause a large class of instructions to be output
31329 differently (such as with rearranged operands). Naturally,
31330 variations in assembler syntax affecting individual insn patterns
31331 ought to be handled by writing conditional output routines in
31334 If this macro is not defined, it is equivalent to a null statement.
31336 -- Macro: PRINT_OPERAND (STREAM, X, CODE)
31337 A C compound statement to output to stdio stream STREAM the
31338 assembler syntax for an instruction operand X. X is an RTL
31341 CODE is a value that can be used to specify one of several ways of
31342 printing the operand. It is used when identical operands must be
31343 printed differently depending on the context. CODE comes from the
31344 `%' specification that was used to request printing of the
31345 operand. If the specification was just `%DIGIT' then CODE is 0;
31346 if the specification was `%LTR DIGIT' then CODE is the ASCII code
31349 If X is a register, this macro should print the register's name.
31350 The names can be found in an array `reg_names' whose type is `char
31351 *[]'. `reg_names' is initialized from `REGISTER_NAMES'.
31353 When the machine description has a specification `%PUNCT' (a `%'
31354 followed by a punctuation character), this macro is called with a
31355 null pointer for X and the punctuation character for CODE.
31357 -- Macro: PRINT_OPERAND_PUNCT_VALID_P (CODE)
31358 A C expression which evaluates to true if CODE is a valid
31359 punctuation character for use in the `PRINT_OPERAND' macro. If
31360 `PRINT_OPERAND_PUNCT_VALID_P' is not defined, it means that no
31361 punctuation characters (except for the standard one, `%') are used
31364 -- Macro: PRINT_OPERAND_ADDRESS (STREAM, X)
31365 A C compound statement to output to stdio stream STREAM the
31366 assembler syntax for an instruction operand that is a memory
31367 reference whose address is X. X is an RTL expression.
31369 On some machines, the syntax for a symbolic address depends on the
31370 section that the address refers to. On these machines, define the
31371 hook `TARGET_ENCODE_SECTION_INFO' to store the information into the
31372 `symbol_ref', and then check for it here. *Note Assembler
31375 -- Macro: DBR_OUTPUT_SEQEND (FILE)
31376 A C statement, to be executed after all slot-filler instructions
31377 have been output. If necessary, call `dbr_sequence_length' to
31378 determine the number of slots filled in a sequence (zero if not
31379 currently outputting a sequence), to decide how many no-ops to
31380 output, or whatever.
31382 Don't define this macro if it has nothing to do, but it is helpful
31383 in reading assembly output if the extent of the delay sequence is
31384 made explicit (e.g. with white space).
31386 Note that output routines for instructions with delay slots must be
31387 prepared to deal with not being output as part of a sequence (i.e. when
31388 the scheduling pass is not run, or when no slot fillers could be
31389 found.) The variable `final_sequence' is null when not processing a
31390 sequence, otherwise it contains the `sequence' rtx being output.
31392 -- Macro: REGISTER_PREFIX
31393 -- Macro: LOCAL_LABEL_PREFIX
31394 -- Macro: USER_LABEL_PREFIX
31395 -- Macro: IMMEDIATE_PREFIX
31396 If defined, C string expressions to be used for the `%R', `%L',
31397 `%U', and `%I' options of `asm_fprintf' (see `final.c'). These
31398 are useful when a single `md' file must support multiple assembler
31399 formats. In that case, the various `tm.h' files can define these
31400 macros differently.
31402 -- Macro: ASM_FPRINTF_EXTENSIONS (FILE, ARGPTR, FORMAT)
31403 If defined this macro should expand to a series of `case'
31404 statements which will be parsed inside the `switch' statement of
31405 the `asm_fprintf' function. This allows targets to define extra
31406 printf formats which may useful when generating their assembler
31407 statements. Note that uppercase letters are reserved for future
31408 generic extensions to asm_fprintf, and so are not available to
31409 target specific code. The output file is given by the parameter
31410 FILE. The varargs input pointer is ARGPTR and the rest of the
31411 format string, starting the character after the one that is being
31412 switched upon, is pointed to by FORMAT.
31414 -- Macro: ASSEMBLER_DIALECT
31415 If your target supports multiple dialects of assembler language
31416 (such as different opcodes), define this macro as a C expression
31417 that gives the numeric index of the assembler language dialect to
31418 use, with zero as the first variant.
31420 If this macro is defined, you may use constructs of the form
31421 `{option0|option1|option2...}'
31422 in the output templates of patterns (*note Output Template::) or
31423 in the first argument of `asm_fprintf'. This construct outputs
31424 `option0', `option1', `option2', etc., if the value of
31425 `ASSEMBLER_DIALECT' is zero, one, two, etc. Any special characters
31426 within these strings retain their usual meaning. If there are
31427 fewer alternatives within the braces than the value of
31428 `ASSEMBLER_DIALECT', the construct outputs nothing.
31430 If you do not define this macro, the characters `{', `|' and `}'
31431 do not have any special meaning when used in templates or operands
31434 Define the macros `REGISTER_PREFIX', `LOCAL_LABEL_PREFIX',
31435 `USER_LABEL_PREFIX' and `IMMEDIATE_PREFIX' if you can express the
31436 variations in assembler language syntax with that mechanism.
31437 Define `ASSEMBLER_DIALECT' and use the `{option0|option1}' syntax
31438 if the syntax variant are larger and involve such things as
31439 different opcodes or operand order.
31441 -- Macro: ASM_OUTPUT_REG_PUSH (STREAM, REGNO)
31442 A C expression to output to STREAM some assembler code which will
31443 push hard register number REGNO onto the stack. The code need not
31444 be optimal, since this macro is used only when profiling.
31446 -- Macro: ASM_OUTPUT_REG_POP (STREAM, REGNO)
31447 A C expression to output to STREAM some assembler code which will
31448 pop hard register number REGNO off of the stack. The code need
31449 not be optimal, since this macro is used only when profiling.
31452 File: gccint.info, Node: Dispatch Tables, Next: Exception Region Output, Prev: Instruction Output, Up: Assembler Format
31454 17.21.8 Output of Dispatch Tables
31455 ---------------------------------
31457 This concerns dispatch tables.
31459 -- Macro: ASM_OUTPUT_ADDR_DIFF_ELT (STREAM, BODY, VALUE, REL)
31460 A C statement to output to the stdio stream STREAM an assembler
31461 pseudo-instruction to generate a difference between two labels.
31462 VALUE and REL are the numbers of two internal labels. The
31463 definitions of these labels are output using
31464 `(*targetm.asm_out.internal_label)', and they must be printed in
31465 the same way here. For example,
31467 fprintf (STREAM, "\t.word L%d-L%d\n",
31470 You must provide this macro on machines where the addresses in a
31471 dispatch table are relative to the table's own address. If
31472 defined, GCC will also use this macro on all machines when
31473 producing PIC. BODY is the body of the `ADDR_DIFF_VEC'; it is
31474 provided so that the mode and flags can be read.
31476 -- Macro: ASM_OUTPUT_ADDR_VEC_ELT (STREAM, VALUE)
31477 This macro should be provided on machines where the addresses in a
31478 dispatch table are absolute.
31480 The definition should be a C statement to output to the stdio
31481 stream STREAM an assembler pseudo-instruction to generate a
31482 reference to a label. VALUE is the number of an internal label
31483 whose definition is output using
31484 `(*targetm.asm_out.internal_label)'. For example,
31486 fprintf (STREAM, "\t.word L%d\n", VALUE)
31488 -- Macro: ASM_OUTPUT_CASE_LABEL (STREAM, PREFIX, NUM, TABLE)
31489 Define this if the label before a jump-table needs to be output
31490 specially. The first three arguments are the same as for
31491 `(*targetm.asm_out.internal_label)'; the fourth argument is the
31492 jump-table which follows (a `jump_insn' containing an `addr_vec'
31493 or `addr_diff_vec').
31495 This feature is used on system V to output a `swbeg' statement for
31498 If this macro is not defined, these labels are output with
31499 `(*targetm.asm_out.internal_label)'.
31501 -- Macro: ASM_OUTPUT_CASE_END (STREAM, NUM, TABLE)
31502 Define this if something special must be output at the end of a
31503 jump-table. The definition should be a C statement to be executed
31504 after the assembler code for the table is written. It should write
31505 the appropriate code to stdio stream STREAM. The argument TABLE
31506 is the jump-table insn, and NUM is the label-number of the
31509 If this macro is not defined, nothing special is output at the end
31512 -- Target Hook: void TARGET_ASM_EMIT_UNWIND_LABEL (STREAM, DECL,
31514 This target hook emits a label at the beginning of each FDE. It
31515 should be defined on targets where FDEs need special labels, and it
31516 should write the appropriate label, for the FDE associated with the
31517 function declaration DECL, to the stdio stream STREAM. The third
31518 argument, FOR_EH, is a boolean: true if this is for an exception
31519 table. The fourth argument, EMPTY, is a boolean: true if this is
31520 a placeholder label for an omitted FDE.
31522 The default is that FDEs are not given nonlocal labels.
31524 -- Target Hook: void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (STREAM)
31525 This target hook emits a label at the beginning of the exception
31526 table. It should be defined on targets where it is desirable for
31527 the table to be broken up according to function.
31529 The default is that no label is emitted.
31531 -- Target Hook: void TARGET_UNWIND_EMIT (FILE * STREAM, rtx INSN)
31532 This target hook emits and assembly directives required to unwind
31533 the given instruction. This is only used when TARGET_UNWIND_INFO
31537 File: gccint.info, Node: Exception Region Output, Next: Alignment Output, Prev: Dispatch Tables, Up: Assembler Format
31539 17.21.9 Assembler Commands for Exception Regions
31540 ------------------------------------------------
31542 This describes commands marking the start and the end of an exception
31545 -- Macro: EH_FRAME_SECTION_NAME
31546 If defined, a C string constant for the name of the section
31547 containing exception handling frame unwind information. If not
31548 defined, GCC will provide a default definition if the target
31549 supports named sections. `crtstuff.c' uses this macro to switch
31550 to the appropriate section.
31552 You should define this symbol if your target supports DWARF 2 frame
31553 unwind information and the default definition does not work.
31555 -- Macro: EH_FRAME_IN_DATA_SECTION
31556 If defined, DWARF 2 frame unwind information will be placed in the
31557 data section even though the target supports named sections. This
31558 might be necessary, for instance, if the system linker does garbage
31559 collection and sections cannot be marked as not to be collected.
31561 Do not define this macro unless `TARGET_ASM_NAMED_SECTION' is also
31564 -- Macro: EH_TABLES_CAN_BE_READ_ONLY
31565 Define this macro to 1 if your target is such that no frame unwind
31566 information encoding used with non-PIC code will ever require a
31567 runtime relocation, but the linker may not support merging
31568 read-only and read-write sections into a single read-write section.
31570 -- Macro: MASK_RETURN_ADDR
31571 An rtx used to mask the return address found via
31572 `RETURN_ADDR_RTX', so that it does not contain any extraneous set
31575 -- Macro: DWARF2_UNWIND_INFO
31576 Define this macro to 0 if your target supports DWARF 2 frame unwind
31577 information, but it does not yet work with exception handling.
31578 Otherwise, if your target supports this information (if it defines
31579 `INCOMING_RETURN_ADDR_RTX' and either `UNALIGNED_INT_ASM_OP' or
31580 `OBJECT_FORMAT_ELF'), GCC will provide a default definition of 1.
31582 If `TARGET_UNWIND_INFO' is defined, the target specific unwinder
31583 will be used in all cases. Defining this macro will enable the
31584 generation of DWARF 2 frame debugging information.
31586 If `TARGET_UNWIND_INFO' is not defined, and this macro is defined
31587 to 1, the DWARF 2 unwinder will be the default exception handling
31588 mechanism; otherwise, the `setjmp'/`longjmp'-based scheme will be
31591 -- Macro: TARGET_UNWIND_INFO
31592 Define this macro if your target has ABI specified unwind tables.
31593 Usually these will be output by `TARGET_UNWIND_EMIT'.
31595 -- Variable: Target Hook bool TARGET_UNWIND_TABLES_DEFAULT
31596 This variable should be set to `true' if the target ABI requires
31597 unwinding tables even when exceptions are not used.
31599 -- Macro: MUST_USE_SJLJ_EXCEPTIONS
31600 This macro need only be defined if `DWARF2_UNWIND_INFO' is
31601 runtime-variable. In that case, `except.h' cannot correctly
31602 determine the corresponding definition of
31603 `MUST_USE_SJLJ_EXCEPTIONS', so the target must provide it directly.
31605 -- Macro: DONT_USE_BUILTIN_SETJMP
31606 Define this macro to 1 if the `setjmp'/`longjmp'-based scheme
31607 should use the `setjmp'/`longjmp' functions from the C library
31608 instead of the `__builtin_setjmp'/`__builtin_longjmp' machinery.
31610 -- Macro: DWARF_CIE_DATA_ALIGNMENT
31611 This macro need only be defined if the target might save registers
31612 in the function prologue at an offset to the stack pointer that is
31613 not aligned to `UNITS_PER_WORD'. The definition should be the
31614 negative minimum alignment if `STACK_GROWS_DOWNWARD' is defined,
31615 and the positive minimum alignment otherwise. *Note SDB and
31616 DWARF::. Only applicable if the target supports DWARF 2 frame
31617 unwind information.
31619 -- Variable: Target Hook bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
31620 Contains the value true if the target should add a zero word onto
31621 the end of a Dwarf-2 frame info section when used for exception
31622 handling. Default value is false if `EH_FRAME_SECTION_NAME' is
31623 defined, and true otherwise.
31625 -- Target Hook: rtx TARGET_DWARF_REGISTER_SPAN (rtx REG)
31626 Given a register, this hook should return a parallel of registers
31627 to represent where to find the register pieces. Define this hook
31628 if the register and its mode are represented in Dwarf in
31629 non-contiguous locations, or if the register should be represented
31630 in more than one register in Dwarf. Otherwise, this hook should
31631 return `NULL_RTX'. If not defined, the default is to return
31634 -- Target Hook: void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree ADDRESS)
31635 If some registers are represented in Dwarf-2 unwind information in
31636 multiple pieces, define this hook to fill in information about the
31637 sizes of those pieces in the table used by the unwinder at runtime.
31638 It will be called by `expand_builtin_init_dwarf_reg_sizes' after
31639 filling in a single size corresponding to each hard register;
31640 ADDRESS is the address of the table.
31642 -- Target Hook: bool TARGET_ASM_TTYPE (rtx SYM)
31643 This hook is used to output a reference from a frame unwinding
31644 table to the type_info object identified by SYM. It should return
31645 `true' if the reference was output. Returning `false' will cause
31646 the reference to be output using the normal Dwarf2 routines.
31648 -- Target Hook: bool TARGET_ARM_EABI_UNWINDER
31649 This hook should be set to `true' on targets that use an ARM EABI
31650 based unwinding library, and `false' on other targets. This
31651 effects the format of unwinding tables, and how the unwinder in
31652 entered after running a cleanup. The default is `false'.
31655 File: gccint.info, Node: Alignment Output, Prev: Exception Region Output, Up: Assembler Format
31657 17.21.10 Assembler Commands for Alignment
31658 -----------------------------------------
31660 This describes commands for alignment.
31662 -- Macro: JUMP_ALIGN (LABEL)
31663 The alignment (log base 2) to put in front of LABEL, which is a
31664 common destination of jumps and has no fallthru incoming edge.
31666 This macro need not be defined if you don't want any special
31667 alignment to be done at such a time. Most machine descriptions do
31668 not currently define the macro.
31670 Unless it's necessary to inspect the LABEL parameter, it is better
31671 to set the variable ALIGN_JUMPS in the target's
31672 `OVERRIDE_OPTIONS'. Otherwise, you should try to honor the user's
31673 selection in ALIGN_JUMPS in a `JUMP_ALIGN' implementation.
31675 -- Macro: LABEL_ALIGN_AFTER_BARRIER (LABEL)
31676 The alignment (log base 2) to put in front of LABEL, which follows
31679 This macro need not be defined if you don't want any special
31680 alignment to be done at such a time. Most machine descriptions do
31681 not currently define the macro.
31683 -- Macro: LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
31684 The maximum number of bytes to skip when applying
31685 `LABEL_ALIGN_AFTER_BARRIER'. This works only if
31686 `ASM_OUTPUT_MAX_SKIP_ALIGN' is defined.
31688 -- Macro: LOOP_ALIGN (LABEL)
31689 The alignment (log base 2) to put in front of LABEL, which follows
31690 a `NOTE_INSN_LOOP_BEG' note.
31692 This macro need not be defined if you don't want any special
31693 alignment to be done at such a time. Most machine descriptions do
31694 not currently define the macro.
31696 Unless it's necessary to inspect the LABEL parameter, it is better
31697 to set the variable `align_loops' in the target's
31698 `OVERRIDE_OPTIONS'. Otherwise, you should try to honor the user's
31699 selection in `align_loops' in a `LOOP_ALIGN' implementation.
31701 -- Macro: LOOP_ALIGN_MAX_SKIP
31702 The maximum number of bytes to skip when applying `LOOP_ALIGN'.
31703 This works only if `ASM_OUTPUT_MAX_SKIP_ALIGN' is defined.
31705 -- Macro: LABEL_ALIGN (LABEL)
31706 The alignment (log base 2) to put in front of LABEL. If
31707 `LABEL_ALIGN_AFTER_BARRIER' / `LOOP_ALIGN' specify a different
31708 alignment, the maximum of the specified values is used.
31710 Unless it's necessary to inspect the LABEL parameter, it is better
31711 to set the variable `align_labels' in the target's
31712 `OVERRIDE_OPTIONS'. Otherwise, you should try to honor the user's
31713 selection in `align_labels' in a `LABEL_ALIGN' implementation.
31715 -- Macro: LABEL_ALIGN_MAX_SKIP
31716 The maximum number of bytes to skip when applying `LABEL_ALIGN'.
31717 This works only if `ASM_OUTPUT_MAX_SKIP_ALIGN' is defined.
31719 -- Macro: ASM_OUTPUT_SKIP (STREAM, NBYTES)
31720 A C statement to output to the stdio stream STREAM an assembler
31721 instruction to advance the location counter by NBYTES bytes.
31722 Those bytes should be zero when loaded. NBYTES will be a C
31723 expression of type `unsigned HOST_WIDE_INT'.
31725 -- Macro: ASM_NO_SKIP_IN_TEXT
31726 Define this macro if `ASM_OUTPUT_SKIP' should not be used in the
31727 text section because it fails to put zeros in the bytes that are
31728 skipped. This is true on many Unix systems, where the pseudo-op
31729 to skip bytes produces no-op instructions rather than zeros when
31730 used in the text section.
31732 -- Macro: ASM_OUTPUT_ALIGN (STREAM, POWER)
31733 A C statement to output to the stdio stream STREAM an assembler
31734 command to advance the location counter to a multiple of 2 to the
31735 POWER bytes. POWER will be a C expression of type `int'.
31737 -- Macro: ASM_OUTPUT_ALIGN_WITH_NOP (STREAM, POWER)
31738 Like `ASM_OUTPUT_ALIGN', except that the "nop" instruction is used
31739 for padding, if necessary.
31741 -- Macro: ASM_OUTPUT_MAX_SKIP_ALIGN (STREAM, POWER, MAX_SKIP)
31742 A C statement to output to the stdio stream STREAM an assembler
31743 command to advance the location counter to a multiple of 2 to the
31744 POWER bytes, but only if MAX_SKIP or fewer bytes are needed to
31745 satisfy the alignment request. POWER and MAX_SKIP will be a C
31746 expression of type `int'.
31749 File: gccint.info, Node: Debugging Info, Next: Floating Point, Prev: Assembler Format, Up: Target Macros
31751 17.22 Controlling Debugging Information Format
31752 ==============================================
31754 This describes how to specify debugging information.
31758 * All Debuggers:: Macros that affect all debugging formats uniformly.
31759 * DBX Options:: Macros enabling specific options in DBX format.
31760 * DBX Hooks:: Hook macros for varying DBX format.
31761 * File Names and DBX:: Macros controlling output of file names in DBX format.
31762 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
31763 * VMS Debug:: Macros for VMS debug format.
31766 File: gccint.info, Node: All Debuggers, Next: DBX Options, Up: Debugging Info
31768 17.22.1 Macros Affecting All Debugging Formats
31769 ----------------------------------------------
31771 These macros affect all debugging formats.
31773 -- Macro: DBX_REGISTER_NUMBER (REGNO)
31774 A C expression that returns the DBX register number for the
31775 compiler register number REGNO. In the default macro provided,
31776 the value of this expression will be REGNO itself. But sometimes
31777 there are some registers that the compiler knows about and DBX
31778 does not, or vice versa. In such cases, some register may need to
31779 have one number in the compiler and another for DBX.
31781 If two registers have consecutive numbers inside GCC, and they can
31782 be used as a pair to hold a multiword value, then they _must_ have
31783 consecutive numbers after renumbering with `DBX_REGISTER_NUMBER'.
31784 Otherwise, debuggers will be unable to access such a pair, because
31785 they expect register pairs to be consecutive in their own
31788 If you find yourself defining `DBX_REGISTER_NUMBER' in way that
31789 does not preserve register pairs, then what you must do instead is
31790 redefine the actual register numbering scheme.
31792 -- Macro: DEBUGGER_AUTO_OFFSET (X)
31793 A C expression that returns the integer offset value for an
31794 automatic variable having address X (an RTL expression). The
31795 default computation assumes that X is based on the frame-pointer
31796 and gives the offset from the frame-pointer. This is required for
31797 targets that produce debugging output for DBX or COFF-style
31798 debugging output for SDB and allow the frame-pointer to be
31799 eliminated when the `-g' options is used.
31801 -- Macro: DEBUGGER_ARG_OFFSET (OFFSET, X)
31802 A C expression that returns the integer offset value for an
31803 argument having address X (an RTL expression). The nominal offset
31806 -- Macro: PREFERRED_DEBUGGING_TYPE
31807 A C expression that returns the type of debugging output GCC should
31808 produce when the user specifies just `-g'. Define this if you
31809 have arranged for GCC to support more than one format of debugging
31810 output. Currently, the allowable values are `DBX_DEBUG',
31811 `SDB_DEBUG', `DWARF_DEBUG', `DWARF2_DEBUG', `XCOFF_DEBUG',
31812 `VMS_DEBUG', and `VMS_AND_DWARF2_DEBUG'.
31814 When the user specifies `-ggdb', GCC normally also uses the value
31815 of this macro to select the debugging output format, but with two
31816 exceptions. If `DWARF2_DEBUGGING_INFO' is defined, GCC uses the
31817 value `DWARF2_DEBUG'. Otherwise, if `DBX_DEBUGGING_INFO' is
31818 defined, GCC uses `DBX_DEBUG'.
31820 The value of this macro only affects the default debugging output;
31821 the user can always get a specific type of output by using
31822 `-gstabs', `-gcoff', `-gdwarf-2', `-gxcoff', or `-gvms'.
31825 File: gccint.info, Node: DBX Options, Next: DBX Hooks, Prev: All Debuggers, Up: Debugging Info
31827 17.22.2 Specific Options for DBX Output
31828 ---------------------------------------
31830 These are specific options for DBX output.
31832 -- Macro: DBX_DEBUGGING_INFO
31833 Define this macro if GCC should produce debugging output for DBX
31834 in response to the `-g' option.
31836 -- Macro: XCOFF_DEBUGGING_INFO
31837 Define this macro if GCC should produce XCOFF format debugging
31838 output in response to the `-g' option. This is a variant of DBX
31841 -- Macro: DEFAULT_GDB_EXTENSIONS
31842 Define this macro to control whether GCC should by default generate
31843 GDB's extended version of DBX debugging information (assuming
31844 DBX-format debugging information is enabled at all). If you don't
31845 define the macro, the default is 1: always generate the extended
31846 information if there is any occasion to.
31848 -- Macro: DEBUG_SYMS_TEXT
31849 Define this macro if all `.stabs' commands should be output while
31850 in the text section.
31852 -- Macro: ASM_STABS_OP
31853 A C string constant, including spacing, naming the assembler
31854 pseudo op to use instead of `"\t.stabs\t"' to define an ordinary
31855 debugging symbol. If you don't define this macro, `"\t.stabs\t"'
31856 is used. This macro applies only to DBX debugging information
31859 -- Macro: ASM_STABD_OP
31860 A C string constant, including spacing, naming the assembler
31861 pseudo op to use instead of `"\t.stabd\t"' to define a debugging
31862 symbol whose value is the current location. If you don't define
31863 this macro, `"\t.stabd\t"' is used. This macro applies only to
31864 DBX debugging information format.
31866 -- Macro: ASM_STABN_OP
31867 A C string constant, including spacing, naming the assembler
31868 pseudo op to use instead of `"\t.stabn\t"' to define a debugging
31869 symbol with no name. If you don't define this macro,
31870 `"\t.stabn\t"' is used. This macro applies only to DBX debugging
31871 information format.
31873 -- Macro: DBX_NO_XREFS
31874 Define this macro if DBX on your system does not support the
31875 construct `xsTAGNAME'. On some systems, this construct is used to
31876 describe a forward reference to a structure named TAGNAME. On
31877 other systems, this construct is not supported at all.
31879 -- Macro: DBX_CONTIN_LENGTH
31880 A symbol name in DBX-format debugging information is normally
31881 continued (split into two separate `.stabs' directives) when it
31882 exceeds a certain length (by default, 80 characters). On some
31883 operating systems, DBX requires this splitting; on others,
31884 splitting must not be done. You can inhibit splitting by defining
31885 this macro with the value zero. You can override the default
31886 splitting-length by defining this macro as an expression for the
31889 -- Macro: DBX_CONTIN_CHAR
31890 Normally continuation is indicated by adding a `\' character to
31891 the end of a `.stabs' string when a continuation follows. To use
31892 a different character instead, define this macro as a character
31893 constant for the character you want to use. Do not define this
31894 macro if backslash is correct for your system.
31896 -- Macro: DBX_STATIC_STAB_DATA_SECTION
31897 Define this macro if it is necessary to go to the data section
31898 before outputting the `.stabs' pseudo-op for a non-global static
31901 -- Macro: DBX_TYPE_DECL_STABS_CODE
31902 The value to use in the "code" field of the `.stabs' directive for
31903 a typedef. The default is `N_LSYM'.
31905 -- Macro: DBX_STATIC_CONST_VAR_CODE
31906 The value to use in the "code" field of the `.stabs' directive for
31907 a static variable located in the text section. DBX format does not
31908 provide any "right" way to do this. The default is `N_FUN'.
31910 -- Macro: DBX_REGPARM_STABS_CODE
31911 The value to use in the "code" field of the `.stabs' directive for
31912 a parameter passed in registers. DBX format does not provide any
31913 "right" way to do this. The default is `N_RSYM'.
31915 -- Macro: DBX_REGPARM_STABS_LETTER
31916 The letter to use in DBX symbol data to identify a symbol as a
31917 parameter passed in registers. DBX format does not customarily
31918 provide any way to do this. The default is `'P''.
31920 -- Macro: DBX_FUNCTION_FIRST
31921 Define this macro if the DBX information for a function and its
31922 arguments should precede the assembler code for the function.
31923 Normally, in DBX format, the debugging information entirely
31924 follows the assembler code.
31926 -- Macro: DBX_BLOCKS_FUNCTION_RELATIVE
31927 Define this macro, with value 1, if the value of a symbol
31928 describing the scope of a block (`N_LBRAC' or `N_RBRAC') should be
31929 relative to the start of the enclosing function. Normally, GCC
31930 uses an absolute address.
31932 -- Macro: DBX_LINES_FUNCTION_RELATIVE
31933 Define this macro, with value 1, if the value of a symbol
31934 indicating the current line number (`N_SLINE') should be relative
31935 to the start of the enclosing function. Normally, GCC uses an
31938 -- Macro: DBX_USE_BINCL
31939 Define this macro if GCC should generate `N_BINCL' and `N_EINCL'
31940 stabs for included header files, as on Sun systems. This macro
31941 also directs GCC to output a type number as a pair of a file
31942 number and a type number within the file. Normally, GCC does not
31943 generate `N_BINCL' or `N_EINCL' stabs, and it outputs a single
31944 number for a type number.
31947 File: gccint.info, Node: DBX Hooks, Next: File Names and DBX, Prev: DBX Options, Up: Debugging Info
31949 17.22.3 Open-Ended Hooks for DBX Format
31950 ---------------------------------------
31952 These are hooks for DBX format.
31954 -- Macro: DBX_OUTPUT_LBRAC (STREAM, NAME)
31955 Define this macro to say how to output to STREAM the debugging
31956 information for the start of a scope level for variable names. The
31957 argument NAME is the name of an assembler symbol (for use with
31958 `assemble_name') whose value is the address where the scope begins.
31960 -- Macro: DBX_OUTPUT_RBRAC (STREAM, NAME)
31961 Like `DBX_OUTPUT_LBRAC', but for the end of a scope level.
31963 -- Macro: DBX_OUTPUT_NFUN (STREAM, LSCOPE_LABEL, DECL)
31964 Define this macro if the target machine requires special handling
31965 to output an `N_FUN' entry for the function DECL.
31967 -- Macro: DBX_OUTPUT_SOURCE_LINE (STREAM, LINE, COUNTER)
31968 A C statement to output DBX debugging information before code for
31969 line number LINE of the current source file to the stdio stream
31970 STREAM. COUNTER is the number of time the macro was invoked,
31971 including the current invocation; it is intended to generate
31972 unique labels in the assembly output.
31974 This macro should not be defined if the default output is correct,
31975 or if it can be made correct by defining
31976 `DBX_LINES_FUNCTION_RELATIVE'.
31978 -- Macro: NO_DBX_FUNCTION_END
31979 Some stabs encapsulation formats (in particular ECOFF), cannot
31980 handle the `.stabs "",N_FUN,,0,0,Lscope-function-1' gdb dbx
31981 extension construct. On those machines, define this macro to turn
31982 this feature off without disturbing the rest of the gdb extensions.
31984 -- Macro: NO_DBX_BNSYM_ENSYM
31985 Some assemblers cannot handle the `.stabd BNSYM/ENSYM,0,0' gdb dbx
31986 extension construct. On those machines, define this macro to turn
31987 this feature off without disturbing the rest of the gdb extensions.
31990 File: gccint.info, Node: File Names and DBX, Next: SDB and DWARF, Prev: DBX Hooks, Up: Debugging Info
31992 17.22.4 File Names in DBX Format
31993 --------------------------------
31995 This describes file names in DBX format.
31997 -- Macro: DBX_OUTPUT_MAIN_SOURCE_FILENAME (STREAM, NAME)
31998 A C statement to output DBX debugging information to the stdio
31999 stream STREAM, which indicates that file NAME is the main source
32000 file--the file specified as the input file for compilation. This
32001 macro is called only once, at the beginning of compilation.
32003 This macro need not be defined if the standard form of output for
32004 DBX debugging information is appropriate.
32006 It may be necessary to refer to a label equal to the beginning of
32007 the text section. You can use `assemble_name (stream,
32008 ltext_label_name)' to do so. If you do this, you must also set
32009 the variable USED_LTEXT_LABEL_NAME to `true'.
32011 -- Macro: NO_DBX_MAIN_SOURCE_DIRECTORY
32012 Define this macro, with value 1, if GCC should not emit an
32013 indication of the current directory for compilation and current
32014 source language at the beginning of the file.
32016 -- Macro: NO_DBX_GCC_MARKER
32017 Define this macro, with value 1, if GCC should not emit an
32018 indication that this object file was compiled by GCC. The default
32019 is to emit an `N_OPT' stab at the beginning of every source file,
32020 with `gcc2_compiled.' for the string and value 0.
32022 -- Macro: DBX_OUTPUT_MAIN_SOURCE_FILE_END (STREAM, NAME)
32023 A C statement to output DBX debugging information at the end of
32024 compilation of the main source file NAME. Output should be
32025 written to the stdio stream STREAM.
32027 If you don't define this macro, nothing special is output at the
32028 end of compilation, which is correct for most machines.
32030 -- Macro: DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
32031 Define this macro _instead of_ defining
32032 `DBX_OUTPUT_MAIN_SOURCE_FILE_END', if what needs to be output at
32033 the end of compilation is a `N_SO' stab with an empty string,
32034 whose value is the highest absolute text address in the file.
32037 File: gccint.info, Node: SDB and DWARF, Next: VMS Debug, Prev: File Names and DBX, Up: Debugging Info
32039 17.22.5 Macros for SDB and DWARF Output
32040 ---------------------------------------
32042 Here are macros for SDB and DWARF output.
32044 -- Macro: SDB_DEBUGGING_INFO
32045 Define this macro if GCC should produce COFF-style debugging output
32046 for SDB in response to the `-g' option.
32048 -- Macro: DWARF2_DEBUGGING_INFO
32049 Define this macro if GCC should produce dwarf version 2 format
32050 debugging output in response to the `-g' option.
32052 -- Target Hook: int TARGET_DWARF_CALLING_CONVENTION (tree
32054 Define this to enable the dwarf attribute
32055 `DW_AT_calling_convention' to be emitted for each function.
32056 Instead of an integer return the enum value for the `DW_CC_'
32059 To support optional call frame debugging information, you must also
32060 define `INCOMING_RETURN_ADDR_RTX' and either set
32061 `RTX_FRAME_RELATED_P' on the prologue insns if you use RTL for the
32062 prologue, or call `dwarf2out_def_cfa' and `dwarf2out_reg_save' as
32063 appropriate from `TARGET_ASM_FUNCTION_PROLOGUE' if you don't.
32065 -- Macro: DWARF2_FRAME_INFO
32066 Define this macro to a nonzero value if GCC should always output
32067 Dwarf 2 frame information. If `DWARF2_UNWIND_INFO' (*note
32068 Exception Region Output:: is nonzero, GCC will output this
32069 information not matter how you define `DWARF2_FRAME_INFO'.
32071 -- Macro: DWARF2_ASM_LINE_DEBUG_INFO
32072 Define this macro to be a nonzero value if the assembler can
32073 generate Dwarf 2 line debug info sections. This will result in
32074 much more compact line number tables, and hence is desirable if it
32077 -- Macro: ASM_OUTPUT_DWARF_DELTA (STREAM, SIZE, LABEL1, LABEL2)
32078 A C statement to issue assembly directives that create a difference
32079 LAB1 minus LAB2, using an integer of the given SIZE.
32081 -- Macro: ASM_OUTPUT_DWARF_OFFSET (STREAM, SIZE, LABEL, SECTION)
32082 A C statement to issue assembly directives that create a
32083 section-relative reference to the given LABEL, using an integer of
32084 the given SIZE. The label is known to be defined in the given
32087 -- Macro: ASM_OUTPUT_DWARF_PCREL (STREAM, SIZE, LABEL)
32088 A C statement to issue assembly directives that create a
32089 self-relative reference to the given LABEL, using an integer of
32092 -- Target Hook: void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *FILE, int
32094 If defined, this target hook is a function which outputs a
32095 DTP-relative reference to the given TLS symbol of the specified
32098 -- Macro: PUT_SDB_...
32099 Define these macros to override the assembler syntax for the
32100 special SDB assembler directives. See `sdbout.c' for a list of
32101 these macros and their arguments. If the standard syntax is used,
32102 you need not define them yourself.
32104 -- Macro: SDB_DELIM
32105 Some assemblers do not support a semicolon as a delimiter, even
32106 between SDB assembler directives. In that case, define this macro
32107 to be the delimiter to use (usually `\n'). It is not necessary to
32108 define a new set of `PUT_SDB_OP' macros if this is the only change
32111 -- Macro: SDB_ALLOW_UNKNOWN_REFERENCES
32112 Define this macro to allow references to unknown structure, union,
32113 or enumeration tags to be emitted. Standard COFF does not allow
32114 handling of unknown references, MIPS ECOFF has support for it.
32116 -- Macro: SDB_ALLOW_FORWARD_REFERENCES
32117 Define this macro to allow references to structure, union, or
32118 enumeration tags that have not yet been seen to be handled. Some
32119 assemblers choke if forward tags are used, while some require it.
32121 -- Macro: SDB_OUTPUT_SOURCE_LINE (STREAM, LINE)
32122 A C statement to output SDB debugging information before code for
32123 line number LINE of the current source file to the stdio stream
32124 STREAM. The default is to emit an `.ln' directive.
32127 File: gccint.info, Node: VMS Debug, Prev: SDB and DWARF, Up: Debugging Info
32129 17.22.6 Macros for VMS Debug Format
32130 -----------------------------------
32132 Here are macros for VMS debug format.
32134 -- Macro: VMS_DEBUGGING_INFO
32135 Define this macro if GCC should produce debugging output for VMS
32136 in response to the `-g' option. The default behavior for VMS is
32137 to generate minimal debug info for a traceback in the absence of
32138 `-g' unless explicitly overridden with `-g0'. This behavior is
32139 controlled by `OPTIMIZATION_OPTIONS' and `OVERRIDE_OPTIONS'.
32142 File: gccint.info, Node: Floating Point, Next: Mode Switching, Prev: Debugging Info, Up: Target Macros
32144 17.23 Cross Compilation and Floating Point
32145 ==========================================
32147 While all modern machines use twos-complement representation for
32148 integers, there are a variety of representations for floating point
32149 numbers. This means that in a cross-compiler the representation of
32150 floating point numbers in the compiled program may be different from
32151 that used in the machine doing the compilation.
32153 Because different representation systems may offer different amounts of
32154 range and precision, all floating point constants must be represented in
32155 the target machine's format. Therefore, the cross compiler cannot
32156 safely use the host machine's floating point arithmetic; it must emulate
32157 the target's arithmetic. To ensure consistency, GCC always uses
32158 emulation to work with floating point values, even when the host and
32159 target floating point formats are identical.
32161 The following macros are provided by `real.h' for the compiler to use.
32162 All parts of the compiler which generate or optimize floating-point
32163 calculations must use these macros. They may evaluate their operands
32164 more than once, so operands must not have side effects.
32166 -- Macro: REAL_VALUE_TYPE
32167 The C data type to be used to hold a floating point value in the
32168 target machine's format. Typically this is a `struct' containing
32169 an array of `HOST_WIDE_INT', but all code should treat it as an
32172 -- Macro: int REAL_VALUES_EQUAL (REAL_VALUE_TYPE X, REAL_VALUE_TYPE Y)
32173 Compares for equality the two values, X and Y. If the target
32174 floating point format supports negative zeroes and/or NaNs,
32175 `REAL_VALUES_EQUAL (-0.0, 0.0)' is true, and `REAL_VALUES_EQUAL
32176 (NaN, NaN)' is false.
32178 -- Macro: int REAL_VALUES_LESS (REAL_VALUE_TYPE X, REAL_VALUE_TYPE Y)
32179 Tests whether X is less than Y.
32181 -- Macro: HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE X)
32182 Truncates X to a signed integer, rounding toward zero.
32184 -- Macro: unsigned HOST_WIDE_INT REAL_VALUE_UNSIGNED_FIX
32185 (REAL_VALUE_TYPE X)
32186 Truncates X to an unsigned integer, rounding toward zero. If X is
32187 negative, returns zero.
32189 -- Macro: REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *STRING, enum
32191 Converts STRING into a floating point number in the target
32192 machine's representation for mode MODE. This routine can handle
32193 both decimal and hexadecimal floating point constants, using the
32194 syntax defined by the C language for both.
32196 -- Macro: int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE X)
32197 Returns 1 if X is negative (including negative zero), 0 otherwise.
32199 -- Macro: int REAL_VALUE_ISINF (REAL_VALUE_TYPE X)
32200 Determines whether X represents infinity (positive or negative).
32202 -- Macro: int REAL_VALUE_ISNAN (REAL_VALUE_TYPE X)
32203 Determines whether X represents a "NaN" (not-a-number).
32205 -- Macro: void REAL_ARITHMETIC (REAL_VALUE_TYPE OUTPUT, enum tree_code
32206 CODE, REAL_VALUE_TYPE X, REAL_VALUE_TYPE Y)
32207 Calculates an arithmetic operation on the two floating point values
32208 X and Y, storing the result in OUTPUT (which must be a variable).
32210 The operation to be performed is specified by CODE. Only the
32211 following codes are supported: `PLUS_EXPR', `MINUS_EXPR',
32212 `MULT_EXPR', `RDIV_EXPR', `MAX_EXPR', `MIN_EXPR'.
32214 If `REAL_ARITHMETIC' is asked to evaluate division by zero and the
32215 target's floating point format cannot represent infinity, it will
32216 call `abort'. Callers should check for this situation first, using
32217 `MODE_HAS_INFINITIES'. *Note Storage Layout::.
32219 -- Macro: REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE X)
32220 Returns the negative of the floating point value X.
32222 -- Macro: REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE X)
32223 Returns the absolute value of X.
32225 -- Macro: REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE MODE,
32226 enum machine_mode X)
32227 Truncates the floating point value X to fit in MODE. The return
32228 value is still a full-size `REAL_VALUE_TYPE', but it has an
32229 appropriate bit pattern to be output as a floating constant whose
32230 precision accords with mode MODE.
32232 -- Macro: void REAL_VALUE_TO_INT (HOST_WIDE_INT LOW, HOST_WIDE_INT
32233 HIGH, REAL_VALUE_TYPE X)
32234 Converts a floating point value X into a double-precision integer
32235 which is then stored into LOW and HIGH. If the value is not
32236 integral, it is truncated.
32238 -- Macro: void REAL_VALUE_FROM_INT (REAL_VALUE_TYPE X, HOST_WIDE_INT
32239 LOW, HOST_WIDE_INT HIGH, enum machine_mode MODE)
32240 Converts a double-precision integer found in LOW and HIGH, into a
32241 floating point value which is then stored into X. The value is
32242 truncated to fit in mode MODE.
32245 File: gccint.info, Node: Mode Switching, Next: Target Attributes, Prev: Floating Point, Up: Target Macros
32247 17.24 Mode Switching Instructions
32248 =================================
32250 The following macros control mode switching optimizations:
32252 -- Macro: OPTIMIZE_MODE_SWITCHING (ENTITY)
32253 Define this macro if the port needs extra instructions inserted
32254 for mode switching in an optimizing compilation.
32256 For an example, the SH4 can perform both single and double
32257 precision floating point operations, but to perform a single
32258 precision operation, the FPSCR PR bit has to be cleared, while for
32259 a double precision operation, this bit has to be set. Changing
32260 the PR bit requires a general purpose register as a scratch
32261 register, hence these FPSCR sets have to be inserted before
32262 reload, i.e. you can't put this into instruction emitting or
32263 `TARGET_MACHINE_DEPENDENT_REORG'.
32265 You can have multiple entities that are mode-switched, and select
32266 at run time which entities actually need it.
32267 `OPTIMIZE_MODE_SWITCHING' should return nonzero for any ENTITY
32268 that needs mode-switching. If you define this macro, you also
32269 have to define `NUM_MODES_FOR_MODE_SWITCHING', `MODE_NEEDED',
32270 `MODE_PRIORITY_TO_MODE' and `EMIT_MODE_SET'. `MODE_AFTER',
32271 `MODE_ENTRY', and `MODE_EXIT' are optional.
32273 -- Macro: NUM_MODES_FOR_MODE_SWITCHING
32274 If you define `OPTIMIZE_MODE_SWITCHING', you have to define this as
32275 initializer for an array of integers. Each initializer element N
32276 refers to an entity that needs mode switching, and specifies the
32277 number of different modes that might need to be set for this
32278 entity. The position of the initializer in the
32279 initializer--starting counting at zero--determines the integer
32280 that is used to refer to the mode-switched entity in question. In
32281 macros that take mode arguments / yield a mode result, modes are
32282 represented as numbers 0 ... N - 1. N is used to specify that no
32283 mode switch is needed / supplied.
32285 -- Macro: MODE_NEEDED (ENTITY, INSN)
32286 ENTITY is an integer specifying a mode-switched entity. If
32287 `OPTIMIZE_MODE_SWITCHING' is defined, you must define this macro to
32288 return an integer value not larger than the corresponding element
32289 in `NUM_MODES_FOR_MODE_SWITCHING', to denote the mode that ENTITY
32290 must be switched into prior to the execution of INSN.
32292 -- Macro: MODE_AFTER (MODE, INSN)
32293 If this macro is defined, it is evaluated for every INSN during
32294 mode switching. It determines the mode that an insn results in (if
32295 different from the incoming mode).
32297 -- Macro: MODE_ENTRY (ENTITY)
32298 If this macro is defined, it is evaluated for every ENTITY that
32299 needs mode switching. It should evaluate to an integer, which is
32300 a mode that ENTITY is assumed to be switched to at function entry.
32301 If `MODE_ENTRY' is defined then `MODE_EXIT' must be defined.
32303 -- Macro: MODE_EXIT (ENTITY)
32304 If this macro is defined, it is evaluated for every ENTITY that
32305 needs mode switching. It should evaluate to an integer, which is
32306 a mode that ENTITY is assumed to be switched to at function exit.
32307 If `MODE_EXIT' is defined then `MODE_ENTRY' must be defined.
32309 -- Macro: MODE_PRIORITY_TO_MODE (ENTITY, N)
32310 This macro specifies the order in which modes for ENTITY are
32311 processed. 0 is the highest priority,
32312 `NUM_MODES_FOR_MODE_SWITCHING[ENTITY] - 1' the lowest. The value
32313 of the macro should be an integer designating a mode for ENTITY.
32314 For any fixed ENTITY, `mode_priority_to_mode' (ENTITY, N) shall be
32315 a bijection in 0 ... `num_modes_for_mode_switching[ENTITY] - 1'.
32317 -- Macro: EMIT_MODE_SET (ENTITY, MODE, HARD_REGS_LIVE)
32318 Generate one or more insns to set ENTITY to MODE. HARD_REG_LIVE
32319 is the set of hard registers live at the point where the insn(s)
32320 are to be inserted.
32323 File: gccint.info, Node: Target Attributes, Next: Emulated TLS, Prev: Mode Switching, Up: Target Macros
32325 17.25 Defining target-specific uses of `__attribute__'
32326 ======================================================
32328 Target-specific attributes may be defined for functions, data and types.
32329 These are described using the following target hooks; they also need to
32330 be documented in `extend.texi'.
32332 -- Target Hook: const struct attribute_spec * TARGET_ATTRIBUTE_TABLE
32333 If defined, this target hook points to an array of `struct
32334 attribute_spec' (defined in `tree.h') specifying the machine
32335 specific attributes for this target and some of the restrictions
32336 on the entities to which these attributes are applied and the
32337 arguments they take.
32339 -- Target Hook: int TARGET_COMP_TYPE_ATTRIBUTES (tree TYPE1, tree
32341 If defined, this target hook is a function which returns zero if
32342 the attributes on TYPE1 and TYPE2 are incompatible, one if they
32343 are compatible, and two if they are nearly compatible (which
32344 causes a warning to be generated). If this is not defined,
32345 machine-specific attributes are supposed always to be compatible.
32347 -- Target Hook: void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree TYPE)
32348 If defined, this target hook is a function which assigns default
32349 attributes to newly defined TYPE.
32351 -- Target Hook: tree TARGET_MERGE_TYPE_ATTRIBUTES (tree TYPE1, tree
32353 Define this target hook if the merging of type attributes needs
32354 special handling. If defined, the result is a list of the combined
32355 `TYPE_ATTRIBUTES' of TYPE1 and TYPE2. It is assumed that
32356 `comptypes' has already been called and returned 1. This function
32357 may call `merge_attributes' to handle machine-independent merging.
32359 -- Target Hook: tree TARGET_MERGE_DECL_ATTRIBUTES (tree OLDDECL, tree
32361 Define this target hook if the merging of decl attributes needs
32362 special handling. If defined, the result is a list of the combined
32363 `DECL_ATTRIBUTES' of OLDDECL and NEWDECL. NEWDECL is a duplicate
32364 declaration of OLDDECL. Examples of when this is needed are when
32365 one attribute overrides another, or when an attribute is nullified
32366 by a subsequent definition. This function may call
32367 `merge_attributes' to handle machine-independent merging.
32369 If the only target-specific handling you require is `dllimport'
32370 for Microsoft Windows targets, you should define the macro
32371 `TARGET_DLLIMPORT_DECL_ATTRIBUTES' to `1'. The compiler will then
32372 define a function called `merge_dllimport_decl_attributes' which
32373 can then be defined as the expansion of
32374 `TARGET_MERGE_DECL_ATTRIBUTES'. You can also add
32375 `handle_dll_attribute' in the attribute table for your port to
32376 perform initial processing of the `dllimport' and `dllexport'
32377 attributes. This is done in `i386/cygwin.h' and `i386/i386.c',
32380 -- Target Hook: bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (tree DECL)
32381 DECL is a variable or function with `__attribute__((dllimport))'
32382 specified. Use this hook if the target needs to add extra
32383 validation checks to `handle_dll_attribute'.
32385 -- Macro: TARGET_DECLSPEC
32386 Define this macro to a nonzero value if you want to treat
32387 `__declspec(X)' as equivalent to `__attribute((X))'. By default,
32388 this behavior is enabled only for targets that define
32389 `TARGET_DLLIMPORT_DECL_ATTRIBUTES'. The current implementation of
32390 `__declspec' is via a built-in macro, but you should not rely on
32391 this implementation detail.
32393 -- Target Hook: void TARGET_INSERT_ATTRIBUTES (tree NODE, tree
32395 Define this target hook if you want to be able to add attributes
32396 to a decl when it is being created. This is normally useful for
32397 back ends which wish to implement a pragma by using the attributes
32398 which correspond to the pragma's effect. The NODE argument is the
32399 decl which is being created. The ATTR_PTR argument is a pointer
32400 to the attribute list for this decl. The list itself should not
32401 be modified, since it may be shared with other decls, but
32402 attributes may be chained on the head of the list and `*ATTR_PTR'
32403 modified to point to the new attributes, or a copy of the list may
32404 be made if further changes are needed.
32406 -- Target Hook: bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree
32408 This target hook returns `true' if it is ok to inline FNDECL into
32409 the current function, despite its having target-specific
32410 attributes, `false' otherwise. By default, if a function has a
32411 target specific attribute attached to it, it will not be inlined.
32413 -- Target Hook: bool TARGET_VALID_OPTION_ATTRIBUTE_P (tree FNDECL,
32414 tree NAME, tree ARGS, int FLAGS)
32415 This hook is called to parse the `attribute(option("..."))', and
32416 it allows the function to set different target machine compile time
32417 options for the current function that might be different than the
32418 options specified on the command line. The hook should return
32419 `true' if the options are valid.
32421 The hook should set the DECL_FUNCTION_SPECIFIC_TARGET field in the
32422 function declaration to hold a pointer to a target specific STRUCT
32423 CL_TARGET_OPTION structure.
32425 -- Target Hook: void TARGET_OPTION_SAVE (struct cl_target_option *PTR)
32426 This hook is called to save any additional target specific
32427 information in the STRUCT CL_TARGET_OPTION structure for function
32428 specific options. *Note Option file format::.
32430 -- Target Hook: void TARGET_OPTION_RESTORE (struct cl_target_option
32432 This hook is called to restore any additional target specific
32433 information in the STRUCT CL_TARGET_OPTION structure for function
32436 -- Target Hook: void TARGET_OPTION_PRINT (struct cl_target_option *PTR)
32437 This hook is called to print any additional target specific
32438 information in the STRUCT CL_TARGET_OPTION structure for function
32441 -- Target Hook: bool TARGET_OPTION_PRAGMA_PARSE (target ARGS)
32442 This target hook parses the options for `#pragma GCC option' to
32443 set the machine specific options for functions that occur later in
32444 the input stream. The options should be the same as handled by the
32445 `TARGET_VALID_OPTION_ATTRIBUTE_P' hook.
32447 -- Target Hook: bool TARGET_CAN_INLINE_P (tree CALLER, tree CALLEE)
32448 This target hook returns `false' if the CALLER function cannot
32449 inline CALLEE, based on target specific information. By default,
32450 inlining is not allowed if the callee function has function
32451 specific target options and the caller does not use the same
32455 File: gccint.info, Node: Emulated TLS, Next: MIPS Coprocessors, Prev: Target Attributes, Up: Target Macros
32457 17.26 Emulating TLS
32458 ===================
32460 For targets whose psABI does not provide Thread Local Storage via
32461 specific relocations and instruction sequences, an emulation layer is
32462 used. A set of target hooks allows this emulation layer to be
32463 configured for the requirements of a particular target. For instance
32464 the psABI may in fact specify TLS support in terms of an emulation
32467 The emulation layer works by creating a control object for every TLS
32468 object. To access the TLS object, a lookup function is provided which,
32469 when given the address of the control object, will return the address
32470 of the current thread's instance of the TLS object.
32472 -- Target Hook: const char * TARGET_EMUTLS_GET_ADDRESS
32473 Contains the name of the helper function that uses a TLS control
32474 object to locate a TLS instance. The default causes libgcc's
32475 emulated TLS helper function to be used.
32477 -- Target Hook: const char * TARGET_EMUTLS_REGISTER_COMMON
32478 Contains the name of the helper function that should be used at
32479 program startup to register TLS objects that are implicitly
32480 initialized to zero. If this is `NULL', all TLS objects will have
32481 explicit initializers. The default causes libgcc's emulated TLS
32482 registration function to be used.
32484 -- Target Hook: const char * TARGET_EMUTLS_VAR_SECTION
32485 Contains the name of the section in which TLS control variables
32486 should be placed. The default of `NULL' allows these to be placed
32489 -- Target Hook: const char * TARGET_EMUTLS_TMPL_SECTION
32490 Contains the name of the section in which TLS initializers should
32491 be placed. The default of `NULL' allows these to be placed in any
32494 -- Target Hook: const char * TARGET_EMUTLS_VAR_PREFIX
32495 Contains the prefix to be prepended to TLS control variable names.
32496 The default of `NULL' uses a target-specific prefix.
32498 -- Target Hook: const char * TARGET_EMUTLS_TMPL_PREFIX
32499 Contains the prefix to be prepended to TLS initializer objects.
32500 The default of `NULL' uses a target-specific prefix.
32502 -- Target Hook: tree TARGET_EMUTLS_VAR_FIELDS (tree TYPE, tree *NAME)
32503 Specifies a function that generates the FIELD_DECLs for a TLS
32504 control object type. TYPE is the RECORD_TYPE the fields are for
32505 and NAME should be filled with the structure tag, if the default of
32506 `__emutls_object' is unsuitable. The default creates a type
32507 suitable for libgcc's emulated TLS function.
32509 -- Target Hook: tree TARGET_EMUTLS_VAR_INIT (tree VAR, tree DECL, tree
32511 Specifies a function that generates the CONSTRUCTOR to initialize a
32512 TLS control object. VAR is the TLS control object, DECL is the
32513 TLS object and TMPL_ADDR is the address of the initializer. The
32514 default initializes libgcc's emulated TLS control object.
32516 -- Target Hook: bool TARGET_EMUTLS_VAR_ALIGN_FIXED
32517 Specifies whether the alignment of TLS control variable objects is
32518 fixed and should not be increased as some backends may do to
32519 optimize single objects. The default is false.
32521 -- Target Hook: bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
32522 Specifies whether a DWARF `DW_OP_form_tls_address' location
32523 descriptor may be used to describe emulated TLS control objects.
32526 File: gccint.info, Node: MIPS Coprocessors, Next: PCH Target, Prev: Emulated TLS, Up: Target Macros
32528 17.27 Defining coprocessor specifics for MIPS targets.
32529 ======================================================
32531 The MIPS specification allows MIPS implementations to have as many as 4
32532 coprocessors, each with as many as 32 private registers. GCC supports
32533 accessing these registers and transferring values between the registers
32534 and memory using asm-ized variables. For example:
32536 register unsigned int cp0count asm ("c0r1");
32541 ("c0r1" is the default name of register 1 in coprocessor 0; alternate
32542 names may be added as described below, or the default names may be
32543 overridden entirely in `SUBTARGET_CONDITIONAL_REGISTER_USAGE'.)
32545 Coprocessor registers are assumed to be epilogue-used; sets to them
32546 will be preserved even if it does not appear that the register is used
32547 again later in the function.
32549 Another note: according to the MIPS spec, coprocessor 1 (if present) is
32550 the FPU. One accesses COP1 registers through standard mips
32551 floating-point support; they are not included in this mechanism.
32553 There is one macro used in defining the MIPS coprocessor interface
32554 which you may want to override in subtargets; it is described below.
32556 -- Macro: ALL_COP_ADDITIONAL_REGISTER_NAMES
32557 A comma-separated list (with leading comma) of pairs describing the
32558 alternate names of coprocessor registers. The format of each
32560 { ALTERNATENAME, REGISTER_NUMBER}
32564 File: gccint.info, Node: PCH Target, Next: C++ ABI, Prev: MIPS Coprocessors, Up: Target Macros
32566 17.28 Parameters for Precompiled Header Validity Checking
32567 =========================================================
32569 -- Target Hook: void *TARGET_GET_PCH_VALIDITY (size_t *SZ)
32570 This hook returns the data needed by `TARGET_PCH_VALID_P' and sets
32571 `*SZ' to the size of the data in bytes.
32573 -- Target Hook: const char *TARGET_PCH_VALID_P (const void *DATA,
32575 This hook checks whether the options used to create a PCH file are
32576 compatible with the current settings. It returns `NULL' if so and
32577 a suitable error message if not. Error messages will be presented
32578 to the user and must be localized using `_(MSG)'.
32580 DATA is the data that was returned by `TARGET_GET_PCH_VALIDITY'
32581 when the PCH file was created and SZ is the size of that data in
32582 bytes. It's safe to assume that the data was created by the same
32583 version of the compiler, so no format checking is needed.
32585 The default definition of `default_pch_valid_p' should be suitable
32588 -- Target Hook: const char *TARGET_CHECK_PCH_TARGET_FLAGS (int
32590 If this hook is nonnull, the default implementation of
32591 `TARGET_PCH_VALID_P' will use it to check for compatible values of
32592 `target_flags'. PCH_FLAGS specifies the value that `target_flags'
32593 had when the PCH file was created. The return value is the same
32594 as for `TARGET_PCH_VALID_P'.
32597 File: gccint.info, Node: C++ ABI, Next: Misc, Prev: PCH Target, Up: Target Macros
32599 17.29 C++ ABI parameters
32600 ========================
32602 -- Target Hook: tree TARGET_CXX_GUARD_TYPE (void)
32603 Define this hook to override the integer type used for guard
32604 variables. These are used to implement one-time construction of
32605 static objects. The default is long_long_integer_type_node.
32607 -- Target Hook: bool TARGET_CXX_GUARD_MASK_BIT (void)
32608 This hook determines how guard variables are used. It should
32609 return `false' (the default) if first byte should be used. A
32610 return value of `true' indicates the least significant bit should
32613 -- Target Hook: tree TARGET_CXX_GET_COOKIE_SIZE (tree TYPE)
32614 This hook returns the size of the cookie to use when allocating an
32615 array whose elements have the indicated TYPE. Assumes that it is
32616 already known that a cookie is needed. The default is `max(sizeof
32617 (size_t), alignof(type))', as defined in section 2.7 of the
32618 IA64/Generic C++ ABI.
32620 -- Target Hook: bool TARGET_CXX_COOKIE_HAS_SIZE (void)
32621 This hook should return `true' if the element size should be
32622 stored in array cookies. The default is to return `false'.
32624 -- Target Hook: int TARGET_CXX_IMPORT_EXPORT_CLASS (tree TYPE, int
32626 If defined by a backend this hook allows the decision made to
32627 export class TYPE to be overruled. Upon entry IMPORT_EXPORT will
32628 contain 1 if the class is going to be exported, -1 if it is going
32629 to be imported and 0 otherwise. This function should return the
32630 modified value and perform any other actions necessary to support
32631 the backend's targeted operating system.
32633 -- Target Hook: bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
32634 This hook should return `true' if constructors and destructors
32635 return the address of the object created/destroyed. The default
32636 is to return `false'.
32638 -- Target Hook: bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
32639 This hook returns true if the key method for a class (i.e., the
32640 method which, if defined in the current translation unit, causes
32641 the virtual table to be emitted) may be an inline function. Under
32642 the standard Itanium C++ ABI the key method may be an inline
32643 function so long as the function is not declared inline in the
32644 class definition. Under some variants of the ABI, an inline
32645 function can never be the key method. The default is to return
32648 -- Target Hook: void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree
32650 DECL is a virtual table, virtual table table, typeinfo object, or
32651 other similar implicit class data object that will be emitted with
32652 external linkage in this translation unit. No ELF visibility has
32653 been explicitly specified. If the target needs to specify a
32654 visibility other than that of the containing class, use this hook
32655 to set `DECL_VISIBILITY' and `DECL_VISIBILITY_SPECIFIED'.
32657 -- Target Hook: bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
32658 This hook returns true (the default) if virtual tables and other
32659 similar implicit class data objects are always COMDAT if they have
32660 external linkage. If this hook returns false, then class data for
32661 classes whose virtual table will be emitted in only one translation
32662 unit will not be COMDAT.
32664 -- Target Hook: bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)
32665 This hook returns true (the default) if the RTTI information for
32666 the basic types which is defined in the C++ runtime should always
32667 be COMDAT, false if it should not be COMDAT.
32669 -- Target Hook: bool TARGET_CXX_USE_AEABI_ATEXIT (void)
32670 This hook returns true if `__aeabi_atexit' (as defined by the ARM
32671 EABI) should be used to register static destructors when
32672 `-fuse-cxa-atexit' is in effect. The default is to return false
32673 to use `__cxa_atexit'.
32675 -- Target Hook: bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)
32676 This hook returns true if the target `atexit' function can be used
32677 in the same manner as `__cxa_atexit' to register C++ static
32678 destructors. This requires that `atexit'-registered functions in
32679 shared libraries are run in the correct order when the libraries
32680 are unloaded. The default is to return false.
32682 -- Target Hook: void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree TYPE)
32683 TYPE is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has
32684 just been defined. Use this hook to make adjustments to the class
32685 (eg, tweak visibility or perform any other required target
32689 File: gccint.info, Node: Misc, Prev: C++ ABI, Up: Target Macros
32691 17.30 Miscellaneous Parameters
32692 ==============================
32694 Here are several miscellaneous parameters.
32696 -- Macro: HAS_LONG_COND_BRANCH
32697 Define this boolean macro to indicate whether or not your
32698 architecture has conditional branches that can span all of memory.
32699 It is used in conjunction with an optimization that partitions
32700 hot and cold basic blocks into separate sections of the
32701 executable. If this macro is set to false, gcc will convert any
32702 conditional branches that attempt to cross between sections into
32703 unconditional branches or indirect jumps.
32705 -- Macro: HAS_LONG_UNCOND_BRANCH
32706 Define this boolean macro to indicate whether or not your
32707 architecture has unconditional branches that can span all of
32708 memory. It is used in conjunction with an optimization that
32709 partitions hot and cold basic blocks into separate sections of the
32710 executable. If this macro is set to false, gcc will convert any
32711 unconditional branches that attempt to cross between sections into
32714 -- Macro: CASE_VECTOR_MODE
32715 An alias for a machine mode name. This is the machine mode that
32716 elements of a jump-table should have.
32718 -- Macro: CASE_VECTOR_SHORTEN_MODE (MIN_OFFSET, MAX_OFFSET, BODY)
32719 Optional: return the preferred mode for an `addr_diff_vec' when
32720 the minimum and maximum offset are known. If you define this, it
32721 enables extra code in branch shortening to deal with
32722 `addr_diff_vec'. To make this work, you also have to define
32723 `INSN_ALIGN' and make the alignment for `addr_diff_vec' explicit.
32724 The BODY argument is provided so that the offset_unsigned and scale
32725 flags can be updated.
32727 -- Macro: CASE_VECTOR_PC_RELATIVE
32728 Define this macro to be a C expression to indicate when jump-tables
32729 should contain relative addresses. You need not define this macro
32730 if jump-tables never contain relative addresses, or jump-tables
32731 should contain relative addresses only when `-fPIC' or `-fPIC' is
32734 -- Macro: CASE_VALUES_THRESHOLD
32735 Define this to be the smallest number of different values for
32736 which it is best to use a jump-table instead of a tree of
32737 conditional branches. The default is four for machines with a
32738 `casesi' instruction and five otherwise. This is best for most
32741 -- Macro: CASE_USE_BIT_TESTS
32742 Define this macro to be a C expression to indicate whether C switch
32743 statements may be implemented by a sequence of bit tests. This is
32744 advantageous on processors that can efficiently implement left
32745 shift of 1 by the number of bits held in a register, but
32746 inappropriate on targets that would require a loop. By default,
32747 this macro returns `true' if the target defines an `ashlsi3'
32748 pattern, and `false' otherwise.
32750 -- Macro: WORD_REGISTER_OPERATIONS
32751 Define this macro if operations between registers with integral
32752 mode smaller than a word are always performed on the entire
32753 register. Most RISC machines have this property and most CISC
32756 -- Macro: LOAD_EXTEND_OP (MEM_MODE)
32757 Define this macro to be a C expression indicating when insns that
32758 read memory in MEM_MODE, an integral mode narrower than a word,
32759 set the bits outside of MEM_MODE to be either the sign-extension
32760 or the zero-extension of the data read. Return `SIGN_EXTEND' for
32761 values of MEM_MODE for which the insn sign-extends, `ZERO_EXTEND'
32762 for which it zero-extends, and `UNKNOWN' for other modes.
32764 This macro is not called with MEM_MODE non-integral or with a width
32765 greater than or equal to `BITS_PER_WORD', so you may return any
32766 value in this case. Do not define this macro if it would always
32767 return `UNKNOWN'. On machines where this macro is defined, you
32768 will normally define it as the constant `SIGN_EXTEND' or
32771 You may return a non-`UNKNOWN' value even if for some hard
32772 registers the sign extension is not performed, if for the
32773 `REGNO_REG_CLASS' of these hard registers
32774 `CANNOT_CHANGE_MODE_CLASS' returns nonzero when the FROM mode is
32775 MEM_MODE and the TO mode is any integral mode larger than this but
32776 not larger than `word_mode'.
32778 You must return `UNKNOWN' if for some hard registers that allow
32779 this mode, `CANNOT_CHANGE_MODE_CLASS' says that they cannot change
32780 to `word_mode', but that they can change to another integral mode
32781 that is larger then MEM_MODE but still smaller than `word_mode'.
32783 -- Macro: SHORT_IMMEDIATES_SIGN_EXTEND
32784 Define this macro if loading short immediate values into registers
32787 -- Macro: FIXUNS_TRUNC_LIKE_FIX_TRUNC
32788 Define this macro if the same instructions that convert a floating
32789 point number to a signed fixed point number also convert validly
32790 to an unsigned one.
32792 -- Target Hook: int TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum
32794 When `-ffast-math' is in effect, GCC tries to optimize divisions
32795 by the same divisor, by turning them into multiplications by the
32796 reciprocal. This target hook specifies the minimum number of
32797 divisions that should be there for GCC to perform the optimization
32798 for a variable of mode MODE. The default implementation returns 3
32799 if the machine has an instruction for the division, and 2 if it
32803 The maximum number of bytes that a single instruction can move
32804 quickly between memory and registers or between two memory
32807 -- Macro: MAX_MOVE_MAX
32808 The maximum number of bytes that a single instruction can move
32809 quickly between memory and registers or between two memory
32810 locations. If this is undefined, the default is `MOVE_MAX'.
32811 Otherwise, it is the constant value that is the largest value that
32812 `MOVE_MAX' can have at run-time.
32814 -- Macro: SHIFT_COUNT_TRUNCATED
32815 A C expression that is nonzero if on this machine the number of
32816 bits actually used for the count of a shift operation is equal to
32817 the number of bits needed to represent the size of the object
32818 being shifted. When this macro is nonzero, the compiler will
32819 assume that it is safe to omit a sign-extend, zero-extend, and
32820 certain bitwise `and' instructions that truncates the count of a
32821 shift operation. On machines that have instructions that act on
32822 bit-fields at variable positions, which may include `bit test'
32823 instructions, a nonzero `SHIFT_COUNT_TRUNCATED' also enables
32824 deletion of truncations of the values that serve as arguments to
32825 bit-field instructions.
32827 If both types of instructions truncate the count (for shifts) and
32828 position (for bit-field operations), or if no variable-position
32829 bit-field instructions exist, you should define this macro.
32831 However, on some machines, such as the 80386 and the 680x0,
32832 truncation only applies to shift operations and not the (real or
32833 pretended) bit-field operations. Define `SHIFT_COUNT_TRUNCATED'
32834 to be zero on such machines. Instead, add patterns to the `md'
32835 file that include the implied truncation of the shift instructions.
32837 You need not define this macro if it would always have the value
32840 -- Target Hook: int TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode
32842 This function describes how the standard shift patterns for MODE
32843 deal with shifts by negative amounts or by more than the width of
32844 the mode. *Note shift patterns::.
32846 On many machines, the shift patterns will apply a mask M to the
32847 shift count, meaning that a fixed-width shift of X by Y is
32848 equivalent to an arbitrary-width shift of X by Y & M. If this is
32849 true for mode MODE, the function should return M, otherwise it
32850 should return 0. A return value of 0 indicates that no particular
32851 behavior is guaranteed.
32853 Note that, unlike `SHIFT_COUNT_TRUNCATED', this function does
32854 _not_ apply to general shift rtxes; it applies only to instructions
32855 that are generated by the named shift patterns.
32857 The default implementation of this function returns
32858 `GET_MODE_BITSIZE (MODE) - 1' if `SHIFT_COUNT_TRUNCATED' and 0
32859 otherwise. This definition is always safe, but if
32860 `SHIFT_COUNT_TRUNCATED' is false, and some shift patterns
32861 nevertheless truncate the shift count, you may get better code by
32864 -- Macro: TRULY_NOOP_TRUNCATION (OUTPREC, INPREC)
32865 A C expression which is nonzero if on this machine it is safe to
32866 "convert" an integer of INPREC bits to one of OUTPREC bits (where
32867 OUTPREC is smaller than INPREC) by merely operating on it as if it
32868 had only OUTPREC bits.
32870 On many machines, this expression can be 1.
32872 When `TRULY_NOOP_TRUNCATION' returns 1 for a pair of sizes for
32873 modes for which `MODES_TIEABLE_P' is 0, suboptimal code can result.
32874 If this is the case, making `TRULY_NOOP_TRUNCATION' return 0 in
32875 such cases may improve things.
32877 -- Target Hook: int TARGET_MODE_REP_EXTENDED (enum machine_mode MODE,
32878 enum machine_mode REP_MODE)
32879 The representation of an integral mode can be such that the values
32880 are always extended to a wider integral mode. Return
32881 `SIGN_EXTEND' if values of MODE are represented in sign-extended
32882 form to REP_MODE. Return `UNKNOWN' otherwise. (Currently, none
32883 of the targets use zero-extended representation this way so unlike
32884 `LOAD_EXTEND_OP', `TARGET_MODE_REP_EXTENDED' is expected to return
32885 either `SIGN_EXTEND' or `UNKNOWN'. Also no target extends MODE to
32886 MODE_REP so that MODE_REP is not the next widest integral mode and
32887 currently we take advantage of this fact.)
32889 Similarly to `LOAD_EXTEND_OP' you may return a non-`UNKNOWN' value
32890 even if the extension is not performed on certain hard registers
32891 as long as for the `REGNO_REG_CLASS' of these hard registers
32892 `CANNOT_CHANGE_MODE_CLASS' returns nonzero.
32894 Note that `TARGET_MODE_REP_EXTENDED' and `LOAD_EXTEND_OP' describe
32895 two related properties. If you define `TARGET_MODE_REP_EXTENDED
32896 (mode, word_mode)' you probably also want to define
32897 `LOAD_EXTEND_OP (mode)' to return the same type of extension.
32899 In order to enforce the representation of `mode',
32900 `TRULY_NOOP_TRUNCATION' should return false when truncating to
32903 -- Macro: STORE_FLAG_VALUE
32904 A C expression describing the value returned by a comparison
32905 operator with an integral mode and stored by a store-flag
32906 instruction (`sCOND') when the condition is true. This
32907 description must apply to _all_ the `sCOND' patterns and all the
32908 comparison operators whose results have a `MODE_INT' mode.
32910 A value of 1 or -1 means that the instruction implementing the
32911 comparison operator returns exactly 1 or -1 when the comparison is
32912 true and 0 when the comparison is false. Otherwise, the value
32913 indicates which bits of the result are guaranteed to be 1 when the
32914 comparison is true. This value is interpreted in the mode of the
32915 comparison operation, which is given by the mode of the first
32916 operand in the `sCOND' pattern. Either the low bit or the sign
32917 bit of `STORE_FLAG_VALUE' be on. Presently, only those bits are
32918 used by the compiler.
32920 If `STORE_FLAG_VALUE' is neither 1 or -1, the compiler will
32921 generate code that depends only on the specified bits. It can also
32922 replace comparison operators with equivalent operations if they
32923 cause the required bits to be set, even if the remaining bits are
32924 undefined. For example, on a machine whose comparison operators
32925 return an `SImode' value and where `STORE_FLAG_VALUE' is defined as
32926 `0x80000000', saying that just the sign bit is relevant, the
32929 (ne:SI (and:SI X (const_int POWER-OF-2)) (const_int 0))
32931 can be converted to
32933 (ashift:SI X (const_int N))
32935 where N is the appropriate shift count to move the bit being
32936 tested into the sign bit.
32938 There is no way to describe a machine that always sets the
32939 low-order bit for a true value, but does not guarantee the value
32940 of any other bits, but we do not know of any machine that has such
32941 an instruction. If you are trying to port GCC to such a machine,
32942 include an instruction to perform a logical-and of the result with
32943 1 in the pattern for the comparison operators and let us know at
32946 Often, a machine will have multiple instructions that obtain a
32947 value from a comparison (or the condition codes). Here are rules
32948 to guide the choice of value for `STORE_FLAG_VALUE', and hence the
32949 instructions to be used:
32951 * Use the shortest sequence that yields a valid definition for
32952 `STORE_FLAG_VALUE'. It is more efficient for the compiler to
32953 "normalize" the value (convert it to, e.g., 1 or 0) than for
32954 the comparison operators to do so because there may be
32955 opportunities to combine the normalization with other
32958 * For equal-length sequences, use a value of 1 or -1, with -1
32959 being slightly preferred on machines with expensive jumps and
32960 1 preferred on other machines.
32962 * As a second choice, choose a value of `0x80000001' if
32963 instructions exist that set both the sign and low-order bits
32964 but do not define the others.
32966 * Otherwise, use a value of `0x80000000'.
32968 Many machines can produce both the value chosen for
32969 `STORE_FLAG_VALUE' and its negation in the same number of
32970 instructions. On those machines, you should also define a pattern
32971 for those cases, e.g., one matching
32973 (set A (neg:M (ne:M B C)))
32975 Some machines can also perform `and' or `plus' operations on
32976 condition code values with less instructions than the corresponding
32977 `sCOND' insn followed by `and' or `plus'. On those machines,
32978 define the appropriate patterns. Use the names `incscc' and
32979 `decscc', respectively, for the patterns which perform `plus' or
32980 `minus' operations on condition code values. See `rs6000.md' for
32981 some examples. The GNU Superoptizer can be used to find such
32982 instruction sequences on other machines.
32984 If this macro is not defined, the default value, 1, is used. You
32985 need not define `STORE_FLAG_VALUE' if the machine has no store-flag
32986 instructions, or if the value generated by these instructions is 1.
32988 -- Macro: FLOAT_STORE_FLAG_VALUE (MODE)
32989 A C expression that gives a nonzero `REAL_VALUE_TYPE' value that is
32990 returned when comparison operators with floating-point results are
32991 true. Define this macro on machines that have comparison
32992 operations that return floating-point values. If there are no
32993 such operations, do not define this macro.
32995 -- Macro: VECTOR_STORE_FLAG_VALUE (MODE)
32996 A C expression that gives a rtx representing the nonzero true
32997 element for vector comparisons. The returned rtx should be valid
32998 for the inner mode of MODE which is guaranteed to be a vector
32999 mode. Define this macro on machines that have vector comparison
33000 operations that return a vector result. If there are no such
33001 operations, do not define this macro. Typically, this macro is
33002 defined as `const1_rtx' or `constm1_rtx'. This macro may return
33003 `NULL_RTX' to prevent the compiler optimizing such vector
33004 comparison operations for the given mode.
33006 -- Macro: CLZ_DEFINED_VALUE_AT_ZERO (MODE, VALUE)
33007 -- Macro: CTZ_DEFINED_VALUE_AT_ZERO (MODE, VALUE)
33008 A C expression that indicates whether the architecture defines a
33009 value for `clz' or `ctz' with a zero operand. A result of `0'
33010 indicates the value is undefined. If the value is defined for
33011 only the RTL expression, the macro should evaluate to `1'; if the
33012 value applies also to the corresponding optab entry (which is
33013 normally the case if it expands directly into the corresponding
33014 RTL), then the macro should evaluate to `2'. In the cases where
33015 the value is defined, VALUE should be set to this value.
33017 If this macro is not defined, the value of `clz' or `ctz' at zero
33018 is assumed to be undefined.
33020 This macro must be defined if the target's expansion for `ffs'
33021 relies on a particular value to get correct results. Otherwise it
33022 is not necessary, though it may be used to optimize some corner
33023 cases, and to provide a default expansion for the `ffs' optab.
33025 Note that regardless of this macro the "definedness" of `clz' and
33026 `ctz' at zero do _not_ extend to the builtin functions visible to
33027 the user. Thus one may be free to adjust the value at will to
33028 match the target expansion of these operations without fear of
33032 An alias for the machine mode for pointers. On most machines,
33033 define this to be the integer mode corresponding to the width of a
33034 hardware pointer; `SImode' on 32-bit machine or `DImode' on 64-bit
33035 machines. On some machines you must define this to be one of the
33036 partial integer modes, such as `PSImode'.
33038 The width of `Pmode' must be at least as large as the value of
33039 `POINTER_SIZE'. If it is not equal, you must define the macro
33040 `POINTERS_EXTEND_UNSIGNED' to specify how pointers are extended to
33043 -- Macro: FUNCTION_MODE
33044 An alias for the machine mode used for memory references to
33045 functions being called, in `call' RTL expressions. On most CISC
33046 machines, where an instruction can begin at any byte address, this
33047 should be `QImode'. On most RISC machines, where all instructions
33048 have fixed size and alignment, this should be a mode with the same
33049 size and alignment as the machine instruction words - typically
33050 `SImode' or `HImode'.
33052 -- Macro: STDC_0_IN_SYSTEM_HEADERS
33053 In normal operation, the preprocessor expands `__STDC__' to the
33054 constant 1, to signify that GCC conforms to ISO Standard C. On
33055 some hosts, like Solaris, the system compiler uses a different
33056 convention, where `__STDC__' is normally 0, but is 1 if the user
33057 specifies strict conformance to the C Standard.
33059 Defining `STDC_0_IN_SYSTEM_HEADERS' makes GNU CPP follows the host
33060 convention when processing system header files, but when
33061 processing user files `__STDC__' will always expand to 1.
33063 -- Macro: NO_IMPLICIT_EXTERN_C
33064 Define this macro if the system header files support C++ as well
33065 as C. This macro inhibits the usual method of using system header
33066 files in C++, which is to pretend that the file's contents are
33067 enclosed in `extern "C" {...}'.
33069 -- Macro: REGISTER_TARGET_PRAGMAS ()
33070 Define this macro if you want to implement any target-specific
33071 pragmas. If defined, it is a C expression which makes a series of
33072 calls to `c_register_pragma' or `c_register_pragma_with_expansion'
33073 for each pragma. The macro may also do any setup required for the
33076 The primary reason to define this macro is to provide
33077 compatibility with other compilers for the same target. In
33078 general, we discourage definition of target-specific pragmas for
33081 If the pragma can be implemented by attributes then you should
33082 consider defining the target hook `TARGET_INSERT_ATTRIBUTES' as
33085 Preprocessor macros that appear on pragma lines are not expanded.
33086 All `#pragma' directives that do not match any registered pragma
33087 are silently ignored, unless the user specifies
33088 `-Wunknown-pragmas'.
33090 -- Function: void c_register_pragma (const char *SPACE, const char
33091 *NAME, void (*CALLBACK) (struct cpp_reader *))
33092 -- Function: void c_register_pragma_with_expansion (const char *SPACE,
33093 const char *NAME, void (*CALLBACK) (struct cpp_reader *))
33094 Each call to `c_register_pragma' or
33095 `c_register_pragma_with_expansion' establishes one pragma. The
33096 CALLBACK routine will be called when the preprocessor encounters a
33099 #pragma [SPACE] NAME ...
33101 SPACE is the case-sensitive namespace of the pragma, or `NULL' to
33102 put the pragma in the global namespace. The callback routine
33103 receives PFILE as its first argument, which can be passed on to
33104 cpplib's functions if necessary. You can lex tokens after the
33105 NAME by calling `pragma_lex'. Tokens that are not read by the
33106 callback will be silently ignored. The end of the line is
33107 indicated by a token of type `CPP_EOF'. Macro expansion occurs on
33108 the arguments of pragmas registered with
33109 `c_register_pragma_with_expansion' but not on the arguments of
33110 pragmas registered with `c_register_pragma'.
33112 Note that the use of `pragma_lex' is specific to the C and C++
33113 compilers. It will not work in the Java or Fortran compilers, or
33114 any other language compilers for that matter. Thus if
33115 `pragma_lex' is going to be called from target-specific code, it
33116 must only be done so when building the C and C++ compilers. This
33117 can be done by defining the variables `c_target_objs' and
33118 `cxx_target_objs' in the target entry in the `config.gcc' file.
33119 These variables should name the target-specific, language-specific
33120 object file which contains the code that uses `pragma_lex'. Note
33121 it will also be necessary to add a rule to the makefile fragment
33122 pointed to by `tmake_file' that shows how to build this object
33125 -- Macro: HANDLE_SYSV_PRAGMA
33126 Define this macro (to a value of 1) if you want the System V style
33127 pragmas `#pragma pack(<n>)' and `#pragma weak <name> [=<value>]'
33128 to be supported by gcc.
33130 The pack pragma specifies the maximum alignment (in bytes) of
33131 fields within a structure, in much the same way as the
33132 `__aligned__' and `__packed__' `__attribute__'s do. A pack value
33133 of zero resets the behavior to the default.
33135 A subtlety for Microsoft Visual C/C++ style bit-field packing
33136 (e.g. -mms-bitfields) for targets that support it: When a
33137 bit-field is inserted into a packed record, the whole size of the
33138 underlying type is used by one or more same-size adjacent
33139 bit-fields (that is, if its long:3, 32 bits is used in the record,
33140 and any additional adjacent long bit-fields are packed into the
33141 same chunk of 32 bits. However, if the size changes, a new field
33142 of that size is allocated).
33144 If both MS bit-fields and `__attribute__((packed))' are used, the
33145 latter will take precedence. If `__attribute__((packed))' is used
33146 on a single field when MS bit-fields are in use, it will take
33147 precedence for that field, but the alignment of the rest of the
33148 structure may affect its placement.
33150 The weak pragma only works if `SUPPORTS_WEAK' and
33151 `ASM_WEAKEN_LABEL' are defined. If enabled it allows the creation
33152 of specifically named weak labels, optionally with a value.
33154 -- Macro: HANDLE_PRAGMA_PACK_PUSH_POP
33155 Define this macro (to a value of 1) if you want to support the
33156 Win32 style pragmas `#pragma pack(push[,N])' and `#pragma
33157 pack(pop)'. The `pack(push,[N])' pragma specifies the maximum
33158 alignment (in bytes) of fields within a structure, in much the
33159 same way as the `__aligned__' and `__packed__' `__attribute__'s
33160 do. A pack value of zero resets the behavior to the default.
33161 Successive invocations of this pragma cause the previous values to
33162 be stacked, so that invocations of `#pragma pack(pop)' will return
33163 to the previous value.
33165 -- Macro: HANDLE_PRAGMA_PACK_WITH_EXPANSION
33166 Define this macro, as well as `HANDLE_SYSV_PRAGMA', if macros
33167 should be expanded in the arguments of `#pragma pack'.
33169 -- Macro: TARGET_DEFAULT_PACK_STRUCT
33170 If your target requires a structure packing default other than 0
33171 (meaning the machine default), define this macro to the necessary
33172 value (in bytes). This must be a value that would also be valid
33173 to use with `#pragma pack()' (that is, a small power of two).
33175 -- Macro: HANDLE_PRAGMA_PUSH_POP_MACRO
33176 Define this macro if you want to support the Win32 style pragmas
33177 `#pragma push_macro(macro-name-as-string)' and `#pragma
33178 pop_macro(macro-name-as-string)'. The `#pragma push_macro(
33179 macro-name-as-string)' pragma saves the named macro and via
33180 `#pragma pop_macro(macro-name-as-string)' it will return to the
33183 -- Macro: DOLLARS_IN_IDENTIFIERS
33184 Define this macro to control use of the character `$' in
33185 identifier names for the C family of languages. 0 means `$' is
33186 not allowed by default; 1 means it is allowed. 1 is the default;
33187 there is no need to define this macro in that case.
33189 -- Macro: NO_DOLLAR_IN_LABEL
33190 Define this macro if the assembler does not accept the character
33191 `$' in label names. By default constructors and destructors in
33192 G++ have `$' in the identifiers. If this macro is defined, `.' is
33195 -- Macro: NO_DOT_IN_LABEL
33196 Define this macro if the assembler does not accept the character
33197 `.' in label names. By default constructors and destructors in G++
33198 have names that use `.'. If this macro is defined, these names
33199 are rewritten to avoid `.'.
33201 -- Macro: INSN_SETS_ARE_DELAYED (INSN)
33202 Define this macro as a C expression that is nonzero if it is safe
33203 for the delay slot scheduler to place instructions in the delay
33204 slot of INSN, even if they appear to use a resource set or
33205 clobbered in INSN. INSN is always a `jump_insn' or an `insn'; GCC
33206 knows that every `call_insn' has this behavior. On machines where
33207 some `insn' or `jump_insn' is really a function call and hence has
33208 this behavior, you should define this macro.
33210 You need not define this macro if it would always return zero.
33212 -- Macro: INSN_REFERENCES_ARE_DELAYED (INSN)
33213 Define this macro as a C expression that is nonzero if it is safe
33214 for the delay slot scheduler to place instructions in the delay
33215 slot of INSN, even if they appear to set or clobber a resource
33216 referenced in INSN. INSN is always a `jump_insn' or an `insn'.
33217 On machines where some `insn' or `jump_insn' is really a function
33218 call and its operands are registers whose use is actually in the
33219 subroutine it calls, you should define this macro. Doing so
33220 allows the delay slot scheduler to move instructions which copy
33221 arguments into the argument registers into the delay slot of INSN.
33223 You need not define this macro if it would always return zero.
33225 -- Macro: MULTIPLE_SYMBOL_SPACES
33226 Define this macro as a C expression that is nonzero if, in some
33227 cases, global symbols from one translation unit may not be bound
33228 to undefined symbols in another translation unit without user
33229 intervention. For instance, under Microsoft Windows symbols must
33230 be explicitly imported from shared libraries (DLLs).
33232 You need not define this macro if it would always evaluate to zero.
33234 -- Target Hook: tree TARGET_MD_ASM_CLOBBERS (tree OUTPUTS, tree
33235 INPUTS, tree CLOBBERS)
33236 This target hook should add to CLOBBERS `STRING_CST' trees for any
33237 hard regs the port wishes to automatically clobber for an asm. It
33238 should return the result of the last `tree_cons' used to add a
33239 clobber. The OUTPUTS, INPUTS and CLOBBER lists are the
33240 corresponding parameters to the asm and may be inspected to avoid
33241 clobbering a register that is an input or output of the asm. You
33242 can use `tree_overlaps_hard_reg_set', declared in `tree.h', to test
33243 for overlap with regards to asm-declared registers.
33245 -- Macro: MATH_LIBRARY
33246 Define this macro as a C string constant for the linker argument
33247 to link in the system math library, or `""' if the target does not
33248 have a separate math library.
33250 You need only define this macro if the default of `"-lm"' is wrong.
33252 -- Macro: LIBRARY_PATH_ENV
33253 Define this macro as a C string constant for the environment
33254 variable that specifies where the linker should look for libraries.
33256 You need only define this macro if the default of `"LIBRARY_PATH"'
33259 -- Macro: TARGET_POSIX_IO
33260 Define this macro if the target supports the following POSIX file
33261 functions, access, mkdir and file locking with fcntl / F_SETLKW.
33262 Defining `TARGET_POSIX_IO' will enable the test coverage code to
33263 use file locking when exiting a program, which avoids race
33264 conditions if the program has forked. It will also create
33265 directories at run-time for cross-profiling.
33267 -- Macro: MAX_CONDITIONAL_EXECUTE
33268 A C expression for the maximum number of instructions to execute
33269 via conditional execution instructions instead of a branch. A
33270 value of `BRANCH_COST'+1 is the default if the machine does not
33271 use cc0, and 1 if it does use cc0.
33273 -- Macro: IFCVT_MODIFY_TESTS (CE_INFO, TRUE_EXPR, FALSE_EXPR)
33274 Used if the target needs to perform machine-dependent
33275 modifications on the conditionals used for turning basic blocks
33276 into conditionally executed code. CE_INFO points to a data
33277 structure, `struct ce_if_block', which contains information about
33278 the currently processed blocks. TRUE_EXPR and FALSE_EXPR are the
33279 tests that are used for converting the then-block and the
33280 else-block, respectively. Set either TRUE_EXPR or FALSE_EXPR to a
33281 null pointer if the tests cannot be converted.
33283 -- Macro: IFCVT_MODIFY_MULTIPLE_TESTS (CE_INFO, BB, TRUE_EXPR,
33285 Like `IFCVT_MODIFY_TESTS', but used when converting more
33286 complicated if-statements into conditions combined by `and' and
33287 `or' operations. BB contains the basic block that contains the
33288 test that is currently being processed and about to be turned into
33291 -- Macro: IFCVT_MODIFY_INSN (CE_INFO, PATTERN, INSN)
33292 A C expression to modify the PATTERN of an INSN that is to be
33293 converted to conditional execution format. CE_INFO points to a
33294 data structure, `struct ce_if_block', which contains information
33295 about the currently processed blocks.
33297 -- Macro: IFCVT_MODIFY_FINAL (CE_INFO)
33298 A C expression to perform any final machine dependent
33299 modifications in converting code to conditional execution. The
33300 involved basic blocks can be found in the `struct ce_if_block'
33301 structure that is pointed to by CE_INFO.
33303 -- Macro: IFCVT_MODIFY_CANCEL (CE_INFO)
33304 A C expression to cancel any machine dependent modifications in
33305 converting code to conditional execution. The involved basic
33306 blocks can be found in the `struct ce_if_block' structure that is
33307 pointed to by CE_INFO.
33309 -- Macro: IFCVT_INIT_EXTRA_FIELDS (CE_INFO)
33310 A C expression to initialize any extra fields in a `struct
33311 ce_if_block' structure, which are defined by the
33312 `IFCVT_EXTRA_FIELDS' macro.
33314 -- Macro: IFCVT_EXTRA_FIELDS
33315 If defined, it should expand to a set of field declarations that
33316 will be added to the `struct ce_if_block' structure. These should
33317 be initialized by the `IFCVT_INIT_EXTRA_FIELDS' macro.
33319 -- Target Hook: void TARGET_MACHINE_DEPENDENT_REORG ()
33320 If non-null, this hook performs a target-specific pass over the
33321 instruction stream. The compiler will run it at all optimization
33322 levels, just before the point at which it normally does
33323 delayed-branch scheduling.
33325 The exact purpose of the hook varies from target to target. Some
33326 use it to do transformations that are necessary for correctness,
33327 such as laying out in-function constant pools or avoiding hardware
33328 hazards. Others use it as an opportunity to do some
33329 machine-dependent optimizations.
33331 You need not implement the hook if it has nothing to do. The
33332 default definition is null.
33334 -- Target Hook: void TARGET_INIT_BUILTINS ()
33335 Define this hook if you have any machine-specific built-in
33336 functions that need to be defined. It should be a function that
33337 performs the necessary setup.
33339 Machine specific built-in functions can be useful to expand
33340 special machine instructions that would otherwise not normally be
33341 generated because they have no equivalent in the source language
33342 (for example, SIMD vector instructions or prefetch instructions).
33344 To create a built-in function, call the function
33345 `lang_hooks.builtin_function' which is defined by the language
33346 front end. You can use any type nodes set up by
33347 `build_common_tree_nodes' and `build_common_tree_nodes_2'; only
33348 language front ends that use those two functions will call
33349 `TARGET_INIT_BUILTINS'.
33351 -- Target Hook: rtx TARGET_EXPAND_BUILTIN (tree EXP, rtx TARGET, rtx
33352 SUBTARGET, enum machine_mode MODE, int IGNORE)
33353 Expand a call to a machine specific built-in function that was set
33354 up by `TARGET_INIT_BUILTINS'. EXP is the expression for the
33355 function call; the result should go to TARGET if that is
33356 convenient, and have mode MODE if that is convenient. SUBTARGET
33357 may be used as the target for computing one of EXP's operands.
33358 IGNORE is nonzero if the value is to be ignored. This function
33359 should return the result of the call to the built-in function.
33361 -- Target Hook: tree TARGET_RESOLVE_OVERLOADED_BUILTIN (tree FNDECL,
33363 Select a replacement for a machine specific built-in function that
33364 was set up by `TARGET_INIT_BUILTINS'. This is done _before_
33365 regular type checking, and so allows the target to implement a
33366 crude form of function overloading. FNDECL is the declaration of
33367 the built-in function. ARGLIST is the list of arguments passed to
33368 the built-in function. The result is a complete expression that
33369 implements the operation, usually another `CALL_EXPR'.
33371 -- Target Hook: tree TARGET_FOLD_BUILTIN (tree FNDECL, tree ARGLIST,
33373 Fold a call to a machine specific built-in function that was set
33374 up by `TARGET_INIT_BUILTINS'. FNDECL is the declaration of the
33375 built-in function. ARGLIST is the list of arguments passed to the
33376 built-in function. The result is another tree containing a
33377 simplified expression for the call's result. If IGNORE is true
33378 the value will be ignored.
33380 -- Target Hook: const char * TARGET_INVALID_WITHIN_DOLOOP (rtx INSN)
33381 Take an instruction in INSN and return NULL if it is valid within a
33382 low-overhead loop, otherwise return a string why doloop could not
33385 Many targets use special registers for low-overhead looping. For
33386 any instruction that clobbers these this function should return a
33387 string indicating the reason why the doloop could not be applied.
33388 By default, the RTL loop optimizer does not use a present doloop
33389 pattern for loops containing function calls or branch on table
33392 -- Macro: MD_CAN_REDIRECT_BRANCH (BRANCH1, BRANCH2)
33393 Take a branch insn in BRANCH1 and another in BRANCH2. Return true
33394 if redirecting BRANCH1 to the destination of BRANCH2 is possible.
33396 On some targets, branches may have a limited range. Optimizing the
33397 filling of delay slots can result in branches being redirected,
33398 and this may in turn cause a branch offset to overflow.
33400 -- Target Hook: bool TARGET_COMMUTATIVE_P (rtx X, OUTER_CODE)
33401 This target hook returns `true' if X is considered to be
33402 commutative. Usually, this is just COMMUTATIVE_P (X), but the HP
33403 PA doesn't consider PLUS to be commutative inside a MEM.
33404 OUTER_CODE is the rtx code of the enclosing rtl, if known,
33405 otherwise it is UNKNOWN.
33407 -- Target Hook: rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx HARD_REG)
33408 When the initial value of a hard register has been copied in a
33409 pseudo register, it is often not necessary to actually allocate
33410 another register to this pseudo register, because the original
33411 hard register or a stack slot it has been saved into can be used.
33412 `TARGET_ALLOCATE_INITIAL_VALUE' is called at the start of register
33413 allocation once for each hard register that had its initial value
33414 copied by using `get_func_hard_reg_initial_val' or
33415 `get_hard_reg_initial_val'. Possible values are `NULL_RTX', if
33416 you don't want to do any special allocation, a `REG' rtx--that
33417 would typically be the hard register itself, if it is known not to
33418 be clobbered--or a `MEM'. If you are returning a `MEM', this is
33419 only a hint for the allocator; it might decide to use another
33420 register anyways. You may use `current_function_leaf_function' in
33421 the hook, functions that use `REG_N_SETS', to determine if the hard
33422 register in question will not be clobbered. The default value of
33423 this hook is `NULL', which disables any special allocation.
33425 -- Target Hook: int TARGET_UNSPEC_MAY_TRAP_P (const_rtx X, unsigned
33427 This target hook returns nonzero if X, an `unspec' or
33428 `unspec_volatile' operation, might cause a trap. Targets can use
33429 this hook to enhance precision of analysis for `unspec' and
33430 `unspec_volatile' operations. You may call `may_trap_p_1' to
33431 analyze inner elements of X in which case FLAGS should be passed
33434 -- Target Hook: void TARGET_SET_CURRENT_FUNCTION (tree DECL)
33435 The compiler invokes this hook whenever it changes its current
33436 function context (`cfun'). You can define this function if the
33437 back end needs to perform any initialization or reset actions on a
33438 per-function basis. For example, it may be used to implement
33439 function attributes that affect register usage or code generation
33440 patterns. The argument DECL is the declaration for the new
33441 function context, and may be null to indicate that the compiler
33442 has left a function context and is returning to processing at the
33443 top level. The default hook function does nothing.
33445 GCC sets `cfun' to a dummy function context during initialization
33446 of some parts of the back end. The hook function is not invoked
33447 in this situation; you need not worry about the hook being invoked
33448 recursively, or when the back end is in a partially-initialized
33451 -- Macro: TARGET_OBJECT_SUFFIX
33452 Define this macro to be a C string representing the suffix for
33453 object files on your target machine. If you do not define this
33454 macro, GCC will use `.o' as the suffix for object files.
33456 -- Macro: TARGET_EXECUTABLE_SUFFIX
33457 Define this macro to be a C string representing the suffix to be
33458 automatically added to executable files on your target machine.
33459 If you do not define this macro, GCC will use the null string as
33460 the suffix for executable files.
33462 -- Macro: COLLECT_EXPORT_LIST
33463 If defined, `collect2' will scan the individual object files
33464 specified on its command line and create an export list for the
33465 linker. Define this macro for systems like AIX, where the linker
33466 discards object files that are not referenced from `main' and uses
33469 -- Macro: MODIFY_JNI_METHOD_CALL (MDECL)
33470 Define this macro to a C expression representing a variant of the
33471 method call MDECL, if Java Native Interface (JNI) methods must be
33472 invoked differently from other methods on your target. For
33473 example, on 32-bit Microsoft Windows, JNI methods must be invoked
33474 using the `stdcall' calling convention and this macro is then
33475 defined as this expression:
33477 build_type_attribute_variant (MDECL,
33479 (get_identifier ("stdcall"),
33482 -- Target Hook: bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
33483 This target hook returns `true' past the point in which new jump
33484 instructions could be created. On machines that require a
33485 register for every jump such as the SHmedia ISA of SH5, this point
33486 would typically be reload, so this target hook should be defined
33487 to a function such as:
33490 cannot_modify_jumps_past_reload_p ()
33492 return (reload_completed || reload_in_progress);
33495 -- Target Hook: int TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
33496 This target hook returns a register class for which branch target
33497 register optimizations should be applied. All registers in this
33498 class should be usable interchangeably. After reload, registers
33499 in this class will be re-allocated and loads will be hoisted out
33500 of loops and be subjected to inter-block scheduling.
33502 -- Target Hook: bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool
33503 AFTER_PROLOGUE_EPILOGUE_GEN)
33504 Branch target register optimization will by default exclude
33505 callee-saved registers that are not already live during the
33506 current function; if this target hook returns true, they will be
33507 included. The target code must than make sure that all target
33508 registers in the class returned by
33509 `TARGET_BRANCH_TARGET_REGISTER_CLASS' that might need saving are
33510 saved. AFTER_PROLOGUE_EPILOGUE_GEN indicates if prologues and
33511 epilogues have already been generated. Note, even if you only
33512 return true when AFTER_PROLOGUE_EPILOGUE_GEN is false, you still
33513 are likely to have to make special provisions in
33514 `INITIAL_ELIMINATION_OFFSET' to reserve space for caller-saved
33517 -- Target Hook: bool TARGET_HAVE_CONDITIONAL_EXECUTION (void)
33518 This target hook returns true if the target supports conditional
33519 execution. This target hook is required only when the target has
33520 several different modes and they have different conditional
33521 execution capability, such as ARM.
33523 -- Macro: POWI_MAX_MULTS
33524 If defined, this macro is interpreted as a signed integer C
33525 expression that specifies the maximum number of floating point
33526 multiplications that should be emitted when expanding
33527 exponentiation by an integer constant inline. When this value is
33528 defined, exponentiation requiring more than this number of
33529 multiplications is implemented by calling the system library's
33530 `pow', `powf' or `powl' routines. The default value places no
33531 upper bound on the multiplication count.
33533 -- Macro: void TARGET_EXTRA_INCLUDES (const char *SYSROOT, const char
33534 *IPREFIX, int STDINC)
33535 This target hook should register any extra include files for the
33536 target. The parameter STDINC indicates if normal include files
33537 are present. The parameter SYSROOT is the system root directory.
33538 The parameter IPREFIX is the prefix for the gcc directory.
33540 -- Macro: void TARGET_EXTRA_PRE_INCLUDES (const char *SYSROOT, const
33541 char *IPREFIX, int STDINC)
33542 This target hook should register any extra include files for the
33543 target before any standard headers. The parameter STDINC
33544 indicates if normal include files are present. The parameter
33545 SYSROOT is the system root directory. The parameter IPREFIX is
33546 the prefix for the gcc directory.
33548 -- Macro: void TARGET_OPTF (char *PATH)
33549 This target hook should register special include paths for the
33550 target. The parameter PATH is the include to register. On Darwin
33551 systems, this is used for Framework includes, which have semantics
33552 that are different from `-I'.
33554 -- Target Hook: bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree FNDECL)
33555 This target hook returns `true' if it is safe to use a local alias
33556 for a virtual function FNDECL when constructing thunks, `false'
33557 otherwise. By default, the hook returns `true' for all functions,
33558 if a target supports aliases (i.e. defines `ASM_OUTPUT_DEF'),
33561 -- Macro: TARGET_FORMAT_TYPES
33562 If defined, this macro is the name of a global variable containing
33563 target-specific format checking information for the `-Wformat'
33564 option. The default is to have no target-specific format checks.
33566 -- Macro: TARGET_N_FORMAT_TYPES
33567 If defined, this macro is the number of entries in
33568 `TARGET_FORMAT_TYPES'.
33570 -- Macro: TARGET_OVERRIDES_FORMAT_ATTRIBUTES
33571 If defined, this macro is the name of a global variable containing
33572 target-specific format overrides for the `-Wformat' option. The
33573 default is to have no target-specific format overrides. If defined,
33574 `TARGET_FORMAT_TYPES' must be defined, too.
33576 -- Macro: TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
33577 If defined, this macro specifies the number of entries in
33578 `TARGET_OVERRIDES_FORMAT_ATTRIBUTES'.
33580 -- Macro: TARGET_OVERRIDES_FORMAT_INIT
33581 If defined, this macro specifies the optional initialization
33582 routine for target specific customizations of the system printf
33583 and scanf formatter settings.
33585 -- Target Hook: bool TARGET_RELAXED_ORDERING
33586 If set to `true', means that the target's memory model does not
33587 guarantee that loads which do not depend on one another will access
33588 main memory in the order of the instruction stream; if ordering is
33589 important, an explicit memory barrier must be used. This is true
33590 of many recent processors which implement a policy of "relaxed,"
33591 "weak," or "release" memory consistency, such as Alpha, PowerPC,
33592 and ia64. The default is `false'.
33594 -- Target Hook: const char *TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN
33595 (tree TYPELIST, tree FUNCDECL, tree VAL)
33596 If defined, this macro returns the diagnostic message when it is
33597 illegal to pass argument VAL to function FUNCDECL with prototype
33600 -- Target Hook: const char * TARGET_INVALID_CONVERSION (tree FROMTYPE,
33602 If defined, this macro returns the diagnostic message when it is
33603 invalid to convert from FROMTYPE to TOTYPE, or `NULL' if validity
33604 should be determined by the front end.
33606 -- Target Hook: const char * TARGET_INVALID_UNARY_OP (int OP, tree
33608 If defined, this macro returns the diagnostic message when it is
33609 invalid to apply operation OP (where unary plus is denoted by
33610 `CONVERT_EXPR') to an operand of type TYPE, or `NULL' if validity
33611 should be determined by the front end.
33613 -- Target Hook: const char * TARGET_INVALID_BINARY_OP (int OP, tree
33615 If defined, this macro returns the diagnostic message when it is
33616 invalid to apply operation OP to operands of types TYPE1 and
33617 TYPE2, or `NULL' if validity should be determined by the front end.
33619 -- Macro: TARGET_USE_JCR_SECTION
33620 This macro determines whether to use the JCR section to register
33621 Java classes. By default, TARGET_USE_JCR_SECTION is defined to 1
33622 if both SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true,
33625 -- Macro: OBJC_JBLEN
33626 This macro determines the size of the objective C jump buffer for
33627 the NeXT runtime. By default, OBJC_JBLEN is defined to an
33630 -- Macro: LIBGCC2_UNWIND_ATTRIBUTE
33631 Define this macro if any target-specific attributes need to be
33632 attached to the functions in `libgcc' that provide low-level
33633 support for call stack unwinding. It is used in declarations in
33634 `unwind-generic.h' and the associated definitions of those
33637 -- Target Hook: void TARGET_UPDATE_STACK_BOUNDARY (void)
33638 Define this macro to update the current function stack boundary if
33641 -- Target Hook: rtx TARGET_GET_DRAP_RTX (void)
33642 Define this macro to an rtx for Dynamic Realign Argument Pointer
33643 if a different argument pointer register is needed to access the
33644 function's argument list when stack is aligned.
33646 -- Target Hook: bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void)
33647 When optimization is disabled, this hook indicates whether or not
33648 arguments should be allocated to stack slots. Normally, GCC
33649 allocates stacks slots for arguments when not optimizing in order
33650 to make debugging easier. However, when a function is declared
33651 with `__attribute__((naked))', there is no stack frame, and the
33652 compiler cannot safely move arguments from the registers in which
33653 they are passed to the stack. Therefore, this hook should return
33654 true in general, but false for naked functions. The default
33655 implementation always returns true.
33658 File: gccint.info, Node: Host Config, Next: Fragments, Prev: Target Macros, Up: Top
33660 18 Host Configuration
33661 *********************
33663 Most details about the machine and system on which the compiler is
33664 actually running are detected by the `configure' script. Some things
33665 are impossible for `configure' to detect; these are described in two
33666 ways, either by macros defined in a file named `xm-MACHINE.h' or by
33667 hook functions in the file specified by the OUT_HOST_HOOK_OBJ variable
33668 in `config.gcc'. (The intention is that very few hosts will need a
33669 header file but nearly every fully supported host will need to override
33672 If you need to define only a few macros, and they have simple
33673 definitions, consider using the `xm_defines' variable in your
33674 `config.gcc' entry instead of creating a host configuration header.
33675 *Note System Config::.
33679 * Host Common:: Things every host probably needs implemented.
33680 * Filesystem:: Your host can't have the letter `a' in filenames?
33681 * Host Misc:: Rare configuration options for hosts.
33684 File: gccint.info, Node: Host Common, Next: Filesystem, Up: Host Config
33689 Some things are just not portable, even between similar operating
33690 systems, and are too difficult for autoconf to detect. They get
33691 implemented using hook functions in the file specified by the
33692 HOST_HOOK_OBJ variable in `config.gcc'.
33694 -- Host Hook: void HOST_HOOKS_EXTRA_SIGNALS (void)
33695 This host hook is used to set up handling for extra signals. The
33696 most common thing to do in this hook is to detect stack overflow.
33698 -- Host Hook: void * HOST_HOOKS_GT_PCH_GET_ADDRESS (size_t SIZE, int
33700 This host hook returns the address of some space that is likely to
33701 be free in some subsequent invocation of the compiler. We intend
33702 to load the PCH data at this address such that the data need not
33703 be relocated. The area should be able to hold SIZE bytes. If the
33704 host uses `mmap', FD is an open file descriptor that can be used
33707 -- Host Hook: int HOST_HOOKS_GT_PCH_USE_ADDRESS (void * ADDRESS,
33708 size_t SIZE, int FD, size_t OFFSET)
33709 This host hook is called when a PCH file is about to be loaded.
33710 We want to load SIZE bytes from FD at OFFSET into memory at
33711 ADDRESS. The given address will be the result of a previous
33712 invocation of `HOST_HOOKS_GT_PCH_GET_ADDRESS'. Return -1 if we
33713 couldn't allocate SIZE bytes at ADDRESS. Return 0 if the memory
33714 is allocated but the data is not loaded. Return 1 if the hook has
33715 performed everything.
33717 If the implementation uses reserved address space, free any
33718 reserved space beyond SIZE, regardless of the return value. If no
33719 PCH will be loaded, this hook may be called with SIZE zero, in
33720 which case all reserved address space should be freed.
33722 Do not try to handle values of ADDRESS that could not have been
33723 returned by this executable; just return -1. Such values usually
33724 indicate an out-of-date PCH file (built by some other GCC
33725 executable), and such a PCH file won't work.
33727 -- Host Hook: size_t HOST_HOOKS_GT_PCH_ALLOC_GRANULARITY (void);
33728 This host hook returns the alignment required for allocating
33729 virtual memory. Usually this is the same as getpagesize, but on
33730 some hosts the alignment for reserving memory differs from the
33731 pagesize for committing memory.
33734 File: gccint.info, Node: Filesystem, Next: Host Misc, Prev: Host Common, Up: Host Config
33736 18.2 Host Filesystem
33737 ====================
33739 GCC needs to know a number of things about the semantics of the host
33740 machine's filesystem. Filesystems with Unix and MS-DOS semantics are
33741 automatically detected. For other systems, you can define the
33742 following macros in `xm-MACHINE.h'.
33744 `HAVE_DOS_BASED_FILE_SYSTEM'
33745 This macro is automatically defined by `system.h' if the host file
33746 system obeys the semantics defined by MS-DOS instead of Unix. DOS
33747 file systems are case insensitive, file specifications may begin
33748 with a drive letter, and both forward slash and backslash (`/' and
33749 `\') are directory separators.
33753 If defined, these macros expand to character constants specifying
33754 separators for directory names within a file specification.
33755 `system.h' will automatically give them appropriate values on Unix
33756 and MS-DOS file systems. If your file system is neither of these,
33757 define one or both appropriately in `xm-MACHINE.h'.
33759 However, operating systems like VMS, where constructing a pathname
33760 is more complicated than just stringing together directory names
33761 separated by a special character, should not define either of these
33765 If defined, this macro should expand to a character constant
33766 specifying the separator for elements of search paths. The default
33767 value is a colon (`:'). DOS-based systems usually, but not
33768 always, use semicolon (`;').
33771 Define this macro if the host system is VMS.
33773 `HOST_OBJECT_SUFFIX'
33774 Define this macro to be a C string representing the suffix for
33775 object files on your host machine. If you do not define this
33776 macro, GCC will use `.o' as the suffix for object files.
33778 `HOST_EXECUTABLE_SUFFIX'
33779 Define this macro to be a C string representing the suffix for
33780 executable files on your host machine. If you do not define this
33781 macro, GCC will use the null string as the suffix for executable
33785 A pathname defined by the host operating system, which can be
33786 opened as a file and written to, but all the information written
33787 is discarded. This is commonly known as a "bit bucket" or "null
33788 device". If you do not define this macro, GCC will use
33789 `/dev/null' as the bit bucket. If the host does not support a bit
33790 bucket, define this macro to an invalid filename.
33792 `UPDATE_PATH_HOST_CANONICALIZE (PATH)'
33793 If defined, a C statement (sans semicolon) that performs
33794 host-dependent canonicalization when a path used in a compilation
33795 driver or preprocessor is canonicalized. PATH is a malloc-ed path
33796 to be canonicalized. If the C statement does canonicalize PATH
33797 into a different buffer, the old path should be freed and the new
33798 buffer should have been allocated with malloc.
33801 Define this macro to be a C string representing the format to use
33802 for constructing the index part of debugging dump file names. The
33803 resultant string must fit in fifteen bytes. The full filename
33804 will be the concatenation of: the prefix of the assembler file
33805 name, the string resulting from applying this format to an index
33806 number, and a string unique to each dump file kind, e.g. `rtl'.
33808 If you do not define this macro, GCC will use `.%02d.'. You should
33809 define this macro if using the default will create an invalid file
33812 `DELETE_IF_ORDINARY'
33813 Define this macro to be a C statement (sans semicolon) that
33814 performs host-dependent removal of ordinary temp files in the
33815 compilation driver.
33817 If you do not define this macro, GCC will use the default version.
33818 You should define this macro if the default version does not
33819 reliably remove the temp file as, for example, on VMS which allows
33820 multiple versions of a file.
33822 `HOST_LACKS_INODE_NUMBERS'
33823 Define this macro if the host filesystem does not report
33824 meaningful inode numbers in struct stat.
33827 File: gccint.info, Node: Host Misc, Prev: Filesystem, Up: Host Config
33833 A C expression for the status code to be returned when the compiler
33834 exits after serious errors. The default is the system-provided
33835 macro `EXIT_FAILURE', or `1' if the system doesn't define that
33836 macro. Define this macro only if these defaults are incorrect.
33838 `SUCCESS_EXIT_CODE'
33839 A C expression for the status code to be returned when the compiler
33840 exits without serious errors. (Warnings are not serious errors.)
33841 The default is the system-provided macro `EXIT_SUCCESS', or `0' if
33842 the system doesn't define that macro. Define this macro only if
33843 these defaults are incorrect.
33846 Define this macro if GCC should use the C implementation of
33847 `alloca' provided by `libiberty.a'. This only affects how some
33848 parts of the compiler itself allocate memory. It does not change
33851 When GCC is built with a compiler other than itself, the C `alloca'
33852 is always used. This is because most other implementations have
33853 serious bugs. You should define this macro only on a system where
33854 no stack-based `alloca' can possibly work. For instance, if a
33855 system has a small limit on the size of the stack, GCC's builtin
33856 `alloca' will not work reliably.
33858 `COLLECT2_HOST_INITIALIZATION'
33859 If defined, a C statement (sans semicolon) that performs
33860 host-dependent initialization when `collect2' is being initialized.
33862 `GCC_DRIVER_HOST_INITIALIZATION'
33863 If defined, a C statement (sans semicolon) that performs
33864 host-dependent initialization when a compilation driver is being
33867 `HOST_LONG_LONG_FORMAT'
33868 If defined, the string used to indicate an argument of type `long
33869 long' to functions like `printf'. The default value is `"ll"'.
33871 In addition, if `configure' generates an incorrect definition of any
33872 of the macros in `auto-host.h', you can override that definition in a
33873 host configuration header. If you need to do this, first see if it is
33874 possible to fix `configure'.
33877 File: gccint.info, Node: Fragments, Next: Collect2, Prev: Host Config, Up: Top
33879 19 Makefile Fragments
33880 *********************
33882 When you configure GCC using the `configure' script, it will construct
33883 the file `Makefile' from the template file `Makefile.in'. When it does
33884 this, it can incorporate makefile fragments from the `config'
33885 directory. These are used to set Makefile parameters that are not
33886 amenable to being calculated by autoconf. The list of fragments to
33887 incorporate is set by `config.gcc' (and occasionally `config.build' and
33888 `config.host'); *Note System Config::.
33890 Fragments are named either `t-TARGET' or `x-HOST', depending on
33891 whether they are relevant to configuring GCC to produce code for a
33892 particular target, or to configuring GCC to run on a particular host.
33893 Here TARGET and HOST are mnemonics which usually have some relationship
33894 to the canonical system name, but no formal connection.
33896 If these files do not exist, it means nothing needs to be added for a
33897 given target or host. Most targets need a few `t-TARGET' fragments,
33898 but needing `x-HOST' fragments is rare.
33902 * Target Fragment:: Writing `t-TARGET' files.
33903 * Host Fragment:: Writing `x-HOST' files.
33906 File: gccint.info, Node: Target Fragment, Next: Host Fragment, Up: Fragments
33908 19.1 Target Makefile Fragments
33909 ==============================
33911 Target makefile fragments can set these Makefile variables.
33914 Compiler flags to use when compiling `libgcc2.c'.
33917 A list of source file names to be compiled or assembled and
33918 inserted into `libgcc.a'.
33920 `Floating Point Emulation'
33921 To have GCC include software floating point libraries in `libgcc.a'
33922 define `FPBIT' and `DPBIT' along with a few rules as follows:
33923 # We want fine grained libraries, so use the new code
33924 # to build the floating point emulation libraries.
33929 fp-bit.c: $(srcdir)/config/fp-bit.c
33930 echo '#define FLOAT' > fp-bit.c
33931 cat $(srcdir)/config/fp-bit.c >> fp-bit.c
33933 dp-bit.c: $(srcdir)/config/fp-bit.c
33934 cat $(srcdir)/config/fp-bit.c > dp-bit.c
33936 You may need to provide additional #defines at the beginning of
33937 `fp-bit.c' and `dp-bit.c' to control target endianness and other
33940 `CRTSTUFF_T_CFLAGS'
33941 Special flags used when compiling `crtstuff.c'. *Note
33944 `CRTSTUFF_T_CFLAGS_S'
33945 Special flags used when compiling `crtstuff.c' for shared linking.
33946 Used if you use `crtbeginS.o' and `crtendS.o' in `EXTRA-PARTS'.
33947 *Note Initialization::.
33950 For some targets, invoking GCC in different ways produces objects
33951 that can not be linked together. For example, for some targets GCC
33952 produces both big and little endian code. For these targets, you
33953 must arrange for multiple versions of `libgcc.a' to be compiled,
33954 one for each set of incompatible options. When GCC invokes the
33955 linker, it arranges to link in the right version of `libgcc.a',
33956 based on the command line options used.
33958 The `MULTILIB_OPTIONS' macro lists the set of options for which
33959 special versions of `libgcc.a' must be built. Write options that
33960 are mutually incompatible side by side, separated by a slash.
33961 Write options that may be used together separated by a space. The
33962 build procedure will build all combinations of compatible options.
33964 For example, if you set `MULTILIB_OPTIONS' to `m68000/m68020
33965 msoft-float', `Makefile' will build special versions of `libgcc.a'
33966 using the following sets of options: `-m68000', `-m68020',
33967 `-msoft-float', `-m68000 -msoft-float', and `-m68020 -msoft-float'.
33969 `MULTILIB_DIRNAMES'
33970 If `MULTILIB_OPTIONS' is used, this variable specifies the
33971 directory names that should be used to hold the various libraries.
33972 Write one element in `MULTILIB_DIRNAMES' for each element in
33973 `MULTILIB_OPTIONS'. If `MULTILIB_DIRNAMES' is not used, the
33974 default value will be `MULTILIB_OPTIONS', with all slashes treated
33977 For example, if `MULTILIB_OPTIONS' is set to `m68000/m68020
33978 msoft-float', then the default value of `MULTILIB_DIRNAMES' is
33979 `m68000 m68020 msoft-float'. You may specify a different value if
33980 you desire a different set of directory names.
33983 Sometimes the same option may be written in two different ways.
33984 If an option is listed in `MULTILIB_OPTIONS', GCC needs to know
33985 about any synonyms. In that case, set `MULTILIB_MATCHES' to a
33986 list of items of the form `option=option' to describe all relevant
33987 synonyms. For example, `m68000=mc68000 m68020=mc68020'.
33989 `MULTILIB_EXCEPTIONS'
33990 Sometimes when there are multiple sets of `MULTILIB_OPTIONS' being
33991 specified, there are combinations that should not be built. In
33992 that case, set `MULTILIB_EXCEPTIONS' to be all of the switch
33993 exceptions in shell case syntax that should not be built.
33995 For example the ARM processor cannot execute both hardware floating
33996 point instructions and the reduced size THUMB instructions at the
33997 same time, so there is no need to build libraries with both of
33998 these options enabled. Therefore `MULTILIB_EXCEPTIONS' is set to:
33999 *mthumb/*mhard-float*
34001 `MULTILIB_EXTRA_OPTS'
34002 Sometimes it is desirable that when building multiple versions of
34003 `libgcc.a' certain options should always be passed on to the
34004 compiler. In that case, set `MULTILIB_EXTRA_OPTS' to be the list
34005 of options to be used for all builds. If you set this, you should
34006 probably set `CRTSTUFF_T_CFLAGS' to a dash followed by it.
34008 `NATIVE_SYSTEM_HEADER_DIR'
34009 If the default location for system headers is not `/usr/include',
34010 you must set this to the directory containing the headers. This
34011 value should match the value of the `SYSTEM_INCLUDE_DIR' macro.
34014 Unfortunately, setting `MULTILIB_EXTRA_OPTS' is not enough, since
34015 it does not affect the build of target libraries, at least not the
34016 build of the default multilib. One possible work-around is to use
34017 `DRIVER_SELF_SPECS' to bring options from the `specs' file as if
34018 they had been passed in the compiler driver command line.
34019 However, you don't want to be adding these options after the
34020 toolchain is installed, so you can instead tweak the `specs' file
34021 that will be used during the toolchain build, while you still
34022 install the original, built-in `specs'. The trick is to set
34023 `SPECS' to some other filename (say `specs.install'), that will
34024 then be created out of the built-in specs, and introduce a
34025 `Makefile' rule to generate the `specs' file that's going to be
34026 used at build time out of your `specs.install'.
34029 These are extra flags to pass to the C compiler. They are used
34030 both when building GCC, and when compiling things with the
34031 just-built GCC. This variable is deprecated and should not be
34035 File: gccint.info, Node: Host Fragment, Prev: Target Fragment, Up: Fragments
34037 19.2 Host Makefile Fragments
34038 ============================
34040 The use of `x-HOST' fragments is discouraged. You should only use it
34041 for makefile dependencies.
34044 File: gccint.info, Node: Collect2, Next: Header Dirs, Prev: Fragments, Up: Top
34049 GCC uses a utility called `collect2' on nearly all systems to arrange
34050 to call various initialization functions at start time.
34052 The program `collect2' works by linking the program once and looking
34053 through the linker output file for symbols with particular names
34054 indicating they are constructor functions. If it finds any, it creates
34055 a new temporary `.c' file containing a table of them, compiles it, and
34056 links the program a second time including that file.
34058 The actual calls to the constructors are carried out by a subroutine
34059 called `__main', which is called (automatically) at the beginning of
34060 the body of `main' (provided `main' was compiled with GNU CC). Calling
34061 `__main' is necessary, even when compiling C code, to allow linking C
34062 and C++ object code together. (If you use `-nostdlib', you get an
34063 unresolved reference to `__main', since it's defined in the standard
34064 GCC library. Include `-lgcc' at the end of your compiler command line
34065 to resolve this reference.)
34067 The program `collect2' is installed as `ld' in the directory where the
34068 passes of the compiler are installed. When `collect2' needs to find
34069 the _real_ `ld', it tries the following file names:
34071 * `real-ld' in the directories listed in the compiler's search
34074 * `real-ld' in the directories listed in the environment variable
34077 * The file specified in the `REAL_LD_FILE_NAME' configuration macro,
34080 * `ld' in the compiler's search directories, except that `collect2'
34081 will not execute itself recursively.
34085 "The compiler's search directories" means all the directories where
34086 `gcc' searches for passes of the compiler. This includes directories
34087 that you specify with `-B'.
34089 Cross-compilers search a little differently:
34091 * `real-ld' in the compiler's search directories.
34093 * `TARGET-real-ld' in `PATH'.
34095 * The file specified in the `REAL_LD_FILE_NAME' configuration macro,
34098 * `ld' in the compiler's search directories.
34100 * `TARGET-ld' in `PATH'.
34102 `collect2' explicitly avoids running `ld' using the file name under
34103 which `collect2' itself was invoked. In fact, it remembers up a list
34104 of such names--in case one copy of `collect2' finds another copy (or
34105 version) of `collect2' installed as `ld' in a second place in the
34108 `collect2' searches for the utilities `nm' and `strip' using the same
34109 algorithm as above for `ld'.
34112 File: gccint.info, Node: Header Dirs, Next: Type Information, Prev: Collect2, Up: Top
34114 21 Standard Header File Directories
34115 ***********************************
34117 `GCC_INCLUDE_DIR' means the same thing for native and cross. It is
34118 where GCC stores its private include files, and also where GCC stores
34119 the fixed include files. A cross compiled GCC runs `fixincludes' on
34120 the header files in `$(tooldir)/include'. (If the cross compilation
34121 header files need to be fixed, they must be installed before GCC is
34122 built. If the cross compilation header files are already suitable for
34123 GCC, nothing special need be done).
34125 `GPLUSPLUS_INCLUDE_DIR' means the same thing for native and cross. It
34126 is where `g++' looks first for header files. The C++ library installs
34127 only target independent header files in that directory.
34129 `LOCAL_INCLUDE_DIR' is used only by native compilers. GCC doesn't
34130 install anything there. It is normally `/usr/local/include'. This is
34131 where local additions to a packaged system should place header files.
34133 `CROSS_INCLUDE_DIR' is used only by cross compilers. GCC doesn't
34134 install anything there.
34136 `TOOL_INCLUDE_DIR' is used for both native and cross compilers. It is
34137 the place for other packages to install header files that GCC will use.
34138 For a cross-compiler, this is the equivalent of `/usr/include'. When
34139 you build a cross-compiler, `fixincludes' processes any header files in
34143 File: gccint.info, Node: Type Information, Next: Plugins, Prev: Header Dirs, Up: Top
34145 22 Memory Management and Type Information
34146 *****************************************
34148 GCC uses some fairly sophisticated memory management techniques, which
34149 involve determining information about GCC's data structures from GCC's
34150 source code and using this information to perform garbage collection and
34151 implement precompiled headers.
34153 A full C parser would be too complicated for this task, so a limited
34154 subset of C is interpreted and special markers are used to determine
34155 what parts of the source to look at. All `struct' and `union'
34156 declarations that define data structures that are allocated under
34157 control of the garbage collector must be marked. All global variables
34158 that hold pointers to garbage-collected memory must also be marked.
34159 Finally, all global variables that need to be saved and restored by a
34160 precompiled header must be marked. (The precompiled header mechanism
34161 can only save static variables if they're scalar. Complex data
34162 structures must be allocated in garbage-collected memory to be saved in
34163 a precompiled header.)
34165 The full format of a marker is
34166 GTY (([OPTION] [(PARAM)], [OPTION] [(PARAM)] ...))
34167 but in most cases no options are needed. The outer double parentheses
34168 are still necessary, though: `GTY(())'. Markers can appear:
34170 * In a structure definition, before the open brace;
34172 * In a global variable declaration, after the keyword `static' or
34175 * In a structure field definition, before the name of the field.
34177 Here are some examples of marking simple data structures and globals.
34184 typedef struct TAG GTY(())
34189 static GTY(()) struct TAG *LIST; /* points to GC memory */
34190 static GTY(()) int COUNTER; /* save counter in a PCH */
34192 The parser understands simple typedefs such as `typedef struct TAG
34193 *NAME;' and `typedef int NAME;'. These don't need to be marked.
34197 * GTY Options:: What goes inside a `GTY(())'.
34198 * GGC Roots:: Making global variables GGC roots.
34199 * Files:: How the generated files work.
34200 * Invoking the garbage collector:: How to invoke the garbage collector.
34203 File: gccint.info, Node: GTY Options, Next: GGC Roots, Up: Type Information
34205 22.1 The Inside of a `GTY(())'
34206 ==============================
34208 Sometimes the C code is not enough to fully describe the type
34209 structure. Extra information can be provided with `GTY' options and
34210 additional markers. Some options take a parameter, which may be either
34211 a string or a type name, depending on the parameter. If an option
34212 takes no parameter, it is acceptable either to omit the parameter
34213 entirely, or to provide an empty string as a parameter. For example,
34214 `GTY ((skip))' and `GTY ((skip ("")))' are equivalent.
34216 When the parameter is a string, often it is a fragment of C code. Four
34217 special escapes may be used in these strings, to refer to pieces of the
34218 data structure being marked:
34221 The current structure.
34224 The structure that immediately contains the current structure.
34227 The outermost structure that contains the current structure.
34230 A partial expression of the form `[i1][i2]...' that indexes the
34231 array item currently being marked.
34233 For instance, suppose that you have a structure of the form
34240 and `b' is a variable of type `struct B'. When marking `b.foo[11]',
34241 `%h' would expand to `b.foo[11]', `%0' and `%1' would both expand to
34242 `b', and `%a' would expand to `[11]'.
34244 As in ordinary C, adjacent strings will be concatenated; this is
34245 helpful when you have a complicated expression.
34246 GTY ((chain_next ("TREE_CODE (&%h.generic) == INTEGER_TYPE"
34247 " ? TYPE_NEXT_VARIANT (&%h.generic)"
34248 " : TREE_CHAIN (&%h.generic)")))
34250 The available options are:
34252 `length ("EXPRESSION")'
34253 There are two places the type machinery will need to be explicitly
34254 told the length of an array. The first case is when a structure
34255 ends in a variable-length array, like this:
34256 struct rtvec_def GTY(()) {
34257 int num_elem; /* number of elements */
34258 rtx GTY ((length ("%h.num_elem"))) elem[1];
34261 In this case, the `length' option is used to override the specified
34262 array length (which should usually be `1'). The parameter of the
34263 option is a fragment of C code that calculates the length.
34265 The second case is when a structure or a global variable contains a
34266 pointer to an array, like this:
34268 GTY ((length ("%h.regno_pointer_align_length"))) regno_decl;
34269 In this case, `regno_decl' has been allocated by writing something
34272 ggc_alloc (x->regno_pointer_align_length * sizeof (tree));
34273 and the `length' provides the length of the field.
34275 This second use of `length' also works on global variables, like:
34276 static GTY((length ("reg_base_value_size")))
34277 rtx *reg_base_value;
34280 If `skip' is applied to a field, the type machinery will ignore it.
34281 This is somewhat dangerous; the only safe use is in a union when
34282 one field really isn't ever used.
34284 `desc ("EXPRESSION")'
34287 The type machinery needs to be told which field of a `union' is
34288 currently active. This is done by giving each field a constant
34289 `tag' value, and then specifying a discriminator using `desc'.
34290 The value of the expression given by `desc' is compared against
34291 each `tag' value, each of which should be different. If no `tag'
34292 is matched, the field marked with `default' is used if there is
34293 one, otherwise no field in the union will be marked.
34295 In the `desc' option, the "current structure" is the union that it
34296 discriminates. Use `%1' to mean the structure containing it.
34297 There are no escapes available to the `tag' option, since it is a
34301 struct tree_binding GTY(())
34303 struct tree_common common;
34304 union tree_binding_u {
34305 tree GTY ((tag ("0"))) scope;
34306 struct cp_binding_level * GTY ((tag ("1"))) level;
34307 } GTY ((desc ("BINDING_HAS_LEVEL_P ((tree)&%0)"))) xscope;
34311 In this example, the value of BINDING_HAS_LEVEL_P when applied to a
34312 `struct tree_binding *' is presumed to be 0 or 1. If 1, the type
34313 mechanism will treat the field `level' as being present and if 0,
34314 will treat the field `scope' as being present.
34318 Sometimes it's convenient to define some data structure to work on
34319 generic pointers (that is, `PTR') and then use it with a specific
34320 type. `param_is' specifies the real type pointed to, and
34321 `use_param' says where in the generic data structure that type
34324 For instance, to have a `htab_t' that points to trees, one would
34325 write the definition of `htab_t' like this:
34326 typedef struct GTY(()) {
34328 void ** GTY ((use_param, ...)) entries;
34331 and then declare variables like this:
34332 static htab_t GTY ((param_is (union tree_node))) ict;
34336 In more complicated cases, the data structure might need to work on
34337 several different types, which might not necessarily all be
34338 pointers. For this, `param1_is' through `param9_is' may be used to
34339 specify the real type of a field identified by `use_param1' through
34343 When a structure contains another structure that is parameterized,
34344 there's no need to do anything special, the inner structure
34345 inherits the parameters of the outer one. When a structure
34346 contains a pointer to a parameterized structure, the type
34347 machinery won't automatically detect this (it could, it just
34348 doesn't yet), so it's necessary to tell it that the pointed-to
34349 structure should use the same parameters as the outer structure.
34350 This is done by marking the pointer with the `use_params' option.
34353 `deletable', when applied to a global variable, indicates that when
34354 garbage collection runs, there's no need to mark anything pointed
34355 to by this variable, it can just be set to `NULL' instead. This
34356 is used to keep a list of free structures around for re-use.
34358 `if_marked ("EXPRESSION")'
34359 Suppose you want some kinds of object to be unique, and so you put
34360 them in a hash table. If garbage collection marks the hash table,
34361 these objects will never be freed, even if the last other
34362 reference to them goes away. GGC has special handling to deal
34363 with this: if you use the `if_marked' option on a global hash
34364 table, GGC will call the routine whose name is the parameter to
34365 the option on each hash table entry. If the routine returns
34366 nonzero, the hash table entry will be marked as usual. If the
34367 routine returns zero, the hash table entry will be deleted.
34369 The routine `ggc_marked_p' can be used to determine if an element
34370 has been marked already; in fact, the usual case is to use
34371 `if_marked ("ggc_marked_p")'.
34373 `mark_hook ("HOOK-ROUTINE-NAME")'
34374 If provided for a structure or union type, the given
34375 HOOK-ROUTINE-NAME (between double-quotes) is the name of a routine
34376 called when the garbage collector has just marked the data as
34377 reachable. This routine should not change the data, or call any ggc
34378 routine. Its only argument is a pointer to the just marked (const)
34379 structure or union.
34382 When applied to a field, `maybe_undef' indicates that it's OK if
34383 the structure that this fields points to is never defined, so long
34384 as this field is always `NULL'. This is used to avoid requiring
34385 backends to define certain optional structures. It doesn't work
34386 with language frontends.
34388 `nested_ptr (TYPE, "TO EXPRESSION", "FROM EXPRESSION")'
34389 The type machinery expects all pointers to point to the start of an
34390 object. Sometimes for abstraction purposes it's convenient to have
34391 a pointer which points inside an object. So long as it's possible
34392 to convert the original object to and from the pointer, such
34393 pointers can still be used. TYPE is the type of the original
34394 object, the TO EXPRESSION returns the pointer given the original
34395 object, and the FROM EXPRESSION returns the original object given
34396 the pointer. The pointer will be available using the `%h' escape.
34398 `chain_next ("EXPRESSION")'
34399 `chain_prev ("EXPRESSION")'
34400 `chain_circular ("EXPRESSION")'
34401 It's helpful for the type machinery to know if objects are often
34402 chained together in long lists; this lets it generate code that
34403 uses less stack space by iterating along the list instead of
34404 recursing down it. `chain_next' is an expression for the next
34405 item in the list, `chain_prev' is an expression for the previous
34406 item. For singly linked lists, use only `chain_next'; for doubly
34407 linked lists, use both. The machinery requires that taking the
34408 next item of the previous item gives the original item.
34409 `chain_circular' is similar to `chain_next', but can be used for
34410 circular single linked lists.
34412 `reorder ("FUNCTION NAME")'
34413 Some data structures depend on the relative ordering of pointers.
34414 If the precompiled header machinery needs to change that ordering,
34415 it will call the function referenced by the `reorder' option,
34416 before changing the pointers in the object that's pointed to by
34417 the field the option applies to. The function must take four
34418 arguments, with the signature
34419 `void *, void *, gt_pointer_operator, void *'. The first
34420 parameter is a pointer to the structure that contains the object
34421 being updated, or the object itself if there is no containing
34422 structure. The second parameter is a cookie that should be
34423 ignored. The third parameter is a routine that, given a pointer,
34424 will update it to its correct new value. The fourth parameter is
34425 a cookie that must be passed to the second parameter.
34427 PCH cannot handle data structures that depend on the absolute
34428 values of pointers. `reorder' functions can be expensive. When
34429 possible, it is better to depend on properties of the data, like
34430 an ID number or the hash of a string instead.
34433 The `special' option is used to mark types that have to be dealt
34434 with by special case machinery. The parameter is the name of the
34435 special case. See `gengtype.c' for further details. Avoid adding
34436 new special cases unless there is no other alternative.
34439 File: gccint.info, Node: GGC Roots, Next: Files, Prev: GTY Options, Up: Type Information
34441 22.2 Marking Roots for the Garbage Collector
34442 ============================================
34444 In addition to keeping track of types, the type machinery also locates
34445 the global variables ("roots") that the garbage collector starts at.
34446 Roots must be declared using one of the following syntaxes:
34448 * `extern GTY(([OPTIONS])) TYPE NAME;'
34450 * `static GTY(([OPTIONS])) TYPE NAME;'
34452 * `GTY(([OPTIONS])) TYPE NAME;'
34453 is _not_ accepted. There should be an `extern' declaration of such a
34454 variable in a header somewhere--mark that, not the definition. Or, if
34455 the variable is only used in one file, make it `static'.
34458 File: gccint.info, Node: Files, Next: Invoking the garbage collector, Prev: GGC Roots, Up: Type Information
34460 22.3 Source Files Containing Type Information
34461 =============================================
34463 Whenever you add `GTY' markers to a source file that previously had
34464 none, or create a new source file containing `GTY' markers, there are
34465 three things you need to do:
34467 1. You need to add the file to the list of source files the type
34468 machinery scans. There are four cases:
34470 a. For a back-end file, this is usually done automatically; if
34471 not, you should add it to `target_gtfiles' in the appropriate
34472 port's entries in `config.gcc'.
34474 b. For files shared by all front ends, add the filename to the
34475 `GTFILES' variable in `Makefile.in'.
34477 c. For files that are part of one front end, add the filename to
34478 the `gtfiles' variable defined in the appropriate
34479 `config-lang.in'. For C, the file is `c-config-lang.in'.
34480 Headers should appear before non-headers in this list.
34482 d. For files that are part of some but not all front ends, add
34483 the filename to the `gtfiles' variable of _all_ the front ends
34486 2. If the file was a header file, you'll need to check that it's
34487 included in the right place to be visible to the generated files.
34488 For a back-end header file, this should be done automatically.
34489 For a front-end header file, it needs to be included by the same
34490 file that includes `gtype-LANG.h'. For other header files, it
34491 needs to be included in `gtype-desc.c', which is a generated file,
34492 so add it to `ifiles' in `open_base_file' in `gengtype.c'.
34494 For source files that aren't header files, the machinery will
34495 generate a header file that should be included in the source file
34496 you just changed. The file will be called `gt-PATH.h' where PATH
34497 is the pathname relative to the `gcc' directory with slashes
34498 replaced by -, so for example the header file to be included in
34499 `cp/parser.c' is called `gt-cp-parser.c'. The generated header
34500 file should be included after everything else in the source file.
34501 Don't forget to mention this file as a dependency in the
34505 For language frontends, there is another file that needs to be included
34506 somewhere. It will be called `gtype-LANG.h', where LANG is the name of
34507 the subdirectory the language is contained in.
34509 Plugins can add additional root tables. Run the `gengtype' utility in
34510 plugin mode as `gengtype -p SOURCE-DIR FILE-LIST PLUGIN*.C' with your
34511 plugin files PLUGIN*.C using `GTY' to generate the corresponding
34512 GT-PLUGIN*.H files. The GCC build tree is needed to be present in that
34516 File: gccint.info, Node: Invoking the garbage collector, Prev: Files, Up: Type Information
34518 22.4 How to invoke the garbage collector
34519 ========================================
34521 The GCC garbage collector GGC is only invoked explicitly. In contrast
34522 with many other garbage collectors, it is not implicitly invoked by
34523 allocation routines when a lot of memory has been consumed. So the only
34524 way to have GGC reclaim storage it to call the `ggc_collect' function
34525 explicitly. This call is an expensive operation, as it may have to scan
34526 the entire heap. Beware that local variables (on the GCC call stack)
34527 are not followed by such an invocation (as many other garbage
34528 collectors do): you should reference all your data from static or
34529 external `GTY'-ed variables, and it is advised to call `ggc_collect'
34530 with a shallow call stack. The GGC is an exact mark and sweep garbage
34531 collector (so it does not scan the call stack for pointers). In
34532 practice GCC passes don't often call `ggc_collect' themselves, because
34533 it is called by the pass manager between passes.
34536 File: gccint.info, Node: Plugins, Next: Funding, Prev: Type Information, Up: Top
34541 23.1 Loading Plugins
34542 ====================
34544 Plugins are supported on platforms that support `-ldl -rdynamic'. They
34545 are loaded by the compiler using `dlopen' and invoked at pre-determined
34546 locations in the compilation process.
34548 Plugins are loaded with
34550 `-fplugin=/path/to/NAME.so' `-fplugin-arg-NAME-<key1>[=<value1>]'
34552 The plugin arguments are parsed by GCC and passed to respective
34553 plugins as key-value pairs. Multiple plugins can be invoked by
34554 specifying multiple `-fplugin' arguments.
34559 Plugins are activated by the compiler at specific events as defined in
34560 `gcc-plugin.h'. For each event of interest, the plugin should call
34561 `register_callback' specifying the name of the event and address of the
34562 callback function that will handle that event.
34564 The header `gcc-plugin.h' must be the first gcc header to be included.
34566 23.2.1 Plugin initialization
34567 ----------------------------
34569 Every plugin should export a function called `plugin_init' that is
34570 called right after the plugin is loaded. This function is responsible
34571 for registering all the callbacks required by the plugin and do any
34572 other required initialization.
34574 This function is called from `compile_file' right before invoking the
34575 parser. The arguments to `plugin_init' are:
34577 * `plugin_info': Plugin invocation information.
34579 * `version': GCC version.
34581 The `plugin_info' struct is defined as follows:
34583 struct plugin_name_args
34585 char *base_name; /* Short name of the plugin
34586 (filename without .so suffix). */
34587 const char *full_name; /* Path to the plugin as specified with
34589 int argc; /* Number of arguments specified with
34590 -fplugin-arg-.... */
34591 struct plugin_argument *argv; /* Array of ARGC key-value pairs. */
34592 const char *version; /* Version string provided by plugin. */
34593 const char *help; /* Help string provided by plugin. */
34596 If initialization fails, `plugin_init' must return a non-zero value.
34597 Otherwise, it should return 0.
34599 The version of the GCC compiler loading the plugin is described by the
34600 following structure:
34602 struct plugin_gcc_version
34604 const char *basever;
34605 const char *datestamp;
34606 const char *devphase;
34607 const char *revision;
34608 const char *configuration_arguments;
34611 The function `plugin_default_version_check' takes two pointers to such
34612 structure and compare them field by field. It can be used by the
34613 plugin's `plugin_init' function.
34615 23.2.2 Plugin callbacks
34616 -----------------------
34618 Callback functions have the following prototype:
34620 /* The prototype for a plugin callback function.
34621 gcc_data - event-specific data provided by GCC
34622 user_data - plugin-specific data provided by the plug-in. */
34623 typedef void (*plugin_callback_func)(void *gcc_data, void *user_data);
34625 Callbacks can be invoked at the following pre-determined events:
34629 PLUGIN_PASS_MANAGER_SETUP, /* To hook into pass manager. */
34630 PLUGIN_FINISH_TYPE, /* After finishing parsing a type. */
34631 PLUGIN_FINISH_UNIT, /* Useful for summary processing. */
34632 PLUGIN_CXX_CP_PRE_GENERICIZE, /* Allows to see low level AST in C++ FE. */
34633 PLUGIN_FINISH, /* Called before GCC exits. */
34634 PLUGIN_INFO, /* Information about the plugin. */
34635 PLUGIN_GGC_START, /* Called at start of GCC Garbage Collection. */
34636 PLUGIN_GGC_MARKING, /* Extend the GGC marking. */
34637 PLUGIN_GGC_END, /* Called at end of GGC. */
34638 PLUGIN_REGISTER_GGC_ROOTS, /* Register an extra GGC root table. */
34639 PLUGIN_ATTRIBUTES, /* Called during attribute registration */
34640 PLUGIN_START_UNIT, /* Called before processing a translation unit. */
34641 PLUGIN_EVENT_LAST /* Dummy event used for indexing callback
34645 To register a callback, the plugin calls `register_callback' with the
34648 * `char *name': Plugin name.
34650 * `enum plugin_event event': The event code.
34652 * `plugin_callback_func callback': The function that handles `event'.
34654 * `void *user_data': Pointer to plugin-specific data.
34656 For the PLUGIN_PASS_MANAGER_SETUP, PLUGIN_INFO, and
34657 PLUGIN_REGISTER_GGC_ROOTS pseudo-events the `callback' should be null,
34658 and the `user_data' is specific.
34660 23.3 Interacting with the pass manager
34661 ======================================
34663 There needs to be a way to add/reorder/remove passes dynamically. This
34664 is useful for both analysis plugins (plugging in after a certain pass
34665 such as CFG or an IPA pass) and optimization plugins.
34667 Basic support for inserting new passes or replacing existing passes is
34668 provided. A plugin registers a new pass with GCC by calling
34669 `register_callback' with the `PLUGIN_PASS_MANAGER_SETUP' event and a
34670 pointer to a `struct plugin_pass' object defined as follows
34672 enum pass_positioning_ops
34674 PASS_POS_INSERT_AFTER, // Insert after the reference pass.
34675 PASS_POS_INSERT_BEFORE, // Insert before the reference pass.
34676 PASS_POS_REPLACE // Replace the reference pass.
34681 struct opt_pass *pass; /* New pass provided by the plugin. */
34682 const char *reference_pass_name; /* Name of the reference pass for hooking
34683 up the new pass. */
34684 int ref_pass_instance_number; /* Insert the pass at the specified
34685 instance number of the reference pass. */
34686 /* Do it for every instance if it is 0. */
34687 enum pass_positioning_ops pos_op; /* how to insert the new pass. */
34691 /* Sample plugin code that registers a new pass. */
34693 plugin_init (struct plugin_name_args *plugin_info,
34694 struct plugin_gcc_version *version)
34696 struct plugin_pass pass_info;
34700 /* Code to fill in the pass_info object with new pass information. */
34704 /* Register the new pass. */
34705 register_callback (plugin_info->base_name, PLUGIN_PASS_MANAGER_SETUP, NULL, &pass_info);
34710 23.4 Interacting with the GCC Garbage Collector
34711 ===============================================
34713 Some plugins may want to be informed when GGC (the GCC Garbage
34714 Collector) is running. They can register callbacks for the
34715 `PLUGIN_GGC_START' and `PLUGIN_GGC_END' events (for which the callback
34716 is called with a null `gcc_data') to be notified of the start or end of
34717 the GCC garbage collection.
34719 Some plugins may need to have GGC mark additional data. This can be
34720 done by registering a callback (called with a null `gcc_data') for the
34721 `PLUGIN_GGC_MARKING' event. Such callbacks can call the `ggc_set_mark'
34722 routine, preferably thru the `ggc_mark' macro (and conversely, these
34723 routines should usually not be used in plugins outside of the
34724 `PLUGIN_GGC_MARKING' event).
34726 Some plugins may need to add extra GGC root tables, e.g. to handle
34727 their own `GTY'-ed data. This can be done with the
34728 `PLUGIN_REGISTER_GGC_ROOTS' pseudo-event with a null callback and the
34729 extra root table as `user_data'. Running the `gengtype -p SOURCE-DIR
34730 FILE-LIST PLUGIN*.C ...' utility generates this extra root table.
34732 You should understand the details of memory management inside GCC
34733 before using `PLUGIN_GGC_MARKING' or `PLUGIN_REGISTER_GGC_ROOTS'.
34735 23.5 Giving information about a plugin
34736 ======================================
34738 A plugin should give some information to the user about itself. This
34739 uses the following structure:
34743 const char *version;
34747 Such a structure is passed as the `user_data' by the plugin's init
34748 routine using `register_callback' with the `PLUGIN_INFO' pseudo-event
34749 and a null callback.
34751 23.6 Registering custom attributes
34752 ==================================
34754 For analysis purposes it is useful to be able to add custom attributes.
34756 The `PLUGIN_ATTRIBUTES' callback is called during attribute
34757 registration. Use the `register_attribute' function to register custom
34760 /* Attribute handler callback */
34762 handle_user_attribute (tree *node, tree name, tree args,
34763 int flags, bool *no_add_attrs)
34768 /* Attribute definition */
34769 static struct attribute_spec user_attr =
34770 { "user", 1, 1, false, false, false, handle_user_attribute };
34772 /* Plugin callback called during attribute registration.
34773 Registered with register_callback (plugin_name, PLUGIN_ATTRIBUTES, register_attributes, NULL)
34776 register_attributes (void *event_data, void *data)
34778 warning (0, G_("Callback to register attributes"));
34779 register_attribute (&user_attr);
34782 23.7 Building GCC plugins
34783 =========================
34785 If plugins are enabled, GCC installs the headers needed to build a
34786 plugin (somehwere in the installation tree, e.g. under `/usr/local').
34787 In particular a `plugin/include' directory is installed, containing all
34788 the header files needed to build plugins.
34790 On most systems, you can query this `plugin' directory by invoking
34791 `gcc -print-file-name=plugin' (replace if needed `gcc' with the
34792 appropriate program path).
34794 The following GNU Makefile excerpt shows how to build a simple plugin:
34797 PLUGIN_SOURCE_FILES= plugin1.c plugin2.c
34798 PLUGIN_OBJECT_FILES= $(patsubst %.c,%.o,$(PLUGIN_SOURCE_FILES))
34799 GCCPLUGINS_DIR:= $(shell $(GCC) -print-file-name=plugin)
34800 CFLAGS+= -I$(GCCPLUGINS_DIR)/include -fPIC -O2
34802 plugin.so: $(PLUGIN_OBJECT_FILES)
34803 $(GCC) -shared $^ -o $
34805 A single source file plugin may be built with `gcc -I`gcc
34806 -print-file-name=plugin`/include -fPIC -shared -O2 plugin.c -o
34807 plugin.so', using backquote shell syntax to query the `plugin'
34810 Plugins needing to use `gengtype' require a GCC build directory for
34811 the same version of GCC that they will be linked against.
34814 File: gccint.info, Node: Funding, Next: GNU Project, Prev: Plugins, Up: Top
34816 Funding Free Software
34817 *********************
34819 If you want to have more free software a few years from now, it makes
34820 sense for you to help encourage people to contribute funds for its
34821 development. The most effective approach known is to encourage
34822 commercial redistributors to donate.
34824 Users of free software systems can boost the pace of development by
34825 encouraging for-a-fee distributors to donate part of their selling price
34826 to free software developers--the Free Software Foundation, and others.
34828 The way to convince distributors to do this is to demand it and expect
34829 it from them. So when you compare distributors, judge them partly by
34830 how much they give to free software development. Show distributors
34831 they must compete to be the one who gives the most.
34833 To make this approach work, you must insist on numbers that you can
34834 compare, such as, "We will donate ten dollars to the Frobnitz project
34835 for each disk sold." Don't be satisfied with a vague promise, such as
34836 "A portion of the profits are donated," since it doesn't give a basis
34839 Even a precise fraction "of the profits from this disk" is not very
34840 meaningful, since creative accounting and unrelated business decisions
34841 can greatly alter what fraction of the sales price counts as profit.
34842 If the price you pay is $50, ten percent of the profit is probably less
34843 than a dollar; it might be a few cents, or nothing at all.
34845 Some redistributors do development work themselves. This is useful
34846 too; but to keep everyone honest, you need to inquire how much they do,
34847 and what kind. Some kinds of development make much more long-term
34848 difference than others. For example, maintaining a separate version of
34849 a program contributes very little; maintaining the standard version of a
34850 program for the whole community contributes much. Easy new ports
34851 contribute little, since someone else would surely do them; difficult
34852 ports such as adding a new CPU to the GNU Compiler Collection
34853 contribute more; major new features or packages contribute the most.
34855 By establishing the idea that supporting further development is "the
34856 proper thing to do" when distributing free software for a fee, we can
34857 assure a steady flow of resources into making more free software.
34859 Copyright (C) 1994 Free Software Foundation, Inc.
34860 Verbatim copying and redistribution of this section is permitted
34861 without royalty; alteration is not permitted.
34864 File: gccint.info, Node: GNU Project, Next: Copying, Prev: Funding, Up: Top
34866 The GNU Project and GNU/Linux
34867 *****************************
34869 The GNU Project was launched in 1984 to develop a complete Unix-like
34870 operating system which is free software: the GNU system. (GNU is a
34871 recursive acronym for "GNU's Not Unix"; it is pronounced "guh-NEW".)
34872 Variants of the GNU operating system, which use the kernel Linux, are
34873 now widely used; though these systems are often referred to as "Linux",
34874 they are more accurately called GNU/Linux systems.
34876 For more information, see:
34877 `http://www.gnu.org/'
34878 `http://www.gnu.org/gnu/linux-and-gnu.html'
34881 File: gccint.info, Node: Copying, Next: GNU Free Documentation License, Prev: GNU Project, Up: Top
34883 GNU General Public License
34884 **************************
34886 Version 3, 29 June 2007
34888 Copyright (C) 2007 Free Software Foundation, Inc. `http://fsf.org/'
34890 Everyone is permitted to copy and distribute verbatim copies of this
34891 license document, but changing it is not allowed.
34896 The GNU General Public License is a free, copyleft license for software
34897 and other kinds of works.
34899 The licenses for most software and other practical works are designed
34900 to take away your freedom to share and change the works. By contrast,
34901 the GNU General Public License is intended to guarantee your freedom to
34902 share and change all versions of a program-to make sure it remains free
34903 software for all its users. We, the Free Software Foundation, use the
34904 GNU General Public License for most of our software; it applies also to
34905 any other work released this way by its authors. You can apply it to
34906 your programs, too.
34908 When we speak of free software, we are referring to freedom, not
34909 price. Our General Public Licenses are designed to make sure that you
34910 have the freedom to distribute copies of free software (and charge for
34911 them if you wish), that you receive source code or can get it if you
34912 want it, that you can change the software or use pieces of it in new
34913 free programs, and that you know you can do these things.
34915 To protect your rights, we need to prevent others from denying you
34916 these rights or asking you to surrender the rights. Therefore, you
34917 have certain responsibilities if you distribute copies of the software,
34918 or if you modify it: responsibilities to respect the freedom of others.
34920 For example, if you distribute copies of such a program, whether
34921 gratis or for a fee, you must pass on to the recipients the same
34922 freedoms that you received. You must make sure that they, too, receive
34923 or can get the source code. And you must show them these terms so they
34926 Developers that use the GNU GPL protect your rights with two steps:
34927 (1) assert copyright on the software, and (2) offer you this License
34928 giving you legal permission to copy, distribute and/or modify it.
34930 For the developers' and authors' protection, the GPL clearly explains
34931 that there is no warranty for this free software. For both users' and
34932 authors' sake, the GPL requires that modified versions be marked as
34933 changed, so that their problems will not be attributed erroneously to
34934 authors of previous versions.
34936 Some devices are designed to deny users access to install or run
34937 modified versions of the software inside them, although the
34938 manufacturer can do so. This is fundamentally incompatible with the
34939 aim of protecting users' freedom to change the software. The
34940 systematic pattern of such abuse occurs in the area of products for
34941 individuals to use, which is precisely where it is most unacceptable.
34942 Therefore, we have designed this version of the GPL to prohibit the
34943 practice for those products. If such problems arise substantially in
34944 other domains, we stand ready to extend this provision to those domains
34945 in future versions of the GPL, as needed to protect the freedom of
34948 Finally, every program is threatened constantly by software patents.
34949 States should not allow patents to restrict development and use of
34950 software on general-purpose computers, but in those that do, we wish to
34951 avoid the special danger that patents applied to a free program could
34952 make it effectively proprietary. To prevent this, the GPL assures that
34953 patents cannot be used to render the program non-free.
34955 The precise terms and conditions for copying, distribution and
34956 modification follow.
34958 TERMS AND CONDITIONS
34959 ====================
34963 "This License" refers to version 3 of the GNU General Public
34966 "Copyright" also means copyright-like laws that apply to other
34967 kinds of works, such as semiconductor masks.
34969 "The Program" refers to any copyrightable work licensed under this
34970 License. Each licensee is addressed as "you". "Licensees" and
34971 "recipients" may be individuals or organizations.
34973 To "modify" a work means to copy from or adapt all or part of the
34974 work in a fashion requiring copyright permission, other than the
34975 making of an exact copy. The resulting work is called a "modified
34976 version" of the earlier work or a work "based on" the earlier work.
34978 A "covered work" means either the unmodified Program or a work
34979 based on the Program.
34981 To "propagate" a work means to do anything with it that, without
34982 permission, would make you directly or secondarily liable for
34983 infringement under applicable copyright law, except executing it
34984 on a computer or modifying a private copy. Propagation includes
34985 copying, distribution (with or without modification), making
34986 available to the public, and in some countries other activities as
34989 To "convey" a work means any kind of propagation that enables other
34990 parties to make or receive copies. Mere interaction with a user
34991 through a computer network, with no transfer of a copy, is not
34994 An interactive user interface displays "Appropriate Legal Notices"
34995 to the extent that it includes a convenient and prominently visible
34996 feature that (1) displays an appropriate copyright notice, and (2)
34997 tells the user that there is no warranty for the work (except to
34998 the extent that warranties are provided), that licensees may
34999 convey the work under this License, and how to view a copy of this
35000 License. If the interface presents a list of user commands or
35001 options, such as a menu, a prominent item in the list meets this
35006 The "source code" for a work means the preferred form of the work
35007 for making modifications to it. "Object code" means any
35008 non-source form of a work.
35010 A "Standard Interface" means an interface that either is an
35011 official standard defined by a recognized standards body, or, in
35012 the case of interfaces specified for a particular programming
35013 language, one that is widely used among developers working in that
35016 The "System Libraries" of an executable work include anything,
35017 other than the work as a whole, that (a) is included in the normal
35018 form of packaging a Major Component, but which is not part of that
35019 Major Component, and (b) serves only to enable use of the work
35020 with that Major Component, or to implement a Standard Interface
35021 for which an implementation is available to the public in source
35022 code form. A "Major Component", in this context, means a major
35023 essential component (kernel, window system, and so on) of the
35024 specific operating system (if any) on which the executable work
35025 runs, or a compiler used to produce the work, or an object code
35026 interpreter used to run it.
35028 The "Corresponding Source" for a work in object code form means all
35029 the source code needed to generate, install, and (for an executable
35030 work) run the object code and to modify the work, including
35031 scripts to control those activities. However, it does not include
35032 the work's System Libraries, or general-purpose tools or generally
35033 available free programs which are used unmodified in performing
35034 those activities but which are not part of the work. For example,
35035 Corresponding Source includes interface definition files
35036 associated with source files for the work, and the source code for
35037 shared libraries and dynamically linked subprograms that the work
35038 is specifically designed to require, such as by intimate data
35039 communication or control flow between those subprograms and other
35042 The Corresponding Source need not include anything that users can
35043 regenerate automatically from other parts of the Corresponding
35046 The Corresponding Source for a work in source code form is that
35049 2. Basic Permissions.
35051 All rights granted under this License are granted for the term of
35052 copyright on the Program, and are irrevocable provided the stated
35053 conditions are met. This License explicitly affirms your unlimited
35054 permission to run the unmodified Program. The output from running
35055 a covered work is covered by this License only if the output,
35056 given its content, constitutes a covered work. This License
35057 acknowledges your rights of fair use or other equivalent, as
35058 provided by copyright law.
35060 You may make, run and propagate covered works that you do not
35061 convey, without conditions so long as your license otherwise
35062 remains in force. You may convey covered works to others for the
35063 sole purpose of having them make modifications exclusively for
35064 you, or provide you with facilities for running those works,
35065 provided that you comply with the terms of this License in
35066 conveying all material for which you do not control copyright.
35067 Those thus making or running the covered works for you must do so
35068 exclusively on your behalf, under your direction and control, on
35069 terms that prohibit them from making any copies of your
35070 copyrighted material outside their relationship with you.
35072 Conveying under any other circumstances is permitted solely under
35073 the conditions stated below. Sublicensing is not allowed; section
35074 10 makes it unnecessary.
35076 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
35078 No covered work shall be deemed part of an effective technological
35079 measure under any applicable law fulfilling obligations under
35080 article 11 of the WIPO copyright treaty adopted on 20 December
35081 1996, or similar laws prohibiting or restricting circumvention of
35084 When you convey a covered work, you waive any legal power to forbid
35085 circumvention of technological measures to the extent such
35086 circumvention is effected by exercising rights under this License
35087 with respect to the covered work, and you disclaim any intention
35088 to limit operation or modification of the work as a means of
35089 enforcing, against the work's users, your or third parties' legal
35090 rights to forbid circumvention of technological measures.
35092 4. Conveying Verbatim Copies.
35094 You may convey verbatim copies of the Program's source code as you
35095 receive it, in any medium, provided that you conspicuously and
35096 appropriately publish on each copy an appropriate copyright notice;
35097 keep intact all notices stating that this License and any
35098 non-permissive terms added in accord with section 7 apply to the
35099 code; keep intact all notices of the absence of any warranty; and
35100 give all recipients a copy of this License along with the Program.
35102 You may charge any price or no price for each copy that you convey,
35103 and you may offer support or warranty protection for a fee.
35105 5. Conveying Modified Source Versions.
35107 You may convey a work based on the Program, or the modifications to
35108 produce it from the Program, in the form of source code under the
35109 terms of section 4, provided that you also meet all of these
35112 a. The work must carry prominent notices stating that you
35113 modified it, and giving a relevant date.
35115 b. The work must carry prominent notices stating that it is
35116 released under this License and any conditions added under
35117 section 7. This requirement modifies the requirement in
35118 section 4 to "keep intact all notices".
35120 c. You must license the entire work, as a whole, under this
35121 License to anyone who comes into possession of a copy. This
35122 License will therefore apply, along with any applicable
35123 section 7 additional terms, to the whole of the work, and all
35124 its parts, regardless of how they are packaged. This License
35125 gives no permission to license the work in any other way, but
35126 it does not invalidate such permission if you have separately
35129 d. If the work has interactive user interfaces, each must display
35130 Appropriate Legal Notices; however, if the Program has
35131 interactive interfaces that do not display Appropriate Legal
35132 Notices, your work need not make them do so.
35134 A compilation of a covered work with other separate and independent
35135 works, which are not by their nature extensions of the covered
35136 work, and which are not combined with it such as to form a larger
35137 program, in or on a volume of a storage or distribution medium, is
35138 called an "aggregate" if the compilation and its resulting
35139 copyright are not used to limit the access or legal rights of the
35140 compilation's users beyond what the individual works permit.
35141 Inclusion of a covered work in an aggregate does not cause this
35142 License to apply to the other parts of the aggregate.
35144 6. Conveying Non-Source Forms.
35146 You may convey a covered work in object code form under the terms
35147 of sections 4 and 5, provided that you also convey the
35148 machine-readable Corresponding Source under the terms of this
35149 License, in one of these ways:
35151 a. Convey the object code in, or embodied in, a physical product
35152 (including a physical distribution medium), accompanied by the
35153 Corresponding Source fixed on a durable physical medium
35154 customarily used for software interchange.
35156 b. Convey the object code in, or embodied in, a physical product
35157 (including a physical distribution medium), accompanied by a
35158 written offer, valid for at least three years and valid for
35159 as long as you offer spare parts or customer support for that
35160 product model, to give anyone who possesses the object code
35161 either (1) a copy of the Corresponding Source for all the
35162 software in the product that is covered by this License, on a
35163 durable physical medium customarily used for software
35164 interchange, for a price no more than your reasonable cost of
35165 physically performing this conveying of source, or (2) access
35166 to copy the Corresponding Source from a network server at no
35169 c. Convey individual copies of the object code with a copy of
35170 the written offer to provide the Corresponding Source. This
35171 alternative is allowed only occasionally and noncommercially,
35172 and only if you received the object code with such an offer,
35173 in accord with subsection 6b.
35175 d. Convey the object code by offering access from a designated
35176 place (gratis or for a charge), and offer equivalent access
35177 to the Corresponding Source in the same way through the same
35178 place at no further charge. You need not require recipients
35179 to copy the Corresponding Source along with the object code.
35180 If the place to copy the object code is a network server, the
35181 Corresponding Source may be on a different server (operated
35182 by you or a third party) that supports equivalent copying
35183 facilities, provided you maintain clear directions next to
35184 the object code saying where to find the Corresponding Source.
35185 Regardless of what server hosts the Corresponding Source, you
35186 remain obligated to ensure that it is available for as long
35187 as needed to satisfy these requirements.
35189 e. Convey the object code using peer-to-peer transmission,
35190 provided you inform other peers where the object code and
35191 Corresponding Source of the work are being offered to the
35192 general public at no charge under subsection 6d.
35195 A separable portion of the object code, whose source code is
35196 excluded from the Corresponding Source as a System Library, need
35197 not be included in conveying the object code work.
35199 A "User Product" is either (1) a "consumer product", which means
35200 any tangible personal property which is normally used for personal,
35201 family, or household purposes, or (2) anything designed or sold for
35202 incorporation into a dwelling. In determining whether a product
35203 is a consumer product, doubtful cases shall be resolved in favor of
35204 coverage. For a particular product received by a particular user,
35205 "normally used" refers to a typical or common use of that class of
35206 product, regardless of the status of the particular user or of the
35207 way in which the particular user actually uses, or expects or is
35208 expected to use, the product. A product is a consumer product
35209 regardless of whether the product has substantial commercial,
35210 industrial or non-consumer uses, unless such uses represent the
35211 only significant mode of use of the product.
35213 "Installation Information" for a User Product means any methods,
35214 procedures, authorization keys, or other information required to
35215 install and execute modified versions of a covered work in that
35216 User Product from a modified version of its Corresponding Source.
35217 The information must suffice to ensure that the continued
35218 functioning of the modified object code is in no case prevented or
35219 interfered with solely because modification has been made.
35221 If you convey an object code work under this section in, or with,
35222 or specifically for use in, a User Product, and the conveying
35223 occurs as part of a transaction in which the right of possession
35224 and use of the User Product is transferred to the recipient in
35225 perpetuity or for a fixed term (regardless of how the transaction
35226 is characterized), the Corresponding Source conveyed under this
35227 section must be accompanied by the Installation Information. But
35228 this requirement does not apply if neither you nor any third party
35229 retains the ability to install modified object code on the User
35230 Product (for example, the work has been installed in ROM).
35232 The requirement to provide Installation Information does not
35233 include a requirement to continue to provide support service,
35234 warranty, or updates for a work that has been modified or
35235 installed by the recipient, or for the User Product in which it
35236 has been modified or installed. Access to a network may be denied
35237 when the modification itself materially and adversely affects the
35238 operation of the network or violates the rules and protocols for
35239 communication across the network.
35241 Corresponding Source conveyed, and Installation Information
35242 provided, in accord with this section must be in a format that is
35243 publicly documented (and with an implementation available to the
35244 public in source code form), and must require no special password
35245 or key for unpacking, reading or copying.
35247 7. Additional Terms.
35249 "Additional permissions" are terms that supplement the terms of
35250 this License by making exceptions from one or more of its
35251 conditions. Additional permissions that are applicable to the
35252 entire Program shall be treated as though they were included in
35253 this License, to the extent that they are valid under applicable
35254 law. If additional permissions apply only to part of the Program,
35255 that part may be used separately under those permissions, but the
35256 entire Program remains governed by this License without regard to
35257 the additional permissions.
35259 When you convey a copy of a covered work, you may at your option
35260 remove any additional permissions from that copy, or from any part
35261 of it. (Additional permissions may be written to require their own
35262 removal in certain cases when you modify the work.) You may place
35263 additional permissions on material, added by you to a covered work,
35264 for which you have or can give appropriate copyright permission.
35266 Notwithstanding any other provision of this License, for material
35267 you add to a covered work, you may (if authorized by the copyright
35268 holders of that material) supplement the terms of this License
35271 a. Disclaiming warranty or limiting liability differently from
35272 the terms of sections 15 and 16 of this License; or
35274 b. Requiring preservation of specified reasonable legal notices
35275 or author attributions in that material or in the Appropriate
35276 Legal Notices displayed by works containing it; or
35278 c. Prohibiting misrepresentation of the origin of that material,
35279 or requiring that modified versions of such material be
35280 marked in reasonable ways as different from the original
35283 d. Limiting the use for publicity purposes of names of licensors
35284 or authors of the material; or
35286 e. Declining to grant rights under trademark law for use of some
35287 trade names, trademarks, or service marks; or
35289 f. Requiring indemnification of licensors and authors of that
35290 material by anyone who conveys the material (or modified
35291 versions of it) with contractual assumptions of liability to
35292 the recipient, for any liability that these contractual
35293 assumptions directly impose on those licensors and authors.
35295 All other non-permissive additional terms are considered "further
35296 restrictions" within the meaning of section 10. If the Program as
35297 you received it, or any part of it, contains a notice stating that
35298 it is governed by this License along with a term that is a further
35299 restriction, you may remove that term. If a license document
35300 contains a further restriction but permits relicensing or
35301 conveying under this License, you may add to a covered work
35302 material governed by the terms of that license document, provided
35303 that the further restriction does not survive such relicensing or
35306 If you add terms to a covered work in accord with this section, you
35307 must place, in the relevant source files, a statement of the
35308 additional terms that apply to those files, or a notice indicating
35309 where to find the applicable terms.
35311 Additional terms, permissive or non-permissive, may be stated in
35312 the form of a separately written license, or stated as exceptions;
35313 the above requirements apply either way.
35317 You may not propagate or modify a covered work except as expressly
35318 provided under this License. Any attempt otherwise to propagate or
35319 modify it is void, and will automatically terminate your rights
35320 under this License (including any patent licenses granted under
35321 the third paragraph of section 11).
35323 However, if you cease all violation of this License, then your
35324 license from a particular copyright holder is reinstated (a)
35325 provisionally, unless and until the copyright holder explicitly
35326 and finally terminates your license, and (b) permanently, if the
35327 copyright holder fails to notify you of the violation by some
35328 reasonable means prior to 60 days after the cessation.
35330 Moreover, your license from a particular copyright holder is
35331 reinstated permanently if the copyright holder notifies you of the
35332 violation by some reasonable means, this is the first time you have
35333 received notice of violation of this License (for any work) from
35334 that copyright holder, and you cure the violation prior to 30 days
35335 after your receipt of the notice.
35337 Termination of your rights under this section does not terminate
35338 the licenses of parties who have received copies or rights from
35339 you under this License. If your rights have been terminated and
35340 not permanently reinstated, you do not qualify to receive new
35341 licenses for the same material under section 10.
35343 9. Acceptance Not Required for Having Copies.
35345 You are not required to accept this License in order to receive or
35346 run a copy of the Program. Ancillary propagation of a covered work
35347 occurring solely as a consequence of using peer-to-peer
35348 transmission to receive a copy likewise does not require
35349 acceptance. However, nothing other than this License grants you
35350 permission to propagate or modify any covered work. These actions
35351 infringe copyright if you do not accept this License. Therefore,
35352 by modifying or propagating a covered work, you indicate your
35353 acceptance of this License to do so.
35355 10. Automatic Licensing of Downstream Recipients.
35357 Each time you convey a covered work, the recipient automatically
35358 receives a license from the original licensors, to run, modify and
35359 propagate that work, subject to this License. You are not
35360 responsible for enforcing compliance by third parties with this
35363 An "entity transaction" is a transaction transferring control of an
35364 organization, or substantially all assets of one, or subdividing an
35365 organization, or merging organizations. If propagation of a
35366 covered work results from an entity transaction, each party to that
35367 transaction who receives a copy of the work also receives whatever
35368 licenses to the work the party's predecessor in interest had or
35369 could give under the previous paragraph, plus a right to
35370 possession of the Corresponding Source of the work from the
35371 predecessor in interest, if the predecessor has it or can get it
35372 with reasonable efforts.
35374 You may not impose any further restrictions on the exercise of the
35375 rights granted or affirmed under this License. For example, you
35376 may not impose a license fee, royalty, or other charge for
35377 exercise of rights granted under this License, and you may not
35378 initiate litigation (including a cross-claim or counterclaim in a
35379 lawsuit) alleging that any patent claim is infringed by making,
35380 using, selling, offering for sale, or importing the Program or any
35385 A "contributor" is a copyright holder who authorizes use under this
35386 License of the Program or a work on which the Program is based.
35387 The work thus licensed is called the contributor's "contributor
35390 A contributor's "essential patent claims" are all patent claims
35391 owned or controlled by the contributor, whether already acquired or
35392 hereafter acquired, that would be infringed by some manner,
35393 permitted by this License, of making, using, or selling its
35394 contributor version, but do not include claims that would be
35395 infringed only as a consequence of further modification of the
35396 contributor version. For purposes of this definition, "control"
35397 includes the right to grant patent sublicenses in a manner
35398 consistent with the requirements of this License.
35400 Each contributor grants you a non-exclusive, worldwide,
35401 royalty-free patent license under the contributor's essential
35402 patent claims, to make, use, sell, offer for sale, import and
35403 otherwise run, modify and propagate the contents of its
35404 contributor version.
35406 In the following three paragraphs, a "patent license" is any
35407 express agreement or commitment, however denominated, not to
35408 enforce a patent (such as an express permission to practice a
35409 patent or covenant not to sue for patent infringement). To
35410 "grant" such a patent license to a party means to make such an
35411 agreement or commitment not to enforce a patent against the party.
35413 If you convey a covered work, knowingly relying on a patent
35414 license, and the Corresponding Source of the work is not available
35415 for anyone to copy, free of charge and under the terms of this
35416 License, through a publicly available network server or other
35417 readily accessible means, then you must either (1) cause the
35418 Corresponding Source to be so available, or (2) arrange to deprive
35419 yourself of the benefit of the patent license for this particular
35420 work, or (3) arrange, in a manner consistent with the requirements
35421 of this License, to extend the patent license to downstream
35422 recipients. "Knowingly relying" means you have actual knowledge
35423 that, but for the patent license, your conveying the covered work
35424 in a country, or your recipient's use of the covered work in a
35425 country, would infringe one or more identifiable patents in that
35426 country that you have reason to believe are valid.
35428 If, pursuant to or in connection with a single transaction or
35429 arrangement, you convey, or propagate by procuring conveyance of, a
35430 covered work, and grant a patent license to some of the parties
35431 receiving the covered work authorizing them to use, propagate,
35432 modify or convey a specific copy of the covered work, then the
35433 patent license you grant is automatically extended to all
35434 recipients of the covered work and works based on it.
35436 A patent license is "discriminatory" if it does not include within
35437 the scope of its coverage, prohibits the exercise of, or is
35438 conditioned on the non-exercise of one or more of the rights that
35439 are specifically granted under this License. You may not convey a
35440 covered work if you are a party to an arrangement with a third
35441 party that is in the business of distributing software, under
35442 which you make payment to the third party based on the extent of
35443 your activity of conveying the work, and under which the third
35444 party grants, to any of the parties who would receive the covered
35445 work from you, a discriminatory patent license (a) in connection
35446 with copies of the covered work conveyed by you (or copies made
35447 from those copies), or (b) primarily for and in connection with
35448 specific products or compilations that contain the covered work,
35449 unless you entered into that arrangement, or that patent license
35450 was granted, prior to 28 March 2007.
35452 Nothing in this License shall be construed as excluding or limiting
35453 any implied license or other defenses to infringement that may
35454 otherwise be available to you under applicable patent law.
35456 12. No Surrender of Others' Freedom.
35458 If conditions are imposed on you (whether by court order,
35459 agreement or otherwise) that contradict the conditions of this
35460 License, they do not excuse you from the conditions of this
35461 License. If you cannot convey a covered work so as to satisfy
35462 simultaneously your obligations under this License and any other
35463 pertinent obligations, then as a consequence you may not convey it
35464 at all. For example, if you agree to terms that obligate you to
35465 collect a royalty for further conveying from those to whom you
35466 convey the Program, the only way you could satisfy both those
35467 terms and this License would be to refrain entirely from conveying
35470 13. Use with the GNU Affero General Public License.
35472 Notwithstanding any other provision of this License, you have
35473 permission to link or combine any covered work with a work licensed
35474 under version 3 of the GNU Affero General Public License into a
35475 single combined work, and to convey the resulting work. The terms
35476 of this License will continue to apply to the part which is the
35477 covered work, but the special requirements of the GNU Affero
35478 General Public License, section 13, concerning interaction through
35479 a network will apply to the combination as such.
35481 14. Revised Versions of this License.
35483 The Free Software Foundation may publish revised and/or new
35484 versions of the GNU General Public License from time to time.
35485 Such new versions will be similar in spirit to the present
35486 version, but may differ in detail to address new problems or
35489 Each version is given a distinguishing version number. If the
35490 Program specifies that a certain numbered version of the GNU
35491 General Public License "or any later version" applies to it, you
35492 have the option of following the terms and conditions either of
35493 that numbered version or of any later version published by the
35494 Free Software Foundation. If the Program does not specify a
35495 version number of the GNU General Public License, you may choose
35496 any version ever published by the Free Software Foundation.
35498 If the Program specifies that a proxy can decide which future
35499 versions of the GNU General Public License can be used, that
35500 proxy's public statement of acceptance of a version permanently
35501 authorizes you to choose that version for the Program.
35503 Later license versions may give you additional or different
35504 permissions. However, no additional obligations are imposed on any
35505 author or copyright holder as a result of your choosing to follow a
35508 15. Disclaimer of Warranty.
35510 THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
35511 APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE
35512 COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS"
35513 WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED,
35514 INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
35515 MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE
35516 RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.
35517 SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL
35518 NECESSARY SERVICING, REPAIR OR CORRECTION.
35520 16. Limitation of Liability.
35522 IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN
35523 WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES
35524 AND/OR CONVEYS THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU
35525 FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR
35526 CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE
35527 THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA
35528 BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
35529 PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER
35530 PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF
35531 THE POSSIBILITY OF SUCH DAMAGES.
35533 17. Interpretation of Sections 15 and 16.
35535 If the disclaimer of warranty and limitation of liability provided
35536 above cannot be given local legal effect according to their terms,
35537 reviewing courts shall apply local law that most closely
35538 approximates an absolute waiver of all civil liability in
35539 connection with the Program, unless a warranty or assumption of
35540 liability accompanies a copy of the Program in return for a fee.
35543 END OF TERMS AND CONDITIONS
35544 ===========================
35546 How to Apply These Terms to Your New Programs
35547 =============================================
35549 If you develop a new program, and you want it to be of the greatest
35550 possible use to the public, the best way to achieve this is to make it
35551 free software which everyone can redistribute and change under these
35554 To do so, attach the following notices to the program. It is safest
35555 to attach them to the start of each source file to most effectively
35556 state the exclusion of warranty; and each file should have at least the
35557 "copyright" line and a pointer to where the full notice is found.
35559 ONE LINE TO GIVE THE PROGRAM'S NAME AND A BRIEF IDEA OF WHAT IT DOES.
35560 Copyright (C) YEAR NAME OF AUTHOR
35562 This program is free software: you can redistribute it and/or modify
35563 it under the terms of the GNU General Public License as published by
35564 the Free Software Foundation, either version 3 of the License, or (at
35565 your option) any later version.
35567 This program is distributed in the hope that it will be useful, but
35568 WITHOUT ANY WARRANTY; without even the implied warranty of
35569 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
35570 General Public License for more details.
35572 You should have received a copy of the GNU General Public License
35573 along with this program. If not, see `http://www.gnu.org/licenses/'.
35575 Also add information on how to contact you by electronic and paper
35578 If the program does terminal interaction, make it output a short
35579 notice like this when it starts in an interactive mode:
35581 PROGRAM Copyright (C) YEAR NAME OF AUTHOR
35582 This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
35583 This is free software, and you are welcome to redistribute it
35584 under certain conditions; type `show c' for details.
35586 The hypothetical commands `show w' and `show c' should show the
35587 appropriate parts of the General Public License. Of course, your
35588 program's commands might be different; for a GUI interface, you would
35589 use an "about box".
35591 You should also get your employer (if you work as a programmer) or
35592 school, if any, to sign a "copyright disclaimer" for the program, if
35593 necessary. For more information on this, and how to apply and follow
35594 the GNU GPL, see `http://www.gnu.org/licenses/'.
35596 The GNU General Public License does not permit incorporating your
35597 program into proprietary programs. If your program is a subroutine
35598 library, you may consider it more useful to permit linking proprietary
35599 applications with the library. If this is what you want to do, use the
35600 GNU Lesser General Public License instead of this License. But first,
35601 please read `http://www.gnu.org/philosophy/why-not-lgpl.html'.
35604 File: gccint.info, Node: GNU Free Documentation License, Next: Contributors, Prev: Copying, Up: Top
35606 GNU Free Documentation License
35607 ******************************
35609 Version 1.2, November 2002
35611 Copyright (C) 2000,2001,2002 Free Software Foundation, Inc.
35612 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
35614 Everyone is permitted to copy and distribute verbatim copies
35615 of this license document, but changing it is not allowed.
35619 The purpose of this License is to make a manual, textbook, or other
35620 functional and useful document "free" in the sense of freedom: to
35621 assure everyone the effective freedom to copy and redistribute it,
35622 with or without modifying it, either commercially or
35623 noncommercially. Secondarily, this License preserves for the
35624 author and publisher a way to get credit for their work, while not
35625 being considered responsible for modifications made by others.
35627 This License is a kind of "copyleft", which means that derivative
35628 works of the document must themselves be free in the same sense.
35629 It complements the GNU General Public License, which is a copyleft
35630 license designed for free software.
35632 We have designed this License in order to use it for manuals for
35633 free software, because free software needs free documentation: a
35634 free program should come with manuals providing the same freedoms
35635 that the software does. But this License is not limited to
35636 software manuals; it can be used for any textual work, regardless
35637 of subject matter or whether it is published as a printed book.
35638 We recommend this License principally for works whose purpose is
35639 instruction or reference.
35641 1. APPLICABILITY AND DEFINITIONS
35643 This License applies to any manual or other work, in any medium,
35644 that contains a notice placed by the copyright holder saying it
35645 can be distributed under the terms of this License. Such a notice
35646 grants a world-wide, royalty-free license, unlimited in duration,
35647 to use that work under the conditions stated herein. The
35648 "Document", below, refers to any such manual or work. Any member
35649 of the public is a licensee, and is addressed as "you". You
35650 accept the license if you copy, modify or distribute the work in a
35651 way requiring permission under copyright law.
35653 A "Modified Version" of the Document means any work containing the
35654 Document or a portion of it, either copied verbatim, or with
35655 modifications and/or translated into another language.
35657 A "Secondary Section" is a named appendix or a front-matter section
35658 of the Document that deals exclusively with the relationship of the
35659 publishers or authors of the Document to the Document's overall
35660 subject (or to related matters) and contains nothing that could
35661 fall directly within that overall subject. (Thus, if the Document
35662 is in part a textbook of mathematics, a Secondary Section may not
35663 explain any mathematics.) The relationship could be a matter of
35664 historical connection with the subject or with related matters, or
35665 of legal, commercial, philosophical, ethical or political position
35668 The "Invariant Sections" are certain Secondary Sections whose
35669 titles are designated, as being those of Invariant Sections, in
35670 the notice that says that the Document is released under this
35671 License. If a section does not fit the above definition of
35672 Secondary then it is not allowed to be designated as Invariant.
35673 The Document may contain zero Invariant Sections. If the Document
35674 does not identify any Invariant Sections then there are none.
35676 The "Cover Texts" are certain short passages of text that are
35677 listed, as Front-Cover Texts or Back-Cover Texts, in the notice
35678 that says that the Document is released under this License. A
35679 Front-Cover Text may be at most 5 words, and a Back-Cover Text may
35680 be at most 25 words.
35682 A "Transparent" copy of the Document means a machine-readable copy,
35683 represented in a format whose specification is available to the
35684 general public, that is suitable for revising the document
35685 straightforwardly with generic text editors or (for images
35686 composed of pixels) generic paint programs or (for drawings) some
35687 widely available drawing editor, and that is suitable for input to
35688 text formatters or for automatic translation to a variety of
35689 formats suitable for input to text formatters. A copy made in an
35690 otherwise Transparent file format whose markup, or absence of
35691 markup, has been arranged to thwart or discourage subsequent
35692 modification by readers is not Transparent. An image format is
35693 not Transparent if used for any substantial amount of text. A
35694 copy that is not "Transparent" is called "Opaque".
35696 Examples of suitable formats for Transparent copies include plain
35697 ASCII without markup, Texinfo input format, LaTeX input format,
35698 SGML or XML using a publicly available DTD, and
35699 standard-conforming simple HTML, PostScript or PDF designed for
35700 human modification. Examples of transparent image formats include
35701 PNG, XCF and JPG. Opaque formats include proprietary formats that
35702 can be read and edited only by proprietary word processors, SGML or
35703 XML for which the DTD and/or processing tools are not generally
35704 available, and the machine-generated HTML, PostScript or PDF
35705 produced by some word processors for output purposes only.
35707 The "Title Page" means, for a printed book, the title page itself,
35708 plus such following pages as are needed to hold, legibly, the
35709 material this License requires to appear in the title page. For
35710 works in formats which do not have any title page as such, "Title
35711 Page" means the text near the most prominent appearance of the
35712 work's title, preceding the beginning of the body of the text.
35714 A section "Entitled XYZ" means a named subunit of the Document
35715 whose title either is precisely XYZ or contains XYZ in parentheses
35716 following text that translates XYZ in another language. (Here XYZ
35717 stands for a specific section name mentioned below, such as
35718 "Acknowledgements", "Dedications", "Endorsements", or "History".)
35719 To "Preserve the Title" of such a section when you modify the
35720 Document means that it remains a section "Entitled XYZ" according
35721 to this definition.
35723 The Document may include Warranty Disclaimers next to the notice
35724 which states that this License applies to the Document. These
35725 Warranty Disclaimers are considered to be included by reference in
35726 this License, but only as regards disclaiming warranties: any other
35727 implication that these Warranty Disclaimers may have is void and
35728 has no effect on the meaning of this License.
35730 2. VERBATIM COPYING
35732 You may copy and distribute the Document in any medium, either
35733 commercially or noncommercially, provided that this License, the
35734 copyright notices, and the license notice saying this License
35735 applies to the Document are reproduced in all copies, and that you
35736 add no other conditions whatsoever to those of this License. You
35737 may not use technical measures to obstruct or control the reading
35738 or further copying of the copies you make or distribute. However,
35739 you may accept compensation in exchange for copies. If you
35740 distribute a large enough number of copies you must also follow
35741 the conditions in section 3.
35743 You may also lend copies, under the same conditions stated above,
35744 and you may publicly display copies.
35746 3. COPYING IN QUANTITY
35748 If you publish printed copies (or copies in media that commonly
35749 have printed covers) of the Document, numbering more than 100, and
35750 the Document's license notice requires Cover Texts, you must
35751 enclose the copies in covers that carry, clearly and legibly, all
35752 these Cover Texts: Front-Cover Texts on the front cover, and
35753 Back-Cover Texts on the back cover. Both covers must also clearly
35754 and legibly identify you as the publisher of these copies. The
35755 front cover must present the full title with all words of the
35756 title equally prominent and visible. You may add other material
35757 on the covers in addition. Copying with changes limited to the
35758 covers, as long as they preserve the title of the Document and
35759 satisfy these conditions, can be treated as verbatim copying in
35762 If the required texts for either cover are too voluminous to fit
35763 legibly, you should put the first ones listed (as many as fit
35764 reasonably) on the actual cover, and continue the rest onto
35767 If you publish or distribute Opaque copies of the Document
35768 numbering more than 100, you must either include a
35769 machine-readable Transparent copy along with each Opaque copy, or
35770 state in or with each Opaque copy a computer-network location from
35771 which the general network-using public has access to download
35772 using public-standard network protocols a complete Transparent
35773 copy of the Document, free of added material. If you use the
35774 latter option, you must take reasonably prudent steps, when you
35775 begin distribution of Opaque copies in quantity, to ensure that
35776 this Transparent copy will remain thus accessible at the stated
35777 location until at least one year after the last time you
35778 distribute an Opaque copy (directly or through your agents or
35779 retailers) of that edition to the public.
35781 It is requested, but not required, that you contact the authors of
35782 the Document well before redistributing any large number of
35783 copies, to give them a chance to provide you with an updated
35784 version of the Document.
35788 You may copy and distribute a Modified Version of the Document
35789 under the conditions of sections 2 and 3 above, provided that you
35790 release the Modified Version under precisely this License, with
35791 the Modified Version filling the role of the Document, thus
35792 licensing distribution and modification of the Modified Version to
35793 whoever possesses a copy of it. In addition, you must do these
35794 things in the Modified Version:
35796 A. Use in the Title Page (and on the covers, if any) a title
35797 distinct from that of the Document, and from those of
35798 previous versions (which should, if there were any, be listed
35799 in the History section of the Document). You may use the
35800 same title as a previous version if the original publisher of
35801 that version gives permission.
35803 B. List on the Title Page, as authors, one or more persons or
35804 entities responsible for authorship of the modifications in
35805 the Modified Version, together with at least five of the
35806 principal authors of the Document (all of its principal
35807 authors, if it has fewer than five), unless they release you
35808 from this requirement.
35810 C. State on the Title page the name of the publisher of the
35811 Modified Version, as the publisher.
35813 D. Preserve all the copyright notices of the Document.
35815 E. Add an appropriate copyright notice for your modifications
35816 adjacent to the other copyright notices.
35818 F. Include, immediately after the copyright notices, a license
35819 notice giving the public permission to use the Modified
35820 Version under the terms of this License, in the form shown in
35821 the Addendum below.
35823 G. Preserve in that license notice the full lists of Invariant
35824 Sections and required Cover Texts given in the Document's
35827 H. Include an unaltered copy of this License.
35829 I. Preserve the section Entitled "History", Preserve its Title,
35830 and add to it an item stating at least the title, year, new
35831 authors, and publisher of the Modified Version as given on
35832 the Title Page. If there is no section Entitled "History" in
35833 the Document, create one stating the title, year, authors,
35834 and publisher of the Document as given on its Title Page,
35835 then add an item describing the Modified Version as stated in
35836 the previous sentence.
35838 J. Preserve the network location, if any, given in the Document
35839 for public access to a Transparent copy of the Document, and
35840 likewise the network locations given in the Document for
35841 previous versions it was based on. These may be placed in
35842 the "History" section. You may omit a network location for a
35843 work that was published at least four years before the
35844 Document itself, or if the original publisher of the version
35845 it refers to gives permission.
35847 K. For any section Entitled "Acknowledgements" or "Dedications",
35848 Preserve the Title of the section, and preserve in the
35849 section all the substance and tone of each of the contributor
35850 acknowledgements and/or dedications given therein.
35852 L. Preserve all the Invariant Sections of the Document,
35853 unaltered in their text and in their titles. Section numbers
35854 or the equivalent are not considered part of the section
35857 M. Delete any section Entitled "Endorsements". Such a section
35858 may not be included in the Modified Version.
35860 N. Do not retitle any existing section to be Entitled
35861 "Endorsements" or to conflict in title with any Invariant
35864 O. Preserve any Warranty Disclaimers.
35866 If the Modified Version includes new front-matter sections or
35867 appendices that qualify as Secondary Sections and contain no
35868 material copied from the Document, you may at your option
35869 designate some or all of these sections as invariant. To do this,
35870 add their titles to the list of Invariant Sections in the Modified
35871 Version's license notice. These titles must be distinct from any
35872 other section titles.
35874 You may add a section Entitled "Endorsements", provided it contains
35875 nothing but endorsements of your Modified Version by various
35876 parties--for example, statements of peer review or that the text
35877 has been approved by an organization as the authoritative
35878 definition of a standard.
35880 You may add a passage of up to five words as a Front-Cover Text,
35881 and a passage of up to 25 words as a Back-Cover Text, to the end
35882 of the list of Cover Texts in the Modified Version. Only one
35883 passage of Front-Cover Text and one of Back-Cover Text may be
35884 added by (or through arrangements made by) any one entity. If the
35885 Document already includes a cover text for the same cover,
35886 previously added by you or by arrangement made by the same entity
35887 you are acting on behalf of, you may not add another; but you may
35888 replace the old one, on explicit permission from the previous
35889 publisher that added the old one.
35891 The author(s) and publisher(s) of the Document do not by this
35892 License give permission to use their names for publicity for or to
35893 assert or imply endorsement of any Modified Version.
35895 5. COMBINING DOCUMENTS
35897 You may combine the Document with other documents released under
35898 this License, under the terms defined in section 4 above for
35899 modified versions, provided that you include in the combination
35900 all of the Invariant Sections of all of the original documents,
35901 unmodified, and list them all as Invariant Sections of your
35902 combined work in its license notice, and that you preserve all
35903 their Warranty Disclaimers.
35905 The combined work need only contain one copy of this License, and
35906 multiple identical Invariant Sections may be replaced with a single
35907 copy. If there are multiple Invariant Sections with the same name
35908 but different contents, make the title of each such section unique
35909 by adding at the end of it, in parentheses, the name of the
35910 original author or publisher of that section if known, or else a
35911 unique number. Make the same adjustment to the section titles in
35912 the list of Invariant Sections in the license notice of the
35915 In the combination, you must combine any sections Entitled
35916 "History" in the various original documents, forming one section
35917 Entitled "History"; likewise combine any sections Entitled
35918 "Acknowledgements", and any sections Entitled "Dedications". You
35919 must delete all sections Entitled "Endorsements."
35921 6. COLLECTIONS OF DOCUMENTS
35923 You may make a collection consisting of the Document and other
35924 documents released under this License, and replace the individual
35925 copies of this License in the various documents with a single copy
35926 that is included in the collection, provided that you follow the
35927 rules of this License for verbatim copying of each of the
35928 documents in all other respects.
35930 You may extract a single document from such a collection, and
35931 distribute it individually under this License, provided you insert
35932 a copy of this License into the extracted document, and follow
35933 this License in all other respects regarding verbatim copying of
35936 7. AGGREGATION WITH INDEPENDENT WORKS
35938 A compilation of the Document or its derivatives with other
35939 separate and independent documents or works, in or on a volume of
35940 a storage or distribution medium, is called an "aggregate" if the
35941 copyright resulting from the compilation is not used to limit the
35942 legal rights of the compilation's users beyond what the individual
35943 works permit. When the Document is included in an aggregate, this
35944 License does not apply to the other works in the aggregate which
35945 are not themselves derivative works of the Document.
35947 If the Cover Text requirement of section 3 is applicable to these
35948 copies of the Document, then if the Document is less than one half
35949 of the entire aggregate, the Document's Cover Texts may be placed
35950 on covers that bracket the Document within the aggregate, or the
35951 electronic equivalent of covers if the Document is in electronic
35952 form. Otherwise they must appear on printed covers that bracket
35953 the whole aggregate.
35957 Translation is considered a kind of modification, so you may
35958 distribute translations of the Document under the terms of section
35959 4. Replacing Invariant Sections with translations requires special
35960 permission from their copyright holders, but you may include
35961 translations of some or all Invariant Sections in addition to the
35962 original versions of these Invariant Sections. You may include a
35963 translation of this License, and all the license notices in the
35964 Document, and any Warranty Disclaimers, provided that you also
35965 include the original English version of this License and the
35966 original versions of those notices and disclaimers. In case of a
35967 disagreement between the translation and the original version of
35968 this License or a notice or disclaimer, the original version will
35971 If a section in the Document is Entitled "Acknowledgements",
35972 "Dedications", or "History", the requirement (section 4) to
35973 Preserve its Title (section 1) will typically require changing the
35978 You may not copy, modify, sublicense, or distribute the Document
35979 except as expressly provided for under this License. Any other
35980 attempt to copy, modify, sublicense or distribute the Document is
35981 void, and will automatically terminate your rights under this
35982 License. However, parties who have received copies, or rights,
35983 from you under this License will not have their licenses
35984 terminated so long as such parties remain in full compliance.
35986 10. FUTURE REVISIONS OF THIS LICENSE
35988 The Free Software Foundation may publish new, revised versions of
35989 the GNU Free Documentation License from time to time. Such new
35990 versions will be similar in spirit to the present version, but may
35991 differ in detail to address new problems or concerns. See
35992 `http://www.gnu.org/copyleft/'.
35994 Each version of the License is given a distinguishing version
35995 number. If the Document specifies that a particular numbered
35996 version of this License "or any later version" applies to it, you
35997 have the option of following the terms and conditions either of
35998 that specified version or of any later version that has been
35999 published (not as a draft) by the Free Software Foundation. If
36000 the Document does not specify a version number of this License,
36001 you may choose any version ever published (not as a draft) by the
36002 Free Software Foundation.
36004 ADDENDUM: How to use this License for your documents
36005 ====================================================
36007 To use this License in a document you have written, include a copy of
36008 the License in the document and put the following copyright and license
36009 notices just after the title page:
36011 Copyright (C) YEAR YOUR NAME.
36012 Permission is granted to copy, distribute and/or modify this document
36013 under the terms of the GNU Free Documentation License, Version 1.2
36014 or any later version published by the Free Software Foundation;
36015 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
36016 Texts. A copy of the license is included in the section entitled ``GNU
36017 Free Documentation License''.
36019 If you have Invariant Sections, Front-Cover Texts and Back-Cover Texts,
36020 replace the "with...Texts." line with this:
36022 with the Invariant Sections being LIST THEIR TITLES, with
36023 the Front-Cover Texts being LIST, and with the Back-Cover Texts
36026 If you have Invariant Sections without Cover Texts, or some other
36027 combination of the three, merge those two alternatives to suit the
36030 If your document contains nontrivial examples of program code, we
36031 recommend releasing these examples in parallel under your choice of
36032 free software license, such as the GNU General Public License, to
36033 permit their use in free software.
36036 File: gccint.info, Node: Contributors, Next: Option Index, Prev: GNU Free Documentation License, Up: Top
36038 Contributors to GCC
36039 *******************
36041 The GCC project would like to thank its many contributors. Without
36042 them the project would not have been nearly as successful as it has
36043 been. Any omissions in this list are accidental. Feel free to contact
36044 <law@redhat.com> or <gerald@pfeifer.com> if you have been left out or
36045 some of your contributions are not listed. Please keep this list in
36046 alphabetical order.
36048 * Analog Devices helped implement the support for complex data types
36051 * John David Anglin for threading-related fixes and improvements to
36052 libstdc++-v3, and the HP-UX port.
36054 * James van Artsdalen wrote the code that makes efficient use of the
36055 Intel 80387 register stack.
36057 * Abramo and Roberto Bagnara for the SysV68 Motorola 3300 Delta
36060 * Alasdair Baird for various bug fixes.
36062 * Giovanni Bajo for analyzing lots of complicated C++ problem
36065 * Peter Barada for his work to improve code generation for new
36068 * Gerald Baumgartner added the signature extension to the C++ front
36071 * Godmar Back for his Java improvements and encouragement.
36073 * Scott Bambrough for help porting the Java compiler.
36075 * Wolfgang Bangerth for processing tons of bug reports.
36077 * Jon Beniston for his Microsoft Windows port of Java.
36079 * Daniel Berlin for better DWARF2 support, faster/better
36080 optimizations, improved alias analysis, plus migrating GCC to
36083 * Geoff Berry for his Java object serialization work and various
36086 * Uros Bizjak for the implementation of x87 math built-in functions
36087 and for various middle end and i386 back end improvements and bug
36090 * Eric Blake for helping to make GCJ and libgcj conform to the
36093 * Janne Blomqvist for contributions to GNU Fortran.
36095 * Segher Boessenkool for various fixes.
36097 * Hans-J. Boehm for his garbage collector, IA-64 libffi port, and
36100 * Neil Booth for work on cpplib, lang hooks, debug hooks and other
36101 miscellaneous clean-ups.
36103 * Steven Bosscher for integrating the GNU Fortran front end into GCC
36104 and for contributing to the tree-ssa branch.
36106 * Eric Botcazou for fixing middle- and backend bugs left and right.
36108 * Per Bothner for his direction via the steering committee and
36109 various improvements to the infrastructure for supporting new
36110 languages. Chill front end implementation. Initial
36111 implementations of cpplib, fix-header, config.guess, libio, and
36112 past C++ library (libg++) maintainer. Dreaming up, designing and
36113 implementing much of GCJ.
36115 * Devon Bowen helped port GCC to the Tahoe.
36117 * Don Bowman for mips-vxworks contributions.
36119 * Dave Brolley for work on cpplib and Chill.
36121 * Paul Brook for work on the ARM architecture and maintaining GNU
36124 * Robert Brown implemented the support for Encore 32000 systems.
36126 * Christian Bruel for improvements to local store elimination.
36128 * Herman A.J. ten Brugge for various fixes.
36130 * Joerg Brunsmann for Java compiler hacking and help with the GCJ
36133 * Joe Buck for his direction via the steering committee.
36135 * Craig Burley for leadership of the G77 Fortran effort.
36137 * Stephan Buys for contributing Doxygen notes for libstdc++.
36139 * Paolo Carlini for libstdc++ work: lots of efficiency improvements
36140 to the C++ strings, streambufs and formatted I/O, hard detective
36141 work on the frustrating localization issues, and keeping up with
36142 the problem reports.
36144 * John Carr for his alias work, SPARC hacking, infrastructure
36145 improvements, previous contributions to the steering committee,
36146 loop optimizations, etc.
36148 * Stephane Carrez for 68HC11 and 68HC12 ports.
36150 * Steve Chamberlain for support for the Renesas SH and H8 processors
36151 and the PicoJava processor, and for GCJ config fixes.
36153 * Glenn Chambers for help with the GCJ FAQ.
36155 * John-Marc Chandonia for various libgcj patches.
36157 * Scott Christley for his Objective-C contributions.
36159 * Eric Christopher for his Java porting help and clean-ups.
36161 * Branko Cibej for more warning contributions.
36163 * The GNU Classpath project for all of their merged runtime code.
36165 * Nick Clifton for arm, mcore, fr30, v850, m32r work, `--help', and
36166 other random hacking.
36168 * Michael Cook for libstdc++ cleanup patches to reduce warnings.
36170 * R. Kelley Cook for making GCC buildable from a read-only directory
36171 as well as other miscellaneous build process and documentation
36174 * Ralf Corsepius for SH testing and minor bug fixing.
36176 * Stan Cox for care and feeding of the x86 port and lots of behind
36177 the scenes hacking.
36179 * Alex Crain provided changes for the 3b1.
36181 * Ian Dall for major improvements to the NS32k port.
36183 * Paul Dale for his work to add uClinux platform support to the m68k
36186 * Dario Dariol contributed the four varieties of sample programs
36187 that print a copy of their source.
36189 * Russell Davidson for fstream and stringstream fixes in libstdc++.
36191 * Bud Davis for work on the G77 and GNU Fortran compilers.
36193 * Mo DeJong for GCJ and libgcj bug fixes.
36195 * DJ Delorie for the DJGPP port, build and libiberty maintenance,
36196 various bug fixes, and the M32C port.
36198 * Arnaud Desitter for helping to debug GNU Fortran.
36200 * Gabriel Dos Reis for contributions to G++, contributions and
36201 maintenance of GCC diagnostics infrastructure, libstdc++-v3,
36202 including `valarray<>', `complex<>', maintaining the numerics
36203 library (including that pesky `<limits>' :-) and keeping
36204 up-to-date anything to do with numbers.
36206 * Ulrich Drepper for his work on glibc, testing of GCC using glibc,
36207 ISO C99 support, CFG dumping support, etc., plus support of the
36208 C++ runtime libraries including for all kinds of C interface
36209 issues, contributing and maintaining `complex<>', sanity checking
36210 and disbursement, configuration architecture, libio maintenance,
36211 and early math work.
36213 * Zdenek Dvorak for a new loop unroller and various fixes.
36215 * Richard Earnshaw for his ongoing work with the ARM.
36217 * David Edelsohn for his direction via the steering committee,
36218 ongoing work with the RS6000/PowerPC port, help cleaning up Haifa
36219 loop changes, doing the entire AIX port of libstdc++ with his bare
36220 hands, and for ensuring GCC properly keeps working on AIX.
36222 * Kevin Ediger for the floating point formatting of num_put::do_put
36225 * Phil Edwards for libstdc++ work including configuration hackery,
36226 documentation maintainer, chief breaker of the web pages, the
36227 occasional iostream bug fix, and work on shared library symbol
36230 * Paul Eggert for random hacking all over GCC.
36232 * Mark Elbrecht for various DJGPP improvements, and for libstdc++
36233 configuration support for locales and fstream-related fixes.
36235 * Vadim Egorov for libstdc++ fixes in strings, streambufs, and
36238 * Christian Ehrhardt for dealing with bug reports.
36240 * Ben Elliston for his work to move the Objective-C runtime into its
36241 own subdirectory and for his work on autoconf.
36243 * Revital Eres for work on the PowerPC 750CL port.
36245 * Marc Espie for OpenBSD support.
36247 * Doug Evans for much of the global optimization framework, arc,
36248 m32r, and SPARC work.
36250 * Christopher Faylor for his work on the Cygwin port and for caring
36251 and feeding the gcc.gnu.org box and saving its users tons of spam.
36253 * Fred Fish for BeOS support and Ada fixes.
36255 * Ivan Fontes Garcia for the Portuguese translation of the GCJ FAQ.
36257 * Peter Gerwinski for various bug fixes and the Pascal front end.
36259 * Kaveh R. Ghazi for his direction via the steering committee,
36260 amazing work to make `-W -Wall -W* -Werror' useful, and
36261 continuously testing GCC on a plethora of platforms. Kaveh
36262 extends his gratitude to the CAIP Center at Rutgers University for
36263 providing him with computing resources to work on Free Software
36264 since the late 1980s.
36266 * John Gilmore for a donation to the FSF earmarked improving GNU
36269 * Judy Goldberg for c++ contributions.
36271 * Torbjorn Granlund for various fixes and the c-torture testsuite,
36272 multiply- and divide-by-constant optimization, improved long long
36273 support, improved leaf function register allocation, and his
36274 direction via the steering committee.
36276 * Anthony Green for his `-Os' contributions and Java front end work.
36278 * Stu Grossman for gdb hacking, allowing GCJ developers to debug
36281 * Michael K. Gschwind contributed the port to the PDP-11.
36283 * Ron Guilmette implemented the `protoize' and `unprotoize' tools,
36284 the support for Dwarf symbolic debugging information, and much of
36285 the support for System V Release 4. He has also worked heavily on
36286 the Intel 386 and 860 support.
36288 * Mostafa Hagog for Swing Modulo Scheduling (SMS) and post reload
36291 * Bruno Haible for improvements in the runtime overhead for EH, new
36292 warnings and assorted bug fixes.
36294 * Andrew Haley for his amazing Java compiler and library efforts.
36296 * Chris Hanson assisted in making GCC work on HP-UX for the 9000
36299 * Michael Hayes for various thankless work he's done trying to get
36300 the c30/c40 ports functional. Lots of loop and unroll
36301 improvements and fixes.
36303 * Dara Hazeghi for wading through myriads of target-specific bug
36306 * Kate Hedstrom for staking the G77 folks with an initial testsuite.
36308 * Richard Henderson for his ongoing SPARC, alpha, ia32, and ia64
36309 work, loop opts, and generally fixing lots of old problems we've
36310 ignored for years, flow rewrite and lots of further stuff,
36311 including reviewing tons of patches.
36313 * Aldy Hernandez for working on the PowerPC port, SIMD support, and
36316 * Nobuyuki Hikichi of Software Research Associates, Tokyo,
36317 contributed the support for the Sony NEWS machine.
36319 * Kazu Hirata for caring and feeding the Renesas H8/300 port and
36322 * Katherine Holcomb for work on GNU Fortran.
36324 * Manfred Hollstein for his ongoing work to keep the m88k alive, lots
36325 of testing and bug fixing, particularly of GCC configury code.
36327 * Steve Holmgren for MachTen patches.
36329 * Jan Hubicka for his x86 port improvements.
36331 * Falk Hueffner for working on C and optimization bug reports.
36333 * Bernardo Innocenti for his m68k work, including merging of
36334 ColdFire improvements and uClinux support.
36336 * Christian Iseli for various bug fixes.
36338 * Kamil Iskra for general m68k hacking.
36340 * Lee Iverson for random fixes and MIPS testing.
36342 * Andreas Jaeger for testing and benchmarking of GCC and various bug
36345 * Jakub Jelinek for his SPARC work and sibling call optimizations as
36346 well as lots of bug fixes and test cases, and for improving the
36349 * Janis Johnson for ia64 testing and fixes, her quality improvement
36350 sidetracks, and web page maintenance.
36352 * Kean Johnston for SCO OpenServer support and various fixes.
36354 * Tim Josling for the sample language treelang based originally on
36355 Richard Kenner's "toy" language.
36357 * Nicolai Josuttis for additional libstdc++ documentation.
36359 * Klaus Kaempf for his ongoing work to make alpha-vms a viable
36362 * Steven G. Kargl for work on GNU Fortran.
36364 * David Kashtan of SRI adapted GCC to VMS.
36366 * Ryszard Kabatek for many, many libstdc++ bug fixes and
36367 optimizations of strings, especially member functions, and for
36370 * Geoffrey Keating for his ongoing work to make the PPC work for
36371 GNU/Linux and his automatic regression tester.
36373 * Brendan Kehoe for his ongoing work with G++ and for a lot of early
36374 work in just about every part of libstdc++.
36376 * Oliver M. Kellogg of Deutsche Aerospace contributed the port to the
36379 * Richard Kenner of the New York University Ultracomputer Research
36380 Laboratory wrote the machine descriptions for the AMD 29000, the
36381 DEC Alpha, the IBM RT PC, and the IBM RS/6000 as well as the
36382 support for instruction attributes. He also made changes to
36383 better support RISC processors including changes to common
36384 subexpression elimination, strength reduction, function calling
36385 sequence handling, and condition code support, in addition to
36386 generalizing the code for frame pointer elimination and delay slot
36387 scheduling. Richard Kenner was also the head maintainer of GCC
36390 * Mumit Khan for various contributions to the Cygwin and Mingw32
36391 ports and maintaining binary releases for Microsoft Windows hosts,
36392 and for massive libstdc++ porting work to Cygwin/Mingw32.
36394 * Robin Kirkham for cpu32 support.
36396 * Mark Klein for PA improvements.
36398 * Thomas Koenig for various bug fixes.
36400 * Bruce Korb for the new and improved fixincludes code.
36402 * Benjamin Kosnik for his G++ work and for leading the libstdc++-v3
36405 * Charles LaBrec contributed the support for the Integrated Solutions
36408 * Asher Langton and Mike Kumbera for contributing Cray pointer
36409 support to GNU Fortran, and for other GNU Fortran improvements.
36411 * Jeff Law for his direction via the steering committee,
36412 coordinating the entire egcs project and GCC 2.95, rolling out
36413 snapshots and releases, handling merges from GCC2, reviewing tons
36414 of patches that might have fallen through the cracks else, and
36415 random but extensive hacking.
36417 * Marc Lehmann for his direction via the steering committee and
36418 helping with analysis and improvements of x86 performance.
36420 * Victor Leikehman for work on GNU Fortran.
36422 * Ted Lemon wrote parts of the RTL reader and printer.
36424 * Kriang Lerdsuwanakij for C++ improvements including template as
36425 template parameter support, and many C++ fixes.
36427 * Warren Levy for tremendous work on libgcj (Java Runtime Library)
36428 and random work on the Java front end.
36430 * Alain Lichnewsky ported GCC to the MIPS CPU.
36432 * Oskar Liljeblad for hacking on AWT and his many Java bug reports
36435 * Robert Lipe for OpenServer support, new testsuites, testing, etc.
36437 * Chen Liqin for various S+core related fixes/improvement, and for
36438 maintaining the S+core port.
36440 * Weiwen Liu for testing and various bug fixes.
36442 * Manuel Lo'pez-Iba'n~ez for improving `-Wconversion' and many other
36443 diagnostics fixes and improvements.
36445 * Dave Love for his ongoing work with the Fortran front end and
36448 * Martin von Lo"wis for internal consistency checking infrastructure,
36449 various C++ improvements including namespace support, and tons of
36450 assistance with libstdc++/compiler merges.
36452 * H.J. Lu for his previous contributions to the steering committee,
36453 many x86 bug reports, prototype patches, and keeping the GNU/Linux
36456 * Greg McGary for random fixes and (someday) bounded pointers.
36458 * Andrew MacLeod for his ongoing work in building a real EH system,
36459 various code generation improvements, work on the global
36462 * Vladimir Makarov for hacking some ugly i960 problems, PowerPC
36463 hacking improvements to compile-time performance, overall
36464 knowledge and direction in the area of instruction scheduling, and
36465 design and implementation of the automaton based instruction
36468 * Bob Manson for his behind the scenes work on dejagnu.
36470 * Philip Martin for lots of libstdc++ string and vector iterator
36471 fixes and improvements, and string clean up and testsuites.
36473 * All of the Mauve project contributors, for Java test code.
36475 * Bryce McKinlay for numerous GCJ and libgcj fixes and improvements.
36477 * Adam Megacz for his work on the Microsoft Windows port of GCJ.
36479 * Michael Meissner for LRS framework, ia32, m32r, v850, m88k, MIPS,
36480 powerpc, haifa, ECOFF debug support, and other assorted hacking.
36482 * Jason Merrill for his direction via the steering committee and
36483 leading the G++ effort.
36485 * Martin Michlmayr for testing GCC on several architectures using the
36486 entire Debian archive.
36488 * David Miller for his direction via the steering committee, lots of
36489 SPARC work, improvements in jump.c and interfacing with the Linux
36492 * Gary Miller ported GCC to Charles River Data Systems machines.
36494 * Alfred Minarik for libstdc++ string and ios bug fixes, and turning
36495 the entire libstdc++ testsuite namespace-compatible.
36497 * Mark Mitchell for his direction via the steering committee,
36498 mountains of C++ work, load/store hoisting out of loops, alias
36499 analysis improvements, ISO C `restrict' support, and serving as
36500 release manager for GCC 3.x.
36502 * Alan Modra for various GNU/Linux bits and testing.
36504 * Toon Moene for his direction via the steering committee, Fortran
36505 maintenance, and his ongoing work to make us make Fortran run fast.
36507 * Jason Molenda for major help in the care and feeding of all the
36508 services on the gcc.gnu.org (formerly egcs.cygnus.com)
36509 machine--mail, web services, ftp services, etc etc. Doing all
36510 this work on scrap paper and the backs of envelopes would have
36513 * Catherine Moore for fixing various ugly problems we have sent her
36514 way, including the haifa bug which was killing the Alpha & PowerPC
36517 * Mike Moreton for his various Java patches.
36519 * David Mosberger-Tang for various Alpha improvements, and for the
36520 initial IA-64 port.
36522 * Stephen Moshier contributed the floating point emulator that
36523 assists in cross-compilation and permits support for floating
36524 point numbers wider than 64 bits and for ISO C99 support.
36526 * Bill Moyer for his behind the scenes work on various issues.
36528 * Philippe De Muyter for his work on the m68k port.
36530 * Joseph S. Myers for his work on the PDP-11 port, format checking
36531 and ISO C99 support, and continuous emphasis on (and contributions
36534 * Nathan Myers for his work on libstdc++-v3: architecture and
36535 authorship through the first three snapshots, including
36536 implementation of locale infrastructure, string, shadow C headers,
36537 and the initial project documentation (DESIGN, CHECKLIST, and so
36538 forth). Later, more work on MT-safe string and shadow headers.
36540 * Felix Natter for documentation on porting libstdc++.
36542 * Nathanael Nerode for cleaning up the configuration/build process.
36544 * NeXT, Inc. donated the front end that supports the Objective-C
36547 * Hans-Peter Nilsson for the CRIS and MMIX ports, improvements to
36548 the search engine setup, various documentation fixes and other
36551 * Geoff Noer for his work on getting cygwin native builds working.
36553 * Diego Novillo for his work on Tree SSA, OpenMP, SPEC performance
36554 tracking web pages, GIMPLE tuples, and assorted fixes.
36556 * David O'Brien for the FreeBSD/alpha, FreeBSD/AMD x86-64,
36557 FreeBSD/ARM, FreeBSD/PowerPC, and FreeBSD/SPARC64 ports and
36558 related infrastructure improvements.
36560 * Alexandre Oliva for various build infrastructure improvements,
36561 scripts and amazing testing work, including keeping libtool issues
36564 * Stefan Olsson for work on mt_alloc.
36566 * Melissa O'Neill for various NeXT fixes.
36568 * Rainer Orth for random MIPS work, including improvements to GCC's
36569 o32 ABI support, improvements to dejagnu's MIPS support, Java
36570 configuration clean-ups and porting work, etc.
36572 * Hartmut Penner for work on the s390 port.
36574 * Paul Petersen wrote the machine description for the Alliant FX/8.
36576 * Alexandre Petit-Bianco for implementing much of the Java compiler
36577 and continued Java maintainership.
36579 * Matthias Pfaller for major improvements to the NS32k port.
36581 * Gerald Pfeifer for his direction via the steering committee,
36582 pointing out lots of problems we need to solve, maintenance of the
36583 web pages, and taking care of documentation maintenance in general.
36585 * Andrew Pinski for processing bug reports by the dozen.
36587 * Ovidiu Predescu for his work on the Objective-C front end and
36590 * Jerry Quinn for major performance improvements in C++ formatted
36593 * Ken Raeburn for various improvements to checker, MIPS ports and
36594 various cleanups in the compiler.
36596 * Rolf W. Rasmussen for hacking on AWT.
36598 * David Reese of Sun Microsystems contributed to the Solaris on
36601 * Volker Reichelt for keeping up with the problem reports.
36603 * Joern Rennecke for maintaining the sh port, loop, regmove & reload
36606 * Loren J. Rittle for improvements to libstdc++-v3 including the
36607 FreeBSD port, threading fixes, thread-related configury changes,
36608 critical threading documentation, and solutions to really tricky
36609 I/O problems, as well as keeping GCC properly working on FreeBSD
36610 and continuous testing.
36612 * Craig Rodrigues for processing tons of bug reports.
36614 * Ola Ro"nnerup for work on mt_alloc.
36616 * Gavin Romig-Koch for lots of behind the scenes MIPS work.
36618 * David Ronis inspired and encouraged Craig to rewrite the G77
36619 documentation in texinfo format by contributing a first pass at a
36620 translation of the old `g77-0.5.16/f/DOC' file.
36622 * Ken Rose for fixes to GCC's delay slot filling code.
36624 * Paul Rubin wrote most of the preprocessor.
36626 * Pe'tur Runo'lfsson for major performance improvements in C++
36627 formatted I/O and large file support in C++ filebuf.
36629 * Chip Salzenberg for libstdc++ patches and improvements to locales,
36630 traits, Makefiles, libio, libtool hackery, and "long long" support.
36632 * Juha Sarlin for improvements to the H8 code generator.
36634 * Greg Satz assisted in making GCC work on HP-UX for the 9000 series
36637 * Roger Sayle for improvements to constant folding and GCC's RTL
36638 optimizers as well as for fixing numerous bugs.
36640 * Bradley Schatz for his work on the GCJ FAQ.
36642 * Peter Schauer wrote the code to allow debugging to work on the
36645 * William Schelter did most of the work on the Intel 80386 support.
36647 * Tobias Schlu"ter for work on GNU Fortran.
36649 * Bernd Schmidt for various code generation improvements and major
36650 work in the reload pass as well a serving as release manager for
36653 * Peter Schmid for constant testing of libstdc++--especially
36654 application testing, going above and beyond what was requested for
36655 the release criteria--and libstdc++ header file tweaks.
36657 * Jason Schroeder for jcf-dump patches.
36659 * Andreas Schwab for his work on the m68k port.
36661 * Lars Segerlund for work on GNU Fortran.
36663 * Joel Sherrill for his direction via the steering committee, RTEMS
36664 contributions and RTEMS testing.
36666 * Nathan Sidwell for many C++ fixes/improvements.
36668 * Jeffrey Siegal for helping RMS with the original design of GCC,
36669 some code which handles the parse tree and RTL data structures,
36670 constant folding and help with the original VAX & m68k ports.
36672 * Kenny Simpson for prompting libstdc++ fixes due to defect reports
36673 from the LWG (thereby keeping GCC in line with updates from the
36676 * Franz Sirl for his ongoing work with making the PPC port stable
36679 * Andrey Slepuhin for assorted AIX hacking.
36681 * Trevor Smigiel for contributing the SPU port.
36683 * Christopher Smith did the port for Convex machines.
36685 * Danny Smith for his major efforts on the Mingw (and Cygwin) ports.
36687 * Randy Smith finished the Sun FPA support.
36689 * Scott Snyder for queue, iterator, istream, and string fixes and
36690 libstdc++ testsuite entries. Also for providing the patch to G77
36691 to add rudimentary support for `INTEGER*1', `INTEGER*2', and
36694 * Brad Spencer for contributions to the GLIBCPP_FORCE_NEW technique.
36696 * Richard Stallman, for writing the original GCC and launching the
36699 * Jan Stein of the Chalmers Computer Society provided support for
36700 Genix, as well as part of the 32000 machine description.
36702 * Nigel Stephens for various mips16 related fixes/improvements.
36704 * Jonathan Stone wrote the machine description for the Pyramid
36707 * Graham Stott for various infrastructure improvements.
36709 * John Stracke for his Java HTTP protocol fixes.
36711 * Mike Stump for his Elxsi port, G++ contributions over the years
36712 and more recently his vxworks contributions
36714 * Jeff Sturm for Java porting help, bug fixes, and encouragement.
36716 * Shigeya Suzuki for this fixes for the bsdi platforms.
36718 * Ian Lance Taylor for his mips16 work, general configury hacking,
36721 * Holger Teutsch provided the support for the Clipper CPU.
36723 * Gary Thomas for his ongoing work to make the PPC work for
36726 * Philipp Thomas for random bug fixes throughout the compiler
36728 * Jason Thorpe for thread support in libstdc++ on NetBSD.
36730 * Kresten Krab Thorup wrote the run time support for the Objective-C
36731 language and the fantastic Java bytecode interpreter.
36733 * Michael Tiemann for random bug fixes, the first instruction
36734 scheduler, initial C++ support, function integration, NS32k, SPARC
36735 and M88k machine description work, delay slot scheduling.
36737 * Andreas Tobler for his work porting libgcj to Darwin.
36739 * Teemu Torma for thread safe exception handling support.
36741 * Leonard Tower wrote parts of the parser, RTL generator, and RTL
36742 definitions, and of the VAX machine description.
36744 * Daniel Towner and Hariharan Sandanagobalane contributed and
36745 maintain the picoChip port.
36747 * Tom Tromey for internationalization support and for his many Java
36748 contributions and libgcj maintainership.
36750 * Lassi Tuura for improvements to config.guess to determine HP
36753 * Petter Urkedal for libstdc++ CXXFLAGS, math, and algorithms fixes.
36755 * Andy Vaught for the design and initial implementation of the GNU
36758 * Brent Verner for work with the libstdc++ cshadow files and their
36759 associated configure steps.
36761 * Todd Vierling for contributions for NetBSD ports.
36763 * Jonathan Wakely for contributing libstdc++ Doxygen notes and XHTML
36766 * Dean Wakerley for converting the install documentation from HTML
36767 to texinfo in time for GCC 3.0.
36769 * Krister Walfridsson for random bug fixes.
36771 * Feng Wang for contributions to GNU Fortran.
36773 * Stephen M. Webb for time and effort on making libstdc++ shadow
36774 files work with the tricky Solaris 8+ headers, and for pushing the
36775 build-time header tree.
36777 * John Wehle for various improvements for the x86 code generator,
36778 related infrastructure improvements to help x86 code generation,
36779 value range propagation and other work, WE32k port.
36781 * Ulrich Weigand for work on the s390 port.
36783 * Zack Weinberg for major work on cpplib and various other bug fixes.
36785 * Matt Welsh for help with Linux Threads support in GCJ.
36787 * Urban Widmark for help fixing java.io.
36789 * Mark Wielaard for new Java library code and his work integrating
36792 * Dale Wiles helped port GCC to the Tahoe.
36794 * Bob Wilson from Tensilica, Inc. for the Xtensa port.
36796 * Jim Wilson for his direction via the steering committee, tackling
36797 hard problems in various places that nobody else wanted to work
36798 on, strength reduction and other loop optimizations.
36800 * Paul Woegerer and Tal Agmon for the CRX port.
36802 * Carlo Wood for various fixes.
36804 * Tom Wood for work on the m88k port.
36806 * Canqun Yang for work on GNU Fortran.
36808 * Masanobu Yuhara of Fujitsu Laboratories implemented the machine
36809 description for the Tron architecture (specifically, the Gmicro).
36811 * Kevin Zachmann helped port GCC to the Tahoe.
36813 * Ayal Zaks for Swing Modulo Scheduling (SMS).
36815 * Xiaoqiang Zhang for work on GNU Fortran.
36817 * Gilles Zunino for help porting Java to Irix.
36820 The following people are recognized for their contributions to GNAT,
36821 the Ada front end of GCC:
36824 * Romain Berrendonner
36874 * Hristian Kirtchev
36917 The following people are recognized for their contributions of new
36918 features, bug reports, testing and integration of classpath/libgcj for
36920 * Lillian Angel for `JTree' implementation and lots Free Swing
36921 additions and bug fixes.
36923 * Wolfgang Baer for `GapContent' bug fixes.
36925 * Anthony Balkissoon for `JList', Free Swing 1.5 updates and mouse
36926 event fixes, lots of Free Swing work including `JTable' editing.
36928 * Stuart Ballard for RMI constant fixes.
36930 * Goffredo Baroncelli for `HTTPURLConnection' fixes.
36932 * Gary Benson for `MessageFormat' fixes.
36934 * Daniel Bonniot for `Serialization' fixes.
36936 * Chris Burdess for lots of gnu.xml and http protocol fixes, `StAX'
36937 and `DOM xml:id' support.
36939 * Ka-Hing Cheung for `TreePath' and `TreeSelection' fixes.
36941 * Archie Cobbs for build fixes, VM interface updates,
36942 `URLClassLoader' updates.
36944 * Kelley Cook for build fixes.
36946 * Martin Cordova for Suggestions for better `SocketTimeoutException'.
36948 * David Daney for `BitSet' bug fixes, `HttpURLConnection' rewrite
36951 * Thomas Fitzsimmons for lots of upgrades to the gtk+ AWT and Cairo
36952 2D support. Lots of imageio framework additions, lots of AWT and
36953 Free Swing bug fixes.
36955 * Jeroen Frijters for `ClassLoader' and nio cleanups, serialization
36956 fixes, better `Proxy' support, bug fixes and IKVM integration.
36958 * Santiago Gala for `AccessControlContext' fixes.
36960 * Nicolas Geoffray for `VMClassLoader' and `AccessController'
36963 * David Gilbert for `basic' and `metal' icon and plaf support and
36964 lots of documenting, Lots of Free Swing and metal theme additions.
36965 `MetalIconFactory' implementation.
36967 * Anthony Green for `MIDI' framework, `ALSA' and `DSSI' providers.
36969 * Andrew Haley for `Serialization' and `URLClassLoader' fixes, gcj
36972 * Kim Ho for `JFileChooser' implementation.
36974 * Andrew John Hughes for `Locale' and net fixes, URI RFC2986
36975 updates, `Serialization' fixes, `Properties' XML support and
36976 generic branch work, VMIntegration guide update.
36978 * Bastiaan Huisman for `TimeZone' bug fixing.
36980 * Andreas Jaeger for mprec updates.
36982 * Paul Jenner for better `-Werror' support.
36984 * Ito Kazumitsu for `NetworkInterface' implementation and updates.
36986 * Roman Kennke for `BoxLayout', `GrayFilter' and `SplitPane', plus
36987 bug fixes all over. Lots of Free Swing work including styled text.
36989 * Simon Kitching for `String' cleanups and optimization suggestions.
36991 * Michael Koch for configuration fixes, `Locale' updates, bug and
36994 * Guilhem Lavaux for configuration, thread and channel fixes and
36995 Kaffe integration. JCL native `Pointer' updates. Logger bug fixes.
36997 * David Lichteblau for JCL support library global/local reference
37000 * Aaron Luchko for JDWP updates and documentation fixes.
37002 * Ziga Mahkovec for `Graphics2D' upgraded to Cairo 0.5 and new regex
37005 * Sven de Marothy for BMP imageio support, CSS and `TextLayout'
37006 fixes. `GtkImage' rewrite, 2D, awt, free swing and date/time fixes
37007 and implementing the Qt4 peers.
37009 * Casey Marshall for crypto algorithm fixes, `FileChannel' lock,
37010 `SystemLogger' and `FileHandler' rotate implementations, NIO
37011 `FileChannel.map' support, security and policy updates.
37013 * Bryce McKinlay for RMI work.
37015 * Audrius Meskauskas for lots of Free Corba, RMI and HTML work plus
37016 testing and documenting.
37018 * Kalle Olavi Niemitalo for build fixes.
37020 * Rainer Orth for build fixes.
37022 * Andrew Overholt for `File' locking fixes.
37024 * Ingo Proetel for `Image', `Logger' and `URLClassLoader' updates.
37026 * Olga Rodimina for `MenuSelectionManager' implementation.
37028 * Jan Roehrich for `BasicTreeUI' and `JTree' fixes.
37030 * Julian Scheid for documentation updates and gjdoc support.
37032 * Christian Schlichtherle for zip fixes and cleanups.
37034 * Robert Schuster for documentation updates and beans fixes,
37035 `TreeNode' enumerations and `ActionCommand' and various fixes, XML
37036 and URL, AWT and Free Swing bug fixes.
37038 * Keith Seitz for lots of JDWP work.
37040 * Christian Thalinger for 64-bit cleanups, Configuration and VM
37041 interface fixes and `CACAO' integration, `fdlibm' updates.
37043 * Gael Thomas for `VMClassLoader' boot packages support suggestions.
37045 * Andreas Tobler for Darwin and Solaris testing and fixing, `Qt4'
37046 support for Darwin/OS X, `Graphics2D' support, `gtk+' updates.
37048 * Dalibor Topic for better `DEBUG' support, build cleanups and Kaffe
37049 integration. `Qt4' build infrastructure, `SHA1PRNG' and
37050 `GdkPixbugDecoder' updates.
37052 * Tom Tromey for Eclipse integration, generics work, lots of bug
37053 fixes and gcj integration including coordinating The Big Merge.
37055 * Mark Wielaard for bug fixes, packaging and release management,
37056 `Clipboard' implementation, system call interrupts and network
37057 timeouts and `GdkPixpufDecoder' fixes.
37060 In addition to the above, all of which also contributed time and
37061 energy in testing GCC, we would like to thank the following for their
37062 contributions to testing:
37064 * Michael Abd-El-Malek
37074 * David Billinghurst
37078 * Stephane Bortzmeyer
37088 * Bradford Castalia
37110 * Charles-Antoine Gauthier
37132 * Kevin B. Hendricks
37136 * Christian Joensson
37144 * Anand Krishnaswamy
37146 * A. O. V. Le Blanc
37210 * Pedro A. M. Vazquez
37220 And finally we'd like to thank everyone who uses the compiler, provides
37221 feedback and generally reminds us why we're doing this work in the first
37225 File: gccint.info, Node: Option Index, Next: Concept Index, Prev: Contributors, Up: Top
37230 GCC's command line options are indexed here without any initial `-' or
37231 `--'. Where an option has both positive and negative forms (such as
37232 `-fOPTION' and `-fno-OPTION'), relevant entries in the manual are
37233 indexed under the most appropriate form; it may sometimes be useful to
37234 look up both forms.
37239 * msoft-float: Soft float library routines.
37243 File: gccint.info, Node: Concept Index, Prev: Option Index, Up: Top
37251 * ! in constraint: Multi-Alternative. (line 47)
37252 * # in constraint: Modifiers. (line 67)
37253 * # in template: Output Template. (line 66)
37254 * #pragma: Misc. (line 381)
37255 * % in constraint: Modifiers. (line 45)
37256 * % in GTY option: GTY Options. (line 18)
37257 * % in template: Output Template. (line 6)
37258 * & in constraint: Modifiers. (line 25)
37259 * ( <1>: Sections. (line 160)
37260 * ( <2>: GIMPLE_CALL. (line 63)
37261 * ( <3>: GIMPLE_ASM. (line 21)
37262 * (: Logical Operators. (line 107)
37263 * (nil): RTL Objects. (line 73)
37264 * * <1>: Host Common. (line 17)
37265 * *: Scheduling. (line 246)
37266 * * in constraint: Modifiers. (line 72)
37267 * * in template: Output Statement. (line 29)
37268 * *gimple_assign_lhs_ptr: GIMPLE_ASSIGN. (line 54)
37269 * *gimple_assign_rhs1_ptr: GIMPLE_ASSIGN. (line 60)
37270 * *gimple_assign_rhs2_ptr: GIMPLE_ASSIGN. (line 67)
37271 * *gimple_call_arg_ptr: GIMPLE_CALL. (line 71)
37272 * *gimple_call_lhs_ptr: GIMPLE_CALL. (line 32)
37273 * *gimple_catch_types_ptr: GIMPLE_CATCH. (line 16)
37274 * *gimple_cdt_location_ptr: GIMPLE_CHANGE_DYNAMIC_TYPE.
37276 * *gimple_cdt_new_type_ptr: GIMPLE_CHANGE_DYNAMIC_TYPE.
37278 * *gimple_eh_filter_types_ptr: GIMPLE_EH_FILTER. (line 15)
37279 * *gimple_omp_critical_name_ptr: GIMPLE_OMP_CRITICAL.
37281 * *gimple_omp_for_clauses_ptr: GIMPLE_OMP_FOR. (line 23)
37282 * *gimple_omp_for_final_ptr: GIMPLE_OMP_FOR. (line 54)
37283 * *gimple_omp_for_incr_ptr: GIMPLE_OMP_FOR. (line 64)
37284 * *gimple_omp_for_index_ptr: GIMPLE_OMP_FOR. (line 34)
37285 * *gimple_omp_for_initial_ptr: GIMPLE_OMP_FOR. (line 44)
37286 * *gimple_omp_parallel_child_fn_ptr: GIMPLE_OMP_PARALLEL.
37288 * *gimple_omp_parallel_clauses_ptr: GIMPLE_OMP_PARALLEL.
37290 * *gimple_omp_parallel_data_arg_ptr: GIMPLE_OMP_PARALLEL.
37292 * *gimple_omp_sections_clauses_ptr: GIMPLE_OMP_SECTIONS.
37294 * *gimple_omp_sections_control_ptr: GIMPLE_OMP_SECTIONS.
37296 * *gimple_omp_single_clauses_ptr: GIMPLE_OMP_SINGLE. (line 17)
37297 * *gimple_op_ptr: Manipulating GIMPLE statements.
37299 * *gimple_ops <1>: Manipulating GIMPLE statements.
37301 * *gimple_ops: Logical Operators. (line 82)
37302 * *gimple_phi_result_ptr: GIMPLE_PHI. (line 22)
37303 * *gsi_stmt_ptr: Sequence iterators. (line 80)
37304 * *TARGET_GET_PCH_VALIDITY: PCH Target. (line 7)
37305 * + in constraint: Modifiers. (line 12)
37306 * -fsection-anchors <1>: Anchored Addresses. (line 6)
37307 * -fsection-anchors: Special Accessors. (line 106)
37308 * /c in RTL dump: Flags. (line 234)
37309 * /f in RTL dump: Flags. (line 242)
37310 * /i in RTL dump: Flags. (line 294)
37311 * /j in RTL dump: Flags. (line 309)
37312 * /s in RTL dump: Flags. (line 258)
37313 * /u in RTL dump: Flags. (line 319)
37314 * /v in RTL dump: Flags. (line 351)
37315 * 0 in constraint: Simple Constraints. (line 120)
37316 * < in constraint: Simple Constraints. (line 48)
37317 * = in constraint: Modifiers. (line 8)
37318 * > in constraint: Simple Constraints. (line 52)
37319 * ? in constraint: Multi-Alternative. (line 41)
37320 * \: Output Template. (line 46)
37321 * __absvdi2: Integer library routines.
37323 * __absvsi2: Integer library routines.
37325 * __addda3: Fixed-point fractional library routines.
37327 * __adddf3: Soft float library routines.
37329 * __adddq3: Fixed-point fractional library routines.
37331 * __addha3: Fixed-point fractional library routines.
37333 * __addhq3: Fixed-point fractional library routines.
37335 * __addqq3: Fixed-point fractional library routines.
37337 * __addsa3: Fixed-point fractional library routines.
37339 * __addsf3: Soft float library routines.
37341 * __addsq3: Fixed-point fractional library routines.
37343 * __addta3: Fixed-point fractional library routines.
37345 * __addtf3: Soft float library routines.
37347 * __adduda3: Fixed-point fractional library routines.
37349 * __addudq3: Fixed-point fractional library routines.
37351 * __adduha3: Fixed-point fractional library routines.
37353 * __adduhq3: Fixed-point fractional library routines.
37355 * __adduqq3: Fixed-point fractional library routines.
37357 * __addusa3: Fixed-point fractional library routines.
37359 * __addusq3: Fixed-point fractional library routines.
37361 * __adduta3: Fixed-point fractional library routines.
37363 * __addvdi3: Integer library routines.
37365 * __addvsi3: Integer library routines.
37367 * __addxf3: Soft float library routines.
37369 * __ashlda3: Fixed-point fractional library routines.
37371 * __ashldi3: Integer library routines.
37373 * __ashldq3: Fixed-point fractional library routines.
37375 * __ashlha3: Fixed-point fractional library routines.
37377 * __ashlhq3: Fixed-point fractional library routines.
37379 * __ashlqq3: Fixed-point fractional library routines.
37381 * __ashlsa3: Fixed-point fractional library routines.
37383 * __ashlsi3: Integer library routines.
37385 * __ashlsq3: Fixed-point fractional library routines.
37387 * __ashlta3: Fixed-point fractional library routines.
37389 * __ashlti3: Integer library routines.
37391 * __ashluda3: Fixed-point fractional library routines.
37393 * __ashludq3: Fixed-point fractional library routines.
37395 * __ashluha3: Fixed-point fractional library routines.
37397 * __ashluhq3: Fixed-point fractional library routines.
37399 * __ashluqq3: Fixed-point fractional library routines.
37401 * __ashlusa3: Fixed-point fractional library routines.
37403 * __ashlusq3: Fixed-point fractional library routines.
37405 * __ashluta3: Fixed-point fractional library routines.
37407 * __ashrda3: Fixed-point fractional library routines.
37409 * __ashrdi3: Integer library routines.
37411 * __ashrdq3: Fixed-point fractional library routines.
37413 * __ashrha3: Fixed-point fractional library routines.
37415 * __ashrhq3: Fixed-point fractional library routines.
37417 * __ashrqq3: Fixed-point fractional library routines.
37419 * __ashrsa3: Fixed-point fractional library routines.
37421 * __ashrsi3: Integer library routines.
37423 * __ashrsq3: Fixed-point fractional library routines.
37425 * __ashrta3: Fixed-point fractional library routines.
37427 * __ashrti3: Integer library routines.
37429 * __bid_adddd3: Decimal float library routines.
37431 * __bid_addsd3: Decimal float library routines.
37433 * __bid_addtd3: Decimal float library routines.
37435 * __bid_divdd3: Decimal float library routines.
37437 * __bid_divsd3: Decimal float library routines.
37439 * __bid_divtd3: Decimal float library routines.
37441 * __bid_eqdd2: Decimal float library routines.
37443 * __bid_eqsd2: Decimal float library routines.
37445 * __bid_eqtd2: Decimal float library routines.
37447 * __bid_extendddtd2: Decimal float library routines.
37449 * __bid_extendddtf: Decimal float library routines.
37451 * __bid_extendddxf: Decimal float library routines.
37453 * __bid_extenddfdd: Decimal float library routines.
37455 * __bid_extenddftd: Decimal float library routines.
37457 * __bid_extendsddd2: Decimal float library routines.
37459 * __bid_extendsddf: Decimal float library routines.
37461 * __bid_extendsdtd2: Decimal float library routines.
37463 * __bid_extendsdtf: Decimal float library routines.
37465 * __bid_extendsdxf: Decimal float library routines.
37467 * __bid_extendsfdd: Decimal float library routines.
37469 * __bid_extendsfsd: Decimal float library routines.
37471 * __bid_extendsftd: Decimal float library routines.
37473 * __bid_extendtftd: Decimal float library routines.
37475 * __bid_extendxftd: Decimal float library routines.
37477 * __bid_fixdddi: Decimal float library routines.
37479 * __bid_fixddsi: Decimal float library routines.
37481 * __bid_fixsddi: Decimal float library routines.
37483 * __bid_fixsdsi: Decimal float library routines.
37485 * __bid_fixtddi: Decimal float library routines.
37487 * __bid_fixtdsi: Decimal float library routines.
37489 * __bid_fixunsdddi: Decimal float library routines.
37491 * __bid_fixunsddsi: Decimal float library routines.
37493 * __bid_fixunssddi: Decimal float library routines.
37495 * __bid_fixunssdsi: Decimal float library routines.
37497 * __bid_fixunstddi: Decimal float library routines.
37499 * __bid_fixunstdsi: Decimal float library routines.
37501 * __bid_floatdidd: Decimal float library routines.
37503 * __bid_floatdisd: Decimal float library routines.
37505 * __bid_floatditd: Decimal float library routines.
37507 * __bid_floatsidd: Decimal float library routines.
37509 * __bid_floatsisd: Decimal float library routines.
37511 * __bid_floatsitd: Decimal float library routines.
37513 * __bid_floatunsdidd: Decimal float library routines.
37515 * __bid_floatunsdisd: Decimal float library routines.
37517 * __bid_floatunsditd: Decimal float library routines.
37519 * __bid_floatunssidd: Decimal float library routines.
37521 * __bid_floatunssisd: Decimal float library routines.
37523 * __bid_floatunssitd: Decimal float library routines.
37525 * __bid_gedd2: Decimal float library routines.
37527 * __bid_gesd2: Decimal float library routines.
37529 * __bid_getd2: Decimal float library routines.
37531 * __bid_gtdd2: Decimal float library routines.
37533 * __bid_gtsd2: Decimal float library routines.
37535 * __bid_gttd2: Decimal float library routines.
37537 * __bid_ledd2: Decimal float library routines.
37539 * __bid_lesd2: Decimal float library routines.
37541 * __bid_letd2: Decimal float library routines.
37543 * __bid_ltdd2: Decimal float library routines.
37545 * __bid_ltsd2: Decimal float library routines.
37547 * __bid_lttd2: Decimal float library routines.
37549 * __bid_muldd3: Decimal float library routines.
37551 * __bid_mulsd3: Decimal float library routines.
37553 * __bid_multd3: Decimal float library routines.
37555 * __bid_nedd2: Decimal float library routines.
37557 * __bid_negdd2: Decimal float library routines.
37559 * __bid_negsd2: Decimal float library routines.
37561 * __bid_negtd2: Decimal float library routines.
37563 * __bid_nesd2: Decimal float library routines.
37565 * __bid_netd2: Decimal float library routines.
37567 * __bid_subdd3: Decimal float library routines.
37569 * __bid_subsd3: Decimal float library routines.
37571 * __bid_subtd3: Decimal float library routines.
37573 * __bid_truncdddf: Decimal float library routines.
37575 * __bid_truncddsd2: Decimal float library routines.
37577 * __bid_truncddsf: Decimal float library routines.
37579 * __bid_truncdfsd: Decimal float library routines.
37581 * __bid_truncsdsf: Decimal float library routines.
37583 * __bid_trunctddd2: Decimal float library routines.
37585 * __bid_trunctddf: Decimal float library routines.
37587 * __bid_trunctdsd2: Decimal float library routines.
37589 * __bid_trunctdsf: Decimal float library routines.
37591 * __bid_trunctdtf: Decimal float library routines.
37593 * __bid_trunctdxf: Decimal float library routines.
37595 * __bid_trunctfdd: Decimal float library routines.
37597 * __bid_trunctfsd: Decimal float library routines.
37599 * __bid_truncxfdd: Decimal float library routines.
37601 * __bid_truncxfsd: Decimal float library routines.
37603 * __bid_unorddd2: Decimal float library routines.
37605 * __bid_unordsd2: Decimal float library routines.
37607 * __bid_unordtd2: Decimal float library routines.
37609 * __bswapdi2: Integer library routines.
37611 * __bswapsi2: Integer library routines.
37613 * __builtin_args_info: Varargs. (line 42)
37614 * __builtin_classify_type: Varargs. (line 76)
37615 * __builtin_next_arg: Varargs. (line 66)
37616 * __builtin_saveregs: Varargs. (line 24)
37617 * __clear_cache: Miscellaneous routines.
37619 * __clzdi2: Integer library routines.
37621 * __clzsi2: Integer library routines.
37623 * __clzti2: Integer library routines.
37625 * __cmpda2: Fixed-point fractional library routines.
37627 * __cmpdf2: Soft float library routines.
37629 * __cmpdi2: Integer library routines.
37631 * __cmpdq2: Fixed-point fractional library routines.
37633 * __cmpha2: Fixed-point fractional library routines.
37635 * __cmphq2: Fixed-point fractional library routines.
37637 * __cmpqq2: Fixed-point fractional library routines.
37639 * __cmpsa2: Fixed-point fractional library routines.
37641 * __cmpsf2: Soft float library routines.
37643 * __cmpsq2: Fixed-point fractional library routines.
37645 * __cmpta2: Fixed-point fractional library routines.
37647 * __cmptf2: Soft float library routines.
37649 * __cmpti2: Integer library routines.
37651 * __cmpuda2: Fixed-point fractional library routines.
37653 * __cmpudq2: Fixed-point fractional library routines.
37655 * __cmpuha2: Fixed-point fractional library routines.
37657 * __cmpuhq2: Fixed-point fractional library routines.
37659 * __cmpuqq2: Fixed-point fractional library routines.
37661 * __cmpusa2: Fixed-point fractional library routines.
37663 * __cmpusq2: Fixed-point fractional library routines.
37665 * __cmputa2: Fixed-point fractional library routines.
37667 * __CTOR_LIST__: Initialization. (line 25)
37668 * __ctzdi2: Integer library routines.
37670 * __ctzsi2: Integer library routines.
37672 * __ctzti2: Integer library routines.
37674 * __divda3: Fixed-point fractional library routines.
37676 * __divdc3: Soft float library routines.
37678 * __divdf3: Soft float library routines.
37680 * __divdi3: Integer library routines.
37682 * __divdq3: Fixed-point fractional library routines.
37684 * __divha3: Fixed-point fractional library routines.
37686 * __divhq3: Fixed-point fractional library routines.
37688 * __divqq3: Fixed-point fractional library routines.
37690 * __divsa3: Fixed-point fractional library routines.
37692 * __divsc3: Soft float library routines.
37694 * __divsf3: Soft float library routines.
37696 * __divsi3: Integer library routines.
37698 * __divsq3: Fixed-point fractional library routines.
37700 * __divta3: Fixed-point fractional library routines.
37702 * __divtc3: Soft float library routines.
37704 * __divtf3: Soft float library routines.
37706 * __divti3: Integer library routines.
37708 * __divxc3: Soft float library routines.
37710 * __divxf3: Soft float library routines.
37712 * __dpd_adddd3: Decimal float library routines.
37714 * __dpd_addsd3: Decimal float library routines.
37716 * __dpd_addtd3: Decimal float library routines.
37718 * __dpd_divdd3: Decimal float library routines.
37720 * __dpd_divsd3: Decimal float library routines.
37722 * __dpd_divtd3: Decimal float library routines.
37724 * __dpd_eqdd2: Decimal float library routines.
37726 * __dpd_eqsd2: Decimal float library routines.
37728 * __dpd_eqtd2: Decimal float library routines.
37730 * __dpd_extendddtd2: Decimal float library routines.
37732 * __dpd_extendddtf: Decimal float library routines.
37734 * __dpd_extendddxf: Decimal float library routines.
37736 * __dpd_extenddfdd: Decimal float library routines.
37738 * __dpd_extenddftd: Decimal float library routines.
37740 * __dpd_extendsddd2: Decimal float library routines.
37742 * __dpd_extendsddf: Decimal float library routines.
37744 * __dpd_extendsdtd2: Decimal float library routines.
37746 * __dpd_extendsdtf: Decimal float library routines.
37748 * __dpd_extendsdxf: Decimal float library routines.
37750 * __dpd_extendsfdd: Decimal float library routines.
37752 * __dpd_extendsfsd: Decimal float library routines.
37754 * __dpd_extendsftd: Decimal float library routines.
37756 * __dpd_extendtftd: Decimal float library routines.
37758 * __dpd_extendxftd: Decimal float library routines.
37760 * __dpd_fixdddi: Decimal float library routines.
37762 * __dpd_fixddsi: Decimal float library routines.
37764 * __dpd_fixsddi: Decimal float library routines.
37766 * __dpd_fixsdsi: Decimal float library routines.
37768 * __dpd_fixtddi: Decimal float library routines.
37770 * __dpd_fixtdsi: Decimal float library routines.
37772 * __dpd_fixunsdddi: Decimal float library routines.
37774 * __dpd_fixunsddsi: Decimal float library routines.
37776 * __dpd_fixunssddi: Decimal float library routines.
37778 * __dpd_fixunssdsi: Decimal float library routines.
37780 * __dpd_fixunstddi: Decimal float library routines.
37782 * __dpd_fixunstdsi: Decimal float library routines.
37784 * __dpd_floatdidd: Decimal float library routines.
37786 * __dpd_floatdisd: Decimal float library routines.
37788 * __dpd_floatditd: Decimal float library routines.
37790 * __dpd_floatsidd: Decimal float library routines.
37792 * __dpd_floatsisd: Decimal float library routines.
37794 * __dpd_floatsitd: Decimal float library routines.
37796 * __dpd_floatunsdidd: Decimal float library routines.
37798 * __dpd_floatunsdisd: Decimal float library routines.
37800 * __dpd_floatunsditd: Decimal float library routines.
37802 * __dpd_floatunssidd: Decimal float library routines.
37804 * __dpd_floatunssisd: Decimal float library routines.
37806 * __dpd_floatunssitd: Decimal float library routines.
37808 * __dpd_gedd2: Decimal float library routines.
37810 * __dpd_gesd2: Decimal float library routines.
37812 * __dpd_getd2: Decimal float library routines.
37814 * __dpd_gtdd2: Decimal float library routines.
37816 * __dpd_gtsd2: Decimal float library routines.
37818 * __dpd_gttd2: Decimal float library routines.
37820 * __dpd_ledd2: Decimal float library routines.
37822 * __dpd_lesd2: Decimal float library routines.
37824 * __dpd_letd2: Decimal float library routines.
37826 * __dpd_ltdd2: Decimal float library routines.
37828 * __dpd_ltsd2: Decimal float library routines.
37830 * __dpd_lttd2: Decimal float library routines.
37832 * __dpd_muldd3: Decimal float library routines.
37834 * __dpd_mulsd3: Decimal float library routines.
37836 * __dpd_multd3: Decimal float library routines.
37838 * __dpd_nedd2: Decimal float library routines.
37840 * __dpd_negdd2: Decimal float library routines.
37842 * __dpd_negsd2: Decimal float library routines.
37844 * __dpd_negtd2: Decimal float library routines.
37846 * __dpd_nesd2: Decimal float library routines.
37848 * __dpd_netd2: Decimal float library routines.
37850 * __dpd_subdd3: Decimal float library routines.
37852 * __dpd_subsd3: Decimal float library routines.
37854 * __dpd_subtd3: Decimal float library routines.
37856 * __dpd_truncdddf: Decimal float library routines.
37858 * __dpd_truncddsd2: Decimal float library routines.
37860 * __dpd_truncddsf: Decimal float library routines.
37862 * __dpd_truncdfsd: Decimal float library routines.
37864 * __dpd_truncsdsf: Decimal float library routines.
37866 * __dpd_trunctddd2: Decimal float library routines.
37868 * __dpd_trunctddf: Decimal float library routines.
37870 * __dpd_trunctdsd2: Decimal float library routines.
37872 * __dpd_trunctdsf: Decimal float library routines.
37874 * __dpd_trunctdtf: Decimal float library routines.
37876 * __dpd_trunctdxf: Decimal float library routines.
37878 * __dpd_trunctfdd: Decimal float library routines.
37880 * __dpd_trunctfsd: Decimal float library routines.
37882 * __dpd_truncxfdd: Decimal float library routines.
37884 * __dpd_truncxfsd: Decimal float library routines.
37886 * __dpd_unorddd2: Decimal float library routines.
37888 * __dpd_unordsd2: Decimal float library routines.
37890 * __dpd_unordtd2: Decimal float library routines.
37892 * __DTOR_LIST__: Initialization. (line 25)
37893 * __eqdf2: Soft float library routines.
37895 * __eqsf2: Soft float library routines.
37897 * __eqtf2: Soft float library routines.
37899 * __extenddftf2: Soft float library routines.
37901 * __extenddfxf2: Soft float library routines.
37903 * __extendsfdf2: Soft float library routines.
37905 * __extendsftf2: Soft float library routines.
37907 * __extendsfxf2: Soft float library routines.
37909 * __ffsdi2: Integer library routines.
37911 * __ffsti2: Integer library routines.
37913 * __fixdfdi: Soft float library routines.
37915 * __fixdfsi: Soft float library routines.
37917 * __fixdfti: Soft float library routines.
37919 * __fixsfdi: Soft float library routines.
37921 * __fixsfsi: Soft float library routines.
37923 * __fixsfti: Soft float library routines.
37925 * __fixtfdi: Soft float library routines.
37927 * __fixtfsi: Soft float library routines.
37929 * __fixtfti: Soft float library routines.
37931 * __fixunsdfdi: Soft float library routines.
37933 * __fixunsdfsi: Soft float library routines.
37935 * __fixunsdfti: Soft float library routines.
37937 * __fixunssfdi: Soft float library routines.
37939 * __fixunssfsi: Soft float library routines.
37941 * __fixunssfti: Soft float library routines.
37943 * __fixunstfdi: Soft float library routines.
37945 * __fixunstfsi: Soft float library routines.
37947 * __fixunstfti: Soft float library routines.
37949 * __fixunsxfdi: Soft float library routines.
37951 * __fixunsxfsi: Soft float library routines.
37953 * __fixunsxfti: Soft float library routines.
37955 * __fixxfdi: Soft float library routines.
37957 * __fixxfsi: Soft float library routines.
37959 * __fixxfti: Soft float library routines.
37961 * __floatdidf: Soft float library routines.
37963 * __floatdisf: Soft float library routines.
37965 * __floatditf: Soft float library routines.
37967 * __floatdixf: Soft float library routines.
37969 * __floatsidf: Soft float library routines.
37971 * __floatsisf: Soft float library routines.
37973 * __floatsitf: Soft float library routines.
37975 * __floatsixf: Soft float library routines.
37977 * __floattidf: Soft float library routines.
37979 * __floattisf: Soft float library routines.
37981 * __floattitf: Soft float library routines.
37983 * __floattixf: Soft float library routines.
37985 * __floatundidf: Soft float library routines.
37987 * __floatundisf: Soft float library routines.
37989 * __floatunditf: Soft float library routines.
37991 * __floatundixf: Soft float library routines.
37993 * __floatunsidf: Soft float library routines.
37995 * __floatunsisf: Soft float library routines.
37997 * __floatunsitf: Soft float library routines.
37999 * __floatunsixf: Soft float library routines.
38001 * __floatuntidf: Soft float library routines.
38003 * __floatuntisf: Soft float library routines.
38005 * __floatuntitf: Soft float library routines.
38007 * __floatuntixf: Soft float library routines.
38009 * __fractdadf: Fixed-point fractional library routines.
38011 * __fractdadi: Fixed-point fractional library routines.
38013 * __fractdadq: Fixed-point fractional library routines.
38015 * __fractdaha2: Fixed-point fractional library routines.
38017 * __fractdahi: Fixed-point fractional library routines.
38019 * __fractdahq: Fixed-point fractional library routines.
38021 * __fractdaqi: Fixed-point fractional library routines.
38023 * __fractdaqq: Fixed-point fractional library routines.
38025 * __fractdasa2: Fixed-point fractional library routines.
38027 * __fractdasf: Fixed-point fractional library routines.
38029 * __fractdasi: Fixed-point fractional library routines.
38031 * __fractdasq: Fixed-point fractional library routines.
38033 * __fractdata2: Fixed-point fractional library routines.
38035 * __fractdati: Fixed-point fractional library routines.
38037 * __fractdauda: Fixed-point fractional library routines.
38039 * __fractdaudq: Fixed-point fractional library routines.
38041 * __fractdauha: Fixed-point fractional library routines.
38043 * __fractdauhq: Fixed-point fractional library routines.
38045 * __fractdauqq: Fixed-point fractional library routines.
38047 * __fractdausa: Fixed-point fractional library routines.
38049 * __fractdausq: Fixed-point fractional library routines.
38051 * __fractdauta: Fixed-point fractional library routines.
38053 * __fractdfda: Fixed-point fractional library routines.
38055 * __fractdfdq: Fixed-point fractional library routines.
38057 * __fractdfha: Fixed-point fractional library routines.
38059 * __fractdfhq: Fixed-point fractional library routines.
38061 * __fractdfqq: Fixed-point fractional library routines.
38063 * __fractdfsa: Fixed-point fractional library routines.
38065 * __fractdfsq: Fixed-point fractional library routines.
38067 * __fractdfta: Fixed-point fractional library routines.
38069 * __fractdfuda: Fixed-point fractional library routines.
38071 * __fractdfudq: Fixed-point fractional library routines.
38073 * __fractdfuha: Fixed-point fractional library routines.
38075 * __fractdfuhq: Fixed-point fractional library routines.
38077 * __fractdfuqq: Fixed-point fractional library routines.
38079 * __fractdfusa: Fixed-point fractional library routines.
38081 * __fractdfusq: Fixed-point fractional library routines.
38083 * __fractdfuta: Fixed-point fractional library routines.
38085 * __fractdida: Fixed-point fractional library routines.
38087 * __fractdidq: Fixed-point fractional library routines.
38089 * __fractdiha: Fixed-point fractional library routines.
38091 * __fractdihq: Fixed-point fractional library routines.
38093 * __fractdiqq: Fixed-point fractional library routines.
38095 * __fractdisa: Fixed-point fractional library routines.
38097 * __fractdisq: Fixed-point fractional library routines.
38099 * __fractdita: Fixed-point fractional library routines.
38101 * __fractdiuda: Fixed-point fractional library routines.
38103 * __fractdiudq: Fixed-point fractional library routines.
38105 * __fractdiuha: Fixed-point fractional library routines.
38107 * __fractdiuhq: Fixed-point fractional library routines.
38109 * __fractdiuqq: Fixed-point fractional library routines.
38111 * __fractdiusa: Fixed-point fractional library routines.
38113 * __fractdiusq: Fixed-point fractional library routines.
38115 * __fractdiuta: Fixed-point fractional library routines.
38117 * __fractdqda: Fixed-point fractional library routines.
38119 * __fractdqdf: Fixed-point fractional library routines.
38121 * __fractdqdi: Fixed-point fractional library routines.
38123 * __fractdqha: Fixed-point fractional library routines.
38125 * __fractdqhi: Fixed-point fractional library routines.
38127 * __fractdqhq2: Fixed-point fractional library routines.
38129 * __fractdqqi: Fixed-point fractional library routines.
38131 * __fractdqqq2: Fixed-point fractional library routines.
38133 * __fractdqsa: Fixed-point fractional library routines.
38135 * __fractdqsf: Fixed-point fractional library routines.
38137 * __fractdqsi: Fixed-point fractional library routines.
38139 * __fractdqsq2: Fixed-point fractional library routines.
38141 * __fractdqta: Fixed-point fractional library routines.
38143 * __fractdqti: Fixed-point fractional library routines.
38145 * __fractdquda: Fixed-point fractional library routines.
38147 * __fractdqudq: Fixed-point fractional library routines.
38149 * __fractdquha: Fixed-point fractional library routines.
38151 * __fractdquhq: Fixed-point fractional library routines.
38153 * __fractdquqq: Fixed-point fractional library routines.
38155 * __fractdqusa: Fixed-point fractional library routines.
38157 * __fractdqusq: Fixed-point fractional library routines.
38159 * __fractdquta: Fixed-point fractional library routines.
38161 * __fracthada2: Fixed-point fractional library routines.
38163 * __fracthadf: Fixed-point fractional library routines.
38165 * __fracthadi: Fixed-point fractional library routines.
38167 * __fracthadq: Fixed-point fractional library routines.
38169 * __fracthahi: Fixed-point fractional library routines.
38171 * __fracthahq: Fixed-point fractional library routines.
38173 * __fracthaqi: Fixed-point fractional library routines.
38175 * __fracthaqq: Fixed-point fractional library routines.
38177 * __fracthasa2: Fixed-point fractional library routines.
38179 * __fracthasf: Fixed-point fractional library routines.
38181 * __fracthasi: Fixed-point fractional library routines.
38183 * __fracthasq: Fixed-point fractional library routines.
38185 * __fracthata2: Fixed-point fractional library routines.
38187 * __fracthati: Fixed-point fractional library routines.
38189 * __fracthauda: Fixed-point fractional library routines.
38191 * __fracthaudq: Fixed-point fractional library routines.
38193 * __fracthauha: Fixed-point fractional library routines.
38195 * __fracthauhq: Fixed-point fractional library routines.
38197 * __fracthauqq: Fixed-point fractional library routines.
38199 * __fracthausa: Fixed-point fractional library routines.
38201 * __fracthausq: Fixed-point fractional library routines.
38203 * __fracthauta: Fixed-point fractional library routines.
38205 * __fracthida: Fixed-point fractional library routines.
38207 * __fracthidq: Fixed-point fractional library routines.
38209 * __fracthiha: Fixed-point fractional library routines.
38211 * __fracthihq: Fixed-point fractional library routines.
38213 * __fracthiqq: Fixed-point fractional library routines.
38215 * __fracthisa: Fixed-point fractional library routines.
38217 * __fracthisq: Fixed-point fractional library routines.
38219 * __fracthita: Fixed-point fractional library routines.
38221 * __fracthiuda: Fixed-point fractional library routines.
38223 * __fracthiudq: Fixed-point fractional library routines.
38225 * __fracthiuha: Fixed-point fractional library routines.
38227 * __fracthiuhq: Fixed-point fractional library routines.
38229 * __fracthiuqq: Fixed-point fractional library routines.
38231 * __fracthiusa: Fixed-point fractional library routines.
38233 * __fracthiusq: Fixed-point fractional library routines.
38235 * __fracthiuta: Fixed-point fractional library routines.
38237 * __fracthqda: Fixed-point fractional library routines.
38239 * __fracthqdf: Fixed-point fractional library routines.
38241 * __fracthqdi: Fixed-point fractional library routines.
38243 * __fracthqdq2: Fixed-point fractional library routines.
38245 * __fracthqha: Fixed-point fractional library routines.
38247 * __fracthqhi: Fixed-point fractional library routines.
38249 * __fracthqqi: Fixed-point fractional library routines.
38251 * __fracthqqq2: Fixed-point fractional library routines.
38253 * __fracthqsa: Fixed-point fractional library routines.
38255 * __fracthqsf: Fixed-point fractional library routines.
38257 * __fracthqsi: Fixed-point fractional library routines.
38259 * __fracthqsq2: Fixed-point fractional library routines.
38261 * __fracthqta: Fixed-point fractional library routines.
38263 * __fracthqti: Fixed-point fractional library routines.
38265 * __fracthquda: Fixed-point fractional library routines.
38267 * __fracthqudq: Fixed-point fractional library routines.
38269 * __fracthquha: Fixed-point fractional library routines.
38271 * __fracthquhq: Fixed-point fractional library routines.
38273 * __fracthquqq: Fixed-point fractional library routines.
38275 * __fracthqusa: Fixed-point fractional library routines.
38277 * __fracthqusq: Fixed-point fractional library routines.
38279 * __fracthquta: Fixed-point fractional library routines.
38281 * __fractqida: Fixed-point fractional library routines.
38283 * __fractqidq: Fixed-point fractional library routines.
38285 * __fractqiha: Fixed-point fractional library routines.
38287 * __fractqihq: Fixed-point fractional library routines.
38289 * __fractqiqq: Fixed-point fractional library routines.
38291 * __fractqisa: Fixed-point fractional library routines.
38293 * __fractqisq: Fixed-point fractional library routines.
38295 * __fractqita: Fixed-point fractional library routines.
38297 * __fractqiuda: Fixed-point fractional library routines.
38299 * __fractqiudq: Fixed-point fractional library routines.
38301 * __fractqiuha: Fixed-point fractional library routines.
38303 * __fractqiuhq: Fixed-point fractional library routines.
38305 * __fractqiuqq: Fixed-point fractional library routines.
38307 * __fractqiusa: Fixed-point fractional library routines.
38309 * __fractqiusq: Fixed-point fractional library routines.
38311 * __fractqiuta: Fixed-point fractional library routines.
38313 * __fractqqda: Fixed-point fractional library routines.
38315 * __fractqqdf: Fixed-point fractional library routines.
38317 * __fractqqdi: Fixed-point fractional library routines.
38319 * __fractqqdq2: Fixed-point fractional library routines.
38321 * __fractqqha: Fixed-point fractional library routines.
38323 * __fractqqhi: Fixed-point fractional library routines.
38325 * __fractqqhq2: Fixed-point fractional library routines.
38327 * __fractqqqi: Fixed-point fractional library routines.
38329 * __fractqqsa: Fixed-point fractional library routines.
38331 * __fractqqsf: Fixed-point fractional library routines.
38333 * __fractqqsi: Fixed-point fractional library routines.
38335 * __fractqqsq2: Fixed-point fractional library routines.
38337 * __fractqqta: Fixed-point fractional library routines.
38339 * __fractqqti: Fixed-point fractional library routines.
38341 * __fractqquda: Fixed-point fractional library routines.
38343 * __fractqqudq: Fixed-point fractional library routines.
38345 * __fractqquha: Fixed-point fractional library routines.
38347 * __fractqquhq: Fixed-point fractional library routines.
38349 * __fractqquqq: Fixed-point fractional library routines.
38351 * __fractqqusa: Fixed-point fractional library routines.
38353 * __fractqqusq: Fixed-point fractional library routines.
38355 * __fractqquta: Fixed-point fractional library routines.
38357 * __fractsada2: Fixed-point fractional library routines.
38359 * __fractsadf: Fixed-point fractional library routines.
38361 * __fractsadi: Fixed-point fractional library routines.
38363 * __fractsadq: Fixed-point fractional library routines.
38365 * __fractsaha2: Fixed-point fractional library routines.
38367 * __fractsahi: Fixed-point fractional library routines.
38369 * __fractsahq: Fixed-point fractional library routines.
38371 * __fractsaqi: Fixed-point fractional library routines.
38373 * __fractsaqq: Fixed-point fractional library routines.
38375 * __fractsasf: Fixed-point fractional library routines.
38377 * __fractsasi: Fixed-point fractional library routines.
38379 * __fractsasq: Fixed-point fractional library routines.
38381 * __fractsata2: Fixed-point fractional library routines.
38383 * __fractsati: Fixed-point fractional library routines.
38385 * __fractsauda: Fixed-point fractional library routines.
38387 * __fractsaudq: Fixed-point fractional library routines.
38389 * __fractsauha: Fixed-point fractional library routines.
38391 * __fractsauhq: Fixed-point fractional library routines.
38393 * __fractsauqq: Fixed-point fractional library routines.
38395 * __fractsausa: Fixed-point fractional library routines.
38397 * __fractsausq: Fixed-point fractional library routines.
38399 * __fractsauta: Fixed-point fractional library routines.
38401 * __fractsfda: Fixed-point fractional library routines.
38403 * __fractsfdq: Fixed-point fractional library routines.
38405 * __fractsfha: Fixed-point fractional library routines.
38407 * __fractsfhq: Fixed-point fractional library routines.
38409 * __fractsfqq: Fixed-point fractional library routines.
38411 * __fractsfsa: Fixed-point fractional library routines.
38413 * __fractsfsq: Fixed-point fractional library routines.
38415 * __fractsfta: Fixed-point fractional library routines.
38417 * __fractsfuda: Fixed-point fractional library routines.
38419 * __fractsfudq: Fixed-point fractional library routines.
38421 * __fractsfuha: Fixed-point fractional library routines.
38423 * __fractsfuhq: Fixed-point fractional library routines.
38425 * __fractsfuqq: Fixed-point fractional library routines.
38427 * __fractsfusa: Fixed-point fractional library routines.
38429 * __fractsfusq: Fixed-point fractional library routines.
38431 * __fractsfuta: Fixed-point fractional library routines.
38433 * __fractsida: Fixed-point fractional library routines.
38435 * __fractsidq: Fixed-point fractional library routines.
38437 * __fractsiha: Fixed-point fractional library routines.
38439 * __fractsihq: Fixed-point fractional library routines.
38441 * __fractsiqq: Fixed-point fractional library routines.
38443 * __fractsisa: Fixed-point fractional library routines.
38445 * __fractsisq: Fixed-point fractional library routines.
38447 * __fractsita: Fixed-point fractional library routines.
38449 * __fractsiuda: Fixed-point fractional library routines.
38451 * __fractsiudq: Fixed-point fractional library routines.
38453 * __fractsiuha: Fixed-point fractional library routines.
38455 * __fractsiuhq: Fixed-point fractional library routines.
38457 * __fractsiuqq: Fixed-point fractional library routines.
38459 * __fractsiusa: Fixed-point fractional library routines.
38461 * __fractsiusq: Fixed-point fractional library routines.
38463 * __fractsiuta: Fixed-point fractional library routines.
38465 * __fractsqda: Fixed-point fractional library routines.
38467 * __fractsqdf: Fixed-point fractional library routines.
38469 * __fractsqdi: Fixed-point fractional library routines.
38471 * __fractsqdq2: Fixed-point fractional library routines.
38473 * __fractsqha: Fixed-point fractional library routines.
38475 * __fractsqhi: Fixed-point fractional library routines.
38477 * __fractsqhq2: Fixed-point fractional library routines.
38479 * __fractsqqi: Fixed-point fractional library routines.
38481 * __fractsqqq2: Fixed-point fractional library routines.
38483 * __fractsqsa: Fixed-point fractional library routines.
38485 * __fractsqsf: Fixed-point fractional library routines.
38487 * __fractsqsi: Fixed-point fractional library routines.
38489 * __fractsqta: Fixed-point fractional library routines.
38491 * __fractsqti: Fixed-point fractional library routines.
38493 * __fractsquda: Fixed-point fractional library routines.
38495 * __fractsqudq: Fixed-point fractional library routines.
38497 * __fractsquha: Fixed-point fractional library routines.
38499 * __fractsquhq: Fixed-point fractional library routines.
38501 * __fractsquqq: Fixed-point fractional library routines.
38503 * __fractsqusa: Fixed-point fractional library routines.
38505 * __fractsqusq: Fixed-point fractional library routines.
38507 * __fractsquta: Fixed-point fractional library routines.
38509 * __fracttada2: Fixed-point fractional library routines.
38511 * __fracttadf: Fixed-point fractional library routines.
38513 * __fracttadi: Fixed-point fractional library routines.
38515 * __fracttadq: Fixed-point fractional library routines.
38517 * __fracttaha2: Fixed-point fractional library routines.
38519 * __fracttahi: Fixed-point fractional library routines.
38521 * __fracttahq: Fixed-point fractional library routines.
38523 * __fracttaqi: Fixed-point fractional library routines.
38525 * __fracttaqq: Fixed-point fractional library routines.
38527 * __fracttasa2: Fixed-point fractional library routines.
38529 * __fracttasf: Fixed-point fractional library routines.
38531 * __fracttasi: Fixed-point fractional library routines.
38533 * __fracttasq: Fixed-point fractional library routines.
38535 * __fracttati: Fixed-point fractional library routines.
38537 * __fracttauda: Fixed-point fractional library routines.
38539 * __fracttaudq: Fixed-point fractional library routines.
38541 * __fracttauha: Fixed-point fractional library routines.
38543 * __fracttauhq: Fixed-point fractional library routines.
38545 * __fracttauqq: Fixed-point fractional library routines.
38547 * __fracttausa: Fixed-point fractional library routines.
38549 * __fracttausq: Fixed-point fractional library routines.
38551 * __fracttauta: Fixed-point fractional library routines.
38553 * __fracttida: Fixed-point fractional library routines.
38555 * __fracttidq: Fixed-point fractional library routines.
38557 * __fracttiha: Fixed-point fractional library routines.
38559 * __fracttihq: Fixed-point fractional library routines.
38561 * __fracttiqq: Fixed-point fractional library routines.
38563 * __fracttisa: Fixed-point fractional library routines.
38565 * __fracttisq: Fixed-point fractional library routines.
38567 * __fracttita: Fixed-point fractional library routines.
38569 * __fracttiuda: Fixed-point fractional library routines.
38571 * __fracttiudq: Fixed-point fractional library routines.
38573 * __fracttiuha: Fixed-point fractional library routines.
38575 * __fracttiuhq: Fixed-point fractional library routines.
38577 * __fracttiuqq: Fixed-point fractional library routines.
38579 * __fracttiusa: Fixed-point fractional library routines.
38581 * __fracttiusq: Fixed-point fractional library routines.
38583 * __fracttiuta: Fixed-point fractional library routines.
38585 * __fractudada: Fixed-point fractional library routines.
38587 * __fractudadf: Fixed-point fractional library routines.
38589 * __fractudadi: Fixed-point fractional library routines.
38591 * __fractudadq: Fixed-point fractional library routines.
38593 * __fractudaha: Fixed-point fractional library routines.
38595 * __fractudahi: Fixed-point fractional library routines.
38597 * __fractudahq: Fixed-point fractional library routines.
38599 * __fractudaqi: Fixed-point fractional library routines.
38601 * __fractudaqq: Fixed-point fractional library routines.
38603 * __fractudasa: Fixed-point fractional library routines.
38605 * __fractudasf: Fixed-point fractional library routines.
38607 * __fractudasi: Fixed-point fractional library routines.
38609 * __fractudasq: Fixed-point fractional library routines.
38611 * __fractudata: Fixed-point fractional library routines.
38613 * __fractudati: Fixed-point fractional library routines.
38615 * __fractudaudq: Fixed-point fractional library routines.
38617 * __fractudauha2: Fixed-point fractional library routines.
38619 * __fractudauhq: Fixed-point fractional library routines.
38621 * __fractudauqq: Fixed-point fractional library routines.
38623 * __fractudausa2: Fixed-point fractional library routines.
38625 * __fractudausq: Fixed-point fractional library routines.
38627 * __fractudauta2: Fixed-point fractional library routines.
38629 * __fractudqda: Fixed-point fractional library routines.
38631 * __fractudqdf: Fixed-point fractional library routines.
38633 * __fractudqdi: Fixed-point fractional library routines.
38635 * __fractudqdq: Fixed-point fractional library routines.
38637 * __fractudqha: Fixed-point fractional library routines.
38639 * __fractudqhi: Fixed-point fractional library routines.
38641 * __fractudqhq: Fixed-point fractional library routines.
38643 * __fractudqqi: Fixed-point fractional library routines.
38645 * __fractudqqq: Fixed-point fractional library routines.
38647 * __fractudqsa: Fixed-point fractional library routines.
38649 * __fractudqsf: Fixed-point fractional library routines.
38651 * __fractudqsi: Fixed-point fractional library routines.
38653 * __fractudqsq: Fixed-point fractional library routines.
38655 * __fractudqta: Fixed-point fractional library routines.
38657 * __fractudqti: Fixed-point fractional library routines.
38659 * __fractudquda: Fixed-point fractional library routines.
38661 * __fractudquha: Fixed-point fractional library routines.
38663 * __fractudquhq2: Fixed-point fractional library routines.
38665 * __fractudquqq2: Fixed-point fractional library routines.
38667 * __fractudqusa: Fixed-point fractional library routines.
38669 * __fractudqusq2: Fixed-point fractional library routines.
38671 * __fractudquta: Fixed-point fractional library routines.
38673 * __fractuhada: Fixed-point fractional library routines.
38675 * __fractuhadf: Fixed-point fractional library routines.
38677 * __fractuhadi: Fixed-point fractional library routines.
38679 * __fractuhadq: Fixed-point fractional library routines.
38681 * __fractuhaha: Fixed-point fractional library routines.
38683 * __fractuhahi: Fixed-point fractional library routines.
38685 * __fractuhahq: Fixed-point fractional library routines.
38687 * __fractuhaqi: Fixed-point fractional library routines.
38689 * __fractuhaqq: Fixed-point fractional library routines.
38691 * __fractuhasa: Fixed-point fractional library routines.
38693 * __fractuhasf: Fixed-point fractional library routines.
38695 * __fractuhasi: Fixed-point fractional library routines.
38697 * __fractuhasq: Fixed-point fractional library routines.
38699 * __fractuhata: Fixed-point fractional library routines.
38701 * __fractuhati: Fixed-point fractional library routines.
38703 * __fractuhauda2: Fixed-point fractional library routines.
38705 * __fractuhaudq: Fixed-point fractional library routines.
38707 * __fractuhauhq: Fixed-point fractional library routines.
38709 * __fractuhauqq: Fixed-point fractional library routines.
38711 * __fractuhausa2: Fixed-point fractional library routines.
38713 * __fractuhausq: Fixed-point fractional library routines.
38715 * __fractuhauta2: Fixed-point fractional library routines.
38717 * __fractuhqda: Fixed-point fractional library routines.
38719 * __fractuhqdf: Fixed-point fractional library routines.
38721 * __fractuhqdi: Fixed-point fractional library routines.
38723 * __fractuhqdq: Fixed-point fractional library routines.
38725 * __fractuhqha: Fixed-point fractional library routines.
38727 * __fractuhqhi: Fixed-point fractional library routines.
38729 * __fractuhqhq: Fixed-point fractional library routines.
38731 * __fractuhqqi: Fixed-point fractional library routines.
38733 * __fractuhqqq: Fixed-point fractional library routines.
38735 * __fractuhqsa: Fixed-point fractional library routines.
38737 * __fractuhqsf: Fixed-point fractional library routines.
38739 * __fractuhqsi: Fixed-point fractional library routines.
38741 * __fractuhqsq: Fixed-point fractional library routines.
38743 * __fractuhqta: Fixed-point fractional library routines.
38745 * __fractuhqti: Fixed-point fractional library routines.
38747 * __fractuhquda: Fixed-point fractional library routines.
38749 * __fractuhqudq2: Fixed-point fractional library routines.
38751 * __fractuhquha: Fixed-point fractional library routines.
38753 * __fractuhquqq2: Fixed-point fractional library routines.
38755 * __fractuhqusa: Fixed-point fractional library routines.
38757 * __fractuhqusq2: Fixed-point fractional library routines.
38759 * __fractuhquta: Fixed-point fractional library routines.
38761 * __fractunsdadi: Fixed-point fractional library routines.
38763 * __fractunsdahi: Fixed-point fractional library routines.
38765 * __fractunsdaqi: Fixed-point fractional library routines.
38767 * __fractunsdasi: Fixed-point fractional library routines.
38769 * __fractunsdati: Fixed-point fractional library routines.
38771 * __fractunsdida: Fixed-point fractional library routines.
38773 * __fractunsdidq: Fixed-point fractional library routines.
38775 * __fractunsdiha: Fixed-point fractional library routines.
38777 * __fractunsdihq: Fixed-point fractional library routines.
38779 * __fractunsdiqq: Fixed-point fractional library routines.
38781 * __fractunsdisa: Fixed-point fractional library routines.
38783 * __fractunsdisq: Fixed-point fractional library routines.
38785 * __fractunsdita: Fixed-point fractional library routines.
38787 * __fractunsdiuda: Fixed-point fractional library routines.
38789 * __fractunsdiudq: Fixed-point fractional library routines.
38791 * __fractunsdiuha: Fixed-point fractional library routines.
38793 * __fractunsdiuhq: Fixed-point fractional library routines.
38795 * __fractunsdiuqq: Fixed-point fractional library routines.
38797 * __fractunsdiusa: Fixed-point fractional library routines.
38799 * __fractunsdiusq: Fixed-point fractional library routines.
38801 * __fractunsdiuta: Fixed-point fractional library routines.
38803 * __fractunsdqdi: Fixed-point fractional library routines.
38805 * __fractunsdqhi: Fixed-point fractional library routines.
38807 * __fractunsdqqi: Fixed-point fractional library routines.
38809 * __fractunsdqsi: Fixed-point fractional library routines.
38811 * __fractunsdqti: Fixed-point fractional library routines.
38813 * __fractunshadi: Fixed-point fractional library routines.
38815 * __fractunshahi: Fixed-point fractional library routines.
38817 * __fractunshaqi: Fixed-point fractional library routines.
38819 * __fractunshasi: Fixed-point fractional library routines.
38821 * __fractunshati: Fixed-point fractional library routines.
38823 * __fractunshida: Fixed-point fractional library routines.
38825 * __fractunshidq: Fixed-point fractional library routines.
38827 * __fractunshiha: Fixed-point fractional library routines.
38829 * __fractunshihq: Fixed-point fractional library routines.
38831 * __fractunshiqq: Fixed-point fractional library routines.
38833 * __fractunshisa: Fixed-point fractional library routines.
38835 * __fractunshisq: Fixed-point fractional library routines.
38837 * __fractunshita: Fixed-point fractional library routines.
38839 * __fractunshiuda: Fixed-point fractional library routines.
38841 * __fractunshiudq: Fixed-point fractional library routines.
38843 * __fractunshiuha: Fixed-point fractional library routines.
38845 * __fractunshiuhq: Fixed-point fractional library routines.
38847 * __fractunshiuqq: Fixed-point fractional library routines.
38849 * __fractunshiusa: Fixed-point fractional library routines.
38851 * __fractunshiusq: Fixed-point fractional library routines.
38853 * __fractunshiuta: Fixed-point fractional library routines.
38855 * __fractunshqdi: Fixed-point fractional library routines.
38857 * __fractunshqhi: Fixed-point fractional library routines.
38859 * __fractunshqqi: Fixed-point fractional library routines.
38861 * __fractunshqsi: Fixed-point fractional library routines.
38863 * __fractunshqti: Fixed-point fractional library routines.
38865 * __fractunsqida: Fixed-point fractional library routines.
38867 * __fractunsqidq: Fixed-point fractional library routines.
38869 * __fractunsqiha: Fixed-point fractional library routines.
38871 * __fractunsqihq: Fixed-point fractional library routines.
38873 * __fractunsqiqq: Fixed-point fractional library routines.
38875 * __fractunsqisa: Fixed-point fractional library routines.
38877 * __fractunsqisq: Fixed-point fractional library routines.
38879 * __fractunsqita: Fixed-point fractional library routines.
38881 * __fractunsqiuda: Fixed-point fractional library routines.
38883 * __fractunsqiudq: Fixed-point fractional library routines.
38885 * __fractunsqiuha: Fixed-point fractional library routines.
38887 * __fractunsqiuhq: Fixed-point fractional library routines.
38889 * __fractunsqiuqq: Fixed-point fractional library routines.
38891 * __fractunsqiusa: Fixed-point fractional library routines.
38893 * __fractunsqiusq: Fixed-point fractional library routines.
38895 * __fractunsqiuta: Fixed-point fractional library routines.
38897 * __fractunsqqdi: Fixed-point fractional library routines.
38899 * __fractunsqqhi: Fixed-point fractional library routines.
38901 * __fractunsqqqi: Fixed-point fractional library routines.
38903 * __fractunsqqsi: Fixed-point fractional library routines.
38905 * __fractunsqqti: Fixed-point fractional library routines.
38907 * __fractunssadi: Fixed-point fractional library routines.
38909 * __fractunssahi: Fixed-point fractional library routines.
38911 * __fractunssaqi: Fixed-point fractional library routines.
38913 * __fractunssasi: Fixed-point fractional library routines.
38915 * __fractunssati: Fixed-point fractional library routines.
38917 * __fractunssida: Fixed-point fractional library routines.
38919 * __fractunssidq: Fixed-point fractional library routines.
38921 * __fractunssiha: Fixed-point fractional library routines.
38923 * __fractunssihq: Fixed-point fractional library routines.
38925 * __fractunssiqq: Fixed-point fractional library routines.
38927 * __fractunssisa: Fixed-point fractional library routines.
38929 * __fractunssisq: Fixed-point fractional library routines.
38931 * __fractunssita: Fixed-point fractional library routines.
38933 * __fractunssiuda: Fixed-point fractional library routines.
38935 * __fractunssiudq: Fixed-point fractional library routines.
38937 * __fractunssiuha: Fixed-point fractional library routines.
38939 * __fractunssiuhq: Fixed-point fractional library routines.
38941 * __fractunssiuqq: Fixed-point fractional library routines.
38943 * __fractunssiusa: Fixed-point fractional library routines.
38945 * __fractunssiusq: Fixed-point fractional library routines.
38947 * __fractunssiuta: Fixed-point fractional library routines.
38949 * __fractunssqdi: Fixed-point fractional library routines.
38951 * __fractunssqhi: Fixed-point fractional library routines.
38953 * __fractunssqqi: Fixed-point fractional library routines.
38955 * __fractunssqsi: Fixed-point fractional library routines.
38957 * __fractunssqti: Fixed-point fractional library routines.
38959 * __fractunstadi: Fixed-point fractional library routines.
38961 * __fractunstahi: Fixed-point fractional library routines.
38963 * __fractunstaqi: Fixed-point fractional library routines.
38965 * __fractunstasi: Fixed-point fractional library routines.
38967 * __fractunstati: Fixed-point fractional library routines.
38969 * __fractunstida: Fixed-point fractional library routines.
38971 * __fractunstidq: Fixed-point fractional library routines.
38973 * __fractunstiha: Fixed-point fractional library routines.
38975 * __fractunstihq: Fixed-point fractional library routines.
38977 * __fractunstiqq: Fixed-point fractional library routines.
38979 * __fractunstisa: Fixed-point fractional library routines.
38981 * __fractunstisq: Fixed-point fractional library routines.
38983 * __fractunstita: Fixed-point fractional library routines.
38985 * __fractunstiuda: Fixed-point fractional library routines.
38987 * __fractunstiudq: Fixed-point fractional library routines.
38989 * __fractunstiuha: Fixed-point fractional library routines.
38991 * __fractunstiuhq: Fixed-point fractional library routines.
38993 * __fractunstiuqq: Fixed-point fractional library routines.
38995 * __fractunstiusa: Fixed-point fractional library routines.
38997 * __fractunstiusq: Fixed-point fractional library routines.
38999 * __fractunstiuta: Fixed-point fractional library routines.
39001 * __fractunsudadi: Fixed-point fractional library routines.
39003 * __fractunsudahi: Fixed-point fractional library routines.
39005 * __fractunsudaqi: Fixed-point fractional library routines.
39007 * __fractunsudasi: Fixed-point fractional library routines.
39009 * __fractunsudati: Fixed-point fractional library routines.
39011 * __fractunsudqdi: Fixed-point fractional library routines.
39013 * __fractunsudqhi: Fixed-point fractional library routines.
39015 * __fractunsudqqi: Fixed-point fractional library routines.
39017 * __fractunsudqsi: Fixed-point fractional library routines.
39019 * __fractunsudqti: Fixed-point fractional library routines.
39021 * __fractunsuhadi: Fixed-point fractional library routines.
39023 * __fractunsuhahi: Fixed-point fractional library routines.
39025 * __fractunsuhaqi: Fixed-point fractional library routines.
39027 * __fractunsuhasi: Fixed-point fractional library routines.
39029 * __fractunsuhati: Fixed-point fractional library routines.
39031 * __fractunsuhqdi: Fixed-point fractional library routines.
39033 * __fractunsuhqhi: Fixed-point fractional library routines.
39035 * __fractunsuhqqi: Fixed-point fractional library routines.
39037 * __fractunsuhqsi: Fixed-point fractional library routines.
39039 * __fractunsuhqti: Fixed-point fractional library routines.
39041 * __fractunsuqqdi: Fixed-point fractional library routines.
39043 * __fractunsuqqhi: Fixed-point fractional library routines.
39045 * __fractunsuqqqi: Fixed-point fractional library routines.
39047 * __fractunsuqqsi: Fixed-point fractional library routines.
39049 * __fractunsuqqti: Fixed-point fractional library routines.
39051 * __fractunsusadi: Fixed-point fractional library routines.
39053 * __fractunsusahi: Fixed-point fractional library routines.
39055 * __fractunsusaqi: Fixed-point fractional library routines.
39057 * __fractunsusasi: Fixed-point fractional library routines.
39059 * __fractunsusati: Fixed-point fractional library routines.
39061 * __fractunsusqdi: Fixed-point fractional library routines.
39063 * __fractunsusqhi: Fixed-point fractional library routines.
39065 * __fractunsusqqi: Fixed-point fractional library routines.
39067 * __fractunsusqsi: Fixed-point fractional library routines.
39069 * __fractunsusqti: Fixed-point fractional library routines.
39071 * __fractunsutadi: Fixed-point fractional library routines.
39073 * __fractunsutahi: Fixed-point fractional library routines.
39075 * __fractunsutaqi: Fixed-point fractional library routines.
39077 * __fractunsutasi: Fixed-point fractional library routines.
39079 * __fractunsutati: Fixed-point fractional library routines.
39081 * __fractuqqda: Fixed-point fractional library routines.
39083 * __fractuqqdf: Fixed-point fractional library routines.
39085 * __fractuqqdi: Fixed-point fractional library routines.
39087 * __fractuqqdq: Fixed-point fractional library routines.
39089 * __fractuqqha: Fixed-point fractional library routines.
39091 * __fractuqqhi: Fixed-point fractional library routines.
39093 * __fractuqqhq: Fixed-point fractional library routines.
39095 * __fractuqqqi: Fixed-point fractional library routines.
39097 * __fractuqqqq: Fixed-point fractional library routines.
39099 * __fractuqqsa: Fixed-point fractional library routines.
39101 * __fractuqqsf: Fixed-point fractional library routines.
39103 * __fractuqqsi: Fixed-point fractional library routines.
39105 * __fractuqqsq: Fixed-point fractional library routines.
39107 * __fractuqqta: Fixed-point fractional library routines.
39109 * __fractuqqti: Fixed-point fractional library routines.
39111 * __fractuqquda: Fixed-point fractional library routines.
39113 * __fractuqqudq2: Fixed-point fractional library routines.
39115 * __fractuqquha: Fixed-point fractional library routines.
39117 * __fractuqquhq2: Fixed-point fractional library routines.
39119 * __fractuqqusa: Fixed-point fractional library routines.
39121 * __fractuqqusq2: Fixed-point fractional library routines.
39123 * __fractuqquta: Fixed-point fractional library routines.
39125 * __fractusada: Fixed-point fractional library routines.
39127 * __fractusadf: Fixed-point fractional library routines.
39129 * __fractusadi: Fixed-point fractional library routines.
39131 * __fractusadq: Fixed-point fractional library routines.
39133 * __fractusaha: Fixed-point fractional library routines.
39135 * __fractusahi: Fixed-point fractional library routines.
39137 * __fractusahq: Fixed-point fractional library routines.
39139 * __fractusaqi: Fixed-point fractional library routines.
39141 * __fractusaqq: Fixed-point fractional library routines.
39143 * __fractusasa: Fixed-point fractional library routines.
39145 * __fractusasf: Fixed-point fractional library routines.
39147 * __fractusasi: Fixed-point fractional library routines.
39149 * __fractusasq: Fixed-point fractional library routines.
39151 * __fractusata: Fixed-point fractional library routines.
39153 * __fractusati: Fixed-point fractional library routines.
39155 * __fractusauda2: Fixed-point fractional library routines.
39157 * __fractusaudq: Fixed-point fractional library routines.
39159 * __fractusauha2: Fixed-point fractional library routines.
39161 * __fractusauhq: Fixed-point fractional library routines.
39163 * __fractusauqq: Fixed-point fractional library routines.
39165 * __fractusausq: Fixed-point fractional library routines.
39167 * __fractusauta2: Fixed-point fractional library routines.
39169 * __fractusqda: Fixed-point fractional library routines.
39171 * __fractusqdf: Fixed-point fractional library routines.
39173 * __fractusqdi: Fixed-point fractional library routines.
39175 * __fractusqdq: Fixed-point fractional library routines.
39177 * __fractusqha: Fixed-point fractional library routines.
39179 * __fractusqhi: Fixed-point fractional library routines.
39181 * __fractusqhq: Fixed-point fractional library routines.
39183 * __fractusqqi: Fixed-point fractional library routines.
39185 * __fractusqqq: Fixed-point fractional library routines.
39187 * __fractusqsa: Fixed-point fractional library routines.
39189 * __fractusqsf: Fixed-point fractional library routines.
39191 * __fractusqsi: Fixed-point fractional library routines.
39193 * __fractusqsq: Fixed-point fractional library routines.
39195 * __fractusqta: Fixed-point fractional library routines.
39197 * __fractusqti: Fixed-point fractional library routines.
39199 * __fractusquda: Fixed-point fractional library routines.
39201 * __fractusqudq2: Fixed-point fractional library routines.
39203 * __fractusquha: Fixed-point fractional library routines.
39205 * __fractusquhq2: Fixed-point fractional library routines.
39207 * __fractusquqq2: Fixed-point fractional library routines.
39209 * __fractusqusa: Fixed-point fractional library routines.
39211 * __fractusquta: Fixed-point fractional library routines.
39213 * __fractutada: Fixed-point fractional library routines.
39215 * __fractutadf: Fixed-point fractional library routines.
39217 * __fractutadi: Fixed-point fractional library routines.
39219 * __fractutadq: Fixed-point fractional library routines.
39221 * __fractutaha: Fixed-point fractional library routines.
39223 * __fractutahi: Fixed-point fractional library routines.
39225 * __fractutahq: Fixed-point fractional library routines.
39227 * __fractutaqi: Fixed-point fractional library routines.
39229 * __fractutaqq: Fixed-point fractional library routines.
39231 * __fractutasa: Fixed-point fractional library routines.
39233 * __fractutasf: Fixed-point fractional library routines.
39235 * __fractutasi: Fixed-point fractional library routines.
39237 * __fractutasq: Fixed-point fractional library routines.
39239 * __fractutata: Fixed-point fractional library routines.
39241 * __fractutati: Fixed-point fractional library routines.
39243 * __fractutauda2: Fixed-point fractional library routines.
39245 * __fractutaudq: Fixed-point fractional library routines.
39247 * __fractutauha2: Fixed-point fractional library routines.
39249 * __fractutauhq: Fixed-point fractional library routines.
39251 * __fractutauqq: Fixed-point fractional library routines.
39253 * __fractutausa2: Fixed-point fractional library routines.
39255 * __fractutausq: Fixed-point fractional library routines.
39257 * __gedf2: Soft float library routines.
39259 * __gesf2: Soft float library routines.
39261 * __getf2: Soft float library routines.
39263 * __gtdf2: Soft float library routines.
39265 * __gtsf2: Soft float library routines.
39267 * __gttf2: Soft float library routines.
39269 * __ledf2: Soft float library routines.
39271 * __lesf2: Soft float library routines.
39273 * __letf2: Soft float library routines.
39275 * __lshrdi3: Integer library routines.
39277 * __lshrsi3: Integer library routines.
39279 * __lshrti3: Integer library routines.
39281 * __lshruda3: Fixed-point fractional library routines.
39283 * __lshrudq3: Fixed-point fractional library routines.
39285 * __lshruha3: Fixed-point fractional library routines.
39287 * __lshruhq3: Fixed-point fractional library routines.
39289 * __lshruqq3: Fixed-point fractional library routines.
39291 * __lshrusa3: Fixed-point fractional library routines.
39293 * __lshrusq3: Fixed-point fractional library routines.
39295 * __lshruta3: Fixed-point fractional library routines.
39297 * __ltdf2: Soft float library routines.
39299 * __ltsf2: Soft float library routines.
39301 * __lttf2: Soft float library routines.
39303 * __main: Collect2. (line 15)
39304 * __moddi3: Integer library routines.
39306 * __modsi3: Integer library routines.
39308 * __modti3: Integer library routines.
39310 * __mulda3: Fixed-point fractional library routines.
39312 * __muldc3: Soft float library routines.
39314 * __muldf3: Soft float library routines.
39316 * __muldi3: Integer library routines.
39318 * __muldq3: Fixed-point fractional library routines.
39320 * __mulha3: Fixed-point fractional library routines.
39322 * __mulhq3: Fixed-point fractional library routines.
39324 * __mulqq3: Fixed-point fractional library routines.
39326 * __mulsa3: Fixed-point fractional library routines.
39328 * __mulsc3: Soft float library routines.
39330 * __mulsf3: Soft float library routines.
39332 * __mulsi3: Integer library routines.
39334 * __mulsq3: Fixed-point fractional library routines.
39336 * __multa3: Fixed-point fractional library routines.
39338 * __multc3: Soft float library routines.
39340 * __multf3: Soft float library routines.
39342 * __multi3: Integer library routines.
39344 * __muluda3: Fixed-point fractional library routines.
39346 * __muludq3: Fixed-point fractional library routines.
39348 * __muluha3: Fixed-point fractional library routines.
39350 * __muluhq3: Fixed-point fractional library routines.
39352 * __muluqq3: Fixed-point fractional library routines.
39354 * __mulusa3: Fixed-point fractional library routines.
39356 * __mulusq3: Fixed-point fractional library routines.
39358 * __muluta3: Fixed-point fractional library routines.
39360 * __mulvdi3: Integer library routines.
39362 * __mulvsi3: Integer library routines.
39364 * __mulxc3: Soft float library routines.
39366 * __mulxf3: Soft float library routines.
39368 * __nedf2: Soft float library routines.
39370 * __negda2: Fixed-point fractional library routines.
39372 * __negdf2: Soft float library routines.
39374 * __negdi2: Integer library routines.
39376 * __negdq2: Fixed-point fractional library routines.
39378 * __negha2: Fixed-point fractional library routines.
39380 * __neghq2: Fixed-point fractional library routines.
39382 * __negqq2: Fixed-point fractional library routines.
39384 * __negsa2: Fixed-point fractional library routines.
39386 * __negsf2: Soft float library routines.
39388 * __negsq2: Fixed-point fractional library routines.
39390 * __negta2: Fixed-point fractional library routines.
39392 * __negtf2: Soft float library routines.
39394 * __negti2: Integer library routines.
39396 * __neguda2: Fixed-point fractional library routines.
39398 * __negudq2: Fixed-point fractional library routines.
39400 * __neguha2: Fixed-point fractional library routines.
39402 * __neguhq2: Fixed-point fractional library routines.
39404 * __neguqq2: Fixed-point fractional library routines.
39406 * __negusa2: Fixed-point fractional library routines.
39408 * __negusq2: Fixed-point fractional library routines.
39410 * __neguta2: Fixed-point fractional library routines.
39412 * __negvdi2: Integer library routines.
39414 * __negvsi2: Integer library routines.
39416 * __negxf2: Soft float library routines.
39418 * __nesf2: Soft float library routines.
39420 * __netf2: Soft float library routines.
39422 * __paritydi2: Integer library routines.
39424 * __paritysi2: Integer library routines.
39426 * __parityti2: Integer library routines.
39428 * __popcountdi2: Integer library routines.
39430 * __popcountsi2: Integer library routines.
39432 * __popcountti2: Integer library routines.
39434 * __powidf2: Soft float library routines.
39436 * __powisf2: Soft float library routines.
39438 * __powitf2: Soft float library routines.
39440 * __powixf2: Soft float library routines.
39442 * __satfractdadq: Fixed-point fractional library routines.
39444 * __satfractdaha2: Fixed-point fractional library routines.
39446 * __satfractdahq: Fixed-point fractional library routines.
39448 * __satfractdaqq: Fixed-point fractional library routines.
39450 * __satfractdasa2: Fixed-point fractional library routines.
39452 * __satfractdasq: Fixed-point fractional library routines.
39454 * __satfractdata2: Fixed-point fractional library routines.
39456 * __satfractdauda: Fixed-point fractional library routines.
39458 * __satfractdaudq: Fixed-point fractional library routines.
39460 * __satfractdauha: Fixed-point fractional library routines.
39462 * __satfractdauhq: Fixed-point fractional library routines.
39464 * __satfractdauqq: Fixed-point fractional library routines.
39466 * __satfractdausa: Fixed-point fractional library routines.
39468 * __satfractdausq: Fixed-point fractional library routines.
39470 * __satfractdauta: Fixed-point fractional library routines.
39472 * __satfractdfda: Fixed-point fractional library routines.
39474 * __satfractdfdq: Fixed-point fractional library routines.
39476 * __satfractdfha: Fixed-point fractional library routines.
39478 * __satfractdfhq: Fixed-point fractional library routines.
39480 * __satfractdfqq: Fixed-point fractional library routines.
39482 * __satfractdfsa: Fixed-point fractional library routines.
39484 * __satfractdfsq: Fixed-point fractional library routines.
39486 * __satfractdfta: Fixed-point fractional library routines.
39488 * __satfractdfuda: Fixed-point fractional library routines.
39490 * __satfractdfudq: Fixed-point fractional library routines.
39492 * __satfractdfuha: Fixed-point fractional library routines.
39494 * __satfractdfuhq: Fixed-point fractional library routines.
39496 * __satfractdfuqq: Fixed-point fractional library routines.
39498 * __satfractdfusa: Fixed-point fractional library routines.
39500 * __satfractdfusq: Fixed-point fractional library routines.
39502 * __satfractdfuta: Fixed-point fractional library routines.
39504 * __satfractdida: Fixed-point fractional library routines.
39506 * __satfractdidq: Fixed-point fractional library routines.
39508 * __satfractdiha: Fixed-point fractional library routines.
39510 * __satfractdihq: Fixed-point fractional library routines.
39512 * __satfractdiqq: Fixed-point fractional library routines.
39514 * __satfractdisa: Fixed-point fractional library routines.
39516 * __satfractdisq: Fixed-point fractional library routines.
39518 * __satfractdita: Fixed-point fractional library routines.
39520 * __satfractdiuda: Fixed-point fractional library routines.
39522 * __satfractdiudq: Fixed-point fractional library routines.
39524 * __satfractdiuha: Fixed-point fractional library routines.
39526 * __satfractdiuhq: Fixed-point fractional library routines.
39528 * __satfractdiuqq: Fixed-point fractional library routines.
39530 * __satfractdiusa: Fixed-point fractional library routines.
39532 * __satfractdiusq: Fixed-point fractional library routines.
39534 * __satfractdiuta: Fixed-point fractional library routines.
39536 * __satfractdqda: Fixed-point fractional library routines.
39538 * __satfractdqha: Fixed-point fractional library routines.
39540 * __satfractdqhq2: Fixed-point fractional library routines.
39542 * __satfractdqqq2: Fixed-point fractional library routines.
39544 * __satfractdqsa: Fixed-point fractional library routines.
39546 * __satfractdqsq2: Fixed-point fractional library routines.
39548 * __satfractdqta: Fixed-point fractional library routines.
39550 * __satfractdquda: Fixed-point fractional library routines.
39552 * __satfractdqudq: Fixed-point fractional library routines.
39554 * __satfractdquha: Fixed-point fractional library routines.
39556 * __satfractdquhq: Fixed-point fractional library routines.
39558 * __satfractdquqq: Fixed-point fractional library routines.
39560 * __satfractdqusa: Fixed-point fractional library routines.
39562 * __satfractdqusq: Fixed-point fractional library routines.
39564 * __satfractdquta: Fixed-point fractional library routines.
39566 * __satfracthada2: Fixed-point fractional library routines.
39568 * __satfracthadq: Fixed-point fractional library routines.
39570 * __satfracthahq: Fixed-point fractional library routines.
39572 * __satfracthaqq: Fixed-point fractional library routines.
39574 * __satfracthasa2: Fixed-point fractional library routines.
39576 * __satfracthasq: Fixed-point fractional library routines.
39578 * __satfracthata2: Fixed-point fractional library routines.
39580 * __satfracthauda: Fixed-point fractional library routines.
39582 * __satfracthaudq: Fixed-point fractional library routines.
39584 * __satfracthauha: Fixed-point fractional library routines.
39586 * __satfracthauhq: Fixed-point fractional library routines.
39588 * __satfracthauqq: Fixed-point fractional library routines.
39590 * __satfracthausa: Fixed-point fractional library routines.
39592 * __satfracthausq: Fixed-point fractional library routines.
39594 * __satfracthauta: Fixed-point fractional library routines.
39596 * __satfracthida: Fixed-point fractional library routines.
39598 * __satfracthidq: Fixed-point fractional library routines.
39600 * __satfracthiha: Fixed-point fractional library routines.
39602 * __satfracthihq: Fixed-point fractional library routines.
39604 * __satfracthiqq: Fixed-point fractional library routines.
39606 * __satfracthisa: Fixed-point fractional library routines.
39608 * __satfracthisq: Fixed-point fractional library routines.
39610 * __satfracthita: Fixed-point fractional library routines.
39612 * __satfracthiuda: Fixed-point fractional library routines.
39614 * __satfracthiudq: Fixed-point fractional library routines.
39616 * __satfracthiuha: Fixed-point fractional library routines.
39618 * __satfracthiuhq: Fixed-point fractional library routines.
39620 * __satfracthiuqq: Fixed-point fractional library routines.
39622 * __satfracthiusa: Fixed-point fractional library routines.
39624 * __satfracthiusq: Fixed-point fractional library routines.
39626 * __satfracthiuta: Fixed-point fractional library routines.
39628 * __satfracthqda: Fixed-point fractional library routines.
39630 * __satfracthqdq2: Fixed-point fractional library routines.
39632 * __satfracthqha: Fixed-point fractional library routines.
39634 * __satfracthqqq2: Fixed-point fractional library routines.
39636 * __satfracthqsa: Fixed-point fractional library routines.
39638 * __satfracthqsq2: Fixed-point fractional library routines.
39640 * __satfracthqta: Fixed-point fractional library routines.
39642 * __satfracthquda: Fixed-point fractional library routines.
39644 * __satfracthqudq: Fixed-point fractional library routines.
39646 * __satfracthquha: Fixed-point fractional library routines.
39648 * __satfracthquhq: Fixed-point fractional library routines.
39650 * __satfracthquqq: Fixed-point fractional library routines.
39652 * __satfracthqusa: Fixed-point fractional library routines.
39654 * __satfracthqusq: Fixed-point fractional library routines.
39656 * __satfracthquta: Fixed-point fractional library routines.
39658 * __satfractqida: Fixed-point fractional library routines.
39660 * __satfractqidq: Fixed-point fractional library routines.
39662 * __satfractqiha: Fixed-point fractional library routines.
39664 * __satfractqihq: Fixed-point fractional library routines.
39666 * __satfractqiqq: Fixed-point fractional library routines.
39668 * __satfractqisa: Fixed-point fractional library routines.
39670 * __satfractqisq: Fixed-point fractional library routines.
39672 * __satfractqita: Fixed-point fractional library routines.
39674 * __satfractqiuda: Fixed-point fractional library routines.
39676 * __satfractqiudq: Fixed-point fractional library routines.
39678 * __satfractqiuha: Fixed-point fractional library routines.
39680 * __satfractqiuhq: Fixed-point fractional library routines.
39682 * __satfractqiuqq: Fixed-point fractional library routines.
39684 * __satfractqiusa: Fixed-point fractional library routines.
39686 * __satfractqiusq: Fixed-point fractional library routines.
39688 * __satfractqiuta: Fixed-point fractional library routines.
39690 * __satfractqqda: Fixed-point fractional library routines.
39692 * __satfractqqdq2: Fixed-point fractional library routines.
39694 * __satfractqqha: Fixed-point fractional library routines.
39696 * __satfractqqhq2: Fixed-point fractional library routines.
39698 * __satfractqqsa: Fixed-point fractional library routines.
39700 * __satfractqqsq2: Fixed-point fractional library routines.
39702 * __satfractqqta: Fixed-point fractional library routines.
39704 * __satfractqquda: Fixed-point fractional library routines.
39706 * __satfractqqudq: Fixed-point fractional library routines.
39708 * __satfractqquha: Fixed-point fractional library routines.
39710 * __satfractqquhq: Fixed-point fractional library routines.
39712 * __satfractqquqq: Fixed-point fractional library routines.
39714 * __satfractqqusa: Fixed-point fractional library routines.
39716 * __satfractqqusq: Fixed-point fractional library routines.
39718 * __satfractqquta: Fixed-point fractional library routines.
39720 * __satfractsada2: Fixed-point fractional library routines.
39722 * __satfractsadq: Fixed-point fractional library routines.
39724 * __satfractsaha2: Fixed-point fractional library routines.
39726 * __satfractsahq: Fixed-point fractional library routines.
39728 * __satfractsaqq: Fixed-point fractional library routines.
39730 * __satfractsasq: Fixed-point fractional library routines.
39732 * __satfractsata2: Fixed-point fractional library routines.
39734 * __satfractsauda: Fixed-point fractional library routines.
39736 * __satfractsaudq: Fixed-point fractional library routines.
39738 * __satfractsauha: Fixed-point fractional library routines.
39740 * __satfractsauhq: Fixed-point fractional library routines.
39742 * __satfractsauqq: Fixed-point fractional library routines.
39744 * __satfractsausa: Fixed-point fractional library routines.
39746 * __satfractsausq: Fixed-point fractional library routines.
39748 * __satfractsauta: Fixed-point fractional library routines.
39750 * __satfractsfda: Fixed-point fractional library routines.
39752 * __satfractsfdq: Fixed-point fractional library routines.
39754 * __satfractsfha: Fixed-point fractional library routines.
39756 * __satfractsfhq: Fixed-point fractional library routines.
39758 * __satfractsfqq: Fixed-point fractional library routines.
39760 * __satfractsfsa: Fixed-point fractional library routines.
39762 * __satfractsfsq: Fixed-point fractional library routines.
39764 * __satfractsfta: Fixed-point fractional library routines.
39766 * __satfractsfuda: Fixed-point fractional library routines.
39768 * __satfractsfudq: Fixed-point fractional library routines.
39770 * __satfractsfuha: Fixed-point fractional library routines.
39772 * __satfractsfuhq: Fixed-point fractional library routines.
39774 * __satfractsfuqq: Fixed-point fractional library routines.
39776 * __satfractsfusa: Fixed-point fractional library routines.
39778 * __satfractsfusq: Fixed-point fractional library routines.
39780 * __satfractsfuta: Fixed-point fractional library routines.
39782 * __satfractsida: Fixed-point fractional library routines.
39784 * __satfractsidq: Fixed-point fractional library routines.
39786 * __satfractsiha: Fixed-point fractional library routines.
39788 * __satfractsihq: Fixed-point fractional library routines.
39790 * __satfractsiqq: Fixed-point fractional library routines.
39792 * __satfractsisa: Fixed-point fractional library routines.
39794 * __satfractsisq: Fixed-point fractional library routines.
39796 * __satfractsita: Fixed-point fractional library routines.
39798 * __satfractsiuda: Fixed-point fractional library routines.
39800 * __satfractsiudq: Fixed-point fractional library routines.
39802 * __satfractsiuha: Fixed-point fractional library routines.
39804 * __satfractsiuhq: Fixed-point fractional library routines.
39806 * __satfractsiuqq: Fixed-point fractional library routines.
39808 * __satfractsiusa: Fixed-point fractional library routines.
39810 * __satfractsiusq: Fixed-point fractional library routines.
39812 * __satfractsiuta: Fixed-point fractional library routines.
39814 * __satfractsqda: Fixed-point fractional library routines.
39816 * __satfractsqdq2: Fixed-point fractional library routines.
39818 * __satfractsqha: Fixed-point fractional library routines.
39820 * __satfractsqhq2: Fixed-point fractional library routines.
39822 * __satfractsqqq2: Fixed-point fractional library routines.
39824 * __satfractsqsa: Fixed-point fractional library routines.
39826 * __satfractsqta: Fixed-point fractional library routines.
39828 * __satfractsquda: Fixed-point fractional library routines.
39830 * __satfractsqudq: Fixed-point fractional library routines.
39832 * __satfractsquha: Fixed-point fractional library routines.
39834 * __satfractsquhq: Fixed-point fractional library routines.
39836 * __satfractsquqq: Fixed-point fractional library routines.
39838 * __satfractsqusa: Fixed-point fractional library routines.
39840 * __satfractsqusq: Fixed-point fractional library routines.
39842 * __satfractsquta: Fixed-point fractional library routines.
39844 * __satfracttada2: Fixed-point fractional library routines.
39846 * __satfracttadq: Fixed-point fractional library routines.
39848 * __satfracttaha2: Fixed-point fractional library routines.
39850 * __satfracttahq: Fixed-point fractional library routines.
39852 * __satfracttaqq: Fixed-point fractional library routines.
39854 * __satfracttasa2: Fixed-point fractional library routines.
39856 * __satfracttasq: Fixed-point fractional library routines.
39858 * __satfracttauda: Fixed-point fractional library routines.
39860 * __satfracttaudq: Fixed-point fractional library routines.
39862 * __satfracttauha: Fixed-point fractional library routines.
39864 * __satfracttauhq: Fixed-point fractional library routines.
39866 * __satfracttauqq: Fixed-point fractional library routines.
39868 * __satfracttausa: Fixed-point fractional library routines.
39870 * __satfracttausq: Fixed-point fractional library routines.
39872 * __satfracttauta: Fixed-point fractional library routines.
39874 * __satfracttida: Fixed-point fractional library routines.
39876 * __satfracttidq: Fixed-point fractional library routines.
39878 * __satfracttiha: Fixed-point fractional library routines.
39880 * __satfracttihq: Fixed-point fractional library routines.
39882 * __satfracttiqq: Fixed-point fractional library routines.
39884 * __satfracttisa: Fixed-point fractional library routines.
39886 * __satfracttisq: Fixed-point fractional library routines.
39888 * __satfracttita: Fixed-point fractional library routines.
39890 * __satfracttiuda: Fixed-point fractional library routines.
39892 * __satfracttiudq: Fixed-point fractional library routines.
39894 * __satfracttiuha: Fixed-point fractional library routines.
39896 * __satfracttiuhq: Fixed-point fractional library routines.
39898 * __satfracttiuqq: Fixed-point fractional library routines.
39900 * __satfracttiusa: Fixed-point fractional library routines.
39902 * __satfracttiusq: Fixed-point fractional library routines.
39904 * __satfracttiuta: Fixed-point fractional library routines.
39906 * __satfractudada: Fixed-point fractional library routines.
39908 * __satfractudadq: Fixed-point fractional library routines.
39910 * __satfractudaha: Fixed-point fractional library routines.
39912 * __satfractudahq: Fixed-point fractional library routines.
39914 * __satfractudaqq: Fixed-point fractional library routines.
39916 * __satfractudasa: Fixed-point fractional library routines.
39918 * __satfractudasq: Fixed-point fractional library routines.
39920 * __satfractudata: Fixed-point fractional library routines.
39922 * __satfractudaudq: Fixed-point fractional library routines.
39924 * __satfractudauha2: Fixed-point fractional library routines.
39926 * __satfractudauhq: Fixed-point fractional library routines.
39928 * __satfractudauqq: Fixed-point fractional library routines.
39930 * __satfractudausa2: Fixed-point fractional library routines.
39932 * __satfractudausq: Fixed-point fractional library routines.
39934 * __satfractudauta2: Fixed-point fractional library routines.
39936 * __satfractudqda: Fixed-point fractional library routines.
39938 * __satfractudqdq: Fixed-point fractional library routines.
39940 * __satfractudqha: Fixed-point fractional library routines.
39942 * __satfractudqhq: Fixed-point fractional library routines.
39944 * __satfractudqqq: Fixed-point fractional library routines.
39946 * __satfractudqsa: Fixed-point fractional library routines.
39948 * __satfractudqsq: Fixed-point fractional library routines.
39950 * __satfractudqta: Fixed-point fractional library routines.
39952 * __satfractudquda: Fixed-point fractional library routines.
39954 * __satfractudquha: Fixed-point fractional library routines.
39956 * __satfractudquhq2: Fixed-point fractional library routines.
39958 * __satfractudquqq2: Fixed-point fractional library routines.
39960 * __satfractudqusa: Fixed-point fractional library routines.
39962 * __satfractudqusq2: Fixed-point fractional library routines.
39964 * __satfractudquta: Fixed-point fractional library routines.
39966 * __satfractuhada: Fixed-point fractional library routines.
39968 * __satfractuhadq: Fixed-point fractional library routines.
39970 * __satfractuhaha: Fixed-point fractional library routines.
39972 * __satfractuhahq: Fixed-point fractional library routines.
39974 * __satfractuhaqq: Fixed-point fractional library routines.
39976 * __satfractuhasa: Fixed-point fractional library routines.
39978 * __satfractuhasq: Fixed-point fractional library routines.
39980 * __satfractuhata: Fixed-point fractional library routines.
39982 * __satfractuhauda2: Fixed-point fractional library routines.
39984 * __satfractuhaudq: Fixed-point fractional library routines.
39986 * __satfractuhauhq: Fixed-point fractional library routines.
39988 * __satfractuhauqq: Fixed-point fractional library routines.
39990 * __satfractuhausa2: Fixed-point fractional library routines.
39992 * __satfractuhausq: Fixed-point fractional library routines.
39994 * __satfractuhauta2: Fixed-point fractional library routines.
39996 * __satfractuhqda: Fixed-point fractional library routines.
39998 * __satfractuhqdq: Fixed-point fractional library routines.
40000 * __satfractuhqha: Fixed-point fractional library routines.
40002 * __satfractuhqhq: Fixed-point fractional library routines.
40004 * __satfractuhqqq: Fixed-point fractional library routines.
40006 * __satfractuhqsa: Fixed-point fractional library routines.
40008 * __satfractuhqsq: Fixed-point fractional library routines.
40010 * __satfractuhqta: Fixed-point fractional library routines.
40012 * __satfractuhquda: Fixed-point fractional library routines.
40014 * __satfractuhqudq2: Fixed-point fractional library routines.
40016 * __satfractuhquha: Fixed-point fractional library routines.
40018 * __satfractuhquqq2: Fixed-point fractional library routines.
40020 * __satfractuhqusa: Fixed-point fractional library routines.
40022 * __satfractuhqusq2: Fixed-point fractional library routines.
40024 * __satfractuhquta: Fixed-point fractional library routines.
40026 * __satfractunsdida: Fixed-point fractional library routines.
40028 * __satfractunsdidq: Fixed-point fractional library routines.
40030 * __satfractunsdiha: Fixed-point fractional library routines.
40032 * __satfractunsdihq: Fixed-point fractional library routines.
40034 * __satfractunsdiqq: Fixed-point fractional library routines.
40036 * __satfractunsdisa: Fixed-point fractional library routines.
40038 * __satfractunsdisq: Fixed-point fractional library routines.
40040 * __satfractunsdita: Fixed-point fractional library routines.
40042 * __satfractunsdiuda: Fixed-point fractional library routines.
40044 * __satfractunsdiudq: Fixed-point fractional library routines.
40046 * __satfractunsdiuha: Fixed-point fractional library routines.
40048 * __satfractunsdiuhq: Fixed-point fractional library routines.
40050 * __satfractunsdiuqq: Fixed-point fractional library routines.
40052 * __satfractunsdiusa: Fixed-point fractional library routines.
40054 * __satfractunsdiusq: Fixed-point fractional library routines.
40056 * __satfractunsdiuta: Fixed-point fractional library routines.
40058 * __satfractunshida: Fixed-point fractional library routines.
40060 * __satfractunshidq: Fixed-point fractional library routines.
40062 * __satfractunshiha: Fixed-point fractional library routines.
40064 * __satfractunshihq: Fixed-point fractional library routines.
40066 * __satfractunshiqq: Fixed-point fractional library routines.
40068 * __satfractunshisa: Fixed-point fractional library routines.
40070 * __satfractunshisq: Fixed-point fractional library routines.
40072 * __satfractunshita: Fixed-point fractional library routines.
40074 * __satfractunshiuda: Fixed-point fractional library routines.
40076 * __satfractunshiudq: Fixed-point fractional library routines.
40078 * __satfractunshiuha: Fixed-point fractional library routines.
40080 * __satfractunshiuhq: Fixed-point fractional library routines.
40082 * __satfractunshiuqq: Fixed-point fractional library routines.
40084 * __satfractunshiusa: Fixed-point fractional library routines.
40086 * __satfractunshiusq: Fixed-point fractional library routines.
40088 * __satfractunshiuta: Fixed-point fractional library routines.
40090 * __satfractunsqida: Fixed-point fractional library routines.
40092 * __satfractunsqidq: Fixed-point fractional library routines.
40094 * __satfractunsqiha: Fixed-point fractional library routines.
40096 * __satfractunsqihq: Fixed-point fractional library routines.
40098 * __satfractunsqiqq: Fixed-point fractional library routines.
40100 * __satfractunsqisa: Fixed-point fractional library routines.
40102 * __satfractunsqisq: Fixed-point fractional library routines.
40104 * __satfractunsqita: Fixed-point fractional library routines.
40106 * __satfractunsqiuda: Fixed-point fractional library routines.
40108 * __satfractunsqiudq: Fixed-point fractional library routines.
40110 * __satfractunsqiuha: Fixed-point fractional library routines.
40112 * __satfractunsqiuhq: Fixed-point fractional library routines.
40114 * __satfractunsqiuqq: Fixed-point fractional library routines.
40116 * __satfractunsqiusa: Fixed-point fractional library routines.
40118 * __satfractunsqiusq: Fixed-point fractional library routines.
40120 * __satfractunsqiuta: Fixed-point fractional library routines.
40122 * __satfractunssida: Fixed-point fractional library routines.
40124 * __satfractunssidq: Fixed-point fractional library routines.
40126 * __satfractunssiha: Fixed-point fractional library routines.
40128 * __satfractunssihq: Fixed-point fractional library routines.
40130 * __satfractunssiqq: Fixed-point fractional library routines.
40132 * __satfractunssisa: Fixed-point fractional library routines.
40134 * __satfractunssisq: Fixed-point fractional library routines.
40136 * __satfractunssita: Fixed-point fractional library routines.
40138 * __satfractunssiuda: Fixed-point fractional library routines.
40140 * __satfractunssiudq: Fixed-point fractional library routines.
40142 * __satfractunssiuha: Fixed-point fractional library routines.
40144 * __satfractunssiuhq: Fixed-point fractional library routines.
40146 * __satfractunssiuqq: Fixed-point fractional library routines.
40148 * __satfractunssiusa: Fixed-point fractional library routines.
40150 * __satfractunssiusq: Fixed-point fractional library routines.
40152 * __satfractunssiuta: Fixed-point fractional library routines.
40154 * __satfractunstida: Fixed-point fractional library routines.
40156 * __satfractunstidq: Fixed-point fractional library routines.
40158 * __satfractunstiha: Fixed-point fractional library routines.
40160 * __satfractunstihq: Fixed-point fractional library routines.
40162 * __satfractunstiqq: Fixed-point fractional library routines.
40164 * __satfractunstisa: Fixed-point fractional library routines.
40166 * __satfractunstisq: Fixed-point fractional library routines.
40168 * __satfractunstita: Fixed-point fractional library routines.
40170 * __satfractunstiuda: Fixed-point fractional library routines.
40172 * __satfractunstiudq: Fixed-point fractional library routines.
40174 * __satfractunstiuha: Fixed-point fractional library routines.
40176 * __satfractunstiuhq: Fixed-point fractional library routines.
40178 * __satfractunstiuqq: Fixed-point fractional library routines.
40180 * __satfractunstiusa: Fixed-point fractional library routines.
40182 * __satfractunstiusq: Fixed-point fractional library routines.
40184 * __satfractunstiuta: Fixed-point fractional library routines.
40186 * __satfractuqqda: Fixed-point fractional library routines.
40188 * __satfractuqqdq: Fixed-point fractional library routines.
40190 * __satfractuqqha: Fixed-point fractional library routines.
40192 * __satfractuqqhq: Fixed-point fractional library routines.
40194 * __satfractuqqqq: Fixed-point fractional library routines.
40196 * __satfractuqqsa: Fixed-point fractional library routines.
40198 * __satfractuqqsq: Fixed-point fractional library routines.
40200 * __satfractuqqta: Fixed-point fractional library routines.
40202 * __satfractuqquda: Fixed-point fractional library routines.
40204 * __satfractuqqudq2: Fixed-point fractional library routines.
40206 * __satfractuqquha: Fixed-point fractional library routines.
40208 * __satfractuqquhq2: Fixed-point fractional library routines.
40210 * __satfractuqqusa: Fixed-point fractional library routines.
40212 * __satfractuqqusq2: Fixed-point fractional library routines.
40214 * __satfractuqquta: Fixed-point fractional library routines.
40216 * __satfractusada: Fixed-point fractional library routines.
40218 * __satfractusadq: Fixed-point fractional library routines.
40220 * __satfractusaha: Fixed-point fractional library routines.
40222 * __satfractusahq: Fixed-point fractional library routines.
40224 * __satfractusaqq: Fixed-point fractional library routines.
40226 * __satfractusasa: Fixed-point fractional library routines.
40228 * __satfractusasq: Fixed-point fractional library routines.
40230 * __satfractusata: Fixed-point fractional library routines.
40232 * __satfractusauda2: Fixed-point fractional library routines.
40234 * __satfractusaudq: Fixed-point fractional library routines.
40236 * __satfractusauha2: Fixed-point fractional library routines.
40238 * __satfractusauhq: Fixed-point fractional library routines.
40240 * __satfractusauqq: Fixed-point fractional library routines.
40242 * __satfractusausq: Fixed-point fractional library routines.
40244 * __satfractusauta2: Fixed-point fractional library routines.
40246 * __satfractusqda: Fixed-point fractional library routines.
40248 * __satfractusqdq: Fixed-point fractional library routines.
40250 * __satfractusqha: Fixed-point fractional library routines.
40252 * __satfractusqhq: Fixed-point fractional library routines.
40254 * __satfractusqqq: Fixed-point fractional library routines.
40256 * __satfractusqsa: Fixed-point fractional library routines.
40258 * __satfractusqsq: Fixed-point fractional library routines.
40260 * __satfractusqta: Fixed-point fractional library routines.
40262 * __satfractusquda: Fixed-point fractional library routines.
40264 * __satfractusqudq2: Fixed-point fractional library routines.
40266 * __satfractusquha: Fixed-point fractional library routines.
40268 * __satfractusquhq2: Fixed-point fractional library routines.
40270 * __satfractusquqq2: Fixed-point fractional library routines.
40272 * __satfractusqusa: Fixed-point fractional library routines.
40274 * __satfractusquta: Fixed-point fractional library routines.
40276 * __satfractutada: Fixed-point fractional library routines.
40278 * __satfractutadq: Fixed-point fractional library routines.
40280 * __satfractutaha: Fixed-point fractional library routines.
40282 * __satfractutahq: Fixed-point fractional library routines.
40284 * __satfractutaqq: Fixed-point fractional library routines.
40286 * __satfractutasa: Fixed-point fractional library routines.
40288 * __satfractutasq: Fixed-point fractional library routines.
40290 * __satfractutata: Fixed-point fractional library routines.
40292 * __satfractutauda2: Fixed-point fractional library routines.
40294 * __satfractutaudq: Fixed-point fractional library routines.
40296 * __satfractutauha2: Fixed-point fractional library routines.
40298 * __satfractutauhq: Fixed-point fractional library routines.
40300 * __satfractutauqq: Fixed-point fractional library routines.
40302 * __satfractutausa2: Fixed-point fractional library routines.
40304 * __satfractutausq: Fixed-point fractional library routines.
40306 * __ssaddda3: Fixed-point fractional library routines.
40308 * __ssadddq3: Fixed-point fractional library routines.
40310 * __ssaddha3: Fixed-point fractional library routines.
40312 * __ssaddhq3: Fixed-point fractional library routines.
40314 * __ssaddqq3: Fixed-point fractional library routines.
40316 * __ssaddsa3: Fixed-point fractional library routines.
40318 * __ssaddsq3: Fixed-point fractional library routines.
40320 * __ssaddta3: Fixed-point fractional library routines.
40322 * __ssashlda3: Fixed-point fractional library routines.
40324 * __ssashldq3: Fixed-point fractional library routines.
40326 * __ssashlha3: Fixed-point fractional library routines.
40328 * __ssashlhq3: Fixed-point fractional library routines.
40330 * __ssashlsa3: Fixed-point fractional library routines.
40332 * __ssashlsq3: Fixed-point fractional library routines.
40334 * __ssashlta3: Fixed-point fractional library routines.
40336 * __ssdivda3: Fixed-point fractional library routines.
40338 * __ssdivdq3: Fixed-point fractional library routines.
40340 * __ssdivha3: Fixed-point fractional library routines.
40342 * __ssdivhq3: Fixed-point fractional library routines.
40344 * __ssdivqq3: Fixed-point fractional library routines.
40346 * __ssdivsa3: Fixed-point fractional library routines.
40348 * __ssdivsq3: Fixed-point fractional library routines.
40350 * __ssdivta3: Fixed-point fractional library routines.
40352 * __ssmulda3: Fixed-point fractional library routines.
40354 * __ssmuldq3: Fixed-point fractional library routines.
40356 * __ssmulha3: Fixed-point fractional library routines.
40358 * __ssmulhq3: Fixed-point fractional library routines.
40360 * __ssmulqq3: Fixed-point fractional library routines.
40362 * __ssmulsa3: Fixed-point fractional library routines.
40364 * __ssmulsq3: Fixed-point fractional library routines.
40366 * __ssmulta3: Fixed-point fractional library routines.
40368 * __ssnegda2: Fixed-point fractional library routines.
40370 * __ssnegdq2: Fixed-point fractional library routines.
40372 * __ssnegha2: Fixed-point fractional library routines.
40374 * __ssneghq2: Fixed-point fractional library routines.
40376 * __ssnegqq2: Fixed-point fractional library routines.
40378 * __ssnegsa2: Fixed-point fractional library routines.
40380 * __ssnegsq2: Fixed-point fractional library routines.
40382 * __ssnegta2: Fixed-point fractional library routines.
40384 * __sssubda3: Fixed-point fractional library routines.
40386 * __sssubdq3: Fixed-point fractional library routines.
40388 * __sssubha3: Fixed-point fractional library routines.
40390 * __sssubhq3: Fixed-point fractional library routines.
40392 * __sssubqq3: Fixed-point fractional library routines.
40394 * __sssubsa3: Fixed-point fractional library routines.
40396 * __sssubsq3: Fixed-point fractional library routines.
40398 * __sssubta3: Fixed-point fractional library routines.
40400 * __subda3: Fixed-point fractional library routines.
40402 * __subdf3: Soft float library routines.
40404 * __subdq3: Fixed-point fractional library routines.
40406 * __subha3: Fixed-point fractional library routines.
40408 * __subhq3: Fixed-point fractional library routines.
40410 * __subqq3: Fixed-point fractional library routines.
40412 * __subsa3: Fixed-point fractional library routines.
40414 * __subsf3: Soft float library routines.
40416 * __subsq3: Fixed-point fractional library routines.
40418 * __subta3: Fixed-point fractional library routines.
40420 * __subtf3: Soft float library routines.
40422 * __subuda3: Fixed-point fractional library routines.
40424 * __subudq3: Fixed-point fractional library routines.
40426 * __subuha3: Fixed-point fractional library routines.
40428 * __subuhq3: Fixed-point fractional library routines.
40430 * __subuqq3: Fixed-point fractional library routines.
40432 * __subusa3: Fixed-point fractional library routines.
40434 * __subusq3: Fixed-point fractional library routines.
40436 * __subuta3: Fixed-point fractional library routines.
40438 * __subvdi3: Integer library routines.
40440 * __subvsi3: Integer library routines.
40442 * __subxf3: Soft float library routines.
40444 * __truncdfsf2: Soft float library routines.
40446 * __trunctfdf2: Soft float library routines.
40448 * __trunctfsf2: Soft float library routines.
40450 * __truncxfdf2: Soft float library routines.
40452 * __truncxfsf2: Soft float library routines.
40454 * __ucmpdi2: Integer library routines.
40456 * __ucmpti2: Integer library routines.
40458 * __udivdi3: Integer library routines.
40460 * __udivmoddi3: Integer library routines.
40462 * __udivsi3: Integer library routines.
40464 * __udivti3: Integer library routines.
40466 * __udivuda3: Fixed-point fractional library routines.
40468 * __udivudq3: Fixed-point fractional library routines.
40470 * __udivuha3: Fixed-point fractional library routines.
40472 * __udivuhq3: Fixed-point fractional library routines.
40474 * __udivuqq3: Fixed-point fractional library routines.
40476 * __udivusa3: Fixed-point fractional library routines.
40478 * __udivusq3: Fixed-point fractional library routines.
40480 * __udivuta3: Fixed-point fractional library routines.
40482 * __umoddi3: Integer library routines.
40484 * __umodsi3: Integer library routines.
40486 * __umodti3: Integer library routines.
40488 * __unorddf2: Soft float library routines.
40490 * __unordsf2: Soft float library routines.
40492 * __unordtf2: Soft float library routines.
40494 * __usadduda3: Fixed-point fractional library routines.
40496 * __usaddudq3: Fixed-point fractional library routines.
40498 * __usadduha3: Fixed-point fractional library routines.
40500 * __usadduhq3: Fixed-point fractional library routines.
40502 * __usadduqq3: Fixed-point fractional library routines.
40504 * __usaddusa3: Fixed-point fractional library routines.
40506 * __usaddusq3: Fixed-point fractional library routines.
40508 * __usadduta3: Fixed-point fractional library routines.
40510 * __usashluda3: Fixed-point fractional library routines.
40512 * __usashludq3: Fixed-point fractional library routines.
40514 * __usashluha3: Fixed-point fractional library routines.
40516 * __usashluhq3: Fixed-point fractional library routines.
40518 * __usashluqq3: Fixed-point fractional library routines.
40520 * __usashlusa3: Fixed-point fractional library routines.
40522 * __usashlusq3: Fixed-point fractional library routines.
40524 * __usashluta3: Fixed-point fractional library routines.
40526 * __usdivuda3: Fixed-point fractional library routines.
40528 * __usdivudq3: Fixed-point fractional library routines.
40530 * __usdivuha3: Fixed-point fractional library routines.
40532 * __usdivuhq3: Fixed-point fractional library routines.
40534 * __usdivuqq3: Fixed-point fractional library routines.
40536 * __usdivusa3: Fixed-point fractional library routines.
40538 * __usdivusq3: Fixed-point fractional library routines.
40540 * __usdivuta3: Fixed-point fractional library routines.
40542 * __usmuluda3: Fixed-point fractional library routines.
40544 * __usmuludq3: Fixed-point fractional library routines.
40546 * __usmuluha3: Fixed-point fractional library routines.
40548 * __usmuluhq3: Fixed-point fractional library routines.
40550 * __usmuluqq3: Fixed-point fractional library routines.
40552 * __usmulusa3: Fixed-point fractional library routines.
40554 * __usmulusq3: Fixed-point fractional library routines.
40556 * __usmuluta3: Fixed-point fractional library routines.
40558 * __usneguda2: Fixed-point fractional library routines.
40560 * __usnegudq2: Fixed-point fractional library routines.
40562 * __usneguha2: Fixed-point fractional library routines.
40564 * __usneguhq2: Fixed-point fractional library routines.
40566 * __usneguqq2: Fixed-point fractional library routines.
40568 * __usnegusa2: Fixed-point fractional library routines.
40570 * __usnegusq2: Fixed-point fractional library routines.
40572 * __usneguta2: Fixed-point fractional library routines.
40574 * __ussubuda3: Fixed-point fractional library routines.
40576 * __ussubudq3: Fixed-point fractional library routines.
40578 * __ussubuha3: Fixed-point fractional library routines.
40580 * __ussubuhq3: Fixed-point fractional library routines.
40582 * __ussubuqq3: Fixed-point fractional library routines.
40584 * __ussubusa3: Fixed-point fractional library routines.
40586 * __ussubusq3: Fixed-point fractional library routines.
40588 * __ussubuta3: Fixed-point fractional library routines.
40590 * abort: Portability. (line 21)
40591 * abs: Arithmetic. (line 195)
40592 * abs and attributes: Expressions. (line 64)
40593 * ABS_EXPR: Expression trees. (line 6)
40594 * absence_set: Processor pipeline description.
40596 * absM2 instruction pattern: Standard Names. (line 452)
40597 * absolute value: Arithmetic. (line 195)
40598 * access to operands: Accessors. (line 6)
40599 * access to special operands: Special Accessors. (line 6)
40600 * accessors: Accessors. (line 6)
40601 * ACCUM_TYPE_SIZE: Type Layout. (line 88)
40602 * ACCUMULATE_OUTGOING_ARGS: Stack Arguments. (line 46)
40603 * ACCUMULATE_OUTGOING_ARGS and stack frames: Function Entry. (line 135)
40604 * ADA_LONG_TYPE_SIZE: Type Layout. (line 26)
40605 * Adding a new GIMPLE statement code: Adding a new GIMPLE statement code.
40607 * ADDITIONAL_REGISTER_NAMES: Instruction Output. (line 15)
40608 * addM3 instruction pattern: Standard Names. (line 216)
40609 * addMODEcc instruction pattern: Standard Names. (line 904)
40610 * addr_diff_vec: Side Effects. (line 302)
40611 * addr_diff_vec, length of: Insn Lengths. (line 26)
40612 * ADDR_EXPR: Expression trees. (line 6)
40613 * addr_vec: Side Effects. (line 297)
40614 * addr_vec, length of: Insn Lengths. (line 26)
40615 * address constraints: Simple Constraints. (line 154)
40616 * address_operand <1>: Simple Constraints. (line 158)
40617 * address_operand: Machine-Independent Predicates.
40619 * addressing modes: Addressing Modes. (line 6)
40620 * ADJUST_FIELD_ALIGN: Storage Layout. (line 201)
40621 * ADJUST_INSN_LENGTH: Insn Lengths. (line 35)
40622 * AGGR_INIT_EXPR: Expression trees. (line 6)
40623 * aggregates as return values: Aggregate Return. (line 6)
40624 * alias: Alias analysis. (line 6)
40625 * ALL_COP_ADDITIONAL_REGISTER_NAMES: MIPS Coprocessors. (line 32)
40626 * ALL_REGS: Register Classes. (line 17)
40627 * allocate_stack instruction pattern: Standard Names. (line 1227)
40628 * alternate entry points: Insns. (line 140)
40629 * anchored addresses: Anchored Addresses. (line 6)
40630 * and: Arithmetic. (line 153)
40631 * and and attributes: Expressions. (line 50)
40632 * and, canonicalization of: Insn Canonicalizations.
40634 * andM3 instruction pattern: Standard Names. (line 222)
40635 * annotations: Annotations. (line 6)
40636 * APPLY_RESULT_SIZE: Scalar Return. (line 95)
40637 * ARG_POINTER_CFA_OFFSET: Frame Layout. (line 194)
40638 * ARG_POINTER_REGNUM: Frame Registers. (line 41)
40639 * ARG_POINTER_REGNUM and virtual registers: Regs and Memory. (line 65)
40640 * arg_pointer_rtx: Frame Registers. (line 85)
40641 * ARGS_GROW_DOWNWARD: Frame Layout. (line 35)
40642 * argument passing: Interface. (line 36)
40643 * arguments in registers: Register Arguments. (line 6)
40644 * arguments on stack: Stack Arguments. (line 6)
40645 * arithmetic library: Soft float library routines.
40647 * arithmetic shift: Arithmetic. (line 168)
40648 * arithmetic shift with signed saturation: Arithmetic. (line 168)
40649 * arithmetic shift with unsigned saturation: Arithmetic. (line 168)
40650 * arithmetic, in RTL: Arithmetic. (line 6)
40651 * ARITHMETIC_TYPE_P: Types. (line 76)
40652 * array: Types. (line 6)
40653 * ARRAY_RANGE_REF: Expression trees. (line 6)
40654 * ARRAY_REF: Expression trees. (line 6)
40655 * ARRAY_TYPE: Types. (line 6)
40656 * AS_NEEDS_DASH_FOR_PIPED_INPUT: Driver. (line 151)
40657 * ashift: Arithmetic. (line 168)
40658 * ashift and attributes: Expressions. (line 64)
40659 * ashiftrt: Arithmetic. (line 185)
40660 * ashiftrt and attributes: Expressions. (line 64)
40661 * ashlM3 instruction pattern: Standard Names. (line 431)
40662 * ashrM3 instruction pattern: Standard Names. (line 441)
40663 * ASM_APP_OFF: File Framework. (line 61)
40664 * ASM_APP_ON: File Framework. (line 54)
40665 * ASM_COMMENT_START: File Framework. (line 49)
40666 * ASM_DECLARE_CLASS_REFERENCE: Label Output. (line 436)
40667 * ASM_DECLARE_CONSTANT_NAME: Label Output. (line 128)
40668 * ASM_DECLARE_FUNCTION_NAME: Label Output. (line 87)
40669 * ASM_DECLARE_FUNCTION_SIZE: Label Output. (line 101)
40670 * ASM_DECLARE_OBJECT_NAME: Label Output. (line 114)
40671 * ASM_DECLARE_REGISTER_GLOBAL: Label Output. (line 143)
40672 * ASM_DECLARE_UNRESOLVED_REFERENCE: Label Output. (line 442)
40673 * ASM_FINAL_SPEC: Driver. (line 144)
40674 * ASM_FINISH_DECLARE_OBJECT: Label Output. (line 151)
40675 * ASM_FORMAT_PRIVATE_NAME: Label Output. (line 354)
40676 * asm_fprintf: Instruction Output. (line 123)
40677 * ASM_FPRINTF_EXTENSIONS: Instruction Output. (line 134)
40678 * ASM_GENERATE_INTERNAL_LABEL: Label Output. (line 338)
40679 * asm_input: Side Effects. (line 284)
40680 * asm_input and /v: Flags. (line 94)
40681 * ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX: Exception Handling. (line 82)
40682 * ASM_NO_SKIP_IN_TEXT: Alignment Output. (line 72)
40683 * asm_noperands: Insns. (line 266)
40684 * asm_operands and /v: Flags. (line 94)
40685 * asm_operands, RTL sharing: Sharing. (line 45)
40686 * asm_operands, usage: Assembler. (line 6)
40687 * ASM_OUTPUT_ADDR_DIFF_ELT: Dispatch Tables. (line 9)
40688 * ASM_OUTPUT_ADDR_VEC_ELT: Dispatch Tables. (line 26)
40689 * ASM_OUTPUT_ALIGN: Alignment Output. (line 79)
40690 * ASM_OUTPUT_ALIGN_WITH_NOP: Alignment Output. (line 84)
40691 * ASM_OUTPUT_ALIGNED_BSS: Uninitialized Data. (line 64)
40692 * ASM_OUTPUT_ALIGNED_COMMON: Uninitialized Data. (line 23)
40693 * ASM_OUTPUT_ALIGNED_DECL_COMMON: Uninitialized Data. (line 31)
40694 * ASM_OUTPUT_ALIGNED_DECL_LOCAL: Uninitialized Data. (line 95)
40695 * ASM_OUTPUT_ALIGNED_LOCAL: Uninitialized Data. (line 87)
40696 * ASM_OUTPUT_ASCII: Data Output. (line 50)
40697 * ASM_OUTPUT_BSS: Uninitialized Data. (line 39)
40698 * ASM_OUTPUT_CASE_END: Dispatch Tables. (line 51)
40699 * ASM_OUTPUT_CASE_LABEL: Dispatch Tables. (line 38)
40700 * ASM_OUTPUT_COMMON: Uninitialized Data. (line 10)
40701 * ASM_OUTPUT_DEBUG_LABEL: Label Output. (line 326)
40702 * ASM_OUTPUT_DEF: Label Output. (line 375)
40703 * ASM_OUTPUT_DEF_FROM_DECLS: Label Output. (line 383)
40704 * ASM_OUTPUT_DWARF_DELTA: SDB and DWARF. (line 42)
40705 * ASM_OUTPUT_DWARF_OFFSET: SDB and DWARF. (line 46)
40706 * ASM_OUTPUT_DWARF_PCREL: SDB and DWARF. (line 52)
40707 * ASM_OUTPUT_EXTERNAL: Label Output. (line 264)
40708 * ASM_OUTPUT_FDESC: Data Output. (line 59)
40709 * ASM_OUTPUT_IDENT: File Framework. (line 83)
40710 * ASM_OUTPUT_INTERNAL_LABEL: Label Output. (line 17)
40711 * ASM_OUTPUT_LABEL: Label Output. (line 9)
40712 * ASM_OUTPUT_LABEL_REF: Label Output. (line 299)
40713 * ASM_OUTPUT_LABELREF: Label Output. (line 285)
40714 * ASM_OUTPUT_LOCAL: Uninitialized Data. (line 74)
40715 * ASM_OUTPUT_MAX_SKIP_ALIGN: Alignment Output. (line 88)
40716 * ASM_OUTPUT_MEASURED_SIZE: Label Output. (line 41)
40717 * ASM_OUTPUT_OPCODE: Instruction Output. (line 21)
40718 * ASM_OUTPUT_POOL_EPILOGUE: Data Output. (line 109)
40719 * ASM_OUTPUT_POOL_PROLOGUE: Data Output. (line 72)
40720 * ASM_OUTPUT_REG_POP: Instruction Output. (line 178)
40721 * ASM_OUTPUT_REG_PUSH: Instruction Output. (line 173)
40722 * ASM_OUTPUT_SIZE_DIRECTIVE: Label Output. (line 35)
40723 * ASM_OUTPUT_SKIP: Alignment Output. (line 66)
40724 * ASM_OUTPUT_SOURCE_FILENAME: File Framework. (line 68)
40725 * ASM_OUTPUT_SPECIAL_POOL_ENTRY: Data Output. (line 84)
40726 * ASM_OUTPUT_SYMBOL_REF: Label Output. (line 292)
40727 * ASM_OUTPUT_TYPE_DIRECTIVE: Label Output. (line 77)
40728 * ASM_OUTPUT_WEAK_ALIAS: Label Output. (line 401)
40729 * ASM_OUTPUT_WEAKREF: Label Output. (line 203)
40730 * ASM_PREFERRED_EH_DATA_FORMAT: Exception Handling. (line 67)
40731 * ASM_SPEC: Driver. (line 136)
40732 * ASM_STABD_OP: DBX Options. (line 36)
40733 * ASM_STABN_OP: DBX Options. (line 43)
40734 * ASM_STABS_OP: DBX Options. (line 29)
40735 * ASM_WEAKEN_DECL: Label Output. (line 195)
40736 * ASM_WEAKEN_LABEL: Label Output. (line 182)
40737 * assemble_name: Label Output. (line 8)
40738 * assemble_name_raw: Label Output. (line 16)
40739 * assembler format: File Framework. (line 6)
40740 * assembler instructions in RTL: Assembler. (line 6)
40741 * ASSEMBLER_DIALECT: Instruction Output. (line 146)
40742 * assigning attribute values to insns: Tagging Insns. (line 6)
40743 * assignment operator: Function Basics. (line 6)
40744 * asterisk in template: Output Statement. (line 29)
40745 * atan2M3 instruction pattern: Standard Names. (line 522)
40746 * attr <1>: Tagging Insns. (line 54)
40747 * attr: Expressions. (line 154)
40748 * attr_flag: Expressions. (line 119)
40749 * attribute expressions: Expressions. (line 6)
40750 * attribute specifications: Attr Example. (line 6)
40751 * attribute specifications example: Attr Example. (line 6)
40752 * ATTRIBUTE_ALIGNED_VALUE: Storage Layout. (line 183)
40753 * attributes: Attributes. (line 6)
40754 * attributes, defining: Defining Attributes.
40756 * attributes, target-specific: Target Attributes. (line 6)
40757 * autoincrement addressing, availability: Portability. (line 21)
40758 * autoincrement/decrement addressing: Simple Constraints. (line 30)
40759 * automata_option: Processor pipeline description.
40761 * automaton based pipeline description: Processor pipeline description.
40763 * automaton based scheduler: Processor pipeline description.
40765 * AVOID_CCMODE_COPIES: Values in Registers.
40767 * backslash: Output Template. (line 46)
40768 * barrier: Insns. (line 160)
40769 * barrier and /f: Flags. (line 125)
40770 * barrier and /v: Flags. (line 44)
40771 * BASE_REG_CLASS: Register Classes. (line 107)
40772 * basic block: Basic Blocks. (line 6)
40773 * basic-block.h: Control Flow. (line 6)
40774 * BASIC_BLOCK: Basic Blocks. (line 19)
40775 * basic_block: Basic Blocks. (line 6)
40776 * BB_HEAD, BB_END: Maintaining the CFG.
40778 * bb_seq: GIMPLE sequences. (line 73)
40779 * bCOND instruction pattern: Standard Names. (line 941)
40780 * BIGGEST_ALIGNMENT: Storage Layout. (line 173)
40781 * BIGGEST_FIELD_ALIGNMENT: Storage Layout. (line 194)
40782 * BImode: Machine Modes. (line 22)
40783 * BIND_EXPR: Expression trees. (line 6)
40784 * BINFO_TYPE: Classes. (line 6)
40785 * bit-fields: Bit-Fields. (line 6)
40786 * BIT_AND_EXPR: Expression trees. (line 6)
40787 * BIT_IOR_EXPR: Expression trees. (line 6)
40788 * BIT_NOT_EXPR: Expression trees. (line 6)
40789 * BIT_XOR_EXPR: Expression trees. (line 6)
40790 * BITFIELD_NBYTES_LIMITED: Storage Layout. (line 382)
40791 * BITS_BIG_ENDIAN: Storage Layout. (line 12)
40792 * BITS_BIG_ENDIAN, effect on sign_extract: Bit-Fields. (line 8)
40793 * BITS_PER_UNIT: Storage Layout. (line 52)
40794 * BITS_PER_WORD: Storage Layout. (line 57)
40795 * bitwise complement: Arithmetic. (line 149)
40796 * bitwise exclusive-or: Arithmetic. (line 163)
40797 * bitwise inclusive-or: Arithmetic. (line 158)
40798 * bitwise logical-and: Arithmetic. (line 153)
40799 * BLKmode: Machine Modes. (line 183)
40800 * BLKmode, and function return values: Calls. (line 23)
40801 * block statement iterators <1>: Maintaining the CFG.
40803 * block statement iterators: Basic Blocks. (line 68)
40804 * BLOCK_FOR_INSN, bb_for_stmt: Maintaining the CFG.
40806 * BLOCK_REG_PADDING: Register Arguments. (line 229)
40807 * blockage instruction pattern: Standard Names. (line 1408)
40808 * Blocks: Blocks. (line 6)
40809 * bool <1>: Exception Region Output.
40811 * bool: Sections. (line 280)
40812 * BOOL_TYPE_SIZE: Type Layout. (line 44)
40813 * BOOLEAN_TYPE: Types. (line 6)
40814 * branch prediction: Profile information.
40816 * BRANCH_COST: Costs. (line 52)
40817 * break_out_memory_refs: Addressing Modes. (line 130)
40818 * BREAK_STMT: Function Bodies. (line 6)
40819 * bsi_commit_edge_inserts: Maintaining the CFG.
40821 * bsi_end_p: Maintaining the CFG.
40823 * bsi_insert_after: Maintaining the CFG.
40825 * bsi_insert_before: Maintaining the CFG.
40827 * bsi_insert_on_edge: Maintaining the CFG.
40829 * bsi_last: Maintaining the CFG.
40831 * bsi_next: Maintaining the CFG.
40833 * bsi_prev: Maintaining the CFG.
40835 * bsi_remove: Maintaining the CFG.
40837 * bsi_start: Maintaining the CFG.
40839 * BSS_SECTION_ASM_OP: Sections. (line 68)
40840 * bswap: Arithmetic. (line 232)
40841 * btruncM2 instruction pattern: Standard Names. (line 540)
40842 * builtin_longjmp instruction pattern: Standard Names. (line 1313)
40843 * builtin_setjmp_receiver instruction pattern: Standard Names.
40845 * builtin_setjmp_setup instruction pattern: Standard Names. (line 1292)
40846 * byte_mode: Machine Modes. (line 336)
40847 * BYTES_BIG_ENDIAN: Storage Layout. (line 24)
40848 * BYTES_BIG_ENDIAN, effect on subreg: Regs and Memory. (line 221)
40849 * C statements for assembler output: Output Statement. (line 6)
40850 * C/C++ Internal Representation: Trees. (line 6)
40851 * C99 math functions, implicit usage: Library Calls. (line 76)
40852 * C_COMMON_OVERRIDE_OPTIONS: Run-time Target. (line 114)
40853 * c_register_pragma: Misc. (line 404)
40854 * c_register_pragma_with_expansion: Misc. (line 406)
40855 * call <1>: Side Effects. (line 86)
40856 * call: Flags. (line 234)
40857 * call instruction pattern: Standard Names. (line 974)
40858 * call usage: Calls. (line 10)
40859 * call, in call_insn: Flags. (line 33)
40860 * call, in mem: Flags. (line 99)
40861 * call-clobbered register: Register Basics. (line 35)
40862 * call-saved register: Register Basics. (line 35)
40863 * call-used register: Register Basics. (line 35)
40864 * CALL_EXPR: Expression trees. (line 6)
40865 * call_insn: Insns. (line 95)
40866 * call_insn and /c: Flags. (line 33)
40867 * call_insn and /f: Flags. (line 125)
40868 * call_insn and /i: Flags. (line 24)
40869 * call_insn and /j: Flags. (line 179)
40870 * call_insn and /s: Flags. (line 49)
40871 * call_insn and /u: Flags. (line 19)
40872 * call_insn and /u or /i: Flags. (line 29)
40873 * call_insn and /v: Flags. (line 44)
40874 * CALL_INSN_FUNCTION_USAGE: Insns. (line 101)
40875 * call_pop instruction pattern: Standard Names. (line 1002)
40876 * CALL_POPS_ARGS: Stack Arguments. (line 130)
40877 * CALL_REALLY_USED_REGISTERS: Register Basics. (line 46)
40878 * CALL_USED_REGISTERS: Register Basics. (line 35)
40879 * call_used_regs: Register Basics. (line 59)
40880 * call_value instruction pattern: Standard Names. (line 994)
40881 * call_value_pop instruction pattern: Standard Names. (line 1002)
40882 * CALLER_SAVE_PROFITABLE: Caller Saves. (line 11)
40883 * calling conventions: Stack and Calling. (line 6)
40884 * calling functions in RTL: Calls. (line 6)
40885 * can_create_pseudo_p: Standard Names. (line 75)
40886 * CAN_DEBUG_WITHOUT_FP: Run-time Target. (line 146)
40887 * CAN_ELIMINATE: Elimination. (line 71)
40888 * can_fallthru: Basic Blocks. (line 57)
40889 * canadian: Configure Terms. (line 6)
40890 * CANNOT_CHANGE_MODE_CLASS: Register Classes. (line 481)
40891 * CANNOT_CHANGE_MODE_CLASS and subreg semantics: Regs and Memory.
40893 * canonicalization of instructions: Insn Canonicalizations.
40895 * CANONICALIZE_COMPARISON: Condition Code. (line 84)
40896 * canonicalize_funcptr_for_compare instruction pattern: Standard Names.
40898 * CASE_USE_BIT_TESTS: Misc. (line 54)
40899 * CASE_VALUES_THRESHOLD: Misc. (line 47)
40900 * CASE_VECTOR_MODE: Misc. (line 27)
40901 * CASE_VECTOR_PC_RELATIVE: Misc. (line 40)
40902 * CASE_VECTOR_SHORTEN_MODE: Misc. (line 31)
40903 * casesi instruction pattern: Standard Names. (line 1082)
40904 * cbranchMODE4 instruction pattern: Standard Names. (line 963)
40905 * cc0: Regs and Memory. (line 307)
40906 * cc0, RTL sharing: Sharing. (line 27)
40907 * cc0_rtx: Regs and Memory. (line 333)
40908 * CC1_SPEC: Driver. (line 118)
40909 * CC1PLUS_SPEC: Driver. (line 126)
40910 * cc_status: Condition Code. (line 8)
40911 * CC_STATUS_MDEP: Condition Code. (line 19)
40912 * CC_STATUS_MDEP_INIT: Condition Code. (line 25)
40913 * CCmode: Machine Modes. (line 176)
40914 * CDImode: Machine Modes. (line 202)
40915 * CEIL_DIV_EXPR: Expression trees. (line 6)
40916 * CEIL_MOD_EXPR: Expression trees. (line 6)
40917 * ceilM2 instruction pattern: Standard Names. (line 556)
40918 * CFA_FRAME_BASE_OFFSET: Frame Layout. (line 226)
40919 * CFG, Control Flow Graph: Control Flow. (line 6)
40920 * cfghooks.h: Maintaining the CFG.
40922 * cgraph_finalize_function: Parsing pass. (line 52)
40923 * chain_circular: GTY Options. (line 196)
40924 * chain_next: GTY Options. (line 196)
40925 * chain_prev: GTY Options. (line 196)
40926 * change_address: Standard Names. (line 47)
40927 * CHANGE_DYNAMIC_TYPE_EXPR: Expression trees. (line 6)
40928 * char <1>: Misc. (line 693)
40929 * char <2>: PCH Target. (line 12)
40930 * char <3>: Sections. (line 272)
40931 * char: GIMPLE_ASM. (line 53)
40932 * CHAR_TYPE_SIZE: Type Layout. (line 39)
40933 * check_stack instruction pattern: Standard Names. (line 1245)
40934 * CHImode: Machine Modes. (line 202)
40935 * class: Classes. (line 6)
40936 * class definitions, register: Register Classes. (line 6)
40937 * class preference constraints: Class Preferences. (line 6)
40938 * CLASS_LIKELY_SPILLED_P: Register Classes. (line 452)
40939 * CLASS_MAX_NREGS: Register Classes. (line 469)
40940 * CLASS_TYPE_P: Types. (line 80)
40941 * classes of RTX codes: RTL Classes. (line 6)
40942 * CLASSTYPE_DECLARED_CLASS: Classes. (line 6)
40943 * CLASSTYPE_HAS_MUTABLE: Classes. (line 80)
40944 * CLASSTYPE_NON_POD_P: Classes. (line 85)
40945 * CLEANUP_DECL: Function Bodies. (line 6)
40946 * CLEANUP_EXPR: Function Bodies. (line 6)
40947 * CLEANUP_POINT_EXPR: Expression trees. (line 6)
40948 * CLEANUP_STMT: Function Bodies. (line 6)
40949 * Cleanups: Cleanups. (line 6)
40950 * CLEAR_BY_PIECES_P: Costs. (line 130)
40951 * clear_cache instruction pattern: Standard Names. (line 1555)
40952 * CLEAR_INSN_CACHE: Trampolines. (line 100)
40953 * CLEAR_RATIO: Costs. (line 121)
40954 * clobber: Side Effects. (line 100)
40955 * clz: Arithmetic. (line 208)
40956 * CLZ_DEFINED_VALUE_AT_ZERO: Misc. (line 319)
40957 * clzM2 instruction pattern: Standard Names. (line 621)
40958 * cmpM instruction pattern: Standard Names. (line 654)
40959 * cmpmemM instruction pattern: Standard Names. (line 769)
40960 * cmpstrM instruction pattern: Standard Names. (line 750)
40961 * cmpstrnM instruction pattern: Standard Names. (line 738)
40962 * code generation RTL sequences: Expander Definitions.
40964 * code iterators in .md files: Code Iterators. (line 6)
40965 * code_label: Insns. (line 119)
40966 * code_label and /i: Flags. (line 59)
40967 * code_label and /v: Flags. (line 44)
40968 * CODE_LABEL_NUMBER: Insns. (line 119)
40969 * codes, RTL expression: RTL Objects. (line 47)
40970 * COImode: Machine Modes. (line 202)
40971 * COLLECT2_HOST_INITIALIZATION: Host Misc. (line 32)
40972 * COLLECT_EXPORT_LIST: Misc. (line 775)
40973 * COLLECT_SHARED_FINI_FUNC: Macros for Initialization.
40975 * COLLECT_SHARED_INIT_FUNC: Macros for Initialization.
40977 * commit_edge_insertions: Maintaining the CFG.
40979 * compare: Arithmetic. (line 43)
40980 * compare, canonicalization of: Insn Canonicalizations.
40982 * comparison_operator: Machine-Independent Predicates.
40984 * compiler passes and files: Passes. (line 6)
40985 * complement, bitwise: Arithmetic. (line 149)
40986 * COMPLEX_CST: Expression trees. (line 6)
40987 * COMPLEX_EXPR: Expression trees. (line 6)
40988 * COMPLEX_TYPE: Types. (line 6)
40989 * COMPONENT_REF: Expression trees. (line 6)
40990 * Compound Expressions: Compound Expressions.
40992 * Compound Lvalues: Compound Lvalues. (line 6)
40993 * COMPOUND_EXPR: Expression trees. (line 6)
40994 * COMPOUND_LITERAL_EXPR: Expression trees. (line 6)
40995 * COMPOUND_LITERAL_EXPR_DECL: Expression trees. (line 608)
40996 * COMPOUND_LITERAL_EXPR_DECL_STMT: Expression trees. (line 608)
40997 * computed jump: Edges. (line 128)
40998 * computing the length of an insn: Insn Lengths. (line 6)
40999 * concat: Regs and Memory. (line 385)
41000 * concatn: Regs and Memory. (line 391)
41001 * cond: Comparisons. (line 90)
41002 * cond and attributes: Expressions. (line 37)
41003 * cond_exec: Side Effects. (line 248)
41004 * COND_EXPR: Expression trees. (line 6)
41005 * condition code register: Regs and Memory. (line 307)
41006 * condition code status: Condition Code. (line 6)
41007 * condition codes: Comparisons. (line 20)
41008 * conditional execution: Conditional Execution.
41010 * Conditional Expressions: Conditional Expressions.
41012 * CONDITIONAL_REGISTER_USAGE: Register Basics. (line 60)
41013 * conditional_trap instruction pattern: Standard Names. (line 1379)
41014 * conditions, in patterns: Patterns. (line 43)
41015 * configuration file <1>: Host Misc. (line 6)
41016 * configuration file: Filesystem. (line 6)
41017 * configure terms: Configure Terms. (line 6)
41018 * CONJ_EXPR: Expression trees. (line 6)
41019 * const: Constants. (line 99)
41020 * CONST0_RTX: Constants. (line 119)
41021 * const0_rtx: Constants. (line 16)
41022 * CONST1_RTX: Constants. (line 119)
41023 * const1_rtx: Constants. (line 16)
41024 * CONST2_RTX: Constants. (line 119)
41025 * const2_rtx: Constants. (line 16)
41026 * CONST_DECL: Declarations. (line 6)
41027 * const_double: Constants. (line 32)
41028 * const_double, RTL sharing: Sharing. (line 29)
41029 * CONST_DOUBLE_LOW: Constants. (line 39)
41030 * CONST_DOUBLE_OK_FOR_CONSTRAINT_P: Old Constraints. (line 69)
41031 * CONST_DOUBLE_OK_FOR_LETTER_P: Old Constraints. (line 54)
41032 * const_double_operand: Machine-Independent Predicates.
41034 * const_fixed: Constants. (line 52)
41035 * const_int: Constants. (line 8)
41036 * const_int and attribute tests: Expressions. (line 47)
41037 * const_int and attributes: Expressions. (line 10)
41038 * const_int, RTL sharing: Sharing. (line 23)
41039 * const_int_operand: Machine-Independent Predicates.
41041 * CONST_OK_FOR_CONSTRAINT_P: Old Constraints. (line 49)
41042 * CONST_OK_FOR_LETTER_P: Old Constraints. (line 40)
41043 * const_string: Constants. (line 71)
41044 * const_string and attributes: Expressions. (line 20)
41045 * const_true_rtx: Constants. (line 26)
41046 * const_vector: Constants. (line 59)
41047 * const_vector, RTL sharing: Sharing. (line 32)
41048 * constant attributes: Constant Attributes.
41050 * constant definitions: Constant Definitions.
41052 * CONSTANT_ADDRESS_P: Addressing Modes. (line 29)
41053 * CONSTANT_ALIGNMENT: Storage Layout. (line 241)
41054 * CONSTANT_P: Addressing Modes. (line 35)
41055 * CONSTANT_POOL_ADDRESS_P: Flags. (line 10)
41056 * CONSTANT_POOL_BEFORE_FUNCTION: Data Output. (line 64)
41057 * constants in constraints: Simple Constraints. (line 60)
41058 * constm1_rtx: Constants. (line 16)
41059 * constraint modifier characters: Modifiers. (line 6)
41060 * constraint, matching: Simple Constraints. (line 132)
41061 * CONSTRAINT_LEN: Old Constraints. (line 12)
41062 * constraint_num: C Constraint Interface.
41064 * constraint_satisfied_p: C Constraint Interface.
41066 * constraints: Constraints. (line 6)
41067 * constraints, defining: Define Constraints. (line 6)
41068 * constraints, defining, obsolete method: Old Constraints. (line 6)
41069 * constraints, machine specific: Machine Constraints.
41071 * constraints, testing: C Constraint Interface.
41073 * CONSTRUCTOR: Expression trees. (line 6)
41074 * constructor: Function Basics. (line 6)
41075 * constructors, automatic calls: Collect2. (line 15)
41076 * constructors, output of: Initialization. (line 6)
41077 * container: Containers. (line 6)
41078 * CONTINUE_STMT: Function Bodies. (line 6)
41079 * contributors: Contributors. (line 6)
41080 * controlling register usage: Register Basics. (line 76)
41081 * controlling the compilation driver: Driver. (line 6)
41082 * conventions, run-time: Interface. (line 6)
41083 * conversions: Conversions. (line 6)
41084 * CONVERT_EXPR: Expression trees. (line 6)
41085 * copy constructor: Function Basics. (line 6)
41086 * copy_rtx: Addressing Modes. (line 182)
41087 * copy_rtx_if_shared: Sharing. (line 64)
41088 * copysignM3 instruction pattern: Standard Names. (line 602)
41089 * cosM2 instruction pattern: Standard Names. (line 481)
41090 * costs of instructions: Costs. (line 6)
41091 * CP_INTEGRAL_TYPE: Types. (line 72)
41092 * cp_namespace_decls: Namespaces. (line 44)
41093 * CP_TYPE_CONST_NON_VOLATILE_P: Types. (line 45)
41094 * CP_TYPE_CONST_P: Types. (line 36)
41095 * CP_TYPE_QUALS: Types. (line 6)
41096 * CP_TYPE_RESTRICT_P: Types. (line 42)
41097 * CP_TYPE_VOLATILE_P: Types. (line 39)
41098 * CPLUSPLUS_CPP_SPEC: Driver. (line 113)
41099 * CPP_SPEC: Driver. (line 106)
41100 * CQImode: Machine Modes. (line 202)
41101 * cross compilation and floating point: Floating Point. (line 6)
41102 * CRT_CALL_STATIC_FUNCTION: Sections. (line 112)
41103 * CRTSTUFF_T_CFLAGS: Target Fragment. (line 35)
41104 * CRTSTUFF_T_CFLAGS_S: Target Fragment. (line 39)
41105 * CSImode: Machine Modes. (line 202)
41106 * CTImode: Machine Modes. (line 202)
41107 * ctz: Arithmetic. (line 216)
41108 * CTZ_DEFINED_VALUE_AT_ZERO: Misc. (line 320)
41109 * ctzM2 instruction pattern: Standard Names. (line 630)
41110 * CUMULATIVE_ARGS: Register Arguments. (line 127)
41111 * current_function_epilogue_delay_list: Function Entry. (line 181)
41112 * current_function_is_leaf: Leaf Functions. (line 51)
41113 * current_function_outgoing_args_size: Stack Arguments. (line 45)
41114 * current_function_pops_args: Function Entry. (line 106)
41115 * current_function_pretend_args_size: Function Entry. (line 112)
41116 * current_function_uses_only_leaf_regs: Leaf Functions. (line 51)
41117 * current_insn_predicate: Conditional Execution.
41119 * DAmode: Machine Modes. (line 152)
41120 * data bypass: Processor pipeline description.
41122 * data dependence delays: Processor pipeline description.
41124 * Data Dependency Analysis: Dependency analysis.
41126 * data structures: Per-Function Data. (line 6)
41127 * DATA_ALIGNMENT: Storage Layout. (line 228)
41128 * DATA_SECTION_ASM_OP: Sections. (line 53)
41129 * DBR_OUTPUT_SEQEND: Instruction Output. (line 107)
41130 * dbr_sequence_length: Instruction Output. (line 106)
41131 * DBX_BLOCKS_FUNCTION_RELATIVE: DBX Options. (line 103)
41132 * DBX_CONTIN_CHAR: DBX Options. (line 66)
41133 * DBX_CONTIN_LENGTH: DBX Options. (line 56)
41134 * DBX_DEBUGGING_INFO: DBX Options. (line 9)
41135 * DBX_FUNCTION_FIRST: DBX Options. (line 97)
41136 * DBX_LINES_FUNCTION_RELATIVE: DBX Options. (line 109)
41137 * DBX_NO_XREFS: DBX Options. (line 50)
41138 * DBX_OUTPUT_LBRAC: DBX Hooks. (line 9)
41139 * DBX_OUTPUT_MAIN_SOURCE_FILE_END: File Names and DBX. (line 34)
41140 * DBX_OUTPUT_MAIN_SOURCE_FILENAME: File Names and DBX. (line 9)
41141 * DBX_OUTPUT_NFUN: DBX Hooks. (line 18)
41142 * DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END: File Names and DBX.
41144 * DBX_OUTPUT_RBRAC: DBX Hooks. (line 15)
41145 * DBX_OUTPUT_SOURCE_LINE: DBX Hooks. (line 22)
41146 * DBX_REGISTER_NUMBER: All Debuggers. (line 9)
41147 * DBX_REGPARM_STABS_CODE: DBX Options. (line 87)
41148 * DBX_REGPARM_STABS_LETTER: DBX Options. (line 92)
41149 * DBX_STATIC_CONST_VAR_CODE: DBX Options. (line 82)
41150 * DBX_STATIC_STAB_DATA_SECTION: DBX Options. (line 73)
41151 * DBX_TYPE_DECL_STABS_CODE: DBX Options. (line 78)
41152 * DBX_USE_BINCL: DBX Options. (line 115)
41153 * DCmode: Machine Modes. (line 197)
41154 * DDmode: Machine Modes. (line 90)
41155 * De Morgan's law: Insn Canonicalizations.
41157 * dead_or_set_p: define_peephole. (line 65)
41158 * DEBUG_SYMS_TEXT: DBX Options. (line 25)
41159 * DEBUGGER_ARG_OFFSET: All Debuggers. (line 37)
41160 * DEBUGGER_AUTO_OFFSET: All Debuggers. (line 28)
41161 * decimal float library: Decimal float library routines.
41163 * DECL_ALIGN: Declarations. (line 6)
41164 * DECL_ANTICIPATED: Function Basics. (line 48)
41165 * DECL_ARGUMENTS: Function Basics. (line 163)
41166 * DECL_ARRAY_DELETE_OPERATOR_P: Function Basics. (line 184)
41167 * DECL_ARTIFICIAL <1>: Function Basics. (line 6)
41168 * DECL_ARTIFICIAL: Working with declarations.
41170 * DECL_ASSEMBLER_NAME: Function Basics. (line 6)
41171 * DECL_ATTRIBUTES: Attributes. (line 22)
41172 * DECL_BASE_CONSTRUCTOR_P: Function Basics. (line 94)
41173 * DECL_CLASS_SCOPE_P: Working with declarations.
41175 * DECL_COMPLETE_CONSTRUCTOR_P: Function Basics. (line 90)
41176 * DECL_COMPLETE_DESTRUCTOR_P: Function Basics. (line 104)
41177 * DECL_CONST_MEMFUNC_P: Function Basics. (line 77)
41178 * DECL_CONSTRUCTOR_P: Function Basics. (line 6)
41179 * DECL_CONTEXT: Namespaces. (line 26)
41180 * DECL_CONV_FN_P: Function Basics. (line 6)
41181 * DECL_COPY_CONSTRUCTOR_P: Function Basics. (line 98)
41182 * DECL_DESTRUCTOR_P: Function Basics. (line 6)
41183 * DECL_EXTERN_C_FUNCTION_P: Function Basics. (line 52)
41184 * DECL_EXTERNAL <1>: Function Basics. (line 38)
41185 * DECL_EXTERNAL: Declarations. (line 6)
41186 * DECL_FUNCTION_MEMBER_P: Function Basics. (line 6)
41187 * DECL_FUNCTION_SCOPE_P: Working with declarations.
41189 * DECL_FUNCTION_SPECIFIC_OPTIMIZATION: Function Basics. (line 6)
41190 * DECL_FUNCTION_SPECIFIC_TARGET: Function Basics. (line 6)
41191 * DECL_GLOBAL_CTOR_P: Function Basics. (line 6)
41192 * DECL_GLOBAL_DTOR_P: Function Basics. (line 6)
41193 * DECL_INITIAL: Declarations. (line 6)
41194 * DECL_LINKONCE_P: Function Basics. (line 6)
41195 * DECL_LOCAL_FUNCTION_P: Function Basics. (line 44)
41196 * DECL_MAIN_P: Function Basics. (line 7)
41197 * DECL_NAME <1>: Function Basics. (line 6)
41198 * DECL_NAME <2>: Working with declarations.
41200 * DECL_NAME: Namespaces. (line 15)
41201 * DECL_NAMESPACE_ALIAS: Namespaces. (line 30)
41202 * DECL_NAMESPACE_SCOPE_P: Working with declarations.
41204 * DECL_NAMESPACE_STD_P: Namespaces. (line 40)
41205 * DECL_NON_THUNK_FUNCTION_P: Function Basics. (line 144)
41206 * DECL_NONCONVERTING_P: Function Basics. (line 86)
41207 * DECL_NONSTATIC_MEMBER_FUNCTION_P: Function Basics. (line 74)
41208 * DECL_OVERLOADED_OPERATOR_P: Function Basics. (line 6)
41209 * DECL_RESULT: Function Basics. (line 168)
41210 * DECL_SIZE: Declarations. (line 6)
41211 * DECL_STATIC_FUNCTION_P: Function Basics. (line 71)
41212 * DECL_STMT: Function Bodies. (line 6)
41213 * DECL_STMT_DECL: Function Bodies. (line 6)
41214 * DECL_THUNK_P: Function Basics. (line 122)
41215 * DECL_VOLATILE_MEMFUNC_P: Function Basics. (line 80)
41216 * declaration: Declarations. (line 6)
41217 * declarations, RTL: RTL Declarations. (line 6)
41218 * DECLARE_LIBRARY_RENAMES: Library Calls. (line 9)
41219 * decrement_and_branch_until_zero instruction pattern: Standard Names.
41221 * def_optype_d: Manipulating GIMPLE statements.
41223 * default: GTY Options. (line 82)
41224 * default_file_start: File Framework. (line 9)
41225 * DEFAULT_GDB_EXTENSIONS: DBX Options. (line 18)
41226 * DEFAULT_PCC_STRUCT_RETURN: Aggregate Return. (line 34)
41227 * DEFAULT_SIGNED_CHAR: Type Layout. (line 154)
41228 * define_address_constraint: Define Constraints. (line 107)
41229 * define_asm_attributes: Tagging Insns. (line 73)
41230 * define_attr: Defining Attributes.
41232 * define_automaton: Processor pipeline description.
41234 * define_bypass: Processor pipeline description.
41236 * define_code_attr: Code Iterators. (line 6)
41237 * define_code_iterator: Code Iterators. (line 6)
41238 * define_cond_exec: Conditional Execution.
41240 * define_constants: Constant Definitions.
41242 * define_constraint: Define Constraints. (line 48)
41243 * define_cpu_unit: Processor pipeline description.
41245 * define_delay: Delay Slots. (line 25)
41246 * define_expand: Expander Definitions.
41248 * define_insn: Patterns. (line 6)
41249 * define_insn example: Example. (line 6)
41250 * define_insn_and_split: Insn Splitting. (line 170)
41251 * define_insn_reservation: Processor pipeline description.
41253 * define_memory_constraint: Define Constraints. (line 88)
41254 * define_mode_attr: Substitutions. (line 6)
41255 * define_mode_iterator: Defining Mode Iterators.
41257 * define_peephole: define_peephole. (line 6)
41258 * define_peephole2: define_peephole2. (line 6)
41259 * define_predicate: Defining Predicates.
41261 * define_query_cpu_unit: Processor pipeline description.
41263 * define_register_constraint: Define Constraints. (line 28)
41264 * define_reservation: Processor pipeline description.
41266 * define_special_predicate: Defining Predicates.
41268 * define_split: Insn Splitting. (line 32)
41269 * defining attributes and their values: Defining Attributes.
41271 * defining constraints: Define Constraints. (line 6)
41272 * defining constraints, obsolete method: Old Constraints. (line 6)
41273 * defining jump instruction patterns: Jump Patterns. (line 6)
41274 * defining looping instruction patterns: Looping Patterns. (line 6)
41275 * defining peephole optimizers: Peephole Definitions.
41277 * defining predicates: Defining Predicates.
41279 * defining RTL sequences for code generation: Expander Definitions.
41281 * delay slots, defining: Delay Slots. (line 6)
41282 * DELAY_SLOTS_FOR_EPILOGUE: Function Entry. (line 163)
41283 * deletable: GTY Options. (line 150)
41284 * DELETE_IF_ORDINARY: Filesystem. (line 79)
41285 * Dependent Patterns: Dependent Patterns. (line 6)
41286 * desc: GTY Options. (line 82)
41287 * destructor: Function Basics. (line 6)
41288 * destructors, output of: Initialization. (line 6)
41289 * deterministic finite state automaton: Processor pipeline description.
41291 * DF_SIZE: Type Layout. (line 130)
41292 * DFmode: Machine Modes. (line 73)
41293 * digits in constraint: Simple Constraints. (line 120)
41294 * DImode: Machine Modes. (line 45)
41295 * DIR_SEPARATOR: Filesystem. (line 18)
41296 * DIR_SEPARATOR_2: Filesystem. (line 19)
41297 * directory options .md: Including Patterns. (line 44)
41298 * disabling certain registers: Register Basics. (line 76)
41299 * dispatch table: Dispatch Tables. (line 8)
41300 * div: Arithmetic. (line 111)
41301 * div and attributes: Expressions. (line 64)
41302 * division: Arithmetic. (line 111)
41303 * divM3 instruction pattern: Standard Names. (line 222)
41304 * divmodM4 instruction pattern: Standard Names. (line 411)
41305 * DO_BODY: Function Bodies. (line 6)
41306 * DO_COND: Function Bodies. (line 6)
41307 * DO_STMT: Function Bodies. (line 6)
41308 * DOLLARS_IN_IDENTIFIERS: Misc. (line 496)
41309 * doloop_begin instruction pattern: Standard Names. (line 1151)
41310 * doloop_end instruction pattern: Standard Names. (line 1130)
41311 * DONE: Expander Definitions.
41313 * DONT_USE_BUILTIN_SETJMP: Exception Region Output.
41315 * DOUBLE_TYPE_SIZE: Type Layout. (line 53)
41316 * DQmode: Machine Modes. (line 115)
41317 * driver: Driver. (line 6)
41318 * DRIVER_SELF_SPECS: Driver. (line 71)
41319 * DUMPFILE_FORMAT: Filesystem. (line 67)
41320 * DWARF2_ASM_LINE_DEBUG_INFO: SDB and DWARF. (line 36)
41321 * DWARF2_DEBUGGING_INFO: SDB and DWARF. (line 13)
41322 * DWARF2_FRAME_INFO: SDB and DWARF. (line 30)
41323 * DWARF2_FRAME_REG_OUT: Frame Registers. (line 133)
41324 * DWARF2_UNWIND_INFO: Exception Region Output.
41326 * DWARF_ALT_FRAME_RETURN_COLUMN: Frame Layout. (line 152)
41327 * DWARF_CIE_DATA_ALIGNMENT: Exception Region Output.
41329 * DWARF_FRAME_REGISTERS: Frame Registers. (line 93)
41330 * DWARF_FRAME_REGNUM: Frame Registers. (line 125)
41331 * DWARF_REG_TO_UNWIND_COLUMN: Frame Registers. (line 117)
41332 * DWARF_ZERO_REG: Frame Layout. (line 163)
41333 * DYNAMIC_CHAIN_ADDRESS: Frame Layout. (line 92)
41334 * E in constraint: Simple Constraints. (line 79)
41335 * earlyclobber operand: Modifiers. (line 25)
41336 * edge: Edges. (line 6)
41337 * edge in the flow graph: Edges. (line 6)
41338 * edge iterators: Edges. (line 15)
41339 * edge splitting: Maintaining the CFG.
41341 * EDGE_ABNORMAL: Edges. (line 128)
41342 * EDGE_ABNORMAL, EDGE_ABNORMAL_CALL: Edges. (line 171)
41343 * EDGE_ABNORMAL, EDGE_EH: Edges. (line 96)
41344 * EDGE_ABNORMAL, EDGE_SIBCALL: Edges. (line 122)
41345 * EDGE_FALLTHRU, force_nonfallthru: Edges. (line 86)
41346 * EDOM, implicit usage: Library Calls. (line 58)
41347 * EH_FRAME_IN_DATA_SECTION: Exception Region Output.
41349 * EH_FRAME_SECTION_NAME: Exception Region Output.
41351 * eh_return instruction pattern: Standard Names. (line 1319)
41352 * EH_RETURN_DATA_REGNO: Exception Handling. (line 7)
41353 * EH_RETURN_HANDLER_RTX: Exception Handling. (line 39)
41354 * EH_RETURN_STACKADJ_RTX: Exception Handling. (line 22)
41355 * EH_TABLES_CAN_BE_READ_ONLY: Exception Region Output.
41357 * EH_USES: Function Entry. (line 158)
41358 * ei_edge: Edges. (line 43)
41359 * ei_end_p: Edges. (line 27)
41360 * ei_last: Edges. (line 23)
41361 * ei_next: Edges. (line 35)
41362 * ei_one_before_end_p: Edges. (line 31)
41363 * ei_prev: Edges. (line 39)
41364 * ei_safe_safe: Edges. (line 47)
41365 * ei_start: Edges. (line 19)
41366 * ELIGIBLE_FOR_EPILOGUE_DELAY: Function Entry. (line 169)
41367 * ELIMINABLE_REGS: Elimination. (line 44)
41368 * ELSE_CLAUSE: Function Bodies. (line 6)
41369 * Embedded C: Fixed-point fractional library routines.
41371 * EMIT_MODE_SET: Mode Switching. (line 74)
41372 * Empty Statements: Empty Statements. (line 6)
41373 * EMPTY_CLASS_EXPR: Function Bodies. (line 6)
41374 * EMPTY_FIELD_BOUNDARY: Storage Layout. (line 295)
41375 * Emulated TLS: Emulated TLS. (line 6)
41376 * ENABLE_EXECUTE_STACK: Trampolines. (line 110)
41377 * enabled: Disable Insn Alternatives.
41379 * ENDFILE_SPEC: Driver. (line 218)
41380 * endianness: Portability. (line 21)
41381 * ENTRY_BLOCK_PTR, EXIT_BLOCK_PTR: Basic Blocks. (line 28)
41382 * enum machine_mode: Machine Modes. (line 6)
41383 * enum reg_class: Register Classes. (line 65)
41384 * ENUMERAL_TYPE: Types. (line 6)
41385 * epilogue: Function Entry. (line 6)
41386 * epilogue instruction pattern: Standard Names. (line 1351)
41387 * EPILOGUE_USES: Function Entry. (line 152)
41388 * eq: Comparisons. (line 52)
41389 * eq and attributes: Expressions. (line 64)
41390 * eq_attr: Expressions. (line 85)
41391 * EQ_EXPR: Expression trees. (line 6)
41392 * equal: Comparisons. (line 52)
41393 * errno, implicit usage: Library Calls. (line 70)
41394 * EXACT_DIV_EXPR: Expression trees. (line 6)
41395 * examining SSA_NAMEs: SSA. (line 218)
41396 * exception handling <1>: Exception Handling. (line 6)
41397 * exception handling: Edges. (line 96)
41398 * exception_receiver instruction pattern: Standard Names. (line 1283)
41399 * exclamation point: Multi-Alternative. (line 47)
41400 * exclusion_set: Processor pipeline description.
41402 * exclusive-or, bitwise: Arithmetic. (line 163)
41403 * EXIT_EXPR: Expression trees. (line 6)
41404 * EXIT_IGNORE_STACK: Function Entry. (line 140)
41405 * expander definitions: Expander Definitions.
41407 * expM2 instruction pattern: Standard Names. (line 497)
41408 * expr_list: Insns. (line 505)
41409 * EXPR_STMT: Function Bodies. (line 6)
41410 * EXPR_STMT_EXPR: Function Bodies. (line 6)
41411 * expression: Expression trees. (line 6)
41412 * expression codes: RTL Objects. (line 47)
41413 * extendMN2 instruction pattern: Standard Names. (line 826)
41414 * extensible constraints: Simple Constraints. (line 163)
41415 * EXTRA_ADDRESS_CONSTRAINT: Old Constraints. (line 123)
41416 * EXTRA_CONSTRAINT: Old Constraints. (line 74)
41417 * EXTRA_CONSTRAINT_STR: Old Constraints. (line 95)
41418 * EXTRA_MEMORY_CONSTRAINT: Old Constraints. (line 100)
41419 * EXTRA_SPECS: Driver. (line 245)
41420 * extv instruction pattern: Standard Names. (line 862)
41421 * extzv instruction pattern: Standard Names. (line 877)
41422 * F in constraint: Simple Constraints. (line 84)
41423 * FAIL: Expander Definitions.
41425 * fall-thru: Edges. (line 69)
41426 * FATAL_EXIT_CODE: Host Misc. (line 6)
41427 * FDL, GNU Free Documentation License: GNU Free Documentation License.
41429 * features, optional, in system conventions: Run-time Target.
41431 * ffs: Arithmetic. (line 202)
41432 * ffsM2 instruction pattern: Standard Names. (line 611)
41433 * FIELD_DECL: Declarations. (line 6)
41434 * file_end_indicate_exec_stack: File Framework. (line 41)
41435 * files and passes of the compiler: Passes. (line 6)
41436 * files, generated: Files. (line 6)
41437 * final_absence_set: Processor pipeline description.
41439 * FINAL_PRESCAN_INSN: Instruction Output. (line 46)
41440 * final_presence_set: Processor pipeline description.
41442 * final_scan_insn: Function Entry. (line 181)
41443 * final_sequence: Instruction Output. (line 117)
41444 * FIND_BASE_TERM: Addressing Modes. (line 110)
41445 * FINI_ARRAY_SECTION_ASM_OP: Sections. (line 105)
41446 * FINI_SECTION_ASM_OP: Sections. (line 90)
41447 * finite state automaton minimization: Processor pipeline description.
41449 * FIRST_PARM_OFFSET: Frame Layout. (line 67)
41450 * FIRST_PARM_OFFSET and virtual registers: Regs and Memory. (line 65)
41451 * FIRST_PSEUDO_REGISTER: Register Basics. (line 9)
41452 * FIRST_STACK_REG: Stack Registers. (line 23)
41453 * FIRST_VIRTUAL_REGISTER: Regs and Memory. (line 51)
41454 * fix: Conversions. (line 66)
41455 * FIX_TRUNC_EXPR: Expression trees. (line 6)
41456 * fix_truncMN2 instruction pattern: Standard Names. (line 813)
41457 * fixed register: Register Basics. (line 15)
41458 * fixed-point fractional library: Fixed-point fractional library routines.
41460 * FIXED_CONVERT_EXPR: Expression trees. (line 6)
41461 * FIXED_CST: Expression trees. (line 6)
41462 * FIXED_POINT_TYPE: Types. (line 6)
41463 * FIXED_REGISTERS: Register Basics. (line 15)
41464 * fixed_regs: Register Basics. (line 59)
41465 * fixMN2 instruction pattern: Standard Names. (line 793)
41466 * FIXUNS_TRUNC_LIKE_FIX_TRUNC: Misc. (line 100)
41467 * fixuns_truncMN2 instruction pattern: Standard Names. (line 817)
41468 * fixunsMN2 instruction pattern: Standard Names. (line 802)
41469 * flags in RTL expression: Flags. (line 6)
41470 * float: Conversions. (line 58)
41471 * FLOAT_EXPR: Expression trees. (line 6)
41472 * float_extend: Conversions. (line 33)
41473 * FLOAT_LIB_COMPARE_RETURNS_BOOL: Library Calls. (line 25)
41474 * FLOAT_STORE_FLAG_VALUE: Misc. (line 301)
41475 * float_truncate: Conversions. (line 53)
41476 * FLOAT_TYPE_SIZE: Type Layout. (line 49)
41477 * FLOAT_WORDS_BIG_ENDIAN: Storage Layout. (line 43)
41478 * FLOAT_WORDS_BIG_ENDIAN, (lack of) effect on subreg: Regs and Memory.
41480 * floating point and cross compilation: Floating Point. (line 6)
41481 * Floating Point Emulation: Target Fragment. (line 15)
41482 * floating point emulation library, US Software GOFAST: Library Calls.
41484 * floatMN2 instruction pattern: Standard Names. (line 785)
41485 * floatunsMN2 instruction pattern: Standard Names. (line 789)
41486 * FLOOR_DIV_EXPR: Expression trees. (line 6)
41487 * FLOOR_MOD_EXPR: Expression trees. (line 6)
41488 * floorM2 instruction pattern: Standard Names. (line 532)
41489 * flow-insensitive alias analysis: Alias analysis. (line 6)
41490 * flow-sensitive alias analysis: Alias analysis. (line 6)
41491 * fmodM3 instruction pattern: Standard Names. (line 463)
41492 * FOR_BODY: Function Bodies. (line 6)
41493 * FOR_COND: Function Bodies. (line 6)
41494 * FOR_EXPR: Function Bodies. (line 6)
41495 * FOR_INIT_STMT: Function Bodies. (line 6)
41496 * FOR_STMT: Function Bodies. (line 6)
41497 * FORCE_CODE_SECTION_ALIGN: Sections. (line 136)
41498 * force_reg: Standard Names. (line 36)
41499 * fract_convert: Conversions. (line 82)
41500 * FRACT_TYPE_SIZE: Type Layout. (line 68)
41501 * fractional types: Fixed-point fractional library routines.
41503 * fractMN2 instruction pattern: Standard Names. (line 835)
41504 * fractunsMN2 instruction pattern: Standard Names. (line 850)
41505 * frame layout: Frame Layout. (line 6)
41506 * FRAME_ADDR_RTX: Frame Layout. (line 116)
41507 * FRAME_GROWS_DOWNWARD: Frame Layout. (line 31)
41508 * FRAME_GROWS_DOWNWARD and virtual registers: Regs and Memory.
41510 * FRAME_POINTER_CFA_OFFSET: Frame Layout. (line 212)
41511 * frame_pointer_needed: Function Entry. (line 34)
41512 * FRAME_POINTER_REGNUM: Frame Registers. (line 14)
41513 * FRAME_POINTER_REGNUM and virtual registers: Regs and Memory.
41515 * FRAME_POINTER_REQUIRED: Elimination. (line 9)
41516 * frame_pointer_rtx: Frame Registers. (line 85)
41517 * frame_related: Flags. (line 242)
41518 * frame_related, in insn, call_insn, jump_insn, barrier, and set: Flags.
41520 * frame_related, in mem: Flags. (line 103)
41521 * frame_related, in reg: Flags. (line 112)
41522 * frame_related, in symbol_ref: Flags. (line 183)
41523 * frequency, count, BB_FREQ_BASE: Profile information.
41525 * ftruncM2 instruction pattern: Standard Names. (line 808)
41526 * function: Functions. (line 6)
41527 * function body: Function Bodies. (line 6)
41528 * function call conventions: Interface. (line 6)
41529 * function entry and exit: Function Entry. (line 6)
41530 * function entry point, alternate function entry point: Edges.
41532 * function-call insns: Calls. (line 6)
41533 * FUNCTION_ARG: Register Arguments. (line 11)
41534 * FUNCTION_ARG_ADVANCE: Register Arguments. (line 186)
41535 * FUNCTION_ARG_BOUNDARY: Register Arguments. (line 239)
41536 * FUNCTION_ARG_OFFSET: Register Arguments. (line 197)
41537 * FUNCTION_ARG_PADDING: Register Arguments. (line 204)
41538 * FUNCTION_ARG_REGNO_P: Register Arguments. (line 244)
41539 * FUNCTION_BOUNDARY: Storage Layout. (line 170)
41540 * FUNCTION_DECL: Functions. (line 6)
41541 * FUNCTION_INCOMING_ARG: Register Arguments. (line 68)
41542 * FUNCTION_MODE: Misc. (line 356)
41543 * FUNCTION_OUTGOING_VALUE: Scalar Return. (line 56)
41544 * FUNCTION_PROFILER: Profiling. (line 9)
41545 * FUNCTION_TYPE: Types. (line 6)
41546 * FUNCTION_VALUE: Scalar Return. (line 52)
41547 * FUNCTION_VALUE_REGNO_P: Scalar Return. (line 69)
41548 * functions, leaf: Leaf Functions. (line 6)
41549 * fundamental type: Types. (line 6)
41550 * g in constraint: Simple Constraints. (line 110)
41551 * G in constraint: Simple Constraints. (line 88)
41552 * garbage collector, invocation: Invoking the garbage collector.
41554 * GCC and portability: Portability. (line 6)
41555 * GCC_DRIVER_HOST_INITIALIZATION: Host Misc. (line 36)
41556 * gcov_type: Profile information.
41558 * ge: Comparisons. (line 72)
41559 * ge and attributes: Expressions. (line 64)
41560 * GE_EXPR: Expression trees. (line 6)
41561 * GEN_ERRNO_RTX: Library Calls. (line 71)
41562 * gencodes: RTL passes. (line 18)
41563 * general_operand: Machine-Independent Predicates.
41565 * GENERAL_REGS: Register Classes. (line 23)
41566 * generated files: Files. (line 6)
41567 * generating assembler output: Output Statement. (line 6)
41568 * generating insns: RTL Template. (line 6)
41569 * GENERIC <1>: GENERIC. (line 6)
41570 * GENERIC <2>: Gimplification pass.
41572 * GENERIC: Parsing pass. (line 6)
41573 * generic predicates: Machine-Independent Predicates.
41575 * genflags: RTL passes. (line 18)
41576 * get_attr: Expressions. (line 80)
41577 * get_attr_length: Insn Lengths. (line 46)
41578 * GET_CLASS_NARROWEST_MODE: Machine Modes. (line 333)
41579 * GET_CODE: RTL Objects. (line 47)
41580 * get_frame_size: Elimination. (line 31)
41581 * get_insns: Insns. (line 34)
41582 * get_last_insn: Insns. (line 34)
41583 * GET_MODE: Machine Modes. (line 280)
41584 * GET_MODE_ALIGNMENT: Machine Modes. (line 320)
41585 * GET_MODE_BITSIZE: Machine Modes. (line 304)
41586 * GET_MODE_CLASS: Machine Modes. (line 294)
41587 * GET_MODE_FBIT: Machine Modes. (line 311)
41588 * GET_MODE_IBIT: Machine Modes. (line 307)
41589 * GET_MODE_MASK: Machine Modes. (line 315)
41590 * GET_MODE_NAME: Machine Modes. (line 291)
41591 * GET_MODE_NUNITS: Machine Modes. (line 329)
41592 * GET_MODE_SIZE: Machine Modes. (line 301)
41593 * GET_MODE_UNIT_SIZE: Machine Modes. (line 323)
41594 * GET_MODE_WIDER_MODE: Machine Modes. (line 297)
41595 * GET_RTX_CLASS: RTL Classes. (line 6)
41596 * GET_RTX_FORMAT: RTL Classes. (line 130)
41597 * GET_RTX_LENGTH: RTL Classes. (line 127)
41598 * geu: Comparisons. (line 72)
41599 * geu and attributes: Expressions. (line 64)
41600 * GGC: Type Information. (line 6)
41601 * ggc_collect: Invoking the garbage collector.
41603 * GIMPLE <1>: GIMPLE. (line 6)
41604 * GIMPLE <2>: Gimplification pass.
41606 * GIMPLE: Parsing pass. (line 14)
41607 * GIMPLE Exception Handling: GIMPLE Exception Handling.
41609 * GIMPLE instruction set: GIMPLE instruction set.
41611 * GIMPLE sequences: GIMPLE sequences. (line 6)
41612 * gimple_addresses_taken: Manipulating GIMPLE statements.
41614 * GIMPLE_ASM: GIMPLE_ASM. (line 6)
41615 * gimple_asm_clear_volatile: GIMPLE_ASM. (line 63)
41616 * gimple_asm_clobber_op: GIMPLE_ASM. (line 46)
41617 * gimple_asm_input_op: GIMPLE_ASM. (line 30)
41618 * gimple_asm_output_op: GIMPLE_ASM. (line 38)
41619 * gimple_asm_set_clobber_op: GIMPLE_ASM. (line 50)
41620 * gimple_asm_set_input_op: GIMPLE_ASM. (line 34)
41621 * gimple_asm_set_output_op: GIMPLE_ASM. (line 42)
41622 * gimple_asm_set_volatile: GIMPLE_ASM. (line 60)
41623 * gimple_asm_volatile_p: GIMPLE_ASM. (line 57)
41624 * GIMPLE_ASSIGN: GIMPLE_ASSIGN. (line 6)
41625 * gimple_assign_cast_p: GIMPLE_ASSIGN. (line 89)
41626 * gimple_assign_lhs: GIMPLE_ASSIGN. (line 51)
41627 * gimple_assign_rhs1: GIMPLE_ASSIGN. (line 57)
41628 * gimple_assign_rhs2: GIMPLE_ASSIGN. (line 64)
41629 * gimple_assign_set_lhs: GIMPLE_ASSIGN. (line 71)
41630 * gimple_assign_set_rhs1: GIMPLE_ASSIGN. (line 74)
41631 * gimple_assign_set_rhs2: GIMPLE_ASSIGN. (line 85)
41632 * gimple_bb: Manipulating GIMPLE statements.
41634 * GIMPLE_BIND: GIMPLE_BIND. (line 6)
41635 * gimple_bind_add_seq: GIMPLE_BIND. (line 36)
41636 * gimple_bind_add_stmt: GIMPLE_BIND. (line 32)
41637 * gimple_bind_append_vars: GIMPLE_BIND. (line 19)
41638 * gimple_bind_block: GIMPLE_BIND. (line 40)
41639 * gimple_bind_body: GIMPLE_BIND. (line 23)
41640 * gimple_bind_set_block: GIMPLE_BIND. (line 45)
41641 * gimple_bind_set_body: GIMPLE_BIND. (line 28)
41642 * gimple_bind_set_vars: GIMPLE_BIND. (line 15)
41643 * gimple_bind_vars: GIMPLE_BIND. (line 12)
41644 * gimple_block: Manipulating GIMPLE statements.
41646 * gimple_build_asm: GIMPLE_ASM. (line 8)
41647 * gimple_build_asm_vec: GIMPLE_ASM. (line 17)
41648 * gimple_build_assign: GIMPLE_ASSIGN. (line 7)
41649 * gimple_build_assign_with_ops: GIMPLE_ASSIGN. (line 30)
41650 * gimple_build_bind: GIMPLE_BIND. (line 8)
41651 * gimple_build_call: GIMPLE_CALL. (line 8)
41652 * gimple_build_call_from_tree: GIMPLE_CALL. (line 16)
41653 * gimple_build_call_vec: GIMPLE_CALL. (line 25)
41654 * gimple_build_catch: GIMPLE_CATCH. (line 8)
41655 * gimple_build_cdt: GIMPLE_CHANGE_DYNAMIC_TYPE.
41657 * gimple_build_cond: GIMPLE_COND. (line 8)
41658 * gimple_build_cond_from_tree: GIMPLE_COND. (line 16)
41659 * gimple_build_eh_filter: GIMPLE_EH_FILTER. (line 8)
41660 * gimple_build_goto: GIMPLE_LABEL. (line 18)
41661 * gimple_build_label: GIMPLE_LABEL. (line 7)
41662 * gimple_build_nop: GIMPLE_NOP. (line 7)
41663 * gimple_build_omp_atomic_load: GIMPLE_OMP_ATOMIC_LOAD.
41665 * gimple_build_omp_atomic_store: GIMPLE_OMP_ATOMIC_STORE.
41667 * gimple_build_omp_continue: GIMPLE_OMP_CONTINUE.
41669 * gimple_build_omp_critical: GIMPLE_OMP_CRITICAL.
41671 * gimple_build_omp_for: GIMPLE_OMP_FOR. (line 9)
41672 * gimple_build_omp_master: GIMPLE_OMP_MASTER. (line 7)
41673 * gimple_build_omp_ordered: GIMPLE_OMP_ORDERED. (line 7)
41674 * gimple_build_omp_parallel: GIMPLE_OMP_PARALLEL.
41676 * gimple_build_omp_return: GIMPLE_OMP_RETURN. (line 7)
41677 * gimple_build_omp_section: GIMPLE_OMP_SECTION. (line 7)
41678 * gimple_build_omp_sections: GIMPLE_OMP_SECTIONS.
41680 * gimple_build_omp_sections_switch: GIMPLE_OMP_SECTIONS.
41682 * gimple_build_omp_single: GIMPLE_OMP_SINGLE. (line 8)
41683 * gimple_build_resx: GIMPLE_RESX. (line 7)
41684 * gimple_build_return: GIMPLE_RETURN. (line 7)
41685 * gimple_build_switch: GIMPLE_SWITCH. (line 8)
41686 * gimple_build_switch_vec: GIMPLE_SWITCH. (line 16)
41687 * gimple_build_try: GIMPLE_TRY. (line 8)
41688 * gimple_build_wce: GIMPLE_WITH_CLEANUP_EXPR.
41690 * GIMPLE_CALL: GIMPLE_CALL. (line 6)
41691 * gimple_call_arg: GIMPLE_CALL. (line 66)
41692 * gimple_call_cannot_inline_p: GIMPLE_CALL. (line 91)
41693 * gimple_call_chain: GIMPLE_CALL. (line 57)
41694 * gimple_call_copy_skip_args: GIMPLE_CALL. (line 98)
41695 * gimple_call_fn: GIMPLE_CALL. (line 38)
41696 * gimple_call_fndecl: GIMPLE_CALL. (line 46)
41697 * gimple_call_lhs: GIMPLE_CALL. (line 29)
41698 * gimple_call_mark_uninlinable: GIMPLE_CALL. (line 88)
41699 * gimple_call_noreturn_p: GIMPLE_CALL. (line 94)
41700 * gimple_call_return_type: GIMPLE_CALL. (line 54)
41701 * gimple_call_set_arg: GIMPLE_CALL. (line 76)
41702 * gimple_call_set_chain: GIMPLE_CALL. (line 60)
41703 * gimple_call_set_fn: GIMPLE_CALL. (line 42)
41704 * gimple_call_set_fndecl: GIMPLE_CALL. (line 51)
41705 * gimple_call_set_lhs: GIMPLE_CALL. (line 35)
41706 * gimple_call_set_tail: GIMPLE_CALL. (line 80)
41707 * gimple_call_tail_p: GIMPLE_CALL. (line 85)
41708 * GIMPLE_CATCH: GIMPLE_CATCH. (line 6)
41709 * gimple_catch_handler: GIMPLE_CATCH. (line 20)
41710 * gimple_catch_set_handler: GIMPLE_CATCH. (line 28)
41711 * gimple_catch_set_types: GIMPLE_CATCH. (line 24)
41712 * gimple_catch_types: GIMPLE_CATCH. (line 13)
41713 * gimple_cdt_location: GIMPLE_CHANGE_DYNAMIC_TYPE.
41715 * gimple_cdt_new_type: GIMPLE_CHANGE_DYNAMIC_TYPE.
41717 * gimple_cdt_set_location: GIMPLE_CHANGE_DYNAMIC_TYPE.
41719 * gimple_cdt_set_new_type: GIMPLE_CHANGE_DYNAMIC_TYPE.
41721 * GIMPLE_CHANGE_DYNAMIC_TYPE: GIMPLE_CHANGE_DYNAMIC_TYPE.
41723 * gimple_code: Manipulating GIMPLE statements.
41725 * GIMPLE_COND: GIMPLE_COND. (line 6)
41726 * gimple_cond_false_label: GIMPLE_COND. (line 60)
41727 * gimple_cond_lhs: GIMPLE_COND. (line 30)
41728 * gimple_cond_make_false: GIMPLE_COND. (line 64)
41729 * gimple_cond_make_true: GIMPLE_COND. (line 67)
41730 * gimple_cond_rhs: GIMPLE_COND. (line 38)
41731 * gimple_cond_set_code: GIMPLE_COND. (line 26)
41732 * gimple_cond_set_false_label: GIMPLE_COND. (line 56)
41733 * gimple_cond_set_lhs: GIMPLE_COND. (line 34)
41734 * gimple_cond_set_rhs: GIMPLE_COND. (line 42)
41735 * gimple_cond_set_true_label: GIMPLE_COND. (line 51)
41736 * gimple_cond_true_label: GIMPLE_COND. (line 46)
41737 * gimple_copy: Manipulating GIMPLE statements.
41739 * GIMPLE_EH_FILTER: GIMPLE_EH_FILTER. (line 6)
41740 * gimple_eh_filter_failure: GIMPLE_EH_FILTER. (line 19)
41741 * gimple_eh_filter_must_not_throw: GIMPLE_EH_FILTER. (line 33)
41742 * gimple_eh_filter_set_failure: GIMPLE_EH_FILTER. (line 29)
41743 * gimple_eh_filter_set_must_not_throw: GIMPLE_EH_FILTER. (line 37)
41744 * gimple_eh_filter_set_types: GIMPLE_EH_FILTER. (line 24)
41745 * gimple_eh_filter_types: GIMPLE_EH_FILTER. (line 12)
41746 * gimple_expr_type: Manipulating GIMPLE statements.
41748 * gimple_goto_dest: GIMPLE_LABEL. (line 21)
41749 * gimple_goto_set_dest: GIMPLE_LABEL. (line 24)
41750 * gimple_has_mem_ops: Manipulating GIMPLE statements.
41752 * gimple_has_ops: Manipulating GIMPLE statements.
41754 * gimple_has_volatile_ops: Manipulating GIMPLE statements.
41756 * GIMPLE_LABEL: GIMPLE_LABEL. (line 6)
41757 * gimple_label_label: GIMPLE_LABEL. (line 11)
41758 * gimple_label_set_label: GIMPLE_LABEL. (line 14)
41759 * gimple_loaded_syms: Manipulating GIMPLE statements.
41761 * gimple_locus: Manipulating GIMPLE statements.
41763 * gimple_locus_empty_p: Manipulating GIMPLE statements.
41765 * gimple_modified_p: Manipulating GIMPLE statements.
41767 * gimple_no_warning_p: Manipulating GIMPLE statements.
41769 * GIMPLE_NOP: GIMPLE_NOP. (line 6)
41770 * gimple_nop_p: GIMPLE_NOP. (line 10)
41771 * gimple_num_ops <1>: Manipulating GIMPLE statements.
41773 * gimple_num_ops: Logical Operators. (line 76)
41774 * GIMPLE_OMP_ATOMIC_LOAD: GIMPLE_OMP_ATOMIC_LOAD.
41776 * gimple_omp_atomic_load_lhs: GIMPLE_OMP_ATOMIC_LOAD.
41778 * gimple_omp_atomic_load_rhs: GIMPLE_OMP_ATOMIC_LOAD.
41780 * gimple_omp_atomic_load_set_lhs: GIMPLE_OMP_ATOMIC_LOAD.
41782 * gimple_omp_atomic_load_set_rhs: GIMPLE_OMP_ATOMIC_LOAD.
41784 * GIMPLE_OMP_ATOMIC_STORE: GIMPLE_OMP_ATOMIC_STORE.
41786 * gimple_omp_atomic_store_set_val: GIMPLE_OMP_ATOMIC_STORE.
41788 * gimple_omp_atomic_store_val: GIMPLE_OMP_ATOMIC_STORE.
41790 * gimple_omp_body: GIMPLE_OMP_PARALLEL.
41792 * GIMPLE_OMP_CONTINUE: GIMPLE_OMP_CONTINUE.
41794 * gimple_omp_continue_control_def: GIMPLE_OMP_CONTINUE.
41796 * gimple_omp_continue_control_def_ptr: GIMPLE_OMP_CONTINUE.
41798 * gimple_omp_continue_control_use: GIMPLE_OMP_CONTINUE.
41800 * gimple_omp_continue_control_use_ptr: GIMPLE_OMP_CONTINUE.
41802 * gimple_omp_continue_set_control_def: GIMPLE_OMP_CONTINUE.
41804 * gimple_omp_continue_set_control_use: GIMPLE_OMP_CONTINUE.
41806 * GIMPLE_OMP_CRITICAL: GIMPLE_OMP_CRITICAL.
41808 * gimple_omp_critical_name: GIMPLE_OMP_CRITICAL.
41810 * gimple_omp_critical_set_name: GIMPLE_OMP_CRITICAL.
41812 * GIMPLE_OMP_FOR: GIMPLE_OMP_FOR. (line 6)
41813 * gimple_omp_for_clauses: GIMPLE_OMP_FOR. (line 20)
41814 * gimple_omp_for_final: GIMPLE_OMP_FOR. (line 51)
41815 * gimple_omp_for_incr: GIMPLE_OMP_FOR. (line 61)
41816 * gimple_omp_for_index: GIMPLE_OMP_FOR. (line 31)
41817 * gimple_omp_for_initial: GIMPLE_OMP_FOR. (line 41)
41818 * gimple_omp_for_pre_body: GIMPLE_OMP_FOR. (line 70)
41819 * gimple_omp_for_set_clauses: GIMPLE_OMP_FOR. (line 27)
41820 * gimple_omp_for_set_cond: GIMPLE_OMP_FOR. (line 80)
41821 * gimple_omp_for_set_final: GIMPLE_OMP_FOR. (line 58)
41822 * gimple_omp_for_set_incr: GIMPLE_OMP_FOR. (line 67)
41823 * gimple_omp_for_set_index: GIMPLE_OMP_FOR. (line 38)
41824 * gimple_omp_for_set_initial: GIMPLE_OMP_FOR. (line 48)
41825 * gimple_omp_for_set_pre_body: GIMPLE_OMP_FOR. (line 75)
41826 * GIMPLE_OMP_MASTER: GIMPLE_OMP_MASTER. (line 6)
41827 * GIMPLE_OMP_ORDERED: GIMPLE_OMP_ORDERED. (line 6)
41828 * GIMPLE_OMP_PARALLEL: GIMPLE_OMP_PARALLEL.
41830 * gimple_omp_parallel_child_fn: GIMPLE_OMP_PARALLEL.
41832 * gimple_omp_parallel_clauses: GIMPLE_OMP_PARALLEL.
41834 * gimple_omp_parallel_combined_p: GIMPLE_OMP_PARALLEL.
41836 * gimple_omp_parallel_data_arg: GIMPLE_OMP_PARALLEL.
41838 * gimple_omp_parallel_set_child_fn: GIMPLE_OMP_PARALLEL.
41840 * gimple_omp_parallel_set_clauses: GIMPLE_OMP_PARALLEL.
41842 * gimple_omp_parallel_set_combined_p: GIMPLE_OMP_PARALLEL.
41844 * gimple_omp_parallel_set_data_arg: GIMPLE_OMP_PARALLEL.
41846 * GIMPLE_OMP_RETURN: GIMPLE_OMP_RETURN. (line 6)
41847 * gimple_omp_return_nowait_p: GIMPLE_OMP_RETURN. (line 14)
41848 * gimple_omp_return_set_nowait: GIMPLE_OMP_RETURN. (line 11)
41849 * GIMPLE_OMP_SECTION: GIMPLE_OMP_SECTION. (line 6)
41850 * gimple_omp_section_last_p: GIMPLE_OMP_SECTION. (line 12)
41851 * gimple_omp_section_set_last: GIMPLE_OMP_SECTION. (line 16)
41852 * GIMPLE_OMP_SECTIONS: GIMPLE_OMP_SECTIONS.
41854 * gimple_omp_sections_clauses: GIMPLE_OMP_SECTIONS.
41856 * gimple_omp_sections_control: GIMPLE_OMP_SECTIONS.
41858 * gimple_omp_sections_set_clauses: GIMPLE_OMP_SECTIONS.
41860 * gimple_omp_sections_set_control: GIMPLE_OMP_SECTIONS.
41862 * gimple_omp_set_body: GIMPLE_OMP_PARALLEL.
41864 * GIMPLE_OMP_SINGLE: GIMPLE_OMP_SINGLE. (line 6)
41865 * gimple_omp_single_clauses: GIMPLE_OMP_SINGLE. (line 14)
41866 * gimple_omp_single_set_clauses: GIMPLE_OMP_SINGLE. (line 21)
41867 * gimple_op <1>: Manipulating GIMPLE statements.
41869 * gimple_op: Logical Operators. (line 79)
41870 * GIMPLE_PHI: GIMPLE_PHI. (line 6)
41871 * gimple_phi_capacity: GIMPLE_PHI. (line 10)
41872 * gimple_phi_num_args: GIMPLE_PHI. (line 14)
41873 * gimple_phi_result: GIMPLE_PHI. (line 19)
41874 * gimple_phi_set_arg: GIMPLE_PHI. (line 33)
41875 * gimple_phi_set_result: GIMPLE_PHI. (line 25)
41876 * GIMPLE_RESX: GIMPLE_RESX. (line 6)
41877 * gimple_resx_region: GIMPLE_RESX. (line 13)
41878 * gimple_resx_set_region: GIMPLE_RESX. (line 16)
41879 * GIMPLE_RETURN: GIMPLE_RETURN. (line 6)
41880 * gimple_return_retval: GIMPLE_RETURN. (line 10)
41881 * gimple_return_set_retval: GIMPLE_RETURN. (line 14)
41882 * gimple_rhs_class: GIMPLE_ASSIGN. (line 46)
41883 * gimple_seq_add_seq: GIMPLE sequences. (line 32)
41884 * gimple_seq_add_stmt: GIMPLE sequences. (line 26)
41885 * gimple_seq_alloc: GIMPLE sequences. (line 62)
41886 * gimple_seq_copy: GIMPLE sequences. (line 67)
41887 * gimple_seq_deep_copy: GIMPLE sequences. (line 37)
41888 * gimple_seq_empty_p: GIMPLE sequences. (line 70)
41889 * gimple_seq_first: GIMPLE sequences. (line 44)
41890 * gimple_seq_init: GIMPLE sequences. (line 59)
41891 * gimple_seq_last: GIMPLE sequences. (line 47)
41892 * gimple_seq_reverse: GIMPLE sequences. (line 40)
41893 * gimple_seq_set_first: GIMPLE sequences. (line 55)
41894 * gimple_seq_set_last: GIMPLE sequences. (line 51)
41895 * gimple_seq_singleton_p: GIMPLE sequences. (line 79)
41896 * gimple_set_block: Manipulating GIMPLE statements.
41898 * gimple_set_def_ops: Manipulating GIMPLE statements.
41900 * gimple_set_has_volatile_ops: Manipulating GIMPLE statements.
41902 * gimple_set_locus: Manipulating GIMPLE statements.
41904 * gimple_set_op: Manipulating GIMPLE statements.
41906 * gimple_set_plf: Manipulating GIMPLE statements.
41908 * gimple_set_use_ops: Manipulating GIMPLE statements.
41910 * gimple_set_vdef_ops: Manipulating GIMPLE statements.
41912 * gimple_set_visited: Manipulating GIMPLE statements.
41914 * gimple_set_vuse_ops: Manipulating GIMPLE statements.
41916 * gimple_statement_base: Tuple representation.
41918 * gimple_statement_with_ops: Tuple representation.
41920 * gimple_stored_syms: Manipulating GIMPLE statements.
41922 * GIMPLE_SWITCH: GIMPLE_SWITCH. (line 6)
41923 * gimple_switch_default_label: GIMPLE_SWITCH. (line 46)
41924 * gimple_switch_index: GIMPLE_SWITCH. (line 31)
41925 * gimple_switch_label: GIMPLE_SWITCH. (line 37)
41926 * gimple_switch_num_labels: GIMPLE_SWITCH. (line 22)
41927 * gimple_switch_set_default_label: GIMPLE_SWITCH. (line 50)
41928 * gimple_switch_set_index: GIMPLE_SWITCH. (line 34)
41929 * gimple_switch_set_label: GIMPLE_SWITCH. (line 42)
41930 * gimple_switch_set_num_labels: GIMPLE_SWITCH. (line 27)
41931 * GIMPLE_TRY: GIMPLE_TRY. (line 6)
41932 * gimple_try_catch_is_cleanup: GIMPLE_TRY. (line 20)
41933 * gimple_try_cleanup: GIMPLE_TRY. (line 27)
41934 * gimple_try_eval: GIMPLE_TRY. (line 23)
41935 * gimple_try_flags: GIMPLE_TRY. (line 16)
41936 * gimple_try_set_catch_is_cleanup: GIMPLE_TRY. (line 32)
41937 * gimple_try_set_cleanup: GIMPLE_TRY. (line 41)
41938 * gimple_try_set_eval: GIMPLE_TRY. (line 36)
41939 * gimple_visited_p: Manipulating GIMPLE statements.
41941 * gimple_wce_cleanup: GIMPLE_WITH_CLEANUP_EXPR.
41943 * gimple_wce_cleanup_eh_only: GIMPLE_WITH_CLEANUP_EXPR.
41945 * gimple_wce_set_cleanup: GIMPLE_WITH_CLEANUP_EXPR.
41947 * gimple_wce_set_cleanup_eh_only: GIMPLE_WITH_CLEANUP_EXPR.
41949 * GIMPLE_WITH_CLEANUP_EXPR: GIMPLE_WITH_CLEANUP_EXPR.
41951 * gimplification <1>: Gimplification pass.
41953 * gimplification: Parsing pass. (line 14)
41954 * gimplifier: Parsing pass. (line 14)
41955 * gimplify_assign: GIMPLE_ASSIGN. (line 19)
41956 * gimplify_expr: Gimplification pass.
41958 * gimplify_function_tree: Gimplification pass.
41960 * GLOBAL_INIT_PRIORITY: Function Basics. (line 6)
41961 * global_regs: Register Basics. (line 59)
41962 * GO_IF_LEGITIMATE_ADDRESS: Addressing Modes. (line 48)
41963 * GO_IF_MODE_DEPENDENT_ADDRESS: Addressing Modes. (line 190)
41964 * GOFAST, floating point emulation library: Library Calls. (line 44)
41965 * gofast_maybe_init_libfuncs: Library Calls. (line 44)
41966 * greater than: Comparisons. (line 60)
41967 * gsi_after_labels: Sequence iterators. (line 76)
41968 * gsi_bb: Sequence iterators. (line 83)
41969 * gsi_commit_edge_inserts: Sequence iterators. (line 194)
41970 * gsi_commit_one_edge_insert: Sequence iterators. (line 190)
41971 * gsi_end_p: Sequence iterators. (line 60)
41972 * gsi_for_stmt: Sequence iterators. (line 157)
41973 * gsi_insert_after: Sequence iterators. (line 147)
41974 * gsi_insert_before: Sequence iterators. (line 136)
41975 * gsi_insert_on_edge: Sequence iterators. (line 174)
41976 * gsi_insert_on_edge_immediate: Sequence iterators. (line 185)
41977 * gsi_insert_seq_after: Sequence iterators. (line 154)
41978 * gsi_insert_seq_before: Sequence iterators. (line 143)
41979 * gsi_insert_seq_on_edge: Sequence iterators. (line 179)
41980 * gsi_last: Sequence iterators. (line 50)
41981 * gsi_last_bb: Sequence iterators. (line 56)
41982 * gsi_link_after: Sequence iterators. (line 115)
41983 * gsi_link_before: Sequence iterators. (line 105)
41984 * gsi_link_seq_after: Sequence iterators. (line 110)
41985 * gsi_link_seq_before: Sequence iterators. (line 99)
41986 * gsi_move_after: Sequence iterators. (line 161)
41987 * gsi_move_before: Sequence iterators. (line 166)
41988 * gsi_move_to_bb_end: Sequence iterators. (line 171)
41989 * gsi_next: Sequence iterators. (line 66)
41990 * gsi_one_before_end_p: Sequence iterators. (line 63)
41991 * gsi_prev: Sequence iterators. (line 69)
41992 * gsi_remove: Sequence iterators. (line 90)
41993 * gsi_replace: Sequence iterators. (line 130)
41994 * gsi_seq: Sequence iterators. (line 86)
41995 * gsi_split_seq_after: Sequence iterators. (line 120)
41996 * gsi_split_seq_before: Sequence iterators. (line 125)
41997 * gsi_start: Sequence iterators. (line 40)
41998 * gsi_start_bb: Sequence iterators. (line 46)
41999 * gsi_stmt: Sequence iterators. (line 72)
42000 * gt: Comparisons. (line 60)
42001 * gt and attributes: Expressions. (line 64)
42002 * GT_EXPR: Expression trees. (line 6)
42003 * gtu: Comparisons. (line 64)
42004 * gtu and attributes: Expressions. (line 64)
42005 * GTY: Type Information. (line 6)
42006 * H in constraint: Simple Constraints. (line 88)
42007 * HAmode: Machine Modes. (line 144)
42008 * HANDLE_PRAGMA_PACK_PUSH_POP: Misc. (line 467)
42009 * HANDLE_PRAGMA_PACK_WITH_EXPANSION: Misc. (line 478)
42010 * HANDLE_PRAGMA_PUSH_POP_MACRO: Misc. (line 488)
42011 * HANDLE_SYSV_PRAGMA: Misc. (line 438)
42012 * HANDLER: Function Bodies. (line 6)
42013 * HANDLER_BODY: Function Bodies. (line 6)
42014 * HANDLER_PARMS: Function Bodies. (line 6)
42015 * hard registers: Regs and Memory. (line 9)
42016 * HARD_FRAME_POINTER_REGNUM: Frame Registers. (line 20)
42017 * HARD_REGNO_CALL_PART_CLOBBERED: Register Basics. (line 53)
42018 * HARD_REGNO_CALLER_SAVE_MODE: Caller Saves. (line 20)
42019 * HARD_REGNO_MODE_OK: Values in Registers.
42021 * HARD_REGNO_NREGS: Values in Registers.
42023 * HARD_REGNO_NREGS_HAS_PADDING: Values in Registers.
42025 * HARD_REGNO_NREGS_WITH_PADDING: Values in Registers.
42027 * HARD_REGNO_RENAME_OK: Values in Registers.
42029 * HAS_INIT_SECTION: Macros for Initialization.
42031 * HAS_LONG_COND_BRANCH: Misc. (line 9)
42032 * HAS_LONG_UNCOND_BRANCH: Misc. (line 18)
42033 * HAVE_DOS_BASED_FILE_SYSTEM: Filesystem. (line 11)
42034 * HAVE_POST_DECREMENT: Addressing Modes. (line 12)
42035 * HAVE_POST_INCREMENT: Addressing Modes. (line 11)
42036 * HAVE_POST_MODIFY_DISP: Addressing Modes. (line 18)
42037 * HAVE_POST_MODIFY_REG: Addressing Modes. (line 24)
42038 * HAVE_PRE_DECREMENT: Addressing Modes. (line 10)
42039 * HAVE_PRE_INCREMENT: Addressing Modes. (line 9)
42040 * HAVE_PRE_MODIFY_DISP: Addressing Modes. (line 17)
42041 * HAVE_PRE_MODIFY_REG: Addressing Modes. (line 23)
42042 * HCmode: Machine Modes. (line 197)
42043 * HFmode: Machine Modes. (line 58)
42044 * high: Constants. (line 109)
42045 * HImode: Machine Modes. (line 29)
42046 * HImode, in insn: Insns. (line 231)
42047 * host configuration: Host Config. (line 6)
42048 * host functions: Host Common. (line 6)
42049 * host hooks: Host Common. (line 6)
42050 * host makefile fragment: Host Fragment. (line 6)
42051 * HOST_BIT_BUCKET: Filesystem. (line 51)
42052 * HOST_EXECUTABLE_SUFFIX: Filesystem. (line 45)
42053 * HOST_HOOKS_EXTRA_SIGNALS: Host Common. (line 12)
42054 * HOST_HOOKS_GT_PCH_ALLOC_GRANULARITY: Host Common. (line 45)
42055 * HOST_HOOKS_GT_PCH_USE_ADDRESS: Host Common. (line 26)
42056 * HOST_LACKS_INODE_NUMBERS: Filesystem. (line 89)
42057 * HOST_LONG_LONG_FORMAT: Host Misc. (line 41)
42058 * HOST_OBJECT_SUFFIX: Filesystem. (line 40)
42059 * HOST_WIDE_INT: Anchored Addresses. (line 33)
42060 * HOT_TEXT_SECTION_NAME: Sections. (line 43)
42061 * HQmode: Machine Modes. (line 107)
42062 * I in constraint: Simple Constraints. (line 71)
42063 * i in constraint: Simple Constraints. (line 60)
42064 * identifier: Identifiers. (line 6)
42065 * IDENTIFIER_LENGTH: Identifiers. (line 20)
42066 * IDENTIFIER_NODE: Identifiers. (line 6)
42067 * IDENTIFIER_OPNAME_P: Identifiers. (line 25)
42068 * IDENTIFIER_POINTER: Identifiers. (line 15)
42069 * IDENTIFIER_TYPENAME_P: Identifiers. (line 31)
42070 * IEEE 754-2008: Decimal float library routines.
42072 * IF_COND: Function Bodies. (line 6)
42073 * if_marked: GTY Options. (line 156)
42074 * IF_STMT: Function Bodies. (line 6)
42075 * if_then_else: Comparisons. (line 80)
42076 * if_then_else and attributes: Expressions. (line 32)
42077 * if_then_else usage: Side Effects. (line 56)
42078 * IFCVT_EXTRA_FIELDS: Misc. (line 627)
42079 * IFCVT_INIT_EXTRA_FIELDS: Misc. (line 622)
42080 * IFCVT_MODIFY_CANCEL: Misc. (line 616)
42081 * IFCVT_MODIFY_FINAL: Misc. (line 610)
42082 * IFCVT_MODIFY_INSN: Misc. (line 604)
42083 * IFCVT_MODIFY_MULTIPLE_TESTS: Misc. (line 597)
42084 * IFCVT_MODIFY_TESTS: Misc. (line 586)
42085 * IMAGPART_EXPR: Expression trees. (line 6)
42086 * Immediate Uses: SSA Operands. (line 274)
42087 * immediate_operand: Machine-Independent Predicates.
42089 * IMMEDIATE_PREFIX: Instruction Output. (line 127)
42090 * in_struct: Flags. (line 258)
42091 * in_struct, in code_label and note: Flags. (line 59)
42092 * in_struct, in insn and jump_insn and call_insn: Flags. (line 49)
42093 * in_struct, in insn, jump_insn and call_insn: Flags. (line 166)
42094 * in_struct, in mem: Flags. (line 70)
42095 * in_struct, in subreg: Flags. (line 205)
42096 * include: Including Patterns. (line 6)
42097 * INCLUDE_DEFAULTS: Driver. (line 430)
42098 * inclusive-or, bitwise: Arithmetic. (line 158)
42099 * INCOMING_FRAME_SP_OFFSET: Frame Layout. (line 183)
42100 * INCOMING_REGNO: Register Basics. (line 91)
42101 * INCOMING_RETURN_ADDR_RTX: Frame Layout. (line 139)
42102 * INCOMING_STACK_BOUNDARY: Storage Layout. (line 165)
42103 * INDEX_REG_CLASS: Register Classes. (line 134)
42104 * indirect_jump instruction pattern: Standard Names. (line 1078)
42105 * indirect_operand: Machine-Independent Predicates.
42107 * INDIRECT_REF: Expression trees. (line 6)
42108 * INIT_ARRAY_SECTION_ASM_OP: Sections. (line 98)
42109 * INIT_CUMULATIVE_ARGS: Register Arguments. (line 149)
42110 * INIT_CUMULATIVE_INCOMING_ARGS: Register Arguments. (line 177)
42111 * INIT_CUMULATIVE_LIBCALL_ARGS: Register Arguments. (line 170)
42112 * INIT_ENVIRONMENT: Driver. (line 369)
42113 * INIT_EXPANDERS: Per-Function Data. (line 39)
42114 * INIT_EXPR: Expression trees. (line 6)
42115 * init_machine_status: Per-Function Data. (line 45)
42116 * init_one_libfunc: Library Calls. (line 15)
42117 * INIT_SECTION_ASM_OP <1>: Macros for Initialization.
42119 * INIT_SECTION_ASM_OP: Sections. (line 82)
42120 * INITIAL_ELIMINATION_OFFSET: Elimination. (line 79)
42121 * INITIAL_FRAME_ADDRESS_RTX: Frame Layout. (line 83)
42122 * INITIAL_FRAME_POINTER_OFFSET: Elimination. (line 32)
42123 * initialization routines: Initialization. (line 6)
42124 * INITIALIZE_TRAMPOLINE: Trampolines. (line 55)
42125 * inlining: Target Attributes. (line 86)
42126 * insert_insn_on_edge: Maintaining the CFG.
42128 * insn: Insns. (line 63)
42129 * insn and /f: Flags. (line 125)
42130 * insn and /j: Flags. (line 175)
42131 * insn and /s: Flags. (line 49)
42132 * insn and /u: Flags. (line 39)
42133 * insn and /v: Flags. (line 44)
42134 * insn attributes: Insn Attributes. (line 6)
42135 * insn canonicalization: Insn Canonicalizations.
42137 * insn includes: Including Patterns. (line 6)
42138 * insn lengths, computing: Insn Lengths. (line 6)
42139 * insn splitting: Insn Splitting. (line 6)
42140 * insn-attr.h: Defining Attributes.
42142 * INSN_ANNULLED_BRANCH_P: Flags. (line 39)
42143 * INSN_CODE: Insns. (line 257)
42144 * INSN_DELETED_P: Flags. (line 44)
42145 * INSN_FROM_TARGET_P: Flags. (line 49)
42146 * insn_list: Insns. (line 505)
42147 * INSN_REFERENCES_ARE_DELAYED: Misc. (line 525)
42148 * INSN_SETS_ARE_DELAYED: Misc. (line 514)
42149 * INSN_UID: Insns. (line 23)
42150 * insns: Insns. (line 6)
42151 * insns, generating: RTL Template. (line 6)
42152 * insns, recognizing: RTL Template. (line 6)
42153 * instruction attributes: Insn Attributes. (line 6)
42154 * instruction latency time: Processor pipeline description.
42156 * instruction patterns: Patterns. (line 6)
42157 * instruction splitting: Insn Splitting. (line 6)
42158 * insv instruction pattern: Standard Names. (line 880)
42159 * int <1>: Run-time Target. (line 56)
42160 * int: Manipulating GIMPLE statements.
42162 * INT_TYPE_SIZE: Type Layout. (line 12)
42163 * INTEGER_CST: Expression trees. (line 6)
42164 * INTEGER_TYPE: Types. (line 6)
42165 * Interdependence of Patterns: Dependent Patterns. (line 6)
42166 * interfacing to GCC output: Interface. (line 6)
42167 * interlock delays: Processor pipeline description.
42169 * intermediate representation lowering: Parsing pass. (line 14)
42170 * INTMAX_TYPE: Type Layout. (line 213)
42171 * introduction: Top. (line 6)
42172 * INVOKE__main: Macros for Initialization.
42174 * ior: Arithmetic. (line 158)
42175 * ior and attributes: Expressions. (line 50)
42176 * ior, canonicalization of: Insn Canonicalizations.
42178 * iorM3 instruction pattern: Standard Names. (line 222)
42179 * IRA_COVER_CLASSES: Register Classes. (line 516)
42180 * IRA_HARD_REGNO_ADD_COST_MULTIPLIER: Allocation Order. (line 37)
42181 * IS_ASM_LOGICAL_LINE_SEPARATOR: Data Output. (line 120)
42182 * is_gimple_omp: GIMPLE_OMP_PARALLEL.
42184 * iterators in .md files: Iterators. (line 6)
42185 * IV analysis on GIMPLE: Scalar evolutions. (line 6)
42186 * IV analysis on RTL: loop-iv. (line 6)
42187 * jump: Flags. (line 309)
42188 * jump instruction pattern: Standard Names. (line 969)
42189 * jump instruction patterns: Jump Patterns. (line 6)
42190 * jump instructions and set: Side Effects. (line 56)
42191 * jump, in call_insn: Flags. (line 179)
42192 * jump, in insn: Flags. (line 175)
42193 * jump, in mem: Flags. (line 79)
42194 * JUMP_ALIGN: Alignment Output. (line 9)
42195 * jump_insn: Insns. (line 73)
42196 * jump_insn and /f: Flags. (line 125)
42197 * jump_insn and /s: Flags. (line 49)
42198 * jump_insn and /u: Flags. (line 39)
42199 * jump_insn and /v: Flags. (line 44)
42200 * JUMP_LABEL: Insns. (line 80)
42201 * JUMP_TABLES_IN_TEXT_SECTION: Sections. (line 142)
42202 * Jumps: Jumps. (line 6)
42203 * LABEL_ALIGN: Alignment Output. (line 52)
42204 * LABEL_ALIGN_AFTER_BARRIER: Alignment Output. (line 22)
42205 * LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP: Alignment Output. (line 30)
42206 * LABEL_ALIGN_MAX_SKIP: Alignment Output. (line 62)
42207 * LABEL_ALT_ENTRY_P: Insns. (line 140)
42208 * LABEL_ALTERNATE_NAME: Edges. (line 180)
42209 * LABEL_DECL: Declarations. (line 6)
42210 * LABEL_KIND: Insns. (line 140)
42211 * LABEL_NUSES: Insns. (line 136)
42212 * LABEL_PRESERVE_P: Flags. (line 59)
42213 * label_ref: Constants. (line 86)
42214 * label_ref and /v: Flags. (line 65)
42215 * label_ref, RTL sharing: Sharing. (line 35)
42216 * LABEL_REF_NONLOCAL_P: Flags. (line 65)
42217 * lang_hooks.gimplify_expr: Gimplification pass.
42219 * lang_hooks.parse_file: Parsing pass. (line 6)
42220 * language-independent intermediate representation: Parsing pass.
42222 * large return values: Aggregate Return. (line 6)
42223 * LARGEST_EXPONENT_IS_NORMAL: Storage Layout. (line 469)
42224 * LAST_STACK_REG: Stack Registers. (line 27)
42225 * LAST_VIRTUAL_REGISTER: Regs and Memory. (line 51)
42226 * lceilMN2: Standard Names. (line 597)
42227 * LCSSA: LCSSA. (line 6)
42228 * LD_FINI_SWITCH: Macros for Initialization.
42230 * LD_INIT_SWITCH: Macros for Initialization.
42232 * LDD_SUFFIX: Macros for Initialization.
42234 * le: Comparisons. (line 76)
42235 * le and attributes: Expressions. (line 64)
42236 * LE_EXPR: Expression trees. (line 6)
42237 * leaf functions: Leaf Functions. (line 6)
42238 * leaf_function_p: Standard Names. (line 1040)
42239 * LEAF_REG_REMAP: Leaf Functions. (line 39)
42240 * LEAF_REGISTERS: Leaf Functions. (line 25)
42241 * left rotate: Arithmetic. (line 190)
42242 * left shift: Arithmetic. (line 168)
42243 * LEGITIMATE_CONSTANT_P: Addressing Modes. (line 205)
42244 * LEGITIMATE_PIC_OPERAND_P: PIC. (line 31)
42245 * LEGITIMIZE_ADDRESS: Addressing Modes. (line 122)
42246 * LEGITIMIZE_RELOAD_ADDRESS: Addressing Modes. (line 145)
42247 * length: GTY Options. (line 50)
42248 * less than: Comparisons. (line 68)
42249 * less than or equal: Comparisons. (line 76)
42250 * leu: Comparisons. (line 76)
42251 * leu and attributes: Expressions. (line 64)
42252 * lfloorMN2: Standard Names. (line 592)
42253 * LIB2FUNCS_EXTRA: Target Fragment. (line 11)
42254 * LIB_SPEC: Driver. (line 170)
42255 * LIBCALL_VALUE: Scalar Return. (line 60)
42256 * libgcc.a: Library Calls. (line 6)
42257 * LIBGCC2_CFLAGS: Target Fragment. (line 8)
42258 * LIBGCC2_HAS_DF_MODE: Type Layout. (line 109)
42259 * LIBGCC2_HAS_TF_MODE: Type Layout. (line 123)
42260 * LIBGCC2_HAS_XF_MODE: Type Layout. (line 117)
42261 * LIBGCC2_LONG_DOUBLE_TYPE_SIZE: Type Layout. (line 103)
42262 * LIBGCC2_UNWIND_ATTRIBUTE: Misc. (line 943)
42263 * LIBGCC2_WORDS_BIG_ENDIAN: Storage Layout. (line 36)
42264 * LIBGCC_SPEC: Driver. (line 178)
42265 * library subroutine names: Library Calls. (line 6)
42266 * LIBRARY_PATH_ENV: Misc. (line 565)
42267 * LIMIT_RELOAD_CLASS: Register Classes. (line 239)
42268 * Linear loop transformations framework: Lambda. (line 6)
42269 * LINK_COMMAND_SPEC: Driver. (line 299)
42270 * LINK_EH_SPEC: Driver. (line 205)
42271 * LINK_ELIMINATE_DUPLICATE_LDIRECTORIES: Driver. (line 309)
42272 * LINK_GCC_C_SEQUENCE_SPEC: Driver. (line 295)
42273 * LINK_LIBGCC_SPECIAL_1: Driver. (line 290)
42274 * LINK_SPEC: Driver. (line 163)
42275 * linkage: Function Basics. (line 6)
42276 * list: Containers. (line 6)
42277 * Liveness representation: Liveness information.
42279 * lo_sum: Arithmetic. (line 24)
42280 * load address instruction: Simple Constraints. (line 154)
42281 * LOAD_EXTEND_OP: Misc. (line 69)
42282 * load_multiple instruction pattern: Standard Names. (line 137)
42283 * LOCAL_ALIGNMENT: Storage Layout. (line 254)
42284 * LOCAL_CLASS_P: Classes. (line 68)
42285 * LOCAL_DECL_ALIGNMENT: Storage Layout. (line 278)
42286 * LOCAL_INCLUDE_DIR: Driver. (line 376)
42287 * LOCAL_LABEL_PREFIX: Instruction Output. (line 125)
42288 * LOCAL_REGNO: Register Basics. (line 105)
42289 * LOG_LINKS: Insns. (line 276)
42290 * Logical Operators: Logical Operators. (line 6)
42291 * logical-and, bitwise: Arithmetic. (line 153)
42292 * logM2 instruction pattern: Standard Names. (line 505)
42293 * LONG_ACCUM_TYPE_SIZE: Type Layout. (line 93)
42294 * LONG_DOUBLE_TYPE_SIZE: Type Layout. (line 58)
42295 * LONG_FRACT_TYPE_SIZE: Type Layout. (line 73)
42296 * LONG_LONG_ACCUM_TYPE_SIZE: Type Layout. (line 98)
42297 * LONG_LONG_FRACT_TYPE_SIZE: Type Layout. (line 78)
42298 * LONG_LONG_TYPE_SIZE: Type Layout. (line 33)
42299 * LONG_TYPE_SIZE: Type Layout. (line 22)
42300 * longjmp and automatic variables: Interface. (line 52)
42301 * Loop analysis: Loop representation.
42303 * Loop manipulation: Loop manipulation. (line 6)
42304 * Loop querying: Loop querying. (line 6)
42305 * Loop representation: Loop representation.
42307 * Loop-closed SSA form: LCSSA. (line 6)
42308 * LOOP_ALIGN: Alignment Output. (line 35)
42309 * LOOP_ALIGN_MAX_SKIP: Alignment Output. (line 48)
42310 * LOOP_EXPR: Expression trees. (line 6)
42311 * looping instruction patterns: Looping Patterns. (line 6)
42312 * lowering, language-dependent intermediate representation: Parsing pass.
42314 * lrintMN2: Standard Names. (line 582)
42315 * lroundMN2: Standard Names. (line 587)
42316 * LSHIFT_EXPR: Expression trees. (line 6)
42317 * lshiftrt: Arithmetic. (line 185)
42318 * lshiftrt and attributes: Expressions. (line 64)
42319 * lshrM3 instruction pattern: Standard Names. (line 441)
42320 * lt: Comparisons. (line 68)
42321 * lt and attributes: Expressions. (line 64)
42322 * LT_EXPR: Expression trees. (line 6)
42323 * LTGT_EXPR: Expression trees. (line 6)
42324 * ltu: Comparisons. (line 68)
42325 * m in constraint: Simple Constraints. (line 17)
42326 * machine attributes: Target Attributes. (line 6)
42327 * machine description macros: Target Macros. (line 6)
42328 * machine descriptions: Machine Desc. (line 6)
42329 * machine mode conversions: Conversions. (line 6)
42330 * machine modes: Machine Modes. (line 6)
42331 * machine specific constraints: Machine Constraints.
42333 * machine-independent predicates: Machine-Independent Predicates.
42335 * machine_mode: Condition Code. (line 157)
42336 * macros, target description: Target Macros. (line 6)
42337 * maddMN4 instruction pattern: Standard Names. (line 364)
42338 * MAKE_DECL_ONE_ONLY: Label Output. (line 218)
42339 * make_phi_node: GIMPLE_PHI. (line 7)
42340 * make_safe_from: Expander Definitions.
42342 * makefile fragment: Fragments. (line 6)
42343 * makefile targets: Makefile. (line 6)
42344 * MALLOC_ABI_ALIGNMENT: Storage Layout. (line 179)
42345 * Manipulating GIMPLE statements: Manipulating GIMPLE statements.
42347 * mark_hook: GTY Options. (line 171)
42348 * marking roots: GGC Roots. (line 6)
42349 * MASK_RETURN_ADDR: Exception Region Output.
42351 * match_dup <1>: define_peephole2. (line 28)
42352 * match_dup: RTL Template. (line 73)
42353 * match_dup and attributes: Insn Lengths. (line 16)
42354 * match_op_dup: RTL Template. (line 163)
42355 * match_operand: RTL Template. (line 16)
42356 * match_operand and attributes: Expressions. (line 55)
42357 * match_operator: RTL Template. (line 95)
42358 * match_par_dup: RTL Template. (line 219)
42359 * match_parallel: RTL Template. (line 172)
42360 * match_scratch <1>: define_peephole2. (line 28)
42361 * match_scratch: RTL Template. (line 58)
42362 * matching constraint: Simple Constraints. (line 132)
42363 * matching operands: Output Template. (line 49)
42364 * math library: Soft float library routines.
42366 * math, in RTL: Arithmetic. (line 6)
42367 * MATH_LIBRARY: Misc. (line 558)
42368 * matherr: Library Calls. (line 58)
42369 * MAX_BITS_PER_WORD: Storage Layout. (line 61)
42370 * MAX_CONDITIONAL_EXECUTE: Misc. (line 580)
42371 * MAX_FIXED_MODE_SIZE: Storage Layout. (line 420)
42372 * MAX_MOVE_MAX: Misc. (line 120)
42373 * MAX_OFILE_ALIGNMENT: Storage Layout. (line 216)
42374 * MAX_REGS_PER_ADDRESS: Addressing Modes. (line 42)
42375 * MAX_STACK_ALIGNMENT: Storage Layout. (line 209)
42376 * maxM3 instruction pattern: Standard Names. (line 234)
42377 * may_trap_p, tree_could_trap_p: Edges. (line 115)
42378 * maybe_undef: GTY Options. (line 179)
42379 * mcount: Profiling. (line 12)
42380 * MD_CAN_REDIRECT_BRANCH: Misc. (line 705)
42381 * MD_EXEC_PREFIX: Driver. (line 330)
42382 * MD_FALLBACK_FRAME_STATE_FOR: Exception Handling. (line 98)
42383 * MD_HANDLE_UNWABI: Exception Handling. (line 118)
42384 * MD_STARTFILE_PREFIX: Driver. (line 358)
42385 * MD_STARTFILE_PREFIX_1: Driver. (line 364)
42386 * MD_UNWIND_SUPPORT: Exception Handling. (line 94)
42387 * mem: Regs and Memory. (line 374)
42388 * mem and /c: Flags. (line 99)
42389 * mem and /f: Flags. (line 103)
42390 * mem and /i: Flags. (line 85)
42391 * mem and /j: Flags. (line 79)
42392 * mem and /s: Flags. (line 70)
42393 * mem and /u: Flags. (line 152)
42394 * mem and /v: Flags. (line 94)
42395 * mem, RTL sharing: Sharing. (line 40)
42396 * MEM_ALIAS_SET: Special Accessors. (line 9)
42397 * MEM_ALIGN: Special Accessors. (line 36)
42398 * MEM_EXPR: Special Accessors. (line 20)
42399 * MEM_IN_STRUCT_P: Flags. (line 70)
42400 * MEM_KEEP_ALIAS_SET_P: Flags. (line 79)
42401 * MEM_NOTRAP_P: Flags. (line 99)
42402 * MEM_OFFSET: Special Accessors. (line 28)
42403 * MEM_POINTER: Flags. (line 103)
42404 * MEM_READONLY_P: Flags. (line 152)
42405 * MEM_SCALAR_P: Flags. (line 85)
42406 * MEM_SIZE: Special Accessors. (line 31)
42407 * MEM_VOLATILE_P: Flags. (line 94)
42408 * MEMBER_TYPE_FORCES_BLK: Storage Layout. (line 400)
42409 * memory reference, nonoffsettable: Simple Constraints. (line 246)
42410 * memory references in constraints: Simple Constraints. (line 17)
42411 * memory_barrier instruction pattern: Standard Names. (line 1413)
42412 * MEMORY_MOVE_COST: Costs. (line 29)
42413 * memory_operand: Machine-Independent Predicates.
42415 * METHOD_TYPE: Types. (line 6)
42416 * MIN_UNITS_PER_WORD: Storage Layout. (line 71)
42417 * MINIMUM_ALIGNMENT: Storage Layout. (line 288)
42418 * MINIMUM_ATOMIC_ALIGNMENT: Storage Layout. (line 187)
42419 * minM3 instruction pattern: Standard Names. (line 234)
42420 * minus: Arithmetic. (line 36)
42421 * minus and attributes: Expressions. (line 64)
42422 * minus, canonicalization of: Insn Canonicalizations.
42424 * MINUS_EXPR: Expression trees. (line 6)
42425 * MIPS coprocessor-definition macros: MIPS Coprocessors. (line 6)
42426 * mod: Arithmetic. (line 131)
42427 * mod and attributes: Expressions. (line 64)
42428 * mode classes: Machine Modes. (line 219)
42429 * mode iterators in .md files: Mode Iterators. (line 6)
42430 * mode switching: Mode Switching. (line 6)
42431 * MODE_ACCUM: Machine Modes. (line 249)
42432 * MODE_AFTER: Mode Switching. (line 49)
42433 * MODE_BASE_REG_CLASS: Register Classes. (line 112)
42434 * MODE_BASE_REG_REG_CLASS: Register Classes. (line 118)
42435 * MODE_CC: Machine Modes. (line 268)
42436 * MODE_CODE_BASE_REG_CLASS: Register Classes. (line 125)
42437 * MODE_COMPLEX_FLOAT: Machine Modes. (line 260)
42438 * MODE_COMPLEX_INT: Machine Modes. (line 257)
42439 * MODE_DECIMAL_FLOAT: Machine Modes. (line 237)
42440 * MODE_ENTRY: Mode Switching. (line 54)
42441 * MODE_EXIT: Mode Switching. (line 60)
42442 * MODE_FLOAT: Machine Modes. (line 233)
42443 * MODE_FRACT: Machine Modes. (line 241)
42444 * MODE_FUNCTION: Machine Modes. (line 264)
42445 * MODE_INT: Machine Modes. (line 225)
42446 * MODE_NEEDED: Mode Switching. (line 42)
42447 * MODE_PARTIAL_INT: Machine Modes. (line 229)
42448 * MODE_PRIORITY_TO_MODE: Mode Switching. (line 66)
42449 * MODE_RANDOM: Machine Modes. (line 273)
42450 * MODE_UACCUM: Machine Modes. (line 253)
42451 * MODE_UFRACT: Machine Modes. (line 245)
42452 * MODES_TIEABLE_P: Values in Registers.
42454 * modifiers in constraints: Modifiers. (line 6)
42455 * MODIFY_EXPR: Expression trees. (line 6)
42456 * MODIFY_JNI_METHOD_CALL: Misc. (line 782)
42457 * MODIFY_TARGET_NAME: Driver. (line 385)
42458 * modM3 instruction pattern: Standard Names. (line 222)
42459 * modulo scheduling: RTL passes. (line 140)
42460 * MOVE_BY_PIECES_P: Costs. (line 110)
42461 * MOVE_MAX: Misc. (line 115)
42462 * MOVE_MAX_PIECES: Costs. (line 116)
42463 * MOVE_RATIO: Costs. (line 97)
42464 * movM instruction pattern: Standard Names. (line 11)
42465 * movmemM instruction pattern: Standard Names. (line 672)
42466 * movmisalignM instruction pattern: Standard Names. (line 126)
42467 * movMODEcc instruction pattern: Standard Names. (line 891)
42468 * movstr instruction pattern: Standard Names. (line 707)
42469 * movstrictM instruction pattern: Standard Names. (line 120)
42470 * msubMN4 instruction pattern: Standard Names. (line 387)
42471 * mulhisi3 instruction pattern: Standard Names. (line 340)
42472 * mulM3 instruction pattern: Standard Names. (line 222)
42473 * mulqihi3 instruction pattern: Standard Names. (line 344)
42474 * mulsidi3 instruction pattern: Standard Names. (line 344)
42475 * mult: Arithmetic. (line 92)
42476 * mult and attributes: Expressions. (line 64)
42477 * mult, canonicalization of: Insn Canonicalizations.
42479 * MULT_EXPR: Expression trees. (line 6)
42480 * MULTILIB_DEFAULTS: Driver. (line 315)
42481 * MULTILIB_DIRNAMES: Target Fragment. (line 64)
42482 * MULTILIB_EXCEPTIONS: Target Fragment. (line 84)
42483 * MULTILIB_EXTRA_OPTS: Target Fragment. (line 96)
42484 * MULTILIB_MATCHES: Target Fragment. (line 77)
42485 * MULTILIB_OPTIONS: Target Fragment. (line 44)
42486 * multiple alternative constraints: Multi-Alternative. (line 6)
42487 * MULTIPLE_SYMBOL_SPACES: Misc. (line 538)
42488 * multiplication: Arithmetic. (line 92)
42489 * multiplication with signed saturation: Arithmetic. (line 92)
42490 * multiplication with unsigned saturation: Arithmetic. (line 92)
42491 * MUST_USE_SJLJ_EXCEPTIONS: Exception Region Output.
42493 * n in constraint: Simple Constraints. (line 65)
42494 * N_REG_CLASSES: Register Classes. (line 76)
42495 * name: Identifiers. (line 6)
42496 * named patterns and conditions: Patterns. (line 47)
42497 * names, pattern: Standard Names. (line 6)
42498 * namespace: Namespaces. (line 6)
42499 * namespace, class, scope: Scopes. (line 6)
42500 * NAMESPACE_DECL <1>: Declarations. (line 6)
42501 * NAMESPACE_DECL: Namespaces. (line 6)
42502 * NATIVE_SYSTEM_HEADER_DIR: Target Fragment. (line 103)
42503 * ne: Comparisons. (line 56)
42504 * ne and attributes: Expressions. (line 64)
42505 * NE_EXPR: Expression trees. (line 6)
42506 * nearbyintM2 instruction pattern: Standard Names. (line 564)
42507 * neg: Arithmetic. (line 81)
42508 * neg and attributes: Expressions. (line 64)
42509 * neg, canonicalization of: Insn Canonicalizations.
42511 * NEGATE_EXPR: Expression trees. (line 6)
42512 * negation: Arithmetic. (line 81)
42513 * negation with signed saturation: Arithmetic. (line 81)
42514 * negation with unsigned saturation: Arithmetic. (line 81)
42515 * negM2 instruction pattern: Standard Names. (line 449)
42516 * nested functions, trampolines for: Trampolines. (line 6)
42517 * nested_ptr: GTY Options. (line 186)
42518 * next_bb, prev_bb, FOR_EACH_BB: Basic Blocks. (line 10)
42519 * next_cc0_user: Jump Patterns. (line 64)
42520 * NEXT_INSN: Insns. (line 30)
42521 * NEXT_OBJC_RUNTIME: Library Calls. (line 94)
42522 * nil: RTL Objects. (line 73)
42523 * NO_DBX_BNSYM_ENSYM: DBX Hooks. (line 39)
42524 * NO_DBX_FUNCTION_END: DBX Hooks. (line 33)
42525 * NO_DBX_GCC_MARKER: File Names and DBX. (line 28)
42526 * NO_DBX_MAIN_SOURCE_DIRECTORY: File Names and DBX. (line 23)
42527 * NO_DOLLAR_IN_LABEL: Misc. (line 502)
42528 * NO_DOT_IN_LABEL: Misc. (line 508)
42529 * NO_FUNCTION_CSE: Costs. (line 200)
42530 * NO_IMPLICIT_EXTERN_C: Misc. (line 376)
42531 * NO_PROFILE_COUNTERS: Profiling. (line 28)
42532 * NO_REGS: Register Classes. (line 17)
42533 * NON_LVALUE_EXPR: Expression trees. (line 6)
42534 * nondeterministic finite state automaton: Processor pipeline description.
42536 * nonimmediate_operand: Machine-Independent Predicates.
42538 * nonlocal goto handler: Edges. (line 171)
42539 * nonlocal_goto instruction pattern: Standard Names. (line 1255)
42540 * nonlocal_goto_receiver instruction pattern: Standard Names.
42542 * nonmemory_operand: Machine-Independent Predicates.
42544 * nonoffsettable memory reference: Simple Constraints. (line 246)
42545 * nop instruction pattern: Standard Names. (line 1073)
42546 * NOP_EXPR: Expression trees. (line 6)
42547 * normal predicates: Predicates. (line 31)
42548 * not: Arithmetic. (line 149)
42549 * not and attributes: Expressions. (line 50)
42550 * not equal: Comparisons. (line 56)
42551 * not, canonicalization of: Insn Canonicalizations.
42553 * note: Insns. (line 168)
42554 * note and /i: Flags. (line 59)
42555 * note and /v: Flags. (line 44)
42556 * NOTE_INSN_BASIC_BLOCK, CODE_LABEL, notes: Basic Blocks. (line 41)
42557 * NOTE_INSN_BLOCK_BEG: Insns. (line 193)
42558 * NOTE_INSN_BLOCK_END: Insns. (line 193)
42559 * NOTE_INSN_DELETED: Insns. (line 183)
42560 * NOTE_INSN_DELETED_LABEL: Insns. (line 188)
42561 * NOTE_INSN_EH_REGION_BEG: Insns. (line 199)
42562 * NOTE_INSN_EH_REGION_END: Insns. (line 199)
42563 * NOTE_INSN_FUNCTION_BEG: Insns. (line 223)
42564 * NOTE_INSN_LOOP_BEG: Insns. (line 207)
42565 * NOTE_INSN_LOOP_CONT: Insns. (line 213)
42566 * NOTE_INSN_LOOP_END: Insns. (line 207)
42567 * NOTE_INSN_LOOP_VTOP: Insns. (line 217)
42568 * NOTE_LINE_NUMBER: Insns. (line 168)
42569 * NOTE_SOURCE_FILE: Insns. (line 168)
42570 * NOTICE_UPDATE_CC: Condition Code. (line 33)
42571 * NUM_MACHINE_MODES: Machine Modes. (line 286)
42572 * NUM_MODES_FOR_MODE_SWITCHING: Mode Switching. (line 30)
42573 * Number of iterations analysis: Number of iterations.
42575 * o in constraint: Simple Constraints. (line 23)
42576 * OBJC_GEN_METHOD_LABEL: Label Output. (line 411)
42577 * OBJC_JBLEN: Misc. (line 938)
42578 * OBJECT_FORMAT_COFF: Macros for Initialization.
42580 * OFFSET_TYPE: Types. (line 6)
42581 * offsettable address: Simple Constraints. (line 23)
42582 * OImode: Machine Modes. (line 51)
42583 * Omega a solver for linear programming problems: Omega. (line 6)
42584 * OMP_ATOMIC: Expression trees. (line 6)
42585 * OMP_CLAUSE: Expression trees. (line 6)
42586 * OMP_CONTINUE: Expression trees. (line 6)
42587 * OMP_CRITICAL: Expression trees. (line 6)
42588 * OMP_FOR: Expression trees. (line 6)
42589 * OMP_MASTER: Expression trees. (line 6)
42590 * OMP_ORDERED: Expression trees. (line 6)
42591 * OMP_PARALLEL: Expression trees. (line 6)
42592 * OMP_RETURN: Expression trees. (line 6)
42593 * OMP_SECTION: Expression trees. (line 6)
42594 * OMP_SECTIONS: Expression trees. (line 6)
42595 * OMP_SINGLE: Expression trees. (line 6)
42596 * one_cmplM2 instruction pattern: Standard Names. (line 651)
42597 * operand access: Accessors. (line 6)
42598 * Operand Access Routines: SSA Operands. (line 119)
42599 * operand constraints: Constraints. (line 6)
42600 * Operand Iterators: SSA Operands. (line 119)
42601 * operand predicates: Predicates. (line 6)
42602 * operand substitution: Output Template. (line 6)
42603 * operands <1>: Patterns. (line 53)
42604 * operands: SSA Operands. (line 6)
42605 * Operands: Operands. (line 6)
42606 * operator predicates: Predicates. (line 6)
42607 * optc-gen.awk: Options. (line 6)
42608 * Optimization infrastructure for GIMPLE: Tree SSA. (line 6)
42609 * OPTIMIZATION_OPTIONS: Run-time Target. (line 120)
42610 * OPTIMIZE_MODE_SWITCHING: Mode Switching. (line 9)
42611 * option specification files: Options. (line 6)
42612 * OPTION_DEFAULT_SPECS: Driver. (line 88)
42613 * optional hardware or system features: Run-time Target. (line 59)
42614 * options, directory search: Including Patterns. (line 44)
42615 * order of register allocation: Allocation Order. (line 6)
42616 * ORDER_REGS_FOR_LOCAL_ALLOC: Allocation Order. (line 23)
42617 * ORDERED_EXPR: Expression trees. (line 6)
42618 * Ordering of Patterns: Pattern Ordering. (line 6)
42619 * ORIGINAL_REGNO: Special Accessors. (line 40)
42620 * other register constraints: Simple Constraints. (line 163)
42621 * OUTGOING_REG_PARM_STACK_SPACE: Stack Arguments. (line 71)
42622 * OUTGOING_REGNO: Register Basics. (line 98)
42623 * output of assembler code: File Framework. (line 6)
42624 * output statements: Output Statement. (line 6)
42625 * output templates: Output Template. (line 6)
42626 * OUTPUT_ADDR_CONST_EXTRA: Data Output. (line 39)
42627 * output_asm_insn: Output Statement. (line 53)
42628 * OUTPUT_QUOTED_STRING: File Framework. (line 76)
42629 * OVERLOAD: Functions. (line 6)
42630 * OVERRIDE_ABI_FORMAT: Register Arguments. (line 140)
42631 * OVERRIDE_OPTIONS: Run-time Target. (line 104)
42632 * OVL_CURRENT: Functions. (line 6)
42633 * OVL_NEXT: Functions. (line 6)
42634 * p in constraint: Simple Constraints. (line 154)
42635 * PAD_VARARGS_DOWN: Register Arguments. (line 221)
42636 * parallel: Side Effects. (line 204)
42637 * param_is: GTY Options. (line 114)
42638 * parameters, c++ abi: C++ ABI. (line 6)
42639 * parameters, miscellaneous: Misc. (line 6)
42640 * parameters, precompiled headers: PCH Target. (line 6)
42641 * paramN_is: GTY Options. (line 132)
42642 * parity: Arithmetic. (line 228)
42643 * parityM2 instruction pattern: Standard Names. (line 645)
42644 * PARM_BOUNDARY: Storage Layout. (line 144)
42645 * PARM_DECL: Declarations. (line 6)
42646 * PARSE_LDD_OUTPUT: Macros for Initialization.
42648 * passes and files of the compiler: Passes. (line 6)
42649 * passing arguments: Interface. (line 36)
42650 * PATH_SEPARATOR: Filesystem. (line 31)
42651 * PATTERN: Insns. (line 247)
42652 * pattern conditions: Patterns. (line 43)
42653 * pattern names: Standard Names. (line 6)
42654 * Pattern Ordering: Pattern Ordering. (line 6)
42655 * patterns: Patterns. (line 6)
42656 * pc: Regs and Memory. (line 361)
42657 * pc and attributes: Insn Lengths. (line 20)
42658 * pc, RTL sharing: Sharing. (line 25)
42659 * PC_REGNUM: Register Basics. (line 112)
42660 * pc_rtx: Regs and Memory. (line 366)
42661 * PCC_BITFIELD_TYPE_MATTERS: Storage Layout. (line 314)
42662 * PCC_STATIC_STRUCT_RETURN: Aggregate Return. (line 64)
42663 * PDImode: Machine Modes. (line 40)
42664 * peephole optimization, RTL representation: Side Effects. (line 238)
42665 * peephole optimizer definitions: Peephole Definitions.
42667 * per-function data: Per-Function Data. (line 6)
42668 * percent sign: Output Template. (line 6)
42669 * PHI nodes: SSA. (line 31)
42670 * phi_arg_d: GIMPLE_PHI. (line 28)
42671 * PHI_ARG_DEF: SSA. (line 71)
42672 * PHI_ARG_EDGE: SSA. (line 68)
42673 * PHI_ARG_ELT: SSA. (line 63)
42674 * PHI_NUM_ARGS: SSA. (line 59)
42675 * PHI_RESULT: SSA. (line 56)
42676 * PIC: PIC. (line 6)
42677 * PIC_OFFSET_TABLE_REG_CALL_CLOBBERED: PIC. (line 26)
42678 * PIC_OFFSET_TABLE_REGNUM: PIC. (line 16)
42679 * pipeline hazard recognizer: Processor pipeline description.
42681 * Plugins: Plugins. (line 6)
42682 * plus: Arithmetic. (line 14)
42683 * plus and attributes: Expressions. (line 64)
42684 * plus, canonicalization of: Insn Canonicalizations.
42686 * PLUS_EXPR: Expression trees. (line 6)
42687 * Pmode: Misc. (line 344)
42688 * pmode_register_operand: Machine-Independent Predicates.
42690 * pointer: Types. (line 6)
42691 * POINTER_PLUS_EXPR: Expression trees. (line 6)
42692 * POINTER_SIZE: Storage Layout. (line 83)
42693 * POINTER_TYPE: Types. (line 6)
42694 * POINTERS_EXTEND_UNSIGNED: Storage Layout. (line 89)
42695 * pop_operand: Machine-Independent Predicates.
42697 * popcount: Arithmetic. (line 224)
42698 * popcountM2 instruction pattern: Standard Names. (line 639)
42699 * portability: Portability. (line 6)
42700 * position independent code: PIC. (line 6)
42701 * post_dec: Incdec. (line 25)
42702 * post_inc: Incdec. (line 30)
42703 * post_modify: Incdec. (line 33)
42704 * POSTDECREMENT_EXPR: Expression trees. (line 6)
42705 * POSTINCREMENT_EXPR: Expression trees. (line 6)
42706 * POWI_MAX_MULTS: Misc. (line 836)
42707 * powM3 instruction pattern: Standard Names. (line 513)
42708 * pragma: Misc. (line 381)
42709 * pre_dec: Incdec. (line 8)
42710 * PRE_GCC3_DWARF_FRAME_REGISTERS: Frame Registers. (line 110)
42711 * pre_inc: Incdec. (line 22)
42712 * pre_modify: Incdec. (line 51)
42713 * PREDECREMENT_EXPR: Expression trees. (line 6)
42714 * predefined macros: Run-time Target. (line 6)
42715 * predicates: Predicates. (line 6)
42716 * predicates and machine modes: Predicates. (line 31)
42717 * predication: Conditional Execution.
42719 * predict.def: Profile information.
42721 * PREFERRED_DEBUGGING_TYPE: All Debuggers. (line 42)
42722 * PREFERRED_OUTPUT_RELOAD_CLASS: Register Classes. (line 231)
42723 * PREFERRED_RELOAD_CLASS: Register Classes. (line 196)
42724 * PREFERRED_STACK_BOUNDARY: Storage Layout. (line 158)
42725 * prefetch: Side Effects. (line 312)
42726 * prefetch instruction pattern: Standard Names. (line 1392)
42727 * PREINCREMENT_EXPR: Expression trees. (line 6)
42728 * presence_set: Processor pipeline description.
42730 * preserving SSA form: SSA. (line 76)
42731 * preserving virtual SSA form: SSA. (line 186)
42732 * prev_active_insn: define_peephole. (line 60)
42733 * prev_cc0_setter: Jump Patterns. (line 64)
42734 * PREV_INSN: Insns. (line 26)
42735 * PRINT_OPERAND: Instruction Output. (line 68)
42736 * PRINT_OPERAND_ADDRESS: Instruction Output. (line 96)
42737 * PRINT_OPERAND_PUNCT_VALID_P: Instruction Output. (line 89)
42738 * processor functional units: Processor pipeline description.
42740 * processor pipeline description: Processor pipeline description.
42742 * product: Arithmetic. (line 92)
42743 * profile feedback: Profile information.
42745 * profile representation: Profile information.
42747 * PROFILE_BEFORE_PROLOGUE: Profiling. (line 35)
42748 * PROFILE_HOOK: Profiling. (line 23)
42749 * profiling, code generation: Profiling. (line 6)
42750 * program counter: Regs and Memory. (line 362)
42751 * prologue: Function Entry. (line 6)
42752 * prologue instruction pattern: Standard Names. (line 1338)
42753 * PROMOTE_FUNCTION_MODE: Storage Layout. (line 123)
42754 * PROMOTE_MODE: Storage Layout. (line 100)
42755 * pseudo registers: Regs and Memory. (line 9)
42756 * PSImode: Machine Modes. (line 32)
42757 * PTRDIFF_TYPE: Type Layout. (line 184)
42758 * PTRMEM_CST: Expression trees. (line 6)
42759 * PTRMEM_CST_CLASS: Expression trees. (line 6)
42760 * PTRMEM_CST_MEMBER: Expression trees. (line 6)
42761 * purge_dead_edges <1>: Maintaining the CFG.
42763 * purge_dead_edges: Edges. (line 104)
42764 * push address instruction: Simple Constraints. (line 154)
42765 * PUSH_ARGS: Stack Arguments. (line 18)
42766 * PUSH_ARGS_REVERSED: Stack Arguments. (line 26)
42767 * push_operand: Machine-Independent Predicates.
42769 * push_reload: Addressing Modes. (line 169)
42770 * PUSH_ROUNDING: Stack Arguments. (line 32)
42771 * pushM1 instruction pattern: Standard Names. (line 209)
42772 * PUT_CODE: RTL Objects. (line 47)
42773 * PUT_MODE: Machine Modes. (line 283)
42774 * PUT_REG_NOTE_KIND: Insns. (line 309)
42775 * PUT_SDB_: SDB and DWARF. (line 63)
42776 * QCmode: Machine Modes. (line 197)
42777 * QFmode: Machine Modes. (line 54)
42778 * QImode: Machine Modes. (line 25)
42779 * QImode, in insn: Insns. (line 231)
42780 * QQmode: Machine Modes. (line 103)
42781 * qualified type: Types. (line 6)
42782 * querying function unit reservations: Processor pipeline description.
42784 * question mark: Multi-Alternative. (line 41)
42785 * quotient: Arithmetic. (line 111)
42786 * r in constraint: Simple Constraints. (line 56)
42787 * RANGE_TEST_NON_SHORT_CIRCUIT: Costs. (line 204)
42788 * RDIV_EXPR: Expression trees. (line 6)
42789 * READONLY_DATA_SECTION_ASM_OP: Sections. (line 63)
42790 * real operands: SSA Operands. (line 6)
42791 * REAL_ARITHMETIC: Floating Point. (line 66)
42792 * REAL_CST: Expression trees. (line 6)
42793 * REAL_LIBGCC_SPEC: Driver. (line 187)
42794 * REAL_NM_FILE_NAME: Macros for Initialization.
42796 * REAL_TYPE: Types. (line 6)
42797 * REAL_VALUE_ABS: Floating Point. (line 82)
42798 * REAL_VALUE_ATOF: Floating Point. (line 50)
42799 * REAL_VALUE_FIX: Floating Point. (line 41)
42800 * REAL_VALUE_FROM_INT: Floating Point. (line 99)
42801 * REAL_VALUE_ISINF: Floating Point. (line 59)
42802 * REAL_VALUE_ISNAN: Floating Point. (line 62)
42803 * REAL_VALUE_NEGATE: Floating Point. (line 79)
42804 * REAL_VALUE_NEGATIVE: Floating Point. (line 56)
42805 * REAL_VALUE_TO_INT: Floating Point. (line 93)
42806 * REAL_VALUE_TO_TARGET_DECIMAL128: Data Output. (line 144)
42807 * REAL_VALUE_TO_TARGET_DECIMAL32: Data Output. (line 142)
42808 * REAL_VALUE_TO_TARGET_DECIMAL64: Data Output. (line 143)
42809 * REAL_VALUE_TO_TARGET_DOUBLE: Data Output. (line 140)
42810 * REAL_VALUE_TO_TARGET_LONG_DOUBLE: Data Output. (line 141)
42811 * REAL_VALUE_TO_TARGET_SINGLE: Data Output. (line 139)
42812 * REAL_VALUE_TRUNCATE: Floating Point. (line 86)
42813 * REAL_VALUE_TYPE: Floating Point. (line 26)
42814 * REAL_VALUE_UNSIGNED_FIX: Floating Point. (line 45)
42815 * REAL_VALUES_EQUAL: Floating Point. (line 32)
42816 * REAL_VALUES_LESS: Floating Point. (line 38)
42817 * REALPART_EXPR: Expression trees. (line 6)
42818 * recog_data.operand: Instruction Output. (line 39)
42819 * recognizing insns: RTL Template. (line 6)
42820 * RECORD_TYPE <1>: Classes. (line 6)
42821 * RECORD_TYPE: Types. (line 6)
42822 * redirect_edge_and_branch: Profile information.
42824 * redirect_edge_and_branch, redirect_jump: Maintaining the CFG.
42826 * reduc_smax_M instruction pattern: Standard Names. (line 240)
42827 * reduc_smin_M instruction pattern: Standard Names. (line 240)
42828 * reduc_splus_M instruction pattern: Standard Names. (line 252)
42829 * reduc_umax_M instruction pattern: Standard Names. (line 246)
42830 * reduc_umin_M instruction pattern: Standard Names. (line 246)
42831 * reduc_uplus_M instruction pattern: Standard Names. (line 258)
42832 * reference: Types. (line 6)
42833 * REFERENCE_TYPE: Types. (line 6)
42834 * reg: Regs and Memory. (line 9)
42835 * reg and /f: Flags. (line 112)
42836 * reg and /i: Flags. (line 107)
42837 * reg and /v: Flags. (line 116)
42838 * reg, RTL sharing: Sharing. (line 17)
42839 * REG_ALLOC_ORDER: Allocation Order. (line 9)
42840 * REG_BR_PRED: Insns. (line 491)
42841 * REG_BR_PROB: Insns. (line 485)
42842 * REG_BR_PROB_BASE, BB_FREQ_BASE, count: Profile information.
42844 * REG_BR_PROB_BASE, EDGE_FREQUENCY: Profile information.
42846 * REG_CC_SETTER: Insns. (line 456)
42847 * REG_CC_USER: Insns. (line 456)
42848 * REG_CLASS_CONTENTS: Register Classes. (line 86)
42849 * reg_class_contents: Register Basics. (line 59)
42850 * REG_CLASS_FROM_CONSTRAINT: Old Constraints. (line 35)
42851 * REG_CLASS_FROM_LETTER: Old Constraints. (line 27)
42852 * REG_CLASS_NAMES: Register Classes. (line 81)
42853 * REG_CROSSING_JUMP: Insns. (line 368)
42854 * REG_DEAD: Insns. (line 320)
42855 * REG_DEAD, REG_UNUSED: Liveness information.
42857 * REG_DEP_ANTI: Insns. (line 478)
42858 * REG_DEP_OUTPUT: Insns. (line 474)
42859 * REG_DEP_TRUE: Insns. (line 471)
42860 * REG_EH_REGION, EDGE_ABNORMAL_CALL: Edges. (line 110)
42861 * REG_EQUAL: Insns. (line 384)
42862 * REG_EQUIV: Insns. (line 384)
42863 * REG_EXPR: Special Accessors. (line 46)
42864 * REG_FRAME_RELATED_EXPR: Insns. (line 497)
42865 * REG_FUNCTION_VALUE_P: Flags. (line 107)
42866 * REG_INC: Insns. (line 336)
42867 * reg_label and /v: Flags. (line 65)
42868 * REG_LABEL_OPERAND: Insns. (line 350)
42869 * REG_LABEL_TARGET: Insns. (line 359)
42870 * reg_names <1>: Instruction Output. (line 80)
42871 * reg_names: Register Basics. (line 59)
42872 * REG_NONNEG: Insns. (line 342)
42873 * REG_NOTE_KIND: Insns. (line 309)
42874 * REG_NOTES: Insns. (line 283)
42875 * REG_OFFSET: Special Accessors. (line 50)
42876 * REG_OK_STRICT: Addressing Modes. (line 67)
42877 * REG_PARM_STACK_SPACE: Stack Arguments. (line 56)
42878 * REG_PARM_STACK_SPACE, and FUNCTION_ARG: Register Arguments.
42880 * REG_POINTER: Flags. (line 112)
42881 * REG_SETJMP: Insns. (line 378)
42882 * REG_UNUSED: Insns. (line 329)
42883 * REG_USERVAR_P: Flags. (line 116)
42884 * regclass_for_constraint: C Constraint Interface.
42886 * register allocation order: Allocation Order. (line 6)
42887 * register class definitions: Register Classes. (line 6)
42888 * register class preference constraints: Class Preferences. (line 6)
42889 * register pairs: Values in Registers.
42891 * Register Transfer Language (RTL): RTL. (line 6)
42892 * register usage: Registers. (line 6)
42893 * REGISTER_MOVE_COST: Costs. (line 10)
42894 * REGISTER_NAMES: Instruction Output. (line 9)
42895 * register_operand: Machine-Independent Predicates.
42897 * REGISTER_PREFIX: Instruction Output. (line 124)
42898 * REGISTER_TARGET_PRAGMAS: Misc. (line 382)
42899 * registers arguments: Register Arguments. (line 6)
42900 * registers in constraints: Simple Constraints. (line 56)
42901 * REGMODE_NATURAL_SIZE: Values in Registers.
42903 * REGNO_MODE_CODE_OK_FOR_BASE_P: Register Classes. (line 170)
42904 * REGNO_MODE_OK_FOR_BASE_P: Register Classes. (line 146)
42905 * REGNO_MODE_OK_FOR_REG_BASE_P: Register Classes. (line 157)
42906 * REGNO_OK_FOR_BASE_P: Register Classes. (line 140)
42907 * REGNO_OK_FOR_INDEX_P: Register Classes. (line 181)
42908 * REGNO_REG_CLASS: Register Classes. (line 101)
42909 * regs_ever_live: Function Entry. (line 21)
42910 * regular expressions: Processor pipeline description.
42912 * relative costs: Costs. (line 6)
42913 * RELATIVE_PREFIX_NOT_LINKDIR: Driver. (line 325)
42914 * reload_completed: Standard Names. (line 1040)
42915 * reload_in instruction pattern: Standard Names. (line 99)
42916 * reload_in_progress: Standard Names. (line 57)
42917 * reload_out instruction pattern: Standard Names. (line 99)
42918 * reloading: RTL passes. (line 191)
42919 * remainder: Arithmetic. (line 131)
42920 * remainderM3 instruction pattern: Standard Names. (line 472)
42921 * reorder: GTY Options. (line 210)
42922 * representation of RTL: RTL. (line 6)
42923 * reservation delays: Processor pipeline description.
42925 * rest_of_decl_compilation: Parsing pass. (line 52)
42926 * rest_of_type_compilation: Parsing pass. (line 52)
42927 * restore_stack_block instruction pattern: Standard Names. (line 1174)
42928 * restore_stack_function instruction pattern: Standard Names.
42930 * restore_stack_nonlocal instruction pattern: Standard Names.
42932 * RESULT_DECL: Declarations. (line 6)
42933 * return: Side Effects. (line 72)
42934 * return instruction pattern: Standard Names. (line 1027)
42935 * return values in registers: Scalar Return. (line 6)
42936 * RETURN_ADDR_IN_PREVIOUS_FRAME: Frame Layout. (line 135)
42937 * RETURN_ADDR_OFFSET: Exception Handling. (line 60)
42938 * RETURN_ADDR_RTX: Frame Layout. (line 124)
42939 * RETURN_ADDRESS_POINTER_REGNUM: Frame Registers. (line 51)
42940 * RETURN_EXPR: Function Bodies. (line 6)
42941 * RETURN_POPS_ARGS: Stack Arguments. (line 90)
42942 * RETURN_STMT: Function Bodies. (line 6)
42943 * return_val: Flags. (line 294)
42944 * return_val, in call_insn: Flags. (line 24)
42945 * return_val, in mem: Flags. (line 85)
42946 * return_val, in reg: Flags. (line 107)
42947 * return_val, in symbol_ref: Flags. (line 220)
42948 * returning aggregate values: Aggregate Return. (line 6)
42949 * returning structures and unions: Interface. (line 10)
42950 * reverse probability: Profile information.
42952 * REVERSE_CONDEXEC_PREDICATES_P: Condition Code. (line 129)
42953 * REVERSE_CONDITION: Condition Code. (line 116)
42954 * REVERSIBLE_CC_MODE: Condition Code. (line 102)
42955 * right rotate: Arithmetic. (line 190)
42956 * right shift: Arithmetic. (line 185)
42957 * rintM2 instruction pattern: Standard Names. (line 572)
42958 * RISC: Processor pipeline description.
42960 * roots, marking: GGC Roots. (line 6)
42961 * rotate: Arithmetic. (line 190)
42962 * rotatert: Arithmetic. (line 190)
42963 * rotlM3 instruction pattern: Standard Names. (line 441)
42964 * rotrM3 instruction pattern: Standard Names. (line 441)
42965 * ROUND_DIV_EXPR: Expression trees. (line 6)
42966 * ROUND_MOD_EXPR: Expression trees. (line 6)
42967 * ROUND_TOWARDS_ZERO: Storage Layout. (line 460)
42968 * ROUND_TYPE_ALIGN: Storage Layout. (line 411)
42969 * roundM2 instruction pattern: Standard Names. (line 548)
42970 * RSHIFT_EXPR: Expression trees. (line 6)
42971 * RTL addition: Arithmetic. (line 14)
42972 * RTL addition with signed saturation: Arithmetic. (line 14)
42973 * RTL addition with unsigned saturation: Arithmetic. (line 14)
42974 * RTL classes: RTL Classes. (line 6)
42975 * RTL comparison: Arithmetic. (line 43)
42976 * RTL comparison operations: Comparisons. (line 6)
42977 * RTL constant expression types: Constants. (line 6)
42978 * RTL constants: Constants. (line 6)
42979 * RTL declarations: RTL Declarations. (line 6)
42980 * RTL difference: Arithmetic. (line 36)
42981 * RTL expression: RTL Objects. (line 6)
42982 * RTL expressions for arithmetic: Arithmetic. (line 6)
42983 * RTL format: RTL Classes. (line 71)
42984 * RTL format characters: RTL Classes. (line 76)
42985 * RTL function-call insns: Calls. (line 6)
42986 * RTL insn template: RTL Template. (line 6)
42987 * RTL integers: RTL Objects. (line 6)
42988 * RTL memory expressions: Regs and Memory. (line 6)
42989 * RTL object types: RTL Objects. (line 6)
42990 * RTL postdecrement: Incdec. (line 6)
42991 * RTL postincrement: Incdec. (line 6)
42992 * RTL predecrement: Incdec. (line 6)
42993 * RTL preincrement: Incdec. (line 6)
42994 * RTL register expressions: Regs and Memory. (line 6)
42995 * RTL representation: RTL. (line 6)
42996 * RTL side effect expressions: Side Effects. (line 6)
42997 * RTL strings: RTL Objects. (line 6)
42998 * RTL structure sharing assumptions: Sharing. (line 6)
42999 * RTL subtraction: Arithmetic. (line 36)
43000 * RTL subtraction with signed saturation: Arithmetic. (line 36)
43001 * RTL subtraction with unsigned saturation: Arithmetic. (line 36)
43002 * RTL sum: Arithmetic. (line 14)
43003 * RTL vectors: RTL Objects. (line 6)
43004 * RTL_CONST_CALL_P: Flags. (line 19)
43005 * RTL_CONST_OR_PURE_CALL_P: Flags. (line 29)
43006 * RTL_LOOPING_CONST_OR_PURE_CALL_P: Flags. (line 33)
43007 * RTL_PURE_CALL_P: Flags. (line 24)
43008 * RTX (See RTL): RTL Objects. (line 6)
43009 * RTX codes, classes of: RTL Classes. (line 6)
43010 * RTX_FRAME_RELATED_P: Flags. (line 125)
43011 * run-time conventions: Interface. (line 6)
43012 * run-time target specification: Run-time Target. (line 6)
43013 * s in constraint: Simple Constraints. (line 92)
43014 * same_type_p: Types. (line 148)
43015 * SAmode: Machine Modes. (line 148)
43016 * sat_fract: Conversions. (line 90)
43017 * satfractMN2 instruction pattern: Standard Names. (line 843)
43018 * satfractunsMN2 instruction pattern: Standard Names. (line 856)
43019 * satisfies_constraint_: C Constraint Interface.
43021 * SAVE_EXPR: Expression trees. (line 6)
43022 * save_stack_block instruction pattern: Standard Names. (line 1174)
43023 * save_stack_function instruction pattern: Standard Names. (line 1174)
43024 * save_stack_nonlocal instruction pattern: Standard Names. (line 1174)
43025 * SBSS_SECTION_ASM_OP: Sections. (line 77)
43026 * Scalar evolutions: Scalar evolutions. (line 6)
43027 * scalars, returned as values: Scalar Return. (line 6)
43028 * SCHED_GROUP_P: Flags. (line 166)
43029 * SCmode: Machine Modes. (line 197)
43030 * sCOND instruction pattern: Standard Names. (line 911)
43031 * scratch: Regs and Memory. (line 298)
43032 * scratch operands: Regs and Memory. (line 298)
43033 * scratch, RTL sharing: Sharing. (line 35)
43034 * scratch_operand: Machine-Independent Predicates.
43036 * SDATA_SECTION_ASM_OP: Sections. (line 58)
43037 * SDB_ALLOW_FORWARD_REFERENCES: SDB and DWARF. (line 81)
43038 * SDB_ALLOW_UNKNOWN_REFERENCES: SDB and DWARF. (line 76)
43039 * SDB_DEBUGGING_INFO: SDB and DWARF. (line 9)
43040 * SDB_DELIM: SDB and DWARF. (line 69)
43041 * SDB_OUTPUT_SOURCE_LINE: SDB and DWARF. (line 86)
43042 * SDmode: Machine Modes. (line 85)
43043 * sdot_prodM instruction pattern: Standard Names. (line 264)
43044 * search options: Including Patterns. (line 44)
43045 * SECONDARY_INPUT_RELOAD_CLASS: Register Classes. (line 335)
43046 * SECONDARY_MEMORY_NEEDED: Register Classes. (line 391)
43047 * SECONDARY_MEMORY_NEEDED_MODE: Register Classes. (line 410)
43048 * SECONDARY_MEMORY_NEEDED_RTX: Register Classes. (line 401)
43049 * SECONDARY_OUTPUT_RELOAD_CLASS: Register Classes. (line 336)
43050 * SECONDARY_RELOAD_CLASS: Register Classes. (line 334)
43051 * SELECT_CC_MODE: Condition Code. (line 68)
43052 * sequence: Side Effects. (line 254)
43053 * Sequence iterators: Sequence iterators. (line 6)
43054 * set: Side Effects. (line 15)
43055 * set and /f: Flags. (line 125)
43056 * SET_ASM_OP: Label Output. (line 378)
43057 * set_attr: Tagging Insns. (line 31)
43058 * set_attr_alternative: Tagging Insns. (line 49)
43059 * set_bb_seq: GIMPLE sequences. (line 76)
43060 * SET_BY_PIECES_P: Costs. (line 145)
43061 * SET_DEST: Side Effects. (line 69)
43062 * SET_IS_RETURN_P: Flags. (line 175)
43063 * SET_LABEL_KIND: Insns. (line 140)
43064 * set_optab_libfunc: Library Calls. (line 15)
43065 * SET_RATIO: Costs. (line 136)
43066 * SET_SRC: Side Effects. (line 69)
43067 * SET_TYPE_STRUCTURAL_EQUALITY: Types. (line 6)
43068 * setmemM instruction pattern: Standard Names. (line 715)
43069 * SETUP_FRAME_ADDRESSES: Frame Layout. (line 102)
43070 * SF_SIZE: Type Layout. (line 129)
43071 * SFmode: Machine Modes. (line 66)
43072 * sharing of RTL components: Sharing. (line 6)
43073 * shift: Arithmetic. (line 168)
43074 * SHIFT_COUNT_TRUNCATED: Misc. (line 127)
43075 * SHLIB_SUFFIX: Macros for Initialization.
43077 * SHORT_ACCUM_TYPE_SIZE: Type Layout. (line 83)
43078 * SHORT_FRACT_TYPE_SIZE: Type Layout. (line 63)
43079 * SHORT_IMMEDIATES_SIGN_EXTEND: Misc. (line 96)
43080 * SHORT_TYPE_SIZE: Type Layout. (line 16)
43081 * sibcall_epilogue instruction pattern: Standard Names. (line 1364)
43082 * sibling call: Edges. (line 122)
43083 * SIBLING_CALL_P: Flags. (line 179)
43084 * sign_extend: Conversions. (line 23)
43085 * sign_extract: Bit-Fields. (line 8)
43086 * sign_extract, canonicalization of: Insn Canonicalizations.
43088 * signed division: Arithmetic. (line 111)
43089 * signed division with signed saturation: Arithmetic. (line 111)
43090 * signed maximum: Arithmetic. (line 136)
43091 * signed minimum: Arithmetic. (line 136)
43092 * SImode: Machine Modes. (line 37)
43093 * simple constraints: Simple Constraints. (line 6)
43094 * sincos math function, implicit usage: Library Calls. (line 84)
43095 * sinM2 instruction pattern: Standard Names. (line 489)
43096 * SIZE_ASM_OP: Label Output. (line 23)
43097 * SIZE_TYPE: Type Layout. (line 168)
43098 * skip: GTY Options. (line 77)
43099 * SLOW_BYTE_ACCESS: Costs. (line 66)
43100 * SLOW_UNALIGNED_ACCESS: Costs. (line 81)
43101 * SMALL_REGISTER_CLASSES: Register Classes. (line 433)
43102 * smax: Arithmetic. (line 136)
43103 * smin: Arithmetic. (line 136)
43104 * sms, swing, software pipelining: RTL passes. (line 140)
43105 * smulM3_highpart instruction pattern: Standard Names. (line 356)
43106 * soft float library: Soft float library routines.
43108 * special: GTY Options. (line 230)
43109 * special predicates: Predicates. (line 31)
43110 * SPECS: Target Fragment. (line 108)
43111 * speed of instructions: Costs. (line 6)
43112 * split_block: Maintaining the CFG.
43114 * splitting instructions: Insn Splitting. (line 6)
43115 * SQmode: Machine Modes. (line 111)
43116 * sqrt: Arithmetic. (line 198)
43117 * sqrtM2 instruction pattern: Standard Names. (line 455)
43118 * square root: Arithmetic. (line 198)
43119 * ss_ashift: Arithmetic. (line 168)
43120 * ss_div: Arithmetic. (line 111)
43121 * ss_minus: Arithmetic. (line 36)
43122 * ss_mult: Arithmetic. (line 92)
43123 * ss_neg: Arithmetic. (line 81)
43124 * ss_plus: Arithmetic. (line 14)
43125 * ss_truncate: Conversions. (line 43)
43126 * SSA: SSA. (line 6)
43127 * SSA_NAME_DEF_STMT: SSA. (line 221)
43128 * SSA_NAME_VERSION: SSA. (line 226)
43129 * ssaddM3 instruction pattern: Standard Names. (line 222)
43130 * ssashlM3 instruction pattern: Standard Names. (line 431)
43131 * ssdivM3 instruction pattern: Standard Names. (line 222)
43132 * ssmaddMN4 instruction pattern: Standard Names. (line 379)
43133 * ssmsubMN4 instruction pattern: Standard Names. (line 403)
43134 * ssmulM3 instruction pattern: Standard Names. (line 222)
43135 * ssnegM2 instruction pattern: Standard Names. (line 449)
43136 * sssubM3 instruction pattern: Standard Names. (line 222)
43137 * ssum_widenM3 instruction pattern: Standard Names. (line 274)
43138 * stack arguments: Stack Arguments. (line 6)
43139 * stack frame layout: Frame Layout. (line 6)
43140 * stack smashing protection: Stack Smashing Protection.
43142 * STACK_ALIGNMENT_NEEDED: Frame Layout. (line 48)
43143 * STACK_BOUNDARY: Storage Layout. (line 150)
43144 * STACK_CHECK_BUILTIN: Stack Checking. (line 32)
43145 * STACK_CHECK_FIXED_FRAME_SIZE: Stack Checking. (line 77)
43146 * STACK_CHECK_MAX_FRAME_SIZE: Stack Checking. (line 68)
43147 * STACK_CHECK_MAX_VAR_SIZE: Stack Checking. (line 84)
43148 * STACK_CHECK_PROBE_INTERVAL: Stack Checking. (line 46)
43149 * STACK_CHECK_PROBE_LOAD: Stack Checking. (line 53)
43150 * STACK_CHECK_PROTECT: Stack Checking. (line 59)
43151 * STACK_CHECK_STATIC_BUILTIN: Stack Checking. (line 39)
43152 * STACK_DYNAMIC_OFFSET: Frame Layout. (line 75)
43153 * STACK_DYNAMIC_OFFSET and virtual registers: Regs and Memory.
43155 * STACK_GROWS_DOWNWARD: Frame Layout. (line 9)
43156 * STACK_PARMS_IN_REG_PARM_AREA: Stack Arguments. (line 81)
43157 * STACK_POINTER_OFFSET: Frame Layout. (line 58)
43158 * STACK_POINTER_OFFSET and virtual registers: Regs and Memory.
43160 * STACK_POINTER_REGNUM: Frame Registers. (line 9)
43161 * STACK_POINTER_REGNUM and virtual registers: Regs and Memory.
43163 * stack_pointer_rtx: Frame Registers. (line 85)
43164 * stack_protect_set instruction pattern: Standard Names. (line 1536)
43165 * stack_protect_test instruction pattern: Standard Names. (line 1546)
43166 * STACK_PUSH_CODE: Frame Layout. (line 17)
43167 * STACK_REGS: Stack Registers. (line 20)
43168 * STACK_SAVEAREA_MODE: Storage Layout. (line 427)
43169 * STACK_SIZE_MODE: Storage Layout. (line 439)
43170 * STACK_SLOT_ALIGNMENT: Storage Layout. (line 265)
43171 * standard pattern names: Standard Names. (line 6)
43172 * STANDARD_INCLUDE_COMPONENT: Driver. (line 425)
43173 * STANDARD_INCLUDE_DIR: Driver. (line 417)
43174 * STANDARD_STARTFILE_PREFIX: Driver. (line 337)
43175 * STANDARD_STARTFILE_PREFIX_1: Driver. (line 344)
43176 * STANDARD_STARTFILE_PREFIX_2: Driver. (line 351)
43177 * STARTFILE_SPEC: Driver. (line 210)
43178 * STARTING_FRAME_OFFSET: Frame Layout. (line 39)
43179 * STARTING_FRAME_OFFSET and virtual registers: Regs and Memory.
43181 * Statement and operand traversals: Statement and operand traversals.
43183 * Statement Sequences: Statement Sequences.
43185 * Statements: Statements. (line 6)
43186 * statements: Function Bodies. (line 6)
43187 * Static profile estimation: Profile information.
43189 * static single assignment: SSA. (line 6)
43190 * STATIC_CHAIN: Frame Registers. (line 77)
43191 * STATIC_CHAIN_INCOMING: Frame Registers. (line 78)
43192 * STATIC_CHAIN_INCOMING_REGNUM: Frame Registers. (line 64)
43193 * STATIC_CHAIN_REGNUM: Frame Registers. (line 63)
43194 * stdarg.h and register arguments: Register Arguments. (line 47)
43195 * STDC_0_IN_SYSTEM_HEADERS: Misc. (line 365)
43196 * STMT_EXPR: Expression trees. (line 6)
43197 * STMT_IS_FULL_EXPR_P: Function Bodies. (line 22)
43198 * storage layout: Storage Layout. (line 6)
43199 * STORE_BY_PIECES_P: Costs. (line 152)
43200 * STORE_FLAG_VALUE: Misc. (line 216)
43201 * store_multiple instruction pattern: Standard Names. (line 160)
43202 * strcpy: Storage Layout. (line 235)
43203 * STRICT_ALIGNMENT: Storage Layout. (line 309)
43204 * strict_low_part: RTL Declarations. (line 9)
43205 * strict_memory_address_p: Addressing Modes. (line 179)
43206 * STRING_CST: Expression trees. (line 6)
43207 * STRING_POOL_ADDRESS_P: Flags. (line 183)
43208 * strlenM instruction pattern: Standard Names. (line 778)
43209 * structure value address: Aggregate Return. (line 6)
43210 * STRUCTURE_SIZE_BOUNDARY: Storage Layout. (line 301)
43211 * structures, returning: Interface. (line 10)
43212 * subM3 instruction pattern: Standard Names. (line 222)
43213 * SUBOBJECT: Function Bodies. (line 6)
43214 * SUBOBJECT_CLEANUP: Function Bodies. (line 6)
43215 * subreg: Regs and Memory. (line 97)
43216 * subreg and /s: Flags. (line 205)
43217 * subreg and /u: Flags. (line 198)
43218 * subreg and /u and /v: Flags. (line 188)
43219 * subreg, in strict_low_part: RTL Declarations. (line 9)
43220 * SUBREG_BYTE: Regs and Memory. (line 289)
43221 * SUBREG_PROMOTED_UNSIGNED_P: Flags. (line 188)
43222 * SUBREG_PROMOTED_UNSIGNED_SET: Flags. (line 198)
43223 * SUBREG_PROMOTED_VAR_P: Flags. (line 205)
43224 * SUBREG_REG: Regs and Memory. (line 289)
43225 * SUCCESS_EXIT_CODE: Host Misc. (line 12)
43226 * SUPPORTS_INIT_PRIORITY: Macros for Initialization.
43228 * SUPPORTS_ONE_ONLY: Label Output. (line 227)
43229 * SUPPORTS_WEAK: Label Output. (line 208)
43230 * SWITCH_BODY: Function Bodies. (line 6)
43231 * SWITCH_COND: Function Bodies. (line 6)
43232 * SWITCH_CURTAILS_COMPILATION: Driver. (line 33)
43233 * SWITCH_STMT: Function Bodies. (line 6)
43234 * SWITCH_TAKES_ARG: Driver. (line 9)
43235 * SWITCHES_NEED_SPACES: Driver. (line 47)
43236 * SYMBOL_FLAG_ANCHOR: Special Accessors. (line 106)
43237 * SYMBOL_FLAG_EXTERNAL: Special Accessors. (line 88)
43238 * SYMBOL_FLAG_FUNCTION: Special Accessors. (line 81)
43239 * SYMBOL_FLAG_HAS_BLOCK_INFO: Special Accessors. (line 102)
43240 * SYMBOL_FLAG_LOCAL: Special Accessors. (line 84)
43241 * SYMBOL_FLAG_SMALL: Special Accessors. (line 93)
43242 * SYMBOL_FLAG_TLS_SHIFT: Special Accessors. (line 97)
43243 * symbol_ref: Constants. (line 76)
43244 * symbol_ref and /f: Flags. (line 183)
43245 * symbol_ref and /i: Flags. (line 220)
43246 * symbol_ref and /u: Flags. (line 10)
43247 * symbol_ref and /v: Flags. (line 224)
43248 * symbol_ref, RTL sharing: Sharing. (line 20)
43249 * SYMBOL_REF_ANCHOR_P: Special Accessors. (line 106)
43250 * SYMBOL_REF_BLOCK: Special Accessors. (line 119)
43251 * SYMBOL_REF_BLOCK_OFFSET: Special Accessors. (line 124)
43252 * SYMBOL_REF_CONSTANT: Special Accessors. (line 67)
43253 * SYMBOL_REF_DATA: Special Accessors. (line 71)
43254 * SYMBOL_REF_DECL: Special Accessors. (line 55)
43255 * SYMBOL_REF_EXTERNAL_P: Special Accessors. (line 88)
43256 * SYMBOL_REF_FLAG: Flags. (line 224)
43257 * SYMBOL_REF_FLAG, in TARGET_ENCODE_SECTION_INFO: Sections. (line 259)
43258 * SYMBOL_REF_FLAGS: Special Accessors. (line 75)
43259 * SYMBOL_REF_FUNCTION_P: Special Accessors. (line 81)
43260 * SYMBOL_REF_HAS_BLOCK_INFO_P: Special Accessors. (line 102)
43261 * SYMBOL_REF_LOCAL_P: Special Accessors. (line 84)
43262 * SYMBOL_REF_SMALL_P: Special Accessors. (line 93)
43263 * SYMBOL_REF_TLS_MODEL: Special Accessors. (line 97)
43264 * SYMBOL_REF_USED: Flags. (line 215)
43265 * SYMBOL_REF_WEAK: Flags. (line 220)
43266 * symbolic label: Sharing. (line 20)
43267 * sync_addMODE instruction pattern: Standard Names. (line 1450)
43268 * sync_andMODE instruction pattern: Standard Names. (line 1450)
43269 * sync_compare_and_swap_ccMODE instruction pattern: Standard Names.
43271 * sync_compare_and_swapMODE instruction pattern: Standard Names.
43273 * sync_iorMODE instruction pattern: Standard Names. (line 1450)
43274 * sync_lock_releaseMODE instruction pattern: Standard Names. (line 1517)
43275 * sync_lock_test_and_setMODE instruction pattern: Standard Names.
43277 * sync_nandMODE instruction pattern: Standard Names. (line 1450)
43278 * sync_new_addMODE instruction pattern: Standard Names. (line 1484)
43279 * sync_new_andMODE instruction pattern: Standard Names. (line 1484)
43280 * sync_new_iorMODE instruction pattern: Standard Names. (line 1484)
43281 * sync_new_nandMODE instruction pattern: Standard Names. (line 1484)
43282 * sync_new_subMODE instruction pattern: Standard Names. (line 1484)
43283 * sync_new_xorMODE instruction pattern: Standard Names. (line 1484)
43284 * sync_old_addMODE instruction pattern: Standard Names. (line 1467)
43285 * sync_old_andMODE instruction pattern: Standard Names. (line 1467)
43286 * sync_old_iorMODE instruction pattern: Standard Names. (line 1467)
43287 * sync_old_nandMODE instruction pattern: Standard Names. (line 1467)
43288 * sync_old_subMODE instruction pattern: Standard Names. (line 1467)
43289 * sync_old_xorMODE instruction pattern: Standard Names. (line 1467)
43290 * sync_subMODE instruction pattern: Standard Names. (line 1450)
43291 * sync_xorMODE instruction pattern: Standard Names. (line 1450)
43292 * SYSROOT_HEADERS_SUFFIX_SPEC: Driver. (line 239)
43293 * SYSROOT_SUFFIX_SPEC: Driver. (line 234)
43294 * SYSTEM_INCLUDE_DIR: Driver. (line 408)
43295 * t-TARGET: Target Fragment. (line 6)
43296 * table jump: Basic Blocks. (line 57)
43297 * tablejump instruction pattern: Standard Names. (line 1102)
43298 * tag: GTY Options. (line 82)
43299 * tagging insns: Tagging Insns. (line 6)
43300 * tail calls: Tail Calls. (line 6)
43301 * TAmode: Machine Modes. (line 156)
43302 * target attributes: Target Attributes. (line 6)
43303 * target description macros: Target Macros. (line 6)
43304 * target functions: Target Structure. (line 6)
43305 * target hooks: Target Structure. (line 6)
43306 * target makefile fragment: Target Fragment. (line 6)
43307 * target specifications: Run-time Target. (line 6)
43308 * TARGET_ADDRESS_COST: Costs. (line 236)
43309 * TARGET_ALIGN_ANON_BITFIELD: Storage Layout. (line 386)
43310 * TARGET_ALLOCATE_INITIAL_VALUE: Misc. (line 720)
43311 * TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS: Misc. (line 959)
43312 * TARGET_ARG_PARTIAL_BYTES: Register Arguments. (line 83)
43313 * TARGET_ARM_EABI_UNWINDER: Exception Region Output.
43315 * TARGET_ASM_ALIGNED_DI_OP: Data Output. (line 10)
43316 * TARGET_ASM_ALIGNED_HI_OP: Data Output. (line 8)
43317 * TARGET_ASM_ALIGNED_SI_OP: Data Output. (line 9)
43318 * TARGET_ASM_ALIGNED_TI_OP: Data Output. (line 11)
43319 * TARGET_ASM_ASSEMBLE_VISIBILITY: Label Output. (line 239)
43320 * TARGET_ASM_BYTE_OP: Data Output. (line 7)
43321 * TARGET_ASM_CAN_OUTPUT_MI_THUNK: Function Entry. (line 237)
43322 * TARGET_ASM_CLOSE_PAREN: Data Output. (line 130)
43323 * TARGET_ASM_CONSTRUCTOR: Macros for Initialization.
43325 * TARGET_ASM_DESTRUCTOR: Macros for Initialization.
43327 * TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL: Dispatch Tables. (line 74)
43328 * TARGET_ASM_EMIT_UNWIND_LABEL: Dispatch Tables. (line 63)
43329 * TARGET_ASM_EXTERNAL_LIBCALL: Label Output. (line 274)
43330 * TARGET_ASM_FILE_END: File Framework. (line 37)
43331 * TARGET_ASM_FILE_START: File Framework. (line 9)
43332 * TARGET_ASM_FILE_START_APP_OFF: File Framework. (line 17)
43333 * TARGET_ASM_FILE_START_FILE_DIRECTIVE: File Framework. (line 31)
43334 * TARGET_ASM_FUNCTION_BEGIN_EPILOGUE: Function Entry. (line 61)
43335 * TARGET_ASM_FUNCTION_END_PROLOGUE: Function Entry. (line 55)
43336 * TARGET_ASM_FUNCTION_EPILOGUE: Function Entry. (line 68)
43337 * TARGET_ASM_FUNCTION_EPILOGUE and trampolines: Trampolines. (line 70)
43338 * TARGET_ASM_FUNCTION_PROLOGUE: Function Entry. (line 11)
43339 * TARGET_ASM_FUNCTION_PROLOGUE and trampolines: Trampolines. (line 70)
43340 * TARGET_ASM_FUNCTION_RODATA_SECTION: Sections. (line 206)
43341 * TARGET_ASM_GLOBALIZE_DECL_NAME: Label Output. (line 174)
43342 * TARGET_ASM_GLOBALIZE_LABEL: Label Output. (line 165)
43343 * TARGET_ASM_INIT_SECTIONS: Sections. (line 151)
43344 * TARGET_ASM_INTEGER: Data Output. (line 27)
43345 * TARGET_ASM_INTERNAL_LABEL: Label Output. (line 309)
43346 * TARGET_ASM_MARK_DECL_PRESERVED: Label Output. (line 280)
43347 * TARGET_ASM_NAMED_SECTION: File Framework. (line 89)
43348 * TARGET_ASM_OPEN_PAREN: Data Output. (line 129)
43349 * TARGET_ASM_OUTPUT_ANCHOR: Anchored Addresses. (line 44)
43350 * TARGET_ASM_OUTPUT_DWARF_DTPREL: SDB and DWARF. (line 58)
43351 * TARGET_ASM_OUTPUT_MI_THUNK: Function Entry. (line 195)
43352 * TARGET_ASM_RECORD_GCC_SWITCHES: File Framework. (line 122)
43353 * TARGET_ASM_RECORD_GCC_SWITCHES_SECTION: File Framework. (line 166)
43354 * TARGET_ASM_SELECT_RTX_SECTION: Sections. (line 214)
43355 * TARGET_ASM_SELECT_SECTION: Sections. (line 172)
43356 * TARGET_ASM_TTYPE: Exception Region Output.
43358 * TARGET_ASM_UNALIGNED_DI_OP: Data Output. (line 14)
43359 * TARGET_ASM_UNALIGNED_HI_OP: Data Output. (line 12)
43360 * TARGET_ASM_UNALIGNED_SI_OP: Data Output. (line 13)
43361 * TARGET_ASM_UNALIGNED_TI_OP: Data Output. (line 15)
43362 * TARGET_ASM_UNIQUE_SECTION: Sections. (line 193)
43363 * TARGET_ATTRIBUTE_TABLE: Target Attributes. (line 11)
43364 * TARGET_BINDS_LOCAL_P: Sections. (line 284)
43365 * TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED: Misc. (line 816)
43366 * TARGET_BRANCH_TARGET_REGISTER_CLASS: Misc. (line 808)
43367 * TARGET_BUILD_BUILTIN_VA_LIST: Register Arguments. (line 264)
43368 * TARGET_BUILTIN_RECIPROCAL: Addressing Modes. (line 240)
43369 * TARGET_BUILTIN_SETJMP_FRAME_VALUE: Frame Layout. (line 109)
43370 * TARGET_C99_FUNCTIONS: Library Calls. (line 77)
43371 * TARGET_CALLEE_COPIES: Register Arguments. (line 115)
43372 * TARGET_CAN_INLINE_P: Target Attributes. (line 126)
43373 * TARGET_CANNOT_FORCE_CONST_MEM: Addressing Modes. (line 221)
43374 * TARGET_CANNOT_MODIFY_JUMPS_P: Misc. (line 795)
43375 * TARGET_CANONICAL_VA_LIST_TYPE: Register Arguments. (line 273)
43376 * TARGET_COMMUTATIVE_P: Misc. (line 713)
43377 * TARGET_COMP_TYPE_ATTRIBUTES: Target Attributes. (line 19)
43378 * TARGET_CPU_CPP_BUILTINS: Run-time Target. (line 9)
43379 * TARGET_CXX_ADJUST_CLASS_AT_DEFINITION: C++ ABI. (line 87)
43380 * TARGET_CXX_CDTOR_RETURNS_THIS: C++ ABI. (line 38)
43381 * TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT: C++ ABI. (line 62)
43382 * TARGET_CXX_COOKIE_HAS_SIZE: C++ ABI. (line 25)
43383 * TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY: C++ ABI. (line 54)
43384 * TARGET_CXX_GET_COOKIE_SIZE: C++ ABI. (line 18)
43385 * TARGET_CXX_GUARD_MASK_BIT: C++ ABI. (line 12)
43386 * TARGET_CXX_GUARD_TYPE: C++ ABI. (line 7)
43387 * TARGET_CXX_IMPORT_EXPORT_CLASS: C++ ABI. (line 30)
43388 * TARGET_CXX_KEY_METHOD_MAY_BE_INLINE: C++ ABI. (line 43)
43389 * TARGET_CXX_LIBRARY_RTTI_COMDAT: C++ ABI. (line 69)
43390 * TARGET_CXX_USE_AEABI_ATEXIT: C++ ABI. (line 74)
43391 * TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT: C++ ABI. (line 80)
43392 * TARGET_DECIMAL_FLOAT_SUPPORTED_P: Storage Layout. (line 513)
43393 * TARGET_DECLSPEC: Target Attributes. (line 64)
43394 * TARGET_DEFAULT_PACK_STRUCT: Misc. (line 482)
43395 * TARGET_DEFAULT_SHORT_ENUMS: Type Layout. (line 160)
43396 * TARGET_DEFERRED_OUTPUT_DEFS: Label Output. (line 393)
43397 * TARGET_DELEGITIMIZE_ADDRESS: Addressing Modes. (line 212)
43398 * TARGET_DLLIMPORT_DECL_ATTRIBUTES: Target Attributes. (line 47)
43399 * TARGET_DWARF_CALLING_CONVENTION: SDB and DWARF. (line 18)
43400 * TARGET_DWARF_HANDLE_FRAME_UNSPEC: Frame Layout. (line 172)
43401 * TARGET_DWARF_REGISTER_SPAN: Exception Region Output.
43403 * TARGET_EDOM: Library Calls. (line 59)
43404 * TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS: Emulated TLS. (line 68)
43405 * TARGET_EMUTLS_GET_ADDRESS: Emulated TLS. (line 19)
43406 * TARGET_EMUTLS_REGISTER_COMMON: Emulated TLS. (line 24)
43407 * TARGET_EMUTLS_TMPL_PREFIX: Emulated TLS. (line 45)
43408 * TARGET_EMUTLS_TMPL_SECTION: Emulated TLS. (line 36)
43409 * TARGET_EMUTLS_VAR_ALIGN_FIXED: Emulated TLS. (line 63)
43410 * TARGET_EMUTLS_VAR_FIELDS: Emulated TLS. (line 49)
43411 * TARGET_EMUTLS_VAR_INIT: Emulated TLS. (line 57)
43412 * TARGET_EMUTLS_VAR_PREFIX: Emulated TLS. (line 41)
43413 * TARGET_EMUTLS_VAR_SECTION: Emulated TLS. (line 31)
43414 * TARGET_ENCODE_SECTION_INFO: Sections. (line 235)
43415 * TARGET_ENCODE_SECTION_INFO and address validation: Addressing Modes.
43417 * TARGET_ENCODE_SECTION_INFO usage: Instruction Output. (line 100)
43418 * TARGET_ENUM_VA_LIST: Scalar Return. (line 84)
43419 * TARGET_EXECUTABLE_SUFFIX: Misc. (line 769)
43420 * TARGET_EXPAND_BUILTIN: Misc. (line 665)
43421 * TARGET_EXPAND_BUILTIN_SAVEREGS: Varargs. (line 92)
43422 * TARGET_EXPAND_TO_RTL_HOOK: Storage Layout. (line 519)
43423 * TARGET_EXPR: Expression trees. (line 6)
43424 * TARGET_EXTRA_INCLUDES: Misc. (line 847)
43425 * TARGET_EXTRA_LIVE_ON_ENTRY: Tail Calls. (line 21)
43426 * TARGET_EXTRA_PRE_INCLUDES: Misc. (line 854)
43427 * TARGET_FIXED_CONDITION_CODE_REGS: Condition Code. (line 142)
43428 * TARGET_FIXED_POINT_SUPPORTED_P: Storage Layout. (line 516)
43429 * target_flags: Run-time Target. (line 52)
43430 * TARGET_FLT_EVAL_METHOD: Type Layout. (line 141)
43431 * TARGET_FN_ABI_VA_LIST: Register Arguments. (line 268)
43432 * TARGET_FOLD_BUILTIN: Misc. (line 685)
43433 * TARGET_FORMAT_TYPES: Misc. (line 874)
43434 * TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P: Target Attributes. (line 86)
43435 * TARGET_FUNCTION_OK_FOR_SIBCALL: Tail Calls. (line 8)
43436 * TARGET_FUNCTION_VALUE: Scalar Return. (line 11)
43437 * TARGET_GET_DRAP_RTX: Misc. (line 954)
43438 * TARGET_GIMPLIFY_VA_ARG_EXPR: Register Arguments. (line 279)
43439 * TARGET_HANDLE_C_OPTION: Run-time Target. (line 78)
43440 * TARGET_HANDLE_OPTION: Run-time Target. (line 61)
43441 * TARGET_HARD_REGNO_SCRATCH_OK: Values in Registers.
43443 * TARGET_HAS_SINCOS: Library Calls. (line 85)
43444 * TARGET_HAVE_CONDITIONAL_EXECUTION: Misc. (line 830)
43445 * TARGET_HAVE_CTORS_DTORS: Macros for Initialization.
43447 * TARGET_HAVE_NAMED_SECTIONS: File Framework. (line 99)
43448 * TARGET_HAVE_SWITCHABLE_BSS_SECTIONS: File Framework. (line 103)
43449 * TARGET_HELP: Run-time Target. (line 140)
43450 * TARGET_IN_SMALL_DATA_P: Sections. (line 276)
43451 * TARGET_INIT_BUILTINS: Misc. (line 647)
43452 * TARGET_INIT_DWARF_REG_SIZES_EXTRA: Exception Region Output.
43454 * TARGET_INIT_LIBFUNCS: Library Calls. (line 16)
43455 * TARGET_INSERT_ATTRIBUTES: Target Attributes. (line 73)
43456 * TARGET_INSTANTIATE_DECLS: Storage Layout. (line 527)
43457 * TARGET_INVALID_BINARY_OP: Misc. (line 927)
43458 * TARGET_INVALID_CONVERSION: Misc. (line 914)
43459 * TARGET_INVALID_UNARY_OP: Misc. (line 920)
43460 * TARGET_IRA_COVER_CLASSES: Register Classes. (line 496)
43461 * TARGET_LIB_INT_CMP_BIASED: Library Calls. (line 35)
43462 * TARGET_LIBGCC_CMP_RETURN_MODE: Storage Layout. (line 448)
43463 * TARGET_LIBGCC_SDATA_SECTION: Sections. (line 123)
43464 * TARGET_LIBGCC_SHIFT_COUNT_MODE: Storage Layout. (line 454)
43465 * TARGET_MACHINE_DEPENDENT_REORG: Misc. (line 632)
43466 * TARGET_MANGLE_DECL_ASSEMBLER_NAME: Sections. (line 225)
43467 * TARGET_MANGLE_TYPE: Storage Layout. (line 531)
43468 * TARGET_MD_ASM_CLOBBERS: Misc. (line 548)
43469 * TARGET_MEM_CONSTRAINT: Addressing Modes. (line 100)
43470 * TARGET_MEM_REF: Expression trees. (line 6)
43471 * TARGET_MERGE_DECL_ATTRIBUTES: Target Attributes. (line 39)
43472 * TARGET_MERGE_TYPE_ATTRIBUTES: Target Attributes. (line 31)
43473 * TARGET_MIN_DIVISIONS_FOR_RECIP_MUL: Misc. (line 106)
43474 * TARGET_MODE_REP_EXTENDED: Misc. (line 191)
43475 * TARGET_MS_BITFIELD_LAYOUT_P: Storage Layout. (line 486)
43476 * TARGET_MUST_PASS_IN_STACK: Register Arguments. (line 62)
43477 * TARGET_MUST_PASS_IN_STACK, and FUNCTION_ARG: Register Arguments.
43479 * TARGET_N_FORMAT_TYPES: Misc. (line 879)
43480 * TARGET_NARROW_VOLATILE_BITFIELD: Storage Layout. (line 392)
43481 * TARGET_OBJECT_SUFFIX: Misc. (line 764)
43482 * TARGET_OBJFMT_CPP_BUILTINS: Run-time Target. (line 46)
43483 * TARGET_OPTF: Misc. (line 861)
43484 * TARGET_OPTION_PRAGMA_PARSE: Target Attributes. (line 120)
43485 * TARGET_OPTION_PRINT: Target Attributes. (line 115)
43486 * TARGET_OPTION_RESTORE: Target Attributes. (line 110)
43487 * TARGET_OPTION_SAVE: Target Attributes. (line 104)
43488 * TARGET_OPTION_TRANSLATE_TABLE: Driver. (line 53)
43489 * TARGET_OS_CPP_BUILTINS: Run-time Target. (line 42)
43490 * TARGET_OVERRIDES_FORMAT_ATTRIBUTES: Misc. (line 883)
43491 * TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT: Misc. (line 889)
43492 * TARGET_OVERRIDES_FORMAT_INIT: Misc. (line 893)
43493 * TARGET_PASS_BY_REFERENCE: Register Arguments. (line 103)
43494 * TARGET_POSIX_IO: Misc. (line 572)
43495 * TARGET_PRETEND_OUTGOING_VARARGS_NAMED: Varargs. (line 152)
43496 * TARGET_PROMOTE_FUNCTION_ARGS: Storage Layout. (line 131)
43497 * TARGET_PROMOTE_FUNCTION_RETURN: Storage Layout. (line 136)
43498 * TARGET_PROMOTE_PROTOTYPES: Stack Arguments. (line 11)
43499 * TARGET_PTRMEMFUNC_VBIT_LOCATION: Type Layout. (line 235)
43500 * TARGET_RELAXED_ORDERING: Misc. (line 898)
43501 * TARGET_RESOLVE_OVERLOADED_BUILTIN: Misc. (line 675)
43502 * TARGET_RETURN_IN_MEMORY: Aggregate Return. (line 16)
43503 * TARGET_RETURN_IN_MSB: Scalar Return. (line 100)
43504 * TARGET_RTX_COSTS: Costs. (line 210)
43505 * TARGET_SCALAR_MODE_SUPPORTED_P: Register Arguments. (line 291)
43506 * TARGET_SCHED_ADJUST_COST: Scheduling. (line 37)
43507 * TARGET_SCHED_ADJUST_PRIORITY: Scheduling. (line 52)
43508 * TARGET_SCHED_CLEAR_SCHED_CONTEXT: Scheduling. (line 261)
43509 * TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK: Scheduling. (line 89)
43510 * TARGET_SCHED_DFA_NEW_CYCLE: Scheduling. (line 205)
43511 * TARGET_SCHED_DFA_POST_CYCLE_ADVANCE: Scheduling. (line 160)
43512 * TARGET_SCHED_DFA_POST_CYCLE_INSN: Scheduling. (line 144)
43513 * TARGET_SCHED_DFA_PRE_CYCLE_ADVANCE: Scheduling. (line 153)
43514 * TARGET_SCHED_DFA_PRE_CYCLE_INSN: Scheduling. (line 132)
43515 * TARGET_SCHED_FINISH: Scheduling. (line 109)
43516 * TARGET_SCHED_FINISH_GLOBAL: Scheduling. (line 126)
43517 * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD: Scheduling.
43519 * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD: Scheduling.
43521 * TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC: Scheduling.
43523 * TARGET_SCHED_FREE_SCHED_CONTEXT: Scheduling. (line 265)
43524 * TARGET_SCHED_GEN_CHECK: Scheduling. (line 309)
43525 * TARGET_SCHED_H_I_D_EXTENDED: Scheduling. (line 241)
43526 * TARGET_SCHED_INIT: Scheduling. (line 99)
43527 * TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN: Scheduling. (line 149)
43528 * TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN: Scheduling. (line 141)
43529 * TARGET_SCHED_INIT_GLOBAL: Scheduling. (line 118)
43530 * TARGET_SCHED_INIT_SCHED_CONTEXT: Scheduling. (line 251)
43531 * TARGET_SCHED_IS_COSTLY_DEPENDENCE: Scheduling. (line 219)
43532 * TARGET_SCHED_ISSUE_RATE: Scheduling. (line 12)
43533 * TARGET_SCHED_NEEDS_BLOCK_P: Scheduling. (line 302)
43534 * TARGET_SCHED_REORDER: Scheduling. (line 60)
43535 * TARGET_SCHED_REORDER2: Scheduling. (line 77)
43536 * TARGET_SCHED_SET_SCHED_CONTEXT: Scheduling. (line 257)
43537 * TARGET_SCHED_SET_SCHED_FLAGS: Scheduling. (line 332)
43538 * TARGET_SCHED_SMS_RES_MII: Scheduling. (line 343)
43539 * TARGET_SCHED_SPECULATE_INSN: Scheduling. (line 291)
43540 * TARGET_SCHED_VARIABLE_ISSUE: Scheduling. (line 24)
43541 * TARGET_SECONDARY_RELOAD: Register Classes. (line 257)
43542 * TARGET_SECTION_TYPE_FLAGS: File Framework. (line 109)
43543 * TARGET_SET_CURRENT_FUNCTION: Misc. (line 747)
43544 * TARGET_SET_DEFAULT_TYPE_ATTRIBUTES: Target Attributes. (line 26)
43545 * TARGET_SETUP_INCOMING_VARARGS: Varargs. (line 101)
43546 * TARGET_SHIFT_TRUNCATION_MASK: Misc. (line 154)
43547 * TARGET_SPLIT_COMPLEX_ARG: Register Arguments. (line 252)
43548 * TARGET_STACK_PROTECT_FAIL: Stack Smashing Protection.
43550 * TARGET_STACK_PROTECT_GUARD: Stack Smashing Protection.
43552 * TARGET_STRICT_ARGUMENT_NAMING: Varargs. (line 137)
43553 * TARGET_STRUCT_VALUE_RTX: Aggregate Return. (line 44)
43554 * TARGET_UNSPEC_MAY_TRAP_P: Misc. (line 739)
43555 * TARGET_UNWIND_EMIT: Dispatch Tables. (line 81)
43556 * TARGET_UNWIND_INFO: Exception Region Output.
43558 * TARGET_UPDATE_STACK_BOUNDARY: Misc. (line 950)
43559 * TARGET_USE_ANCHORS_FOR_SYMBOL_P: Anchored Addresses. (line 55)
43560 * TARGET_USE_BLOCKS_FOR_CONSTANT_P: Addressing Modes. (line 233)
43561 * TARGET_USE_JCR_SECTION: Misc. (line 932)
43562 * TARGET_USE_LOCAL_THUNK_ALIAS_P: Misc. (line 867)
43563 * TARGET_USES_WEAK_UNWIND_INFO: Exception Handling. (line 129)
43564 * TARGET_VALID_DLLIMPORT_ATTRIBUTE_P: Target Attributes. (line 59)
43565 * TARGET_VALID_OPTION_ATTRIBUTE_P: Target Attributes. (line 93)
43566 * TARGET_VALID_POINTER_MODE: Register Arguments. (line 285)
43567 * TARGET_VECTOR_MODE_SUPPORTED_P: Register Arguments. (line 303)
43568 * TARGET_VECTOR_OPAQUE_P: Storage Layout. (line 479)
43569 * TARGET_VECTORIZE_BUILTIN_CONVERSION: Addressing Modes. (line 300)
43570 * TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD: Addressing Modes. (line 249)
43571 * TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN: Addressing Modes. (line 275)
43572 * TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD: Addressing Modes. (line 287)
43573 * TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION: Addressing Modes.
43575 * TARGET_VERSION: Run-time Target. (line 91)
43576 * TARGET_VTABLE_DATA_ENTRY_DISTANCE: Type Layout. (line 288)
43577 * TARGET_VTABLE_ENTRY_ALIGN: Type Layout. (line 282)
43578 * TARGET_VTABLE_USES_DESCRIPTORS: Type Layout. (line 271)
43579 * TARGET_WEAK_NOT_IN_ARCHIVE_TOC: Label Output. (line 245)
43580 * targetm: Target Structure. (line 7)
43581 * targets, makefile: Makefile. (line 6)
43582 * TCmode: Machine Modes. (line 197)
43583 * TDmode: Machine Modes. (line 94)
43584 * TEMPLATE_DECL: Declarations. (line 6)
43585 * Temporaries: Temporaries. (line 6)
43586 * termination routines: Initialization. (line 6)
43587 * testing constraints: C Constraint Interface.
43589 * TEXT_SECTION_ASM_OP: Sections. (line 38)
43590 * TF_SIZE: Type Layout. (line 132)
43591 * TFmode: Machine Modes. (line 98)
43592 * THEN_CLAUSE: Function Bodies. (line 6)
43593 * THREAD_MODEL_SPEC: Driver. (line 225)
43594 * THROW_EXPR: Expression trees. (line 6)
43595 * THUNK_DECL: Declarations. (line 6)
43596 * THUNK_DELTA: Declarations. (line 6)
43597 * TImode: Machine Modes. (line 48)
43598 * TImode, in insn: Insns. (line 231)
43599 * tm.h macros: Target Macros. (line 6)
43600 * TQFmode: Machine Modes. (line 62)
43601 * TQmode: Machine Modes. (line 119)
43602 * TRAMPOLINE_ADJUST_ADDRESS: Trampolines. (line 62)
43603 * TRAMPOLINE_ALIGNMENT: Trampolines. (line 49)
43604 * TRAMPOLINE_SECTION: Trampolines. (line 40)
43605 * TRAMPOLINE_SIZE: Trampolines. (line 45)
43606 * TRAMPOLINE_TEMPLATE: Trampolines. (line 29)
43607 * trampolines for nested functions: Trampolines. (line 6)
43608 * TRANSFER_FROM_TRAMPOLINE: Trampolines. (line 124)
43609 * trap instruction pattern: Standard Names. (line 1374)
43610 * tree <1>: Macros and Functions.
43612 * tree: Tree overview. (line 6)
43613 * Tree SSA: Tree SSA. (line 6)
43614 * tree_code <1>: GIMPLE_OMP_FOR. (line 83)
43615 * tree_code <2>: GIMPLE_COND. (line 21)
43616 * tree_code <3>: GIMPLE_ASSIGN. (line 41)
43617 * tree_code: Manipulating GIMPLE statements.
43619 * TREE_CODE: Tree overview. (line 6)
43620 * TREE_FILENAME: Working with declarations.
43622 * tree_int_cst_equal: Expression trees. (line 6)
43623 * TREE_INT_CST_HIGH: Expression trees. (line 6)
43624 * TREE_INT_CST_LOW: Expression trees. (line 6)
43625 * tree_int_cst_lt: Expression trees. (line 6)
43626 * TREE_LINENO: Working with declarations.
43628 * TREE_LIST: Containers. (line 6)
43629 * TREE_OPERAND: Expression trees. (line 6)
43630 * TREE_PUBLIC: Function Basics. (line 6)
43631 * TREE_PURPOSE: Containers. (line 6)
43632 * TREE_STRING_LENGTH: Expression trees. (line 6)
43633 * TREE_STRING_POINTER: Expression trees. (line 6)
43634 * TREE_TYPE <1>: Expression trees. (line 6)
43635 * TREE_TYPE <2>: Function Basics. (line 171)
43636 * TREE_TYPE <3>: Working with declarations.
43638 * TREE_TYPE: Types. (line 6)
43639 * TREE_VALUE: Containers. (line 6)
43640 * TREE_VEC: Containers. (line 6)
43641 * TREE_VEC_ELT: Containers. (line 6)
43642 * TREE_VEC_LENGTH: Containers. (line 6)
43643 * Trees: Trees. (line 6)
43644 * TRULY_NOOP_TRUNCATION: Misc. (line 177)
43645 * TRUNC_DIV_EXPR: Expression trees. (line 6)
43646 * TRUNC_MOD_EXPR: Expression trees. (line 6)
43647 * truncate: Conversions. (line 38)
43648 * truncMN2 instruction pattern: Standard Names. (line 821)
43649 * TRUTH_AND_EXPR: Expression trees. (line 6)
43650 * TRUTH_ANDIF_EXPR: Expression trees. (line 6)
43651 * TRUTH_NOT_EXPR: Expression trees. (line 6)
43652 * TRUTH_OR_EXPR: Expression trees. (line 6)
43653 * TRUTH_ORIF_EXPR: Expression trees. (line 6)
43654 * TRUTH_XOR_EXPR: Expression trees. (line 6)
43655 * TRY_BLOCK: Function Bodies. (line 6)
43656 * TRY_HANDLERS: Function Bodies. (line 6)
43657 * TRY_STMTS: Function Bodies. (line 6)
43658 * tstM instruction pattern: Standard Names. (line 661)
43659 * Tuple specific accessors: Tuple specific accessors.
43661 * tuples: Tuple representation.
43663 * type: Types. (line 6)
43664 * type declaration: Declarations. (line 6)
43665 * TYPE_ALIGN: Types. (line 6)
43666 * TYPE_ARG_TYPES: Types. (line 6)
43667 * TYPE_ASM_OP: Label Output. (line 55)
43668 * TYPE_ATTRIBUTES: Attributes. (line 25)
43669 * TYPE_BINFO: Classes. (line 6)
43670 * TYPE_BUILT_IN: Types. (line 83)
43671 * TYPE_CANONICAL: Types. (line 6)
43672 * TYPE_CONTEXT: Types. (line 6)
43673 * TYPE_DECL: Declarations. (line 6)
43674 * TYPE_FIELDS <1>: Classes. (line 6)
43675 * TYPE_FIELDS: Types. (line 6)
43676 * TYPE_HAS_ARRAY_NEW_OPERATOR: Classes. (line 91)
43677 * TYPE_HAS_DEFAULT_CONSTRUCTOR: Classes. (line 76)
43678 * TYPE_HAS_MUTABLE_P: Classes. (line 81)
43679 * TYPE_HAS_NEW_OPERATOR: Classes. (line 88)
43680 * TYPE_MAIN_VARIANT: Types. (line 6)
43681 * TYPE_MAX_VALUE: Types. (line 6)
43682 * TYPE_METHOD_BASETYPE: Types. (line 6)
43683 * TYPE_METHODS: Classes. (line 6)
43684 * TYPE_MIN_VALUE: Types. (line 6)
43685 * TYPE_NAME: Types. (line 6)
43686 * TYPE_NOTHROW_P: Function Basics. (line 180)
43687 * TYPE_OFFSET_BASETYPE: Types. (line 6)
43688 * TYPE_OPERAND_FMT: Label Output. (line 66)
43689 * TYPE_OVERLOADS_ARRAY_REF: Classes. (line 99)
43690 * TYPE_OVERLOADS_ARROW: Classes. (line 102)
43691 * TYPE_OVERLOADS_CALL_EXPR: Classes. (line 95)
43692 * TYPE_POLYMORPHIC_P: Classes. (line 72)
43693 * TYPE_PRECISION: Types. (line 6)
43694 * TYPE_PTR_P: Types. (line 89)
43695 * TYPE_PTRFN_P: Types. (line 93)
43696 * TYPE_PTRMEM_P: Types. (line 6)
43697 * TYPE_PTROB_P: Types. (line 96)
43698 * TYPE_PTROBV_P: Types. (line 6)
43699 * TYPE_QUAL_CONST: Types. (line 6)
43700 * TYPE_QUAL_RESTRICT: Types. (line 6)
43701 * TYPE_QUAL_VOLATILE: Types. (line 6)
43702 * TYPE_RAISES_EXCEPTIONS: Function Basics. (line 175)
43703 * TYPE_SIZE: Types. (line 6)
43704 * TYPE_STRUCTURAL_EQUALITY_P: Types. (line 6)
43705 * TYPE_UNQUALIFIED: Types. (line 6)
43706 * TYPE_VFIELD: Classes. (line 6)
43707 * TYPENAME_TYPE: Types. (line 6)
43708 * TYPENAME_TYPE_FULLNAME: Types. (line 6)
43709 * TYPEOF_TYPE: Types. (line 6)
43710 * UDAmode: Machine Modes. (line 168)
43711 * udiv: Arithmetic. (line 125)
43712 * udivM3 instruction pattern: Standard Names. (line 222)
43713 * udivmodM4 instruction pattern: Standard Names. (line 428)
43714 * udot_prodM instruction pattern: Standard Names. (line 265)
43715 * UDQmode: Machine Modes. (line 136)
43716 * UHAmode: Machine Modes. (line 160)
43717 * UHQmode: Machine Modes. (line 128)
43718 * UINTMAX_TYPE: Type Layout. (line 224)
43719 * umaddMN4 instruction pattern: Standard Names. (line 375)
43720 * umax: Arithmetic. (line 144)
43721 * umaxM3 instruction pattern: Standard Names. (line 222)
43722 * umin: Arithmetic. (line 144)
43723 * uminM3 instruction pattern: Standard Names. (line 222)
43724 * umod: Arithmetic. (line 131)
43725 * umodM3 instruction pattern: Standard Names. (line 222)
43726 * umsubMN4 instruction pattern: Standard Names. (line 399)
43727 * umulhisi3 instruction pattern: Standard Names. (line 347)
43728 * umulM3_highpart instruction pattern: Standard Names. (line 361)
43729 * umulqihi3 instruction pattern: Standard Names. (line 347)
43730 * umulsidi3 instruction pattern: Standard Names. (line 347)
43731 * unchanging: Flags. (line 319)
43732 * unchanging, in call_insn: Flags. (line 19)
43733 * unchanging, in jump_insn, call_insn and insn: Flags. (line 39)
43734 * unchanging, in mem: Flags. (line 152)
43735 * unchanging, in subreg: Flags. (line 188)
43736 * unchanging, in symbol_ref: Flags. (line 10)
43737 * UNEQ_EXPR: Expression trees. (line 6)
43738 * UNGE_EXPR: Expression trees. (line 6)
43739 * UNGT_EXPR: Expression trees. (line 6)
43740 * UNION_TYPE <1>: Classes. (line 6)
43741 * UNION_TYPE: Types. (line 6)
43742 * unions, returning: Interface. (line 10)
43743 * UNITS_PER_SIMD_WORD: Storage Layout. (line 77)
43744 * UNITS_PER_WORD: Storage Layout. (line 67)
43745 * UNKNOWN_TYPE: Types. (line 6)
43746 * UNLE_EXPR: Expression trees. (line 6)
43747 * UNLIKELY_EXECUTED_TEXT_SECTION_NAME: Sections. (line 49)
43748 * UNLT_EXPR: Expression trees. (line 6)
43749 * UNORDERED_EXPR: Expression trees. (line 6)
43750 * unshare_all_rtl: Sharing. (line 58)
43751 * unsigned division: Arithmetic. (line 125)
43752 * unsigned division with unsigned saturation: Arithmetic. (line 125)
43753 * unsigned greater than: Comparisons. (line 64)
43754 * unsigned less than: Comparisons. (line 68)
43755 * unsigned minimum and maximum: Arithmetic. (line 144)
43756 * unsigned_fix: Conversions. (line 77)
43757 * unsigned_float: Conversions. (line 62)
43758 * unsigned_fract_convert: Conversions. (line 97)
43759 * unsigned_sat_fract: Conversions. (line 103)
43760 * unspec: Side Effects. (line 287)
43761 * unspec_volatile: Side Effects. (line 287)
43762 * untyped_call instruction pattern: Standard Names. (line 1012)
43763 * untyped_return instruction pattern: Standard Names. (line 1062)
43764 * UPDATE_PATH_HOST_CANONICALIZE (PATH): Filesystem. (line 59)
43765 * update_ssa: SSA. (line 76)
43766 * update_stmt <1>: SSA Operands. (line 6)
43767 * update_stmt: Manipulating GIMPLE statements.
43769 * update_stmt_if_modified: Manipulating GIMPLE statements.
43771 * UQQmode: Machine Modes. (line 123)
43772 * US Software GOFAST, floating point emulation library: Library Calls.
43774 * us_ashift: Arithmetic. (line 168)
43775 * us_minus: Arithmetic. (line 36)
43776 * us_mult: Arithmetic. (line 92)
43777 * us_neg: Arithmetic. (line 81)
43778 * us_plus: Arithmetic. (line 14)
43779 * US_SOFTWARE_GOFAST: Library Calls. (line 45)
43780 * us_truncate: Conversions. (line 48)
43781 * usaddM3 instruction pattern: Standard Names. (line 222)
43782 * USAmode: Machine Modes. (line 164)
43783 * usashlM3 instruction pattern: Standard Names. (line 431)
43784 * usdivM3 instruction pattern: Standard Names. (line 222)
43785 * use: Side Effects. (line 162)
43786 * USE_C_ALLOCA: Host Misc. (line 19)
43787 * USE_LD_AS_NEEDED: Driver. (line 198)
43788 * USE_LOAD_POST_DECREMENT: Costs. (line 165)
43789 * USE_LOAD_POST_INCREMENT: Costs. (line 160)
43790 * USE_LOAD_PRE_DECREMENT: Costs. (line 175)
43791 * USE_LOAD_PRE_INCREMENT: Costs. (line 170)
43792 * use_optype_d: Manipulating GIMPLE statements.
43794 * use_param: GTY Options. (line 114)
43795 * use_paramN: GTY Options. (line 132)
43796 * use_params: GTY Options. (line 140)
43797 * USE_SELECT_SECTION_FOR_FUNCTIONS: Sections. (line 185)
43798 * USE_STORE_POST_DECREMENT: Costs. (line 185)
43799 * USE_STORE_POST_INCREMENT: Costs. (line 180)
43800 * USE_STORE_PRE_DECREMENT: Costs. (line 195)
43801 * USE_STORE_PRE_INCREMENT: Costs. (line 190)
43802 * used: Flags. (line 337)
43803 * used, in symbol_ref: Flags. (line 215)
43804 * USER_LABEL_PREFIX: Instruction Output. (line 126)
43805 * USING_DECL: Declarations. (line 6)
43806 * USING_STMT: Function Bodies. (line 6)
43807 * usmaddMN4 instruction pattern: Standard Names. (line 383)
43808 * usmsubMN4 instruction pattern: Standard Names. (line 407)
43809 * usmulhisi3 instruction pattern: Standard Names. (line 351)
43810 * usmulM3 instruction pattern: Standard Names. (line 222)
43811 * usmulqihi3 instruction pattern: Standard Names. (line 351)
43812 * usmulsidi3 instruction pattern: Standard Names. (line 351)
43813 * usnegM2 instruction pattern: Standard Names. (line 449)
43814 * USQmode: Machine Modes. (line 132)
43815 * ussubM3 instruction pattern: Standard Names. (line 222)
43816 * usum_widenM3 instruction pattern: Standard Names. (line 275)
43817 * UTAmode: Machine Modes. (line 172)
43818 * UTQmode: Machine Modes. (line 140)
43819 * V in constraint: Simple Constraints. (line 43)
43820 * VA_ARG_EXPR: Expression trees. (line 6)
43821 * values, returned by functions: Scalar Return. (line 6)
43822 * VAR_DECL <1>: Expression trees. (line 6)
43823 * VAR_DECL: Declarations. (line 6)
43824 * varargs implementation: Varargs. (line 6)
43825 * variable: Declarations. (line 6)
43826 * vashlM3 instruction pattern: Standard Names. (line 445)
43827 * vashrM3 instruction pattern: Standard Names. (line 445)
43828 * vec_concat: Vector Operations. (line 25)
43829 * vec_duplicate: Vector Operations. (line 30)
43830 * VEC_EXTRACT_EVEN_EXPR: Expression trees. (line 6)
43831 * vec_extract_evenM instruction pattern: Standard Names. (line 176)
43832 * VEC_EXTRACT_ODD_EXPR: Expression trees. (line 6)
43833 * vec_extract_oddM instruction pattern: Standard Names. (line 183)
43834 * vec_extractM instruction pattern: Standard Names. (line 171)
43835 * vec_initM instruction pattern: Standard Names. (line 204)
43836 * VEC_INTERLEAVE_HIGH_EXPR: Expression trees. (line 6)
43837 * vec_interleave_highM instruction pattern: Standard Names. (line 190)
43838 * VEC_INTERLEAVE_LOW_EXPR: Expression trees. (line 6)
43839 * vec_interleave_lowM instruction pattern: Standard Names. (line 197)
43840 * VEC_LSHIFT_EXPR: Expression trees. (line 6)
43841 * vec_merge: Vector Operations. (line 11)
43842 * VEC_PACK_FIX_TRUNC_EXPR: Expression trees. (line 6)
43843 * VEC_PACK_SAT_EXPR: Expression trees. (line 6)
43844 * vec_pack_sfix_trunc_M instruction pattern: Standard Names. (line 302)
43845 * vec_pack_ssat_M instruction pattern: Standard Names. (line 295)
43846 * VEC_PACK_TRUNC_EXPR: Expression trees. (line 6)
43847 * vec_pack_trunc_M instruction pattern: Standard Names. (line 288)
43848 * vec_pack_ufix_trunc_M instruction pattern: Standard Names. (line 302)
43849 * vec_pack_usat_M instruction pattern: Standard Names. (line 295)
43850 * VEC_RSHIFT_EXPR: Expression trees. (line 6)
43851 * vec_select: Vector Operations. (line 19)
43852 * vec_setM instruction pattern: Standard Names. (line 166)
43853 * vec_shl_M instruction pattern: Standard Names. (line 282)
43854 * vec_shr_M instruction pattern: Standard Names. (line 282)
43855 * VEC_UNPACK_FLOAT_HI_EXPR: Expression trees. (line 6)
43856 * VEC_UNPACK_FLOAT_LO_EXPR: Expression trees. (line 6)
43857 * VEC_UNPACK_HI_EXPR: Expression trees. (line 6)
43858 * VEC_UNPACK_LO_EXPR: Expression trees. (line 6)
43859 * vec_unpacks_float_hi_M instruction pattern: Standard Names.
43861 * vec_unpacks_float_lo_M instruction pattern: Standard Names.
43863 * vec_unpacks_hi_M instruction pattern: Standard Names. (line 309)
43864 * vec_unpacks_lo_M instruction pattern: Standard Names. (line 309)
43865 * vec_unpacku_float_hi_M instruction pattern: Standard Names.
43867 * vec_unpacku_float_lo_M instruction pattern: Standard Names.
43869 * vec_unpacku_hi_M instruction pattern: Standard Names. (line 317)
43870 * vec_unpacku_lo_M instruction pattern: Standard Names. (line 317)
43871 * VEC_WIDEN_MULT_HI_EXPR: Expression trees. (line 6)
43872 * VEC_WIDEN_MULT_LO_EXPR: Expression trees. (line 6)
43873 * vec_widen_smult_hi_M instruction pattern: Standard Names. (line 333)
43874 * vec_widen_smult_lo_M instruction pattern: Standard Names. (line 333)
43875 * vec_widen_umult_hi_M instruction pattern: Standard Names. (line 333)
43876 * vec_widen_umult_lo__M instruction pattern: Standard Names. (line 333)
43877 * vector: Containers. (line 6)
43878 * vector operations: Vector Operations. (line 6)
43879 * VECTOR_CST: Expression trees. (line 6)
43880 * VECTOR_STORE_FLAG_VALUE: Misc. (line 308)
43881 * virtual operands: SSA Operands. (line 6)
43882 * VIRTUAL_INCOMING_ARGS_REGNUM: Regs and Memory. (line 59)
43883 * VIRTUAL_OUTGOING_ARGS_REGNUM: Regs and Memory. (line 87)
43884 * VIRTUAL_STACK_DYNAMIC_REGNUM: Regs and Memory. (line 78)
43885 * VIRTUAL_STACK_VARS_REGNUM: Regs and Memory. (line 69)
43886 * VLIW: Processor pipeline description.
43888 * vlshrM3 instruction pattern: Standard Names. (line 445)
43889 * VMS: Filesystem. (line 37)
43890 * VMS_DEBUGGING_INFO: VMS Debug. (line 9)
43891 * VOID_TYPE: Types. (line 6)
43892 * VOIDmode: Machine Modes. (line 190)
43893 * volatil: Flags. (line 351)
43894 * volatil, in insn, call_insn, jump_insn, code_label, barrier, and note: Flags.
43896 * volatil, in label_ref and reg_label: Flags. (line 65)
43897 * volatil, in mem, asm_operands, and asm_input: Flags. (line 94)
43898 * volatil, in reg: Flags. (line 116)
43899 * volatil, in subreg: Flags. (line 188)
43900 * volatil, in symbol_ref: Flags. (line 224)
43901 * volatile memory references: Flags. (line 352)
43902 * voptype_d: Manipulating GIMPLE statements.
43904 * voting between constraint alternatives: Class Preferences. (line 6)
43905 * vrotlM3 instruction pattern: Standard Names. (line 445)
43906 * vrotrM3 instruction pattern: Standard Names. (line 445)
43907 * walk_dominator_tree: SSA. (line 256)
43908 * walk_gimple_op: Statement and operand traversals.
43910 * walk_gimple_seq: Statement and operand traversals.
43912 * walk_gimple_stmt: Statement and operand traversals.
43914 * walk_use_def_chains: SSA. (line 232)
43915 * WCHAR_TYPE: Type Layout. (line 192)
43916 * WCHAR_TYPE_SIZE: Type Layout. (line 200)
43917 * which_alternative: Output Statement. (line 59)
43918 * WHILE_BODY: Function Bodies. (line 6)
43919 * WHILE_COND: Function Bodies. (line 6)
43920 * WHILE_STMT: Function Bodies. (line 6)
43921 * WIDEST_HARDWARE_FP_SIZE: Type Layout. (line 147)
43922 * WINT_TYPE: Type Layout. (line 205)
43923 * word_mode: Machine Modes. (line 336)
43924 * WORD_REGISTER_OPERATIONS: Misc. (line 63)
43925 * WORD_SWITCH_TAKES_ARG: Driver. (line 20)
43926 * WORDS_BIG_ENDIAN: Storage Layout. (line 29)
43927 * WORDS_BIG_ENDIAN, effect on subreg: Regs and Memory. (line 217)
43928 * X in constraint: Simple Constraints. (line 114)
43929 * x-HOST: Host Fragment. (line 6)
43930 * XCmode: Machine Modes. (line 197)
43931 * XCOFF_DEBUGGING_INFO: DBX Options. (line 13)
43932 * XEXP: Accessors. (line 6)
43933 * XF_SIZE: Type Layout. (line 131)
43934 * XFmode: Machine Modes. (line 79)
43935 * XINT: Accessors. (line 6)
43936 * xm-MACHINE.h <1>: Host Misc. (line 6)
43937 * xm-MACHINE.h: Filesystem. (line 6)
43938 * xor: Arithmetic. (line 163)
43939 * xor, canonicalization of: Insn Canonicalizations.
43941 * xorM3 instruction pattern: Standard Names. (line 222)
43942 * XSTR: Accessors. (line 6)
43943 * XVEC: Accessors. (line 41)
43944 * XVECEXP: Accessors. (line 48)
43945 * XVECLEN: Accessors. (line 44)
43946 * XWINT: Accessors. (line 6)
43947 * zero_extend: Conversions. (line 28)
43948 * zero_extendMN2 instruction pattern: Standard Names. (line 831)
43949 * zero_extract: Bit-Fields. (line 30)
43950 * zero_extract, canonicalization of: Insn Canonicalizations.
43957 Node: Contributing
\7f5148
43958 Node: Portability
\7f5889
43959 Node: Interface
\7f7677
43960 Node: Libgcc
\7f10717
43961 Node: Integer library routines
\7f12558
43962 Node: Soft float library routines
\7f19397
43963 Node: Decimal float library routines
\7f31334
43964 Node: Fixed-point fractional library routines
\7f47091
43965 Node: Exception handling routines
\7f147489
43966 Node: Miscellaneous routines
\7f148596
43967 Node: Languages
\7f148979
43968 Node: Source Tree
\7f150526
43969 Node: Configure Terms
\7f151145
43970 Node: Top Level
\7f154103
43971 Node: gcc Directory
\7f156873
43972 Node: Subdirectories
\7f157842
43973 Node: Configuration
\7f159692
43974 Node: Config Fragments
\7f160412
43975 Node: System Config
\7f161641
43976 Node: Configuration Files
\7f162577
43977 Node: Build
\7f165152
43978 Node: Makefile
\7f165564
43979 Ref: Makefile-Footnote-1
\7f172282
43980 Ref: Makefile-Footnote-2
\7f172427
43981 Node: Library Files
\7f172499
43982 Node: Headers
\7f173061
43983 Node: Documentation
\7f175144
43984 Node: Texinfo Manuals
\7f176003
43985 Node: Man Page Generation
\7f178341
43986 Node: Miscellaneous Docs
\7f180256
43987 Node: Front End
\7f181555
43988 Node: Front End Directory
\7f185256
43989 Node: Front End Config
\7f190446
43990 Node: Back End
\7f193360
43991 Node: Testsuites
\7f197037
43992 Node: Test Idioms
\7f197901
43993 Node: Test Directives
\7f201302
43994 Node: Ada Tests
\7f213366
43995 Node: C Tests
\7f214658
43996 Node: libgcj Tests
\7f219013
43997 Node: gcov Testing
\7f220145
43998 Node: profopt Testing
\7f223129
43999 Node: compat Testing
\7f224572
44000 Node: Torture Tests
\7f228816
44001 Node: Options
\7f230448
44002 Node: Option file format
\7f230889
44003 Node: Option properties
\7f233638
44004 Node: Passes
\7f239694
44005 Node: Parsing pass
\7f240436
44006 Node: Gimplification pass
\7f243964
44007 Node: Pass manager
\7f245791
44008 Node: Tree-SSA passes
\7f247274
44009 Node: RTL passes
\7f269105
44010 Node: Trees
\7f281690
44011 Node: Deficiencies
\7f284416
44012 Node: Tree overview
\7f284653
44013 Node: Macros and Functions
\7f288776
44014 Node: Identifiers
\7f288922
44015 Node: Containers
\7f290447
44016 Node: Types
\7f291602
44017 Node: Scopes
\7f307305
44018 Node: Namespaces
\7f308067
44019 Node: Classes
\7f310879
44020 Node: Declarations
\7f315636
44021 Node: Working with declarations
\7f316131
44022 Node: Internal structure
\7f322588
44023 Node: Current structure hierarchy
\7f322970
44024 Node: Adding new DECL node types
\7f325062
44025 Node: Functions
\7f329133
44026 Node: Function Basics
\7f331536
44027 Node: Function Bodies
\7f339266
44028 Node: Attributes
\7f350508
44029 Node: Expression trees
\7f351749
44031 Node: RTL Objects
\7f396457
44032 Node: RTL Classes
\7f400331
44033 Node: Accessors
\7f405283
44034 Node: Special Accessors
\7f407677
44035 Node: Flags
\7f412895
44036 Node: Machine Modes
\7f428763
44037 Node: Constants
\7f441079
44038 Node: Regs and Memory
\7f447108
44039 Node: Arithmetic
\7f465009
44040 Node: Comparisons
\7f474529
44041 Node: Bit-Fields
\7f478821
44042 Node: Vector Operations
\7f480373
44043 Node: Conversions
\7f481999
44044 Node: RTL Declarations
\7f486497
44045 Node: Side Effects
\7f487318
44046 Node: Incdec
\7f503641
44047 Node: Assembler
\7f506976
44048 Node: Insns
\7f508508
44049 Node: Calls
\7f532397
44050 Node: Sharing
\7f534990
44051 Node: Reading RTL
\7f538100
44052 Node: GENERIC
\7f539090
44053 Node: Statements
\7f540727
44054 Node: Blocks
\7f541172
44055 Node: Statement Sequences
\7f542425
44056 Node: Empty Statements
\7f542758
44057 Node: Jumps
\7f543332
44058 Node: Cleanups
\7f543985
44059 Node: GIMPLE
\7f545738
44060 Node: Tuple representation
\7f549359
44061 Node: GIMPLE instruction set
\7f558014
44062 Node: GIMPLE Exception Handling
\7f559682
44063 Node: Temporaries
\7f561597
44064 Ref: Temporaries-Footnote-1
\7f562916
44065 Node: Operands
\7f562979
44066 Node: Compound Expressions
\7f563753
44067 Node: Compound Lvalues
\7f563987
44068 Node: Conditional Expressions
\7f564753
44069 Node: Logical Operators
\7f565423
44070 Node: Manipulating GIMPLE statements
\7f571514
44071 Node: Tuple specific accessors
\7f577442
44072 Node: `GIMPLE_ASM'
\7f578275
44073 Node: `GIMPLE_ASSIGN'
\7f580880
44074 Node: `GIMPLE_BIND'
\7f584826
44075 Node: `GIMPLE_CALL'
\7f586633
44076 Node: `GIMPLE_CATCH'
\7f590892
44077 Node: `GIMPLE_CHANGE_DYNAMIC_TYPE'
\7f592050
44078 Node: `GIMPLE_COND'
\7f593383
44079 Node: `GIMPLE_EH_FILTER'
\7f596189
44080 Node: `GIMPLE_LABEL'
\7f597675
44081 Node: `GIMPLE_NOP'
\7f598650
44082 Node: `GIMPLE_OMP_ATOMIC_LOAD'
\7f599019
44083 Node: `GIMPLE_OMP_ATOMIC_STORE'
\7f599929
44084 Node: `GIMPLE_OMP_CONTINUE'
\7f600568
44085 Node: `GIMPLE_OMP_CRITICAL'
\7f601918
44086 Node: `GIMPLE_OMP_FOR'
\7f602854
44087 Node: `GIMPLE_OMP_MASTER'
\7f606364
44088 Node: `GIMPLE_OMP_ORDERED'
\7f606747
44089 Node: `GIMPLE_OMP_PARALLEL'
\7f607147
44090 Node: `GIMPLE_OMP_RETURN'
\7f609916
44091 Node: `GIMPLE_OMP_SECTION'
\7f610566
44092 Node: `GIMPLE_OMP_SECTIONS'
\7f611232
44093 Node: `GIMPLE_OMP_SINGLE'
\7f612836
44094 Node: `GIMPLE_PHI'
\7f613772
44095 Node: `GIMPLE_RESX'
\7f615185
44096 Node: `GIMPLE_RETURN'
\7f615904
44097 Node: `GIMPLE_SWITCH'
\7f616472
44098 Node: `GIMPLE_TRY'
\7f618602
44099 Node: `GIMPLE_WITH_CLEANUP_EXPR'
\7f620392
44100 Node: GIMPLE sequences
\7f621275
44101 Node: Sequence iterators
\7f624481
44102 Node: Adding a new GIMPLE statement code
\7f632936
44103 Node: Statement and operand traversals
\7f634216
44104 Node: Tree SSA
\7f636826
44105 Node: Annotations
\7f638555
44106 Node: SSA Operands
\7f639081
44108 Node: Alias analysis
\7f665903
44109 Node: Loop Analysis and Representation
\7f673359
44110 Node: Loop representation
\7f674540
44111 Node: Loop querying
\7f681460
44112 Node: Loop manipulation
\7f684293
44113 Node: LCSSA
\7f686661
44114 Node: Scalar evolutions
\7f688733
44115 Node: loop-iv
\7f691977
44116 Node: Number of iterations
\7f693903
44117 Node: Dependency analysis
\7f696712
44118 Node: Lambda
\7f703080
44119 Node: Omega
\7f704750
44120 Node: Control Flow
\7f706315
44121 Node: Basic Blocks
\7f707315
44122 Node: Edges
\7f711883
44123 Node: Profile information
\7f720445
44124 Node: Maintaining the CFG
\7f725131
44125 Node: Liveness information
\7f732013
44126 Node: Machine Desc
\7f734140
44127 Node: Overview
\7f736608
44128 Node: Patterns
\7f738649
44129 Node: Example
\7f742087
44130 Node: RTL Template
\7f743522
44131 Node: Output Template
\7f754177
44132 Node: Output Statement
\7f758143
44133 Node: Predicates
\7f762105
44134 Node: Machine-Independent Predicates
\7f765023
44135 Node: Defining Predicates
\7f769655
44136 Node: Constraints
\7f775620
44137 Node: Simple Constraints
\7f776868
44138 Node: Multi-Alternative
\7f789074
44139 Node: Class Preferences
\7f791915
44140 Node: Modifiers
\7f792807
44141 Node: Machine Constraints
\7f796939
44142 Node: Disable Insn Alternatives
\7f829662
44143 Node: Define Constraints
\7f832555
44144 Node: C Constraint Interface
\7f839335
44145 Node: Standard Names
\7f842976
44146 Ref: shift patterns
\7f861904
44147 Ref: prologue instruction pattern
\7f902922
44148 Ref: epilogue instruction pattern
\7f903415
44149 Node: Pattern Ordering
\7f912958
44150 Node: Dependent Patterns
\7f914194
44151 Node: Jump Patterns
\7f917008
44152 Node: Looping Patterns
\7f922704
44153 Node: Insn Canonicalizations
\7f927432
44154 Node: Expander Definitions
\7f931816
44155 Node: Insn Splitting
\7f939934
44156 Node: Including Patterns
\7f949537
44157 Node: Peephole Definitions
\7f951317
44158 Node: define_peephole
\7f952570
44159 Node: define_peephole2
\7f958901
44160 Node: Insn Attributes
\7f961968
44161 Node: Defining Attributes
\7f963074
44162 Node: Expressions
\7f965594
44163 Node: Tagging Insns
\7f972196
44164 Node: Attr Example
\7f976549
44165 Node: Insn Lengths
\7f978923
44166 Node: Constant Attributes
\7f981982
44167 Node: Delay Slots
\7f983151
44168 Node: Processor pipeline description
\7f986375
44169 Ref: Processor pipeline description-Footnote-1
\7f1003741
44170 Node: Conditional Execution
\7f1004063
44171 Node: Constant Definitions
\7f1006916
44172 Node: Iterators
\7f1008511
44173 Node: Mode Iterators
\7f1008958
44174 Node: Defining Mode Iterators
\7f1009936
44175 Node: Substitutions
\7f1011430
44176 Node: Examples
\7f1013671
44177 Node: Code Iterators
\7f1015119
44178 Node: Target Macros
\7f1017376
44179 Node: Target Structure
\7f1020399
44180 Node: Driver
\7f1021668
44181 Node: Run-time Target
\7f1045349
44182 Node: Per-Function Data
\7f1052473
44183 Node: Storage Layout
\7f1055236
44184 Node: Type Layout
\7f1080650
44185 Node: Registers
\7f1093607
44186 Node: Register Basics
\7f1094581
44187 Node: Allocation Order
\7f1100148
44188 Node: Values in Registers
\7f1102169
44189 Node: Leaf Functions
\7f1109658
44190 Node: Stack Registers
\7f1112516
44191 Node: Register Classes
\7f1113632
44192 Node: Old Constraints
\7f1140344
44193 Node: Stack and Calling
\7f1147495
44194 Node: Frame Layout
\7f1148029
44195 Node: Exception Handling
\7f1158875
44196 Node: Stack Checking
\7f1165253
44197 Node: Frame Registers
\7f1169640
44198 Node: Elimination
\7f1176246
44199 Node: Stack Arguments
\7f1180277
44200 Node: Register Arguments
\7f1187080
44201 Node: Scalar Return
\7f1202533
44202 Node: Aggregate Return
\7f1208079
44203 Node: Caller Saves
\7f1211738
44204 Node: Function Entry
\7f1212916
44205 Node: Profiling
\7f1225531
44206 Node: Tail Calls
\7f1227230
44207 Node: Stack Smashing Protection
\7f1228597
44208 Node: Varargs
\7f1229709
44209 Node: Trampolines
\7f1237669
44210 Node: Library Calls
\7f1244335
44211 Node: Addressing Modes
\7f1249185
44212 Node: Anchored Addresses
\7f1265103
44213 Node: Condition Code
\7f1267764
44214 Node: Costs
\7f1276053
44215 Node: Scheduling
\7f1289152
44216 Node: Sections
\7f1307713
44217 Node: PIC
\7f1322363
44218 Node: Assembler Format
\7f1324353
44219 Node: File Framework
\7f1325491
44220 Ref: TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
\7f1330397
44221 Node: Data Output
\7f1333663
44222 Node: Uninitialized Data
\7f1341422
44223 Node: Label Output
\7f1346493
44224 Node: Initialization
\7f1368160
44225 Node: Macros for Initialization
\7f1374122
44226 Node: Instruction Output
\7f1380574
44227 Node: Dispatch Tables
\7f1389568
44228 Node: Exception Region Output
\7f1393363
44229 Node: Alignment Output
\7f1399123
44230 Node: Debugging Info
\7f1403286
44231 Node: All Debuggers
\7f1403956
44232 Node: DBX Options
\7f1406811
44233 Node: DBX Hooks
\7f1412260
44234 Node: File Names and DBX
\7f1414186
44235 Node: SDB and DWARF
\7f1416297
44236 Node: VMS Debug
\7f1420289
44237 Node: Floating Point
\7f1420859
44238 Node: Mode Switching
\7f1425682
44239 Node: Target Attributes
\7f1429608
44240 Node: Emulated TLS
\7f1436372
44241 Node: MIPS Coprocessors
\7f1439762
44242 Node: PCH Target
\7f1441331
44243 Node: C++ ABI
\7f1442852
44244 Node: Misc
\7f1447471
44245 Ref: TARGET_SHIFT_TRUNCATION_MASK
\7f1454842
44246 Node: Host Config
\7f1496097
44247 Node: Host Common
\7f1497165
44248 Node: Filesystem
\7f1499544
44249 Node: Host Misc
\7f1503659
44250 Node: Fragments
\7f1505798
44251 Node: Target Fragment
\7f1506993
44252 Node: Host Fragment
\7f1512883
44253 Node: Collect2
\7f1513123
44254 Node: Header Dirs
\7f1515666
44255 Node: Type Information
\7f1517089
44256 Node: GTY Options
\7f1519380
44257 Node: GGC Roots
\7f1530060
44258 Node: Files
\7f1530780
44259 Node: Invoking the garbage collector
\7f1533530
44260 Node: Plugins
\7f1534583
44261 Node: Funding
\7f1544948
44262 Node: GNU Project
\7f1547435
44263 Node: Copying
\7f1548084
44264 Node: GNU Free Documentation License
\7f1585615
44265 Node: Contributors
\7f1608024
44266 Node: Option Index
\7f1644354
44267 Node: Concept Index
\7f1644939