extend.texi: Document how GCC estimates and relies on the size of an asm.
[gcc.git] / gcc / doc / extend.texi
1 @c Copyright (C) 1988,1989,1992,1993,1994,1996,1998,1999,2000,2001,2002, 2003
2 @c Free Software Foundation, Inc.
3 @c This is part of the GCC manual.
4 @c For copying conditions, see the file gcc.texi.
5
6 @node C Implementation
7 @chapter C Implementation-defined behavior
8 @cindex implementation-defined behavior, C language
9
10 A conforming implementation of ISO C is required to document its
11 choice of behavior in each of the areas that are designated
12 ``implementation defined.'' The following lists all such areas,
13 along with the section number from the ISO/IEC 9899:1999 standard.
14
15 @menu
16 * Translation implementation::
17 * Environment implementation::
18 * Identifiers implementation::
19 * Characters implementation::
20 * Integers implementation::
21 * Floating point implementation::
22 * Arrays and pointers implementation::
23 * Hints implementation::
24 * Structures unions enumerations and bit-fields implementation::
25 * Qualifiers implementation::
26 * Preprocessing directives implementation::
27 * Library functions implementation::
28 * Architecture implementation::
29 * Locale-specific behavior implementation::
30 @end menu
31
32 @node Translation implementation
33 @section Translation
34
35 @itemize @bullet
36 @item
37 @cite{How a diagnostic is identified (3.10, 5.1.1.3).}
38
39 Diagnostics consist of all the output sent to stderr by GCC.
40
41 @item
42 @cite{Whether each nonempty sequence of white-space characters other than
43 new-line is retained or replaced by one space character in translation
44 phase 3 (5.1.1.2).}
45 @end itemize
46
47 @node Environment implementation
48 @section Environment
49
50 The behavior of these points are dependent on the implementation
51 of the C library, and are not defined by GCC itself.
52
53 @node Identifiers implementation
54 @section Identifiers
55
56 @itemize @bullet
57 @item
58 @cite{Which additional multibyte characters may appear in identifiers
59 and their correspondence to universal character names (6.4.2).}
60
61 @item
62 @cite{The number of significant initial characters in an identifier
63 (5.2.4.1, 6.4.2).}
64
65 For internal names, all characters are significant. For external names,
66 the number of significant characters are defined by the linker; for
67 almost all targets, all characters are significant.
68
69 @end itemize
70
71 @node Characters implementation
72 @section Characters
73
74 @itemize @bullet
75 @item
76 @cite{The number of bits in a byte (3.6).}
77
78 @item
79 @cite{The values of the members of the execution character set (5.2.1).}
80
81 @item
82 @cite{The unique value of the member of the execution character set produced
83 for each of the standard alphabetic escape sequences (5.2.2).}
84
85 @item
86 @cite{The value of a @code{char} object into which has been stored any
87 character other than a member of the basic execution character set (6.2.5).}
88
89 @item
90 @cite{Which of @code{signed char} or @code{unsigned char} has the same range,
91 representation, and behavior as ``plain'' @code{char} (6.2.5, 6.3.1.1).}
92
93 @item
94 @cite{The mapping of members of the source character set (in character
95 constants and string literals) to members of the execution character
96 set (6.4.4.4, 5.1.1.2).}
97
98 @item
99 @cite{The value of an integer character constant containing more than one
100 character or containing a character or escape sequence that does not map
101 to a single-byte execution character (6.4.4.4).}
102
103 @item
104 @cite{The value of a wide character constant containing more than one
105 multibyte character, or containing a multibyte character or escape
106 sequence not represented in the extended execution character set (6.4.4.4).}
107
108 @item
109 @cite{The current locale used to convert a wide character constant consisting
110 of a single multibyte character that maps to a member of the extended
111 execution character set into a corresponding wide character code (6.4.4.4).}
112
113 @item
114 @cite{The current locale used to convert a wide string literal into
115 corresponding wide character codes (6.4.5).}
116
117 @item
118 @cite{The value of a string literal containing a multibyte character or escape
119 sequence not represented in the execution character set (6.4.5).}
120 @end itemize
121
122 @node Integers implementation
123 @section Integers
124
125 @itemize @bullet
126 @item
127 @cite{Any extended integer types that exist in the implementation (6.2.5).}
128
129 @item
130 @cite{Whether signed integer types are represented using sign and magnitude,
131 two's complement, or one's complement, and whether the extraordinary value
132 is a trap representation or an ordinary value (6.2.6.2).}
133
134 GCC supports only two's complement integer types, and all bit patterns
135 are ordinary values.
136
137 @item
138 @cite{The rank of any extended integer type relative to another extended
139 integer type with the same precision (6.3.1.1).}
140
141 @item
142 @cite{The result of, or the signal raised by, converting an integer to a
143 signed integer type when the value cannot be represented in an object of
144 that type (6.3.1.3).}
145
146 @item
147 @cite{The results of some bitwise operations on signed integers (6.5).}
148 @end itemize
149
150 @node Floating point implementation
151 @section Floating point
152
153 @itemize @bullet
154 @item
155 @cite{The accuracy of the floating-point operations and of the library
156 functions in @code{<math.h>} and @code{<complex.h>} that return floating-point
157 results (5.2.4.2.2).}
158
159 @item
160 @cite{The rounding behaviors characterized by non-standard values
161 of @code{FLT_ROUNDS} @gol
162 (5.2.4.2.2).}
163
164 @item
165 @cite{The evaluation methods characterized by non-standard negative
166 values of @code{FLT_EVAL_METHOD} (5.2.4.2.2).}
167
168 @item
169 @cite{The direction of rounding when an integer is converted to a
170 floating-point number that cannot exactly represent the original
171 value (6.3.1.4).}
172
173 @item
174 @cite{The direction of rounding when a floating-point number is
175 converted to a narrower floating-point number (6.3.1.5).}
176
177 @item
178 @cite{How the nearest representable value or the larger or smaller
179 representable value immediately adjacent to the nearest representable
180 value is chosen for certain floating constants (6.4.4.2).}
181
182 @item
183 @cite{Whether and how floating expressions are contracted when not
184 disallowed by the @code{FP_CONTRACT} pragma (6.5).}
185
186 @item
187 @cite{The default state for the @code{FENV_ACCESS} pragma (7.6.1).}
188
189 @item
190 @cite{Additional floating-point exceptions, rounding modes, environments,
191 and classifications, and their macro names (7.6, 7.12).}
192
193 @item
194 @cite{The default state for the @code{FP_CONTRACT} pragma (7.12.2).}
195
196 @item
197 @cite{Whether the ``inexact'' floating-point exception can be raised
198 when the rounded result actually does equal the mathematical result
199 in an IEC 60559 conformant implementation (F.9).}
200
201 @item
202 @cite{Whether the ``underflow'' (and ``inexact'') floating-point
203 exception can be raised when a result is tiny but not inexact in an
204 IEC 60559 conformant implementation (F.9).}
205
206 @end itemize
207
208 @node Arrays and pointers implementation
209 @section Arrays and pointers
210
211 @itemize @bullet
212 @item
213 @cite{The result of converting a pointer to an integer or
214 vice versa (6.3.2.3).}
215
216 A cast from pointer to integer discards most-significant bits if the
217 pointer representation is larger than the integer type,
218 sign-extends@footnote{Future versions of GCC may zero-extend, or use
219 a target-defined @code{ptr_extend} pattern. Do not rely on sign extension.}
220 if the pointer representation is smaller than the integer type, otherwise
221 the bits are unchanged.
222 @c ??? We've always claimed that pointers were unsigned entities.
223 @c Shouldn't we therefore be doing zero-extension? If so, the bug
224 @c is in convert_to_integer, where we call type_for_size and request
225 @c a signed integral type. On the other hand, it might be most useful
226 @c for the target if we extend according to POINTERS_EXTEND_UNSIGNED.
227
228 A cast from integer to pointer discards most-significant bits if the
229 pointer representation is smaller than the integer type, extends according
230 to the signedness of the integer type if the pointer representation
231 is larger than the integer type, otherwise the bits are unchanged.
232
233 When casting from pointer to integer and back again, the resulting
234 pointer must reference the same object as the original pointer, otherwise
235 the behavior is undefined. That is, one may not use integer arithmetic to
236 avoid the undefined behavior of pointer arithmetic as proscribed in 6.5.6/8.
237
238 @item
239 @cite{The size of the result of subtracting two pointers to elements
240 of the same array (6.5.6).}
241
242 @end itemize
243
244 @node Hints implementation
245 @section Hints
246
247 @itemize @bullet
248 @item
249 @cite{The extent to which suggestions made by using the @code{register}
250 storage-class specifier are effective (6.7.1).}
251
252 The @code{register} specifier affects code generation only in these ways:
253
254 @itemize @bullet
255 @item
256 When used as part of the register variable extension, see
257 @ref{Explicit Reg Vars}.
258
259 @item
260 When @option{-O0} is in use, the compiler allocates distinct stack
261 memory for all variables that do not have the @code{register}
262 storage-class specifier; if @code{register} is specified, the variable
263 may have a shorter lifespan than the code would indicate and may never
264 be placed in memory.
265
266 @item
267 On some rare x86 targets, @code{setjmp} doesn't save the registers in
268 all circumstances. In those cases, GCC doesn't allocate any variables
269 in registers unless they are marked @code{register}.
270
271 @end itemize
272
273 @item
274 @cite{The extent to which suggestions made by using the inline function
275 specifier are effective (6.7.4).}
276
277 GCC will not inline any functions if the @option{-fno-inline} option is
278 used or if @option{-O0} is used. Otherwise, GCC may still be unable to
279 inline a function for many reasons; the @option{-Winline} option may be
280 used to determine if a function has not been inlined and why not.
281
282 @end itemize
283
284 @node Structures unions enumerations and bit-fields implementation
285 @section Structures, unions, enumerations, and bit-fields
286
287 @itemize @bullet
288 @item
289 @cite{Whether a ``plain'' int bit-field is treated as a @code{signed int}
290 bit-field or as an @code{unsigned int} bit-field (6.7.2, 6.7.2.1).}
291
292 @item
293 @cite{Allowable bit-field types other than @code{_Bool}, @code{signed int},
294 and @code{unsigned int} (6.7.2.1).}
295
296 @item
297 @cite{Whether a bit-field can straddle a storage-unit boundary (6.7.2.1).}
298
299 @item
300 @cite{The order of allocation of bit-fields within a unit (6.7.2.1).}
301
302 @item
303 @cite{The alignment of non-bit-field members of structures (6.7.2.1).}
304
305 @item
306 @cite{The integer type compatible with each enumerated type (6.7.2.2).}
307
308 @end itemize
309
310 @node Qualifiers implementation
311 @section Qualifiers
312
313 @itemize @bullet
314 @item
315 @cite{What constitutes an access to an object that has volatile-qualified
316 type (6.7.3).}
317
318 @end itemize
319
320 @node Preprocessing directives implementation
321 @section Preprocessing directives
322
323 @itemize @bullet
324 @item
325 @cite{How sequences in both forms of header names are mapped to headers
326 or external source file names (6.4.7).}
327
328 @item
329 @cite{Whether the value of a character constant in a constant expression
330 that controls conditional inclusion matches the value of the same character
331 constant in the execution character set (6.10.1).}
332
333 @item
334 @cite{Whether the value of a single-character character constant in a
335 constant expression that controls conditional inclusion may have a
336 negative value (6.10.1).}
337
338 @item
339 @cite{The places that are searched for an included @samp{<>} delimited
340 header, and how the places are specified or the header is
341 identified (6.10.2).}
342
343 @item
344 @cite{How the named source file is searched for in an included @samp{""}
345 delimited header (6.10.2).}
346
347 @item
348 @cite{The method by which preprocessing tokens (possibly resulting from
349 macro expansion) in a @code{#include} directive are combined into a header
350 name (6.10.2).}
351
352 @item
353 @cite{The nesting limit for @code{#include} processing (6.10.2).}
354
355 GCC imposes a limit of 200 nested @code{#include}s.
356
357 @item
358 @cite{Whether the @samp{#} operator inserts a @samp{\} character before
359 the @samp{\} character that begins a universal character name in a
360 character constant or string literal (6.10.3.2).}
361
362 @item
363 @cite{The behavior on each recognized non-@code{STDC #pragma}
364 directive (6.10.6).}
365
366 @item
367 @cite{The definitions for @code{__DATE__} and @code{__TIME__} when
368 respectively, the date and time of translation are not available (6.10.8).}
369
370 If the date and time are not available, @code{__DATE__} expands to
371 @code{@w{"??? ?? ????"}} and @code{__TIME__} expands to
372 @code{"??:??:??"}.
373
374 @end itemize
375
376 @node Library functions implementation
377 @section Library functions
378
379 The behavior of these points are dependent on the implementation
380 of the C library, and are not defined by GCC itself.
381
382 @node Architecture implementation
383 @section Architecture
384
385 @itemize @bullet
386 @item
387 @cite{The values or expressions assigned to the macros specified in the
388 headers @code{<float.h>}, @code{<limits.h>}, and @code{<stdint.h>}
389 (5.2.4.2, 7.18.2, 7.18.3).}
390
391 @item
392 @cite{The number, order, and encoding of bytes in any object
393 (when not explicitly specified in this International Standard) (6.2.6.1).}
394
395 @item
396 @cite{The value of the result of the sizeof operator (6.5.3.4).}
397
398 @end itemize
399
400 @node Locale-specific behavior implementation
401 @section Locale-specific behavior
402
403 The behavior of these points are dependent on the implementation
404 of the C library, and are not defined by GCC itself.
405
406 @node C Extensions
407 @chapter Extensions to the C Language Family
408 @cindex extensions, C language
409 @cindex C language extensions
410
411 @opindex pedantic
412 GNU C provides several language features not found in ISO standard C@.
413 (The @option{-pedantic} option directs GCC to print a warning message if
414 any of these features is used.) To test for the availability of these
415 features in conditional compilation, check for a predefined macro
416 @code{__GNUC__}, which is always defined under GCC@.
417
418 These extensions are available in C and Objective-C@. Most of them are
419 also available in C++. @xref{C++ Extensions,,Extensions to the
420 C++ Language}, for extensions that apply @emph{only} to C++.
421
422 Some features that are in ISO C99 but not C89 or C++ are also, as
423 extensions, accepted by GCC in C89 mode and in C++.
424
425 @menu
426 * Statement Exprs:: Putting statements and declarations inside expressions.
427 * Local Labels:: Labels local to a block.
428 * Labels as Values:: Getting pointers to labels, and computed gotos.
429 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
430 * Constructing Calls:: Dispatching a call to another function.
431 * Typeof:: @code{typeof}: referring to the type of an expression.
432 * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
433 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
434 * Long Long:: Double-word integers---@code{long long int}.
435 * Complex:: Data types for complex numbers.
436 * Hex Floats:: Hexadecimal floating-point constants.
437 * Zero Length:: Zero-length arrays.
438 * Variable Length:: Arrays whose length is computed at run time.
439 * Empty Structures:: Structures with no members.
440 * Variadic Macros:: Macros with a variable number of arguments.
441 * Escaped Newlines:: Slightly looser rules for escaped newlines.
442 * Subscripting:: Any array can be subscripted, even if not an lvalue.
443 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
444 * Initializers:: Non-constant initializers.
445 * Compound Literals:: Compound literals give structures, unions
446 or arrays as values.
447 * Designated Inits:: Labeling elements of initializers.
448 * Cast to Union:: Casting to union type from any member of the union.
449 * Case Ranges:: `case 1 ... 9' and such.
450 * Mixed Declarations:: Mixing declarations and code.
451 * Function Attributes:: Declaring that functions have no side effects,
452 or that they can never return.
453 * Attribute Syntax:: Formal syntax for attributes.
454 * Function Prototypes:: Prototype declarations and old-style definitions.
455 * C++ Comments:: C++ comments are recognized.
456 * Dollar Signs:: Dollar sign is allowed in identifiers.
457 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
458 * Variable Attributes:: Specifying attributes of variables.
459 * Type Attributes:: Specifying attributes of types.
460 * Alignment:: Inquiring about the alignment of a type or variable.
461 * Inline:: Defining inline functions (as fast as macros).
462 * Extended Asm:: Assembler instructions with C expressions as operands.
463 (With them you can define ``built-in'' functions.)
464 * Constraints:: Constraints for asm operands
465 * Asm Labels:: Specifying the assembler name to use for a C symbol.
466 * Explicit Reg Vars:: Defining variables residing in specified registers.
467 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
468 * Incomplete Enums:: @code{enum foo;}, with details to follow.
469 * Function Names:: Printable strings which are the name of the current
470 function.
471 * Return Address:: Getting the return or frame address of a function.
472 * Vector Extensions:: Using vector instructions through built-in functions.
473 * Other Builtins:: Other built-in functions.
474 * Target Builtins:: Built-in functions specific to particular targets.
475 * Pragmas:: Pragmas accepted by GCC.
476 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
477 * Thread-Local:: Per-thread variables.
478 @end menu
479
480 @node Statement Exprs
481 @section Statements and Declarations in Expressions
482 @cindex statements inside expressions
483 @cindex declarations inside expressions
484 @cindex expressions containing statements
485 @cindex macros, statements in expressions
486
487 @c the above section title wrapped and causes an underfull hbox.. i
488 @c changed it from "within" to "in". --mew 4feb93
489 A compound statement enclosed in parentheses may appear as an expression
490 in GNU C@. This allows you to use loops, switches, and local variables
491 within an expression.
492
493 Recall that a compound statement is a sequence of statements surrounded
494 by braces; in this construct, parentheses go around the braces. For
495 example:
496
497 @example
498 (@{ int y = foo (); int z;
499 if (y > 0) z = y;
500 else z = - y;
501 z; @})
502 @end example
503
504 @noindent
505 is a valid (though slightly more complex than necessary) expression
506 for the absolute value of @code{foo ()}.
507
508 The last thing in the compound statement should be an expression
509 followed by a semicolon; the value of this subexpression serves as the
510 value of the entire construct. (If you use some other kind of statement
511 last within the braces, the construct has type @code{void}, and thus
512 effectively no value.)
513
514 This feature is especially useful in making macro definitions ``safe'' (so
515 that they evaluate each operand exactly once). For example, the
516 ``maximum'' function is commonly defined as a macro in standard C as
517 follows:
518
519 @example
520 #define max(a,b) ((a) > (b) ? (a) : (b))
521 @end example
522
523 @noindent
524 @cindex side effects, macro argument
525 But this definition computes either @var{a} or @var{b} twice, with bad
526 results if the operand has side effects. In GNU C, if you know the
527 type of the operands (here let's assume @code{int}), you can define
528 the macro safely as follows:
529
530 @example
531 #define maxint(a,b) \
532 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
533 @end example
534
535 Embedded statements are not allowed in constant expressions, such as
536 the value of an enumeration constant, the width of a bit-field, or
537 the initial value of a static variable.
538
539 If you don't know the type of the operand, you can still do this, but you
540 must use @code{typeof} (@pxref{Typeof}).
541
542 In G++, the result value of a statement expression undergoes array and
543 function pointer decay, and is returned by value to the enclosing
544 expression. For instance, if @code{A} is a class, then
545
546 @smallexample
547 A a;
548
549 (@{a;@}).Foo ()
550 @end smallexample
551
552 @noindent
553 will construct a temporary @code{A} object to hold the result of the
554 statement expression, and that will be used to invoke @code{Foo}.
555 Therefore the @code{this} pointer observed by @code{Foo} will not be the
556 address of @code{a}.
557
558 Any temporaries created within a statement within a statement expression
559 will be destroyed at the statement's end. This makes statement
560 expressions inside macros slightly different from function calls. In
561 the latter case temporaries introduced during argument evaluation will
562 be destroyed at the end of the statement that includes the function
563 call. In the statement expression case they will be destroyed during
564 the statement expression. For instance,
565
566 @smallexample
567 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
568 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
569
570 void foo ()
571 @{
572 macro (X ());
573 function (X ());
574 @}
575 @end smallexample
576
577 @noindent
578 will have different places where temporaries are destroyed. For the
579 @code{macro} case, the temporary @code{X} will be destroyed just after
580 the initialization of @code{b}. In the @code{function} case that
581 temporary will be destroyed when the function returns.
582
583 These considerations mean that it is probably a bad idea to use
584 statement-expressions of this form in header files that are designed to
585 work with C++. (Note that some versions of the GNU C Library contained
586 header files using statement-expression that lead to precisely this
587 bug.)
588
589 @node Local Labels
590 @section Locally Declared Labels
591 @cindex local labels
592 @cindex macros, local labels
593
594 GCC allows you to declare @dfn{local labels} in any nested block
595 scope. A local label is just like an ordinary label, but you can
596 only reference it (with a @code{goto} statement, or by taking its
597 address) within the block in which it was declared.
598
599 A local label declaration looks like this:
600
601 @example
602 __label__ @var{label};
603 @end example
604
605 @noindent
606 or
607
608 @example
609 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
610 @end example
611
612 Local label declarations must come at the beginning of the block,
613 before any ordinary declarations or statements.
614
615 The label declaration defines the label @emph{name}, but does not define
616 the label itself. You must do this in the usual way, with
617 @code{@var{label}:}, within the statements of the statement expression.
618
619 The local label feature is useful for complex macros. If a macro
620 contains nested loops, a @code{goto} can be useful for breaking out of
621 them. However, an ordinary label whose scope is the whole function
622 cannot be used: if the macro can be expanded several times in one
623 function, the label will be multiply defined in that function. A
624 local label avoids this problem. For example:
625
626 @example
627 #define SEARCH(value, array, target) \
628 do @{ \
629 __label__ found; \
630 typeof (target) _SEARCH_target = (target); \
631 typeof (*(array)) *_SEARCH_array = (array); \
632 int i, j; \
633 int value; \
634 for (i = 0; i < max; i++) \
635 for (j = 0; j < max; j++) \
636 if (_SEARCH_array[i][j] == _SEARCH_target) \
637 @{ (value) = i; goto found; @} \
638 (value) = -1; \
639 found:; \
640 @} while (0)
641 @end example
642
643 This could also be written using a statement-expression:
644
645 @example
646 #define SEARCH(array, target) \
647 (@{ \
648 __label__ found; \
649 typeof (target) _SEARCH_target = (target); \
650 typeof (*(array)) *_SEARCH_array = (array); \
651 int i, j; \
652 int value; \
653 for (i = 0; i < max; i++) \
654 for (j = 0; j < max; j++) \
655 if (_SEARCH_array[i][j] == _SEARCH_target) \
656 @{ value = i; goto found; @} \
657 value = -1; \
658 found: \
659 value; \
660 @})
661 @end example
662
663 Local label declarations also make the labels they declare visible to
664 nested functions, if there are any. @xref{Nested Functions}, for details.
665
666 @node Labels as Values
667 @section Labels as Values
668 @cindex labels as values
669 @cindex computed gotos
670 @cindex goto with computed label
671 @cindex address of a label
672
673 You can get the address of a label defined in the current function
674 (or a containing function) with the unary operator @samp{&&}. The
675 value has type @code{void *}. This value is a constant and can be used
676 wherever a constant of that type is valid. For example:
677
678 @example
679 void *ptr;
680 /* @r{@dots{}} */
681 ptr = &&foo;
682 @end example
683
684 To use these values, you need to be able to jump to one. This is done
685 with the computed goto statement@footnote{The analogous feature in
686 Fortran is called an assigned goto, but that name seems inappropriate in
687 C, where one can do more than simply store label addresses in label
688 variables.}, @code{goto *@var{exp};}. For example,
689
690 @example
691 goto *ptr;
692 @end example
693
694 @noindent
695 Any expression of type @code{void *} is allowed.
696
697 One way of using these constants is in initializing a static array that
698 will serve as a jump table:
699
700 @example
701 static void *array[] = @{ &&foo, &&bar, &&hack @};
702 @end example
703
704 Then you can select a label with indexing, like this:
705
706 @example
707 goto *array[i];
708 @end example
709
710 @noindent
711 Note that this does not check whether the subscript is in bounds---array
712 indexing in C never does that.
713
714 Such an array of label values serves a purpose much like that of the
715 @code{switch} statement. The @code{switch} statement is cleaner, so
716 use that rather than an array unless the problem does not fit a
717 @code{switch} statement very well.
718
719 Another use of label values is in an interpreter for threaded code.
720 The labels within the interpreter function can be stored in the
721 threaded code for super-fast dispatching.
722
723 You may not use this mechanism to jump to code in a different function.
724 If you do that, totally unpredictable things will happen. The best way to
725 avoid this is to store the label address only in automatic variables and
726 never pass it as an argument.
727
728 An alternate way to write the above example is
729
730 @example
731 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
732 &&hack - &&foo @};
733 goto *(&&foo + array[i]);
734 @end example
735
736 @noindent
737 This is more friendly to code living in shared libraries, as it reduces
738 the number of dynamic relocations that are needed, and by consequence,
739 allows the data to be read-only.
740
741 @node Nested Functions
742 @section Nested Functions
743 @cindex nested functions
744 @cindex downward funargs
745 @cindex thunks
746
747 A @dfn{nested function} is a function defined inside another function.
748 (Nested functions are not supported for GNU C++.) The nested function's
749 name is local to the block where it is defined. For example, here we
750 define a nested function named @code{square}, and call it twice:
751
752 @example
753 @group
754 foo (double a, double b)
755 @{
756 double square (double z) @{ return z * z; @}
757
758 return square (a) + square (b);
759 @}
760 @end group
761 @end example
762
763 The nested function can access all the variables of the containing
764 function that are visible at the point of its definition. This is
765 called @dfn{lexical scoping}. For example, here we show a nested
766 function which uses an inherited variable named @code{offset}:
767
768 @example
769 @group
770 bar (int *array, int offset, int size)
771 @{
772 int access (int *array, int index)
773 @{ return array[index + offset]; @}
774 int i;
775 /* @r{@dots{}} */
776 for (i = 0; i < size; i++)
777 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
778 @}
779 @end group
780 @end example
781
782 Nested function definitions are permitted within functions in the places
783 where variable definitions are allowed; that is, in any block, before
784 the first statement in the block.
785
786 It is possible to call the nested function from outside the scope of its
787 name by storing its address or passing the address to another function:
788
789 @example
790 hack (int *array, int size)
791 @{
792 void store (int index, int value)
793 @{ array[index] = value; @}
794
795 intermediate (store, size);
796 @}
797 @end example
798
799 Here, the function @code{intermediate} receives the address of
800 @code{store} as an argument. If @code{intermediate} calls @code{store},
801 the arguments given to @code{store} are used to store into @code{array}.
802 But this technique works only so long as the containing function
803 (@code{hack}, in this example) does not exit.
804
805 If you try to call the nested function through its address after the
806 containing function has exited, all hell will break loose. If you try
807 to call it after a containing scope level has exited, and if it refers
808 to some of the variables that are no longer in scope, you may be lucky,
809 but it's not wise to take the risk. If, however, the nested function
810 does not refer to anything that has gone out of scope, you should be
811 safe.
812
813 GCC implements taking the address of a nested function using a technique
814 called @dfn{trampolines}. A paper describing them is available as
815
816 @noindent
817 @uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}.
818
819 A nested function can jump to a label inherited from a containing
820 function, provided the label was explicitly declared in the containing
821 function (@pxref{Local Labels}). Such a jump returns instantly to the
822 containing function, exiting the nested function which did the
823 @code{goto} and any intermediate functions as well. Here is an example:
824
825 @example
826 @group
827 bar (int *array, int offset, int size)
828 @{
829 __label__ failure;
830 int access (int *array, int index)
831 @{
832 if (index > size)
833 goto failure;
834 return array[index + offset];
835 @}
836 int i;
837 /* @r{@dots{}} */
838 for (i = 0; i < size; i++)
839 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
840 /* @r{@dots{}} */
841 return 0;
842
843 /* @r{Control comes here from @code{access}
844 if it detects an error.} */
845 failure:
846 return -1;
847 @}
848 @end group
849 @end example
850
851 A nested function always has internal linkage. Declaring one with
852 @code{extern} is erroneous. If you need to declare the nested function
853 before its definition, use @code{auto} (which is otherwise meaningless
854 for function declarations).
855
856 @example
857 bar (int *array, int offset, int size)
858 @{
859 __label__ failure;
860 auto int access (int *, int);
861 /* @r{@dots{}} */
862 int access (int *array, int index)
863 @{
864 if (index > size)
865 goto failure;
866 return array[index + offset];
867 @}
868 /* @r{@dots{}} */
869 @}
870 @end example
871
872 @node Constructing Calls
873 @section Constructing Function Calls
874 @cindex constructing calls
875 @cindex forwarding calls
876
877 Using the built-in functions described below, you can record
878 the arguments a function received, and call another function
879 with the same arguments, without knowing the number or types
880 of the arguments.
881
882 You can also record the return value of that function call,
883 and later return that value, without knowing what data type
884 the function tried to return (as long as your caller expects
885 that data type).
886
887 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
888 This built-in function returns a pointer to data
889 describing how to perform a call with the same arguments as were passed
890 to the current function.
891
892 The function saves the arg pointer register, structure value address,
893 and all registers that might be used to pass arguments to a function
894 into a block of memory allocated on the stack. Then it returns the
895 address of that block.
896 @end deftypefn
897
898 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
899 This built-in function invokes @var{function}
900 with a copy of the parameters described by @var{arguments}
901 and @var{size}.
902
903 The value of @var{arguments} should be the value returned by
904 @code{__builtin_apply_args}. The argument @var{size} specifies the size
905 of the stack argument data, in bytes.
906
907 This function returns a pointer to data describing
908 how to return whatever value was returned by @var{function}. The data
909 is saved in a block of memory allocated on the stack.
910
911 It is not always simple to compute the proper value for @var{size}. The
912 value is used by @code{__builtin_apply} to compute the amount of data
913 that should be pushed on the stack and copied from the incoming argument
914 area.
915 @end deftypefn
916
917 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
918 This built-in function returns the value described by @var{result} from
919 the containing function. You should specify, for @var{result}, a value
920 returned by @code{__builtin_apply}.
921 @end deftypefn
922
923 @node Typeof
924 @section Referring to a Type with @code{typeof}
925 @findex typeof
926 @findex sizeof
927 @cindex macros, types of arguments
928
929 Another way to refer to the type of an expression is with @code{typeof}.
930 The syntax of using of this keyword looks like @code{sizeof}, but the
931 construct acts semantically like a type name defined with @code{typedef}.
932
933 There are two ways of writing the argument to @code{typeof}: with an
934 expression or with a type. Here is an example with an expression:
935
936 @example
937 typeof (x[0](1))
938 @end example
939
940 @noindent
941 This assumes that @code{x} is an array of pointers to functions;
942 the type described is that of the values of the functions.
943
944 Here is an example with a typename as the argument:
945
946 @example
947 typeof (int *)
948 @end example
949
950 @noindent
951 Here the type described is that of pointers to @code{int}.
952
953 If you are writing a header file that must work when included in ISO C
954 programs, write @code{__typeof__} instead of @code{typeof}.
955 @xref{Alternate Keywords}.
956
957 A @code{typeof}-construct can be used anywhere a typedef name could be
958 used. For example, you can use it in a declaration, in a cast, or inside
959 of @code{sizeof} or @code{typeof}.
960
961 @code{typeof} is often useful in conjunction with the
962 statements-within-expressions feature. Here is how the two together can
963 be used to define a safe ``maximum'' macro that operates on any
964 arithmetic type and evaluates each of its arguments exactly once:
965
966 @example
967 #define max(a,b) \
968 (@{ typeof (a) _a = (a); \
969 typeof (b) _b = (b); \
970 _a > _b ? _a : _b; @})
971 @end example
972
973 @cindex underscores in variables in macros
974 @cindex @samp{_} in variables in macros
975 @cindex local variables in macros
976 @cindex variables, local, in macros
977 @cindex macros, local variables in
978
979 The reason for using names that start with underscores for the local
980 variables is to avoid conflicts with variable names that occur within the
981 expressions that are substituted for @code{a} and @code{b}. Eventually we
982 hope to design a new form of declaration syntax that allows you to declare
983 variables whose scopes start only after their initializers; this will be a
984 more reliable way to prevent such conflicts.
985
986 @noindent
987 Some more examples of the use of @code{typeof}:
988
989 @itemize @bullet
990 @item
991 This declares @code{y} with the type of what @code{x} points to.
992
993 @example
994 typeof (*x) y;
995 @end example
996
997 @item
998 This declares @code{y} as an array of such values.
999
1000 @example
1001 typeof (*x) y[4];
1002 @end example
1003
1004 @item
1005 This declares @code{y} as an array of pointers to characters:
1006
1007 @example
1008 typeof (typeof (char *)[4]) y;
1009 @end example
1010
1011 @noindent
1012 It is equivalent to the following traditional C declaration:
1013
1014 @example
1015 char *y[4];
1016 @end example
1017
1018 To see the meaning of the declaration using @code{typeof}, and why it
1019 might be a useful way to write, let's rewrite it with these macros:
1020
1021 @example
1022 #define pointer(T) typeof(T *)
1023 #define array(T, N) typeof(T [N])
1024 @end example
1025
1026 @noindent
1027 Now the declaration can be rewritten this way:
1028
1029 @example
1030 array (pointer (char), 4) y;
1031 @end example
1032
1033 @noindent
1034 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
1035 pointers to @code{char}.
1036 @end itemize
1037
1038 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
1039 a more limited extension which permitted one to write
1040
1041 @example
1042 typedef @var{T} = @var{expr};
1043 @end example
1044
1045 @noindent
1046 with the effect of declaring @var{T} to have the type of the expression
1047 @var{expr}. This extension does not work with GCC 3 (versions between
1048 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
1049 relies on it should be rewritten to use @code{typeof}:
1050
1051 @example
1052 typedef typeof(@var{expr}) @var{T};
1053 @end example
1054
1055 @noindent
1056 This will work with all versions of GCC@.
1057
1058 @node Lvalues
1059 @section Generalized Lvalues
1060 @cindex compound expressions as lvalues
1061 @cindex expressions, compound, as lvalues
1062 @cindex conditional expressions as lvalues
1063 @cindex expressions, conditional, as lvalues
1064 @cindex casts as lvalues
1065 @cindex generalized lvalues
1066 @cindex lvalues, generalized
1067 @cindex extensions, @code{?:}
1068 @cindex @code{?:} extensions
1069
1070 Compound expressions, conditional expressions and casts are allowed as
1071 lvalues provided their operands are lvalues. This means that you can take
1072 their addresses or store values into them.
1073
1074 Standard C++ allows compound expressions and conditional expressions
1075 as lvalues, and permits casts to reference type, so use of this
1076 extension is not supported for C++ code.
1077
1078 For example, a compound expression can be assigned, provided the last
1079 expression in the sequence is an lvalue. These two expressions are
1080 equivalent:
1081
1082 @example
1083 (a, b) += 5
1084 a, (b += 5)
1085 @end example
1086
1087 Similarly, the address of the compound expression can be taken. These two
1088 expressions are equivalent:
1089
1090 @example
1091 &(a, b)
1092 a, &b
1093 @end example
1094
1095 A conditional expression is a valid lvalue if its type is not void and the
1096 true and false branches are both valid lvalues. For example, these two
1097 expressions are equivalent:
1098
1099 @example
1100 (a ? b : c) = 5
1101 (a ? b = 5 : (c = 5))
1102 @end example
1103
1104 A cast is a valid lvalue if its operand is an lvalue. A simple
1105 assignment whose left-hand side is a cast works by converting the
1106 right-hand side first to the specified type, then to the type of the
1107 inner left-hand side expression. After this is stored, the value is
1108 converted back to the specified type to become the value of the
1109 assignment. Thus, if @code{a} has type @code{char *}, the following two
1110 expressions are equivalent:
1111
1112 @example
1113 (int)a = 5
1114 (int)(a = (char *)(int)5)
1115 @end example
1116
1117 An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1118 performs the arithmetic using the type resulting from the cast, and then
1119 continues as in the previous case. Therefore, these two expressions are
1120 equivalent:
1121
1122 @example
1123 (int)a += 5
1124 (int)(a = (char *)(int) ((int)a + 5))
1125 @end example
1126
1127 You cannot take the address of an lvalue cast, because the use of its
1128 address would not work out coherently. Suppose that @code{&(int)f} were
1129 permitted, where @code{f} has type @code{float}. Then the following
1130 statement would try to store an integer bit-pattern where a floating
1131 point number belongs:
1132
1133 @example
1134 *&(int)f = 1;
1135 @end example
1136
1137 This is quite different from what @code{(int)f = 1} would do---that
1138 would convert 1 to floating point and store it. Rather than cause this
1139 inconsistency, we think it is better to prohibit use of @samp{&} on a cast.
1140
1141 If you really do want an @code{int *} pointer with the address of
1142 @code{f}, you can simply write @code{(int *)&f}.
1143
1144 @node Conditionals
1145 @section Conditionals with Omitted Operands
1146 @cindex conditional expressions, extensions
1147 @cindex omitted middle-operands
1148 @cindex middle-operands, omitted
1149 @cindex extensions, @code{?:}
1150 @cindex @code{?:} extensions
1151
1152 The middle operand in a conditional expression may be omitted. Then
1153 if the first operand is nonzero, its value is the value of the conditional
1154 expression.
1155
1156 Therefore, the expression
1157
1158 @example
1159 x ? : y
1160 @end example
1161
1162 @noindent
1163 has the value of @code{x} if that is nonzero; otherwise, the value of
1164 @code{y}.
1165
1166 This example is perfectly equivalent to
1167
1168 @example
1169 x ? x : y
1170 @end example
1171
1172 @cindex side effect in ?:
1173 @cindex ?: side effect
1174 @noindent
1175 In this simple case, the ability to omit the middle operand is not
1176 especially useful. When it becomes useful is when the first operand does,
1177 or may (if it is a macro argument), contain a side effect. Then repeating
1178 the operand in the middle would perform the side effect twice. Omitting
1179 the middle operand uses the value already computed without the undesirable
1180 effects of recomputing it.
1181
1182 @node Long Long
1183 @section Double-Word Integers
1184 @cindex @code{long long} data types
1185 @cindex double-word arithmetic
1186 @cindex multiprecision arithmetic
1187 @cindex @code{LL} integer suffix
1188 @cindex @code{ULL} integer suffix
1189
1190 ISO C99 supports data types for integers that are at least 64 bits wide,
1191 and as an extension GCC supports them in C89 mode and in C++.
1192 Simply write @code{long long int} for a signed integer, or
1193 @code{unsigned long long int} for an unsigned integer. To make an
1194 integer constant of type @code{long long int}, add the suffix @samp{LL}
1195 to the integer. To make an integer constant of type @code{unsigned long
1196 long int}, add the suffix @samp{ULL} to the integer.
1197
1198 You can use these types in arithmetic like any other integer types.
1199 Addition, subtraction, and bitwise boolean operations on these types
1200 are open-coded on all types of machines. Multiplication is open-coded
1201 if the machine supports fullword-to-doubleword a widening multiply
1202 instruction. Division and shifts are open-coded only on machines that
1203 provide special support. The operations that are not open-coded use
1204 special library routines that come with GCC@.
1205
1206 There may be pitfalls when you use @code{long long} types for function
1207 arguments, unless you declare function prototypes. If a function
1208 expects type @code{int} for its argument, and you pass a value of type
1209 @code{long long int}, confusion will result because the caller and the
1210 subroutine will disagree about the number of bytes for the argument.
1211 Likewise, if the function expects @code{long long int} and you pass
1212 @code{int}. The best way to avoid such problems is to use prototypes.
1213
1214 @node Complex
1215 @section Complex Numbers
1216 @cindex complex numbers
1217 @cindex @code{_Complex} keyword
1218 @cindex @code{__complex__} keyword
1219
1220 ISO C99 supports complex floating data types, and as an extension GCC
1221 supports them in C89 mode and in C++, and supports complex integer data
1222 types which are not part of ISO C99. You can declare complex types
1223 using the keyword @code{_Complex}. As an extension, the older GNU
1224 keyword @code{__complex__} is also supported.
1225
1226 For example, @samp{_Complex double x;} declares @code{x} as a
1227 variable whose real part and imaginary part are both of type
1228 @code{double}. @samp{_Complex short int y;} declares @code{y} to
1229 have real and imaginary parts of type @code{short int}; this is not
1230 likely to be useful, but it shows that the set of complex types is
1231 complete.
1232
1233 To write a constant with a complex data type, use the suffix @samp{i} or
1234 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
1235 has type @code{_Complex float} and @code{3i} has type
1236 @code{_Complex int}. Such a constant always has a pure imaginary
1237 value, but you can form any complex value you like by adding one to a
1238 real constant. This is a GNU extension; if you have an ISO C99
1239 conforming C library (such as GNU libc), and want to construct complex
1240 constants of floating type, you should include @code{<complex.h>} and
1241 use the macros @code{I} or @code{_Complex_I} instead.
1242
1243 @cindex @code{__real__} keyword
1244 @cindex @code{__imag__} keyword
1245 To extract the real part of a complex-valued expression @var{exp}, write
1246 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
1247 extract the imaginary part. This is a GNU extension; for values of
1248 floating type, you should use the ISO C99 functions @code{crealf},
1249 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
1250 @code{cimagl}, declared in @code{<complex.h>} and also provided as
1251 built-in functions by GCC@.
1252
1253 @cindex complex conjugation
1254 The operator @samp{~} performs complex conjugation when used on a value
1255 with a complex type. This is a GNU extension; for values of
1256 floating type, you should use the ISO C99 functions @code{conjf},
1257 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
1258 provided as built-in functions by GCC@.
1259
1260 GCC can allocate complex automatic variables in a noncontiguous
1261 fashion; it's even possible for the real part to be in a register while
1262 the imaginary part is on the stack (or vice-versa). Only the DWARF2
1263 debug info format can represent this, so use of DWARF2 is recommended.
1264 If you are using the stabs debug info format, GCC describes a noncontiguous
1265 complex variable as if it were two separate variables of noncomplex type.
1266 If the variable's actual name is @code{foo}, the two fictitious
1267 variables are named @code{foo$real} and @code{foo$imag}. You can
1268 examine and set these two fictitious variables with your debugger.
1269
1270 @node Hex Floats
1271 @section Hex Floats
1272 @cindex hex floats
1273
1274 ISO C99 supports floating-point numbers written not only in the usual
1275 decimal notation, such as @code{1.55e1}, but also numbers such as
1276 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1277 supports this in C89 mode (except in some cases when strictly
1278 conforming) and in C++. In that format the
1279 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1280 mandatory. The exponent is a decimal number that indicates the power of
1281 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1282 @tex
1283 $1 {15\over16}$,
1284 @end tex
1285 @ifnottex
1286 1 15/16,
1287 @end ifnottex
1288 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1289 is the same as @code{1.55e1}.
1290
1291 Unlike for floating-point numbers in the decimal notation the exponent
1292 is always required in the hexadecimal notation. Otherwise the compiler
1293 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1294 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1295 extension for floating-point constants of type @code{float}.
1296
1297 @node Zero Length
1298 @section Arrays of Length Zero
1299 @cindex arrays of length zero
1300 @cindex zero-length arrays
1301 @cindex length-zero arrays
1302 @cindex flexible array members
1303
1304 Zero-length arrays are allowed in GNU C@. They are very useful as the
1305 last element of a structure which is really a header for a variable-length
1306 object:
1307
1308 @example
1309 struct line @{
1310 int length;
1311 char contents[0];
1312 @};
1313
1314 struct line *thisline = (struct line *)
1315 malloc (sizeof (struct line) + this_length);
1316 thisline->length = this_length;
1317 @end example
1318
1319 In ISO C90, you would have to give @code{contents} a length of 1, which
1320 means either you waste space or complicate the argument to @code{malloc}.
1321
1322 In ISO C99, you would use a @dfn{flexible array member}, which is
1323 slightly different in syntax and semantics:
1324
1325 @itemize @bullet
1326 @item
1327 Flexible array members are written as @code{contents[]} without
1328 the @code{0}.
1329
1330 @item
1331 Flexible array members have incomplete type, and so the @code{sizeof}
1332 operator may not be applied. As a quirk of the original implementation
1333 of zero-length arrays, @code{sizeof} evaluates to zero.
1334
1335 @item
1336 Flexible array members may only appear as the last member of a
1337 @code{struct} that is otherwise non-empty.
1338
1339 @item
1340 A structure containing a flexible array member, or a union containing
1341 such a structure (possibly recursively), may not be a member of a
1342 structure or an element of an array. (However, these uses are
1343 permitted by GCC as extensions.)
1344 @end itemize
1345
1346 GCC versions before 3.0 allowed zero-length arrays to be statically
1347 initialized, as if they were flexible arrays. In addition to those
1348 cases that were useful, it also allowed initializations in situations
1349 that would corrupt later data. Non-empty initialization of zero-length
1350 arrays is now treated like any case where there are more initializer
1351 elements than the array holds, in that a suitable warning about "excess
1352 elements in array" is given, and the excess elements (all of them, in
1353 this case) are ignored.
1354
1355 Instead GCC allows static initialization of flexible array members.
1356 This is equivalent to defining a new structure containing the original
1357 structure followed by an array of sufficient size to contain the data.
1358 I.e.@: in the following, @code{f1} is constructed as if it were declared
1359 like @code{f2}.
1360
1361 @example
1362 struct f1 @{
1363 int x; int y[];
1364 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1365
1366 struct f2 @{
1367 struct f1 f1; int data[3];
1368 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1369 @end example
1370
1371 @noindent
1372 The convenience of this extension is that @code{f1} has the desired
1373 type, eliminating the need to consistently refer to @code{f2.f1}.
1374
1375 This has symmetry with normal static arrays, in that an array of
1376 unknown size is also written with @code{[]}.
1377
1378 Of course, this extension only makes sense if the extra data comes at
1379 the end of a top-level object, as otherwise we would be overwriting
1380 data at subsequent offsets. To avoid undue complication and confusion
1381 with initialization of deeply nested arrays, we simply disallow any
1382 non-empty initialization except when the structure is the top-level
1383 object. For example:
1384
1385 @example
1386 struct foo @{ int x; int y[]; @};
1387 struct bar @{ struct foo z; @};
1388
1389 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1390 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1391 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1392 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1393 @end example
1394
1395 @node Empty Structures
1396 @section Structures With No Members
1397 @cindex empty structures
1398 @cindex zero-size structures
1399
1400 GCC permits a C structure to have no members:
1401
1402 @example
1403 struct empty @{
1404 @};
1405 @end example
1406
1407 The structure will have size zero. In C++, empty structures are part
1408 of the language. G++ treats empty structures as if they had a single
1409 member of type @code{char}.
1410
1411 @node Variable Length
1412 @section Arrays of Variable Length
1413 @cindex variable-length arrays
1414 @cindex arrays of variable length
1415 @cindex VLAs
1416
1417 Variable-length automatic arrays are allowed in ISO C99, and as an
1418 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1419 implementation of variable-length arrays does not yet conform in detail
1420 to the ISO C99 standard.) These arrays are
1421 declared like any other automatic arrays, but with a length that is not
1422 a constant expression. The storage is allocated at the point of
1423 declaration and deallocated when the brace-level is exited. For
1424 example:
1425
1426 @example
1427 FILE *
1428 concat_fopen (char *s1, char *s2, char *mode)
1429 @{
1430 char str[strlen (s1) + strlen (s2) + 1];
1431 strcpy (str, s1);
1432 strcat (str, s2);
1433 return fopen (str, mode);
1434 @}
1435 @end example
1436
1437 @cindex scope of a variable length array
1438 @cindex variable-length array scope
1439 @cindex deallocating variable length arrays
1440 Jumping or breaking out of the scope of the array name deallocates the
1441 storage. Jumping into the scope is not allowed; you get an error
1442 message for it.
1443
1444 @cindex @code{alloca} vs variable-length arrays
1445 You can use the function @code{alloca} to get an effect much like
1446 variable-length arrays. The function @code{alloca} is available in
1447 many other C implementations (but not in all). On the other hand,
1448 variable-length arrays are more elegant.
1449
1450 There are other differences between these two methods. Space allocated
1451 with @code{alloca} exists until the containing @emph{function} returns.
1452 The space for a variable-length array is deallocated as soon as the array
1453 name's scope ends. (If you use both variable-length arrays and
1454 @code{alloca} in the same function, deallocation of a variable-length array
1455 will also deallocate anything more recently allocated with @code{alloca}.)
1456
1457 You can also use variable-length arrays as arguments to functions:
1458
1459 @example
1460 struct entry
1461 tester (int len, char data[len][len])
1462 @{
1463 /* @r{@dots{}} */
1464 @}
1465 @end example
1466
1467 The length of an array is computed once when the storage is allocated
1468 and is remembered for the scope of the array in case you access it with
1469 @code{sizeof}.
1470
1471 If you want to pass the array first and the length afterward, you can
1472 use a forward declaration in the parameter list---another GNU extension.
1473
1474 @example
1475 struct entry
1476 tester (int len; char data[len][len], int len)
1477 @{
1478 /* @r{@dots{}} */
1479 @}
1480 @end example
1481
1482 @cindex parameter forward declaration
1483 The @samp{int len} before the semicolon is a @dfn{parameter forward
1484 declaration}, and it serves the purpose of making the name @code{len}
1485 known when the declaration of @code{data} is parsed.
1486
1487 You can write any number of such parameter forward declarations in the
1488 parameter list. They can be separated by commas or semicolons, but the
1489 last one must end with a semicolon, which is followed by the ``real''
1490 parameter declarations. Each forward declaration must match a ``real''
1491 declaration in parameter name and data type. ISO C99 does not support
1492 parameter forward declarations.
1493
1494 @node Variadic Macros
1495 @section Macros with a Variable Number of Arguments.
1496 @cindex variable number of arguments
1497 @cindex macro with variable arguments
1498 @cindex rest argument (in macro)
1499 @cindex variadic macros
1500
1501 In the ISO C standard of 1999, a macro can be declared to accept a
1502 variable number of arguments much as a function can. The syntax for
1503 defining the macro is similar to that of a function. Here is an
1504 example:
1505
1506 @smallexample
1507 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1508 @end smallexample
1509
1510 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1511 such a macro, it represents the zero or more tokens until the closing
1512 parenthesis that ends the invocation, including any commas. This set of
1513 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1514 wherever it appears. See the CPP manual for more information.
1515
1516 GCC has long supported variadic macros, and used a different syntax that
1517 allowed you to give a name to the variable arguments just like any other
1518 argument. Here is an example:
1519
1520 @example
1521 #define debug(format, args...) fprintf (stderr, format, args)
1522 @end example
1523
1524 This is in all ways equivalent to the ISO C example above, but arguably
1525 more readable and descriptive.
1526
1527 GNU CPP has two further variadic macro extensions, and permits them to
1528 be used with either of the above forms of macro definition.
1529
1530 In standard C, you are not allowed to leave the variable argument out
1531 entirely; but you are allowed to pass an empty argument. For example,
1532 this invocation is invalid in ISO C, because there is no comma after
1533 the string:
1534
1535 @example
1536 debug ("A message")
1537 @end example
1538
1539 GNU CPP permits you to completely omit the variable arguments in this
1540 way. In the above examples, the compiler would complain, though since
1541 the expansion of the macro still has the extra comma after the format
1542 string.
1543
1544 To help solve this problem, CPP behaves specially for variable arguments
1545 used with the token paste operator, @samp{##}. If instead you write
1546
1547 @smallexample
1548 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1549 @end smallexample
1550
1551 and if the variable arguments are omitted or empty, the @samp{##}
1552 operator causes the preprocessor to remove the comma before it. If you
1553 do provide some variable arguments in your macro invocation, GNU CPP
1554 does not complain about the paste operation and instead places the
1555 variable arguments after the comma. Just like any other pasted macro
1556 argument, these arguments are not macro expanded.
1557
1558 @node Escaped Newlines
1559 @section Slightly Looser Rules for Escaped Newlines
1560 @cindex escaped newlines
1561 @cindex newlines (escaped)
1562
1563 Recently, the preprocessor has relaxed its treatment of escaped
1564 newlines. Previously, the newline had to immediately follow a
1565 backslash. The current implementation allows whitespace in the form
1566 of spaces, horizontal and vertical tabs, and form feeds between the
1567 backslash and the subsequent newline. The preprocessor issues a
1568 warning, but treats it as a valid escaped newline and combines the two
1569 lines to form a single logical line. This works within comments and
1570 tokens, as well as between tokens. Comments are @emph{not} treated as
1571 whitespace for the purposes of this relaxation, since they have not
1572 yet been replaced with spaces.
1573
1574 @node Subscripting
1575 @section Non-Lvalue Arrays May Have Subscripts
1576 @cindex subscripting
1577 @cindex arrays, non-lvalue
1578
1579 @cindex subscripting and function values
1580 In ISO C99, arrays that are not lvalues still decay to pointers, and
1581 may be subscripted, although they may not be modified or used after
1582 the next sequence point and the unary @samp{&} operator may not be
1583 applied to them. As an extension, GCC allows such arrays to be
1584 subscripted in C89 mode, though otherwise they do not decay to
1585 pointers outside C99 mode. For example,
1586 this is valid in GNU C though not valid in C89:
1587
1588 @example
1589 @group
1590 struct foo @{int a[4];@};
1591
1592 struct foo f();
1593
1594 bar (int index)
1595 @{
1596 return f().a[index];
1597 @}
1598 @end group
1599 @end example
1600
1601 @node Pointer Arith
1602 @section Arithmetic on @code{void}- and Function-Pointers
1603 @cindex void pointers, arithmetic
1604 @cindex void, size of pointer to
1605 @cindex function pointers, arithmetic
1606 @cindex function, size of pointer to
1607
1608 In GNU C, addition and subtraction operations are supported on pointers to
1609 @code{void} and on pointers to functions. This is done by treating the
1610 size of a @code{void} or of a function as 1.
1611
1612 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1613 and on function types, and returns 1.
1614
1615 @opindex Wpointer-arith
1616 The option @option{-Wpointer-arith} requests a warning if these extensions
1617 are used.
1618
1619 @node Initializers
1620 @section Non-Constant Initializers
1621 @cindex initializers, non-constant
1622 @cindex non-constant initializers
1623
1624 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1625 automatic variable are not required to be constant expressions in GNU C@.
1626 Here is an example of an initializer with run-time varying elements:
1627
1628 @example
1629 foo (float f, float g)
1630 @{
1631 float beat_freqs[2] = @{ f-g, f+g @};
1632 /* @r{@dots{}} */
1633 @}
1634 @end example
1635
1636 @node Compound Literals
1637 @section Compound Literals
1638 @cindex constructor expressions
1639 @cindex initializations in expressions
1640 @cindex structures, constructor expression
1641 @cindex expressions, constructor
1642 @cindex compound literals
1643 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1644
1645 ISO C99 supports compound literals. A compound literal looks like
1646 a cast containing an initializer. Its value is an object of the
1647 type specified in the cast, containing the elements specified in
1648 the initializer; it is an lvalue. As an extension, GCC supports
1649 compound literals in C89 mode and in C++.
1650
1651 Usually, the specified type is a structure. Assume that
1652 @code{struct foo} and @code{structure} are declared as shown:
1653
1654 @example
1655 struct foo @{int a; char b[2];@} structure;
1656 @end example
1657
1658 @noindent
1659 Here is an example of constructing a @code{struct foo} with a compound literal:
1660
1661 @example
1662 structure = ((struct foo) @{x + y, 'a', 0@});
1663 @end example
1664
1665 @noindent
1666 This is equivalent to writing the following:
1667
1668 @example
1669 @{
1670 struct foo temp = @{x + y, 'a', 0@};
1671 structure = temp;
1672 @}
1673 @end example
1674
1675 You can also construct an array. If all the elements of the compound literal
1676 are (made up of) simple constant expressions, suitable for use in
1677 initializers of objects of static storage duration, then the compound
1678 literal can be coerced to a pointer to its first element and used in
1679 such an initializer, as shown here:
1680
1681 @example
1682 char **foo = (char *[]) @{ "x", "y", "z" @};
1683 @end example
1684
1685 Compound literals for scalar types and union types are is
1686 also allowed, but then the compound literal is equivalent
1687 to a cast.
1688
1689 As a GNU extension, GCC allows initialization of objects with static storage
1690 duration by compound literals (which is not possible in ISO C99, because
1691 the initializer is not a constant).
1692 It is handled as if the object was initialized only with the bracket
1693 enclosed list if compound literal's and object types match.
1694 The initializer list of the compound literal must be constant.
1695 If the object being initialized has array type of unknown size, the size is
1696 determined by compound literal size.
1697
1698 @example
1699 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1700 static int y[] = (int []) @{1, 2, 3@};
1701 static int z[] = (int [3]) @{1@};
1702 @end example
1703
1704 @noindent
1705 The above lines are equivalent to the following:
1706 @example
1707 static struct foo x = @{1, 'a', 'b'@};
1708 static int y[] = @{1, 2, 3@};
1709 static int z[] = @{1, 0, 0@};
1710 @end example
1711
1712 @node Designated Inits
1713 @section Designated Initializers
1714 @cindex initializers with labeled elements
1715 @cindex labeled elements in initializers
1716 @cindex case labels in initializers
1717 @cindex designated initializers
1718
1719 Standard C89 requires the elements of an initializer to appear in a fixed
1720 order, the same as the order of the elements in the array or structure
1721 being initialized.
1722
1723 In ISO C99 you can give the elements in any order, specifying the array
1724 indices or structure field names they apply to, and GNU C allows this as
1725 an extension in C89 mode as well. This extension is not
1726 implemented in GNU C++.
1727
1728 To specify an array index, write
1729 @samp{[@var{index}] =} before the element value. For example,
1730
1731 @example
1732 int a[6] = @{ [4] = 29, [2] = 15 @};
1733 @end example
1734
1735 @noindent
1736 is equivalent to
1737
1738 @example
1739 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1740 @end example
1741
1742 @noindent
1743 The index values must be constant expressions, even if the array being
1744 initialized is automatic.
1745
1746 An alternative syntax for this which has been obsolete since GCC 2.5 but
1747 GCC still accepts is to write @samp{[@var{index}]} before the element
1748 value, with no @samp{=}.
1749
1750 To initialize a range of elements to the same value, write
1751 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1752 extension. For example,
1753
1754 @example
1755 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1756 @end example
1757
1758 @noindent
1759 If the value in it has side-effects, the side-effects will happen only once,
1760 not for each initialized field by the range initializer.
1761
1762 @noindent
1763 Note that the length of the array is the highest value specified
1764 plus one.
1765
1766 In a structure initializer, specify the name of a field to initialize
1767 with @samp{.@var{fieldname} =} before the element value. For example,
1768 given the following structure,
1769
1770 @example
1771 struct point @{ int x, y; @};
1772 @end example
1773
1774 @noindent
1775 the following initialization
1776
1777 @example
1778 struct point p = @{ .y = yvalue, .x = xvalue @};
1779 @end example
1780
1781 @noindent
1782 is equivalent to
1783
1784 @example
1785 struct point p = @{ xvalue, yvalue @};
1786 @end example
1787
1788 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1789 @samp{@var{fieldname}:}, as shown here:
1790
1791 @example
1792 struct point p = @{ y: yvalue, x: xvalue @};
1793 @end example
1794
1795 @cindex designators
1796 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1797 @dfn{designator}. You can also use a designator (or the obsolete colon
1798 syntax) when initializing a union, to specify which element of the union
1799 should be used. For example,
1800
1801 @example
1802 union foo @{ int i; double d; @};
1803
1804 union foo f = @{ .d = 4 @};
1805 @end example
1806
1807 @noindent
1808 will convert 4 to a @code{double} to store it in the union using
1809 the second element. By contrast, casting 4 to type @code{union foo}
1810 would store it into the union as the integer @code{i}, since it is
1811 an integer. (@xref{Cast to Union}.)
1812
1813 You can combine this technique of naming elements with ordinary C
1814 initialization of successive elements. Each initializer element that
1815 does not have a designator applies to the next consecutive element of the
1816 array or structure. For example,
1817
1818 @example
1819 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1820 @end example
1821
1822 @noindent
1823 is equivalent to
1824
1825 @example
1826 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1827 @end example
1828
1829 Labeling the elements of an array initializer is especially useful
1830 when the indices are characters or belong to an @code{enum} type.
1831 For example:
1832
1833 @example
1834 int whitespace[256]
1835 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1836 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1837 @end example
1838
1839 @cindex designator lists
1840 You can also write a series of @samp{.@var{fieldname}} and
1841 @samp{[@var{index}]} designators before an @samp{=} to specify a
1842 nested subobject to initialize; the list is taken relative to the
1843 subobject corresponding to the closest surrounding brace pair. For
1844 example, with the @samp{struct point} declaration above:
1845
1846 @smallexample
1847 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1848 @end smallexample
1849
1850 @noindent
1851 If the same field is initialized multiple times, it will have value from
1852 the last initialization. If any such overridden initialization has
1853 side-effect, it is unspecified whether the side-effect happens or not.
1854 Currently, gcc will discard them and issue a warning.
1855
1856 @node Case Ranges
1857 @section Case Ranges
1858 @cindex case ranges
1859 @cindex ranges in case statements
1860
1861 You can specify a range of consecutive values in a single @code{case} label,
1862 like this:
1863
1864 @example
1865 case @var{low} ... @var{high}:
1866 @end example
1867
1868 @noindent
1869 This has the same effect as the proper number of individual @code{case}
1870 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1871
1872 This feature is especially useful for ranges of ASCII character codes:
1873
1874 @example
1875 case 'A' ... 'Z':
1876 @end example
1877
1878 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1879 it may be parsed wrong when you use it with integer values. For example,
1880 write this:
1881
1882 @example
1883 case 1 ... 5:
1884 @end example
1885
1886 @noindent
1887 rather than this:
1888
1889 @example
1890 case 1...5:
1891 @end example
1892
1893 @node Cast to Union
1894 @section Cast to a Union Type
1895 @cindex cast to a union
1896 @cindex union, casting to a
1897
1898 A cast to union type is similar to other casts, except that the type
1899 specified is a union type. You can specify the type either with
1900 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1901 a constructor though, not a cast, and hence does not yield an lvalue like
1902 normal casts. (@xref{Compound Literals}.)
1903
1904 The types that may be cast to the union type are those of the members
1905 of the union. Thus, given the following union and variables:
1906
1907 @example
1908 union foo @{ int i; double d; @};
1909 int x;
1910 double y;
1911 @end example
1912
1913 @noindent
1914 both @code{x} and @code{y} can be cast to type @code{union foo}.
1915
1916 Using the cast as the right-hand side of an assignment to a variable of
1917 union type is equivalent to storing in a member of the union:
1918
1919 @example
1920 union foo u;
1921 /* @r{@dots{}} */
1922 u = (union foo) x @equiv{} u.i = x
1923 u = (union foo) y @equiv{} u.d = y
1924 @end example
1925
1926 You can also use the union cast as a function argument:
1927
1928 @example
1929 void hack (union foo);
1930 /* @r{@dots{}} */
1931 hack ((union foo) x);
1932 @end example
1933
1934 @node Mixed Declarations
1935 @section Mixed Declarations and Code
1936 @cindex mixed declarations and code
1937 @cindex declarations, mixed with code
1938 @cindex code, mixed with declarations
1939
1940 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1941 within compound statements. As an extension, GCC also allows this in
1942 C89 mode. For example, you could do:
1943
1944 @example
1945 int i;
1946 /* @r{@dots{}} */
1947 i++;
1948 int j = i + 2;
1949 @end example
1950
1951 Each identifier is visible from where it is declared until the end of
1952 the enclosing block.
1953
1954 @node Function Attributes
1955 @section Declaring Attributes of Functions
1956 @cindex function attributes
1957 @cindex declaring attributes of functions
1958 @cindex functions that never return
1959 @cindex functions that have no side effects
1960 @cindex functions in arbitrary sections
1961 @cindex functions that behave like malloc
1962 @cindex @code{volatile} applied to function
1963 @cindex @code{const} applied to function
1964 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1965 @cindex functions with non-null pointer arguments
1966 @cindex functions that are passed arguments in registers on the 386
1967 @cindex functions that pop the argument stack on the 386
1968 @cindex functions that do not pop the argument stack on the 386
1969
1970 In GNU C, you declare certain things about functions called in your program
1971 which help the compiler optimize function calls and check your code more
1972 carefully.
1973
1974 The keyword @code{__attribute__} allows you to specify special
1975 attributes when making a declaration. This keyword is followed by an
1976 attribute specification inside double parentheses. The following
1977 attributes are currently defined for functions on all targets:
1978 @code{noreturn}, @code{noinline}, @code{always_inline},
1979 @code{pure}, @code{const}, @code{nothrow},
1980 @code{format}, @code{format_arg}, @code{no_instrument_function},
1981 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1982 @code{unused}, @code{deprecated}, @code{weak}, @code{malloc},
1983 @code{alias}, @code{warn_unused_result} and @code{nonnull}. Several other
1984 attributes are defined for functions on particular target systems. Other
1985 attributes, including @code{section} are supported for variables declarations
1986 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1987
1988 You may also specify attributes with @samp{__} preceding and following
1989 each keyword. This allows you to use them in header files without
1990 being concerned about a possible macro of the same name. For example,
1991 you may use @code{__noreturn__} instead of @code{noreturn}.
1992
1993 @xref{Attribute Syntax}, for details of the exact syntax for using
1994 attributes.
1995
1996 @table @code
1997 @cindex @code{noreturn} function attribute
1998 @item noreturn
1999 A few standard library functions, such as @code{abort} and @code{exit},
2000 cannot return. GCC knows this automatically. Some programs define
2001 their own functions that never return. You can declare them
2002 @code{noreturn} to tell the compiler this fact. For example,
2003
2004 @smallexample
2005 @group
2006 void fatal () __attribute__ ((noreturn));
2007
2008 void
2009 fatal (/* @r{@dots{}} */)
2010 @{
2011 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2012 exit (1);
2013 @}
2014 @end group
2015 @end smallexample
2016
2017 The @code{noreturn} keyword tells the compiler to assume that
2018 @code{fatal} cannot return. It can then optimize without regard to what
2019 would happen if @code{fatal} ever did return. This makes slightly
2020 better code. More importantly, it helps avoid spurious warnings of
2021 uninitialized variables.
2022
2023 Do not assume that registers saved by the calling function are
2024 restored before calling the @code{noreturn} function.
2025
2026 It does not make sense for a @code{noreturn} function to have a return
2027 type other than @code{void}.
2028
2029 The attribute @code{noreturn} is not implemented in GCC versions
2030 earlier than 2.5. An alternative way to declare that a function does
2031 not return, which works in the current version and in some older
2032 versions, is as follows:
2033
2034 @smallexample
2035 typedef void voidfn ();
2036
2037 volatile voidfn fatal;
2038 @end smallexample
2039
2040 @cindex @code{noinline} function attribute
2041 @item noinline
2042 This function attribute prevents a function from being considered for
2043 inlining.
2044
2045 @cindex @code{always_inline} function attribute
2046 @item always_inline
2047 Generally, functions are not inlined unless optimization is specified.
2048 For functions declared inline, this attribute inlines the function even
2049 if no optimization level was specified.
2050
2051 @cindex @code{pure} function attribute
2052 @item pure
2053 Many functions have no effects except the return value and their
2054 return value depends only on the parameters and/or global variables.
2055 Such a function can be subject
2056 to common subexpression elimination and loop optimization just as an
2057 arithmetic operator would be. These functions should be declared
2058 with the attribute @code{pure}. For example,
2059
2060 @smallexample
2061 int square (int) __attribute__ ((pure));
2062 @end smallexample
2063
2064 @noindent
2065 says that the hypothetical function @code{square} is safe to call
2066 fewer times than the program says.
2067
2068 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2069 Interesting non-pure functions are functions with infinite loops or those
2070 depending on volatile memory or other system resource, that may change between
2071 two consecutive calls (such as @code{feof} in a multithreading environment).
2072
2073 The attribute @code{pure} is not implemented in GCC versions earlier
2074 than 2.96.
2075 @cindex @code{const} function attribute
2076 @item const
2077 Many functions do not examine any values except their arguments, and
2078 have no effects except the return value. Basically this is just slightly
2079 more strict class than the @code{pure} attribute above, since function is not
2080 allowed to read global memory.
2081
2082 @cindex pointer arguments
2083 Note that a function that has pointer arguments and examines the data
2084 pointed to must @emph{not} be declared @code{const}. Likewise, a
2085 function that calls a non-@code{const} function usually must not be
2086 @code{const}. It does not make sense for a @code{const} function to
2087 return @code{void}.
2088
2089 The attribute @code{const} is not implemented in GCC versions earlier
2090 than 2.5. An alternative way to declare that a function has no side
2091 effects, which works in the current version and in some older versions,
2092 is as follows:
2093
2094 @smallexample
2095 typedef int intfn ();
2096
2097 extern const intfn square;
2098 @end smallexample
2099
2100 This approach does not work in GNU C++ from 2.6.0 on, since the language
2101 specifies that the @samp{const} must be attached to the return value.
2102
2103 @cindex @code{nothrow} function attribute
2104 @item nothrow
2105 The @code{nothrow} attribute is used to inform the compiler that a
2106 function cannot throw an exception. For example, most functions in
2107 the standard C library can be guaranteed not to throw an exception
2108 with the notable exceptions of @code{qsort} and @code{bsearch} that
2109 take function pointer arguments. The @code{nothrow} attribute is not
2110 implemented in GCC versions earlier than 3.2.
2111
2112 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2113 @cindex @code{format} function attribute
2114 @opindex Wformat
2115 The @code{format} attribute specifies that a function takes @code{printf},
2116 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2117 should be type-checked against a format string. For example, the
2118 declaration:
2119
2120 @smallexample
2121 extern int
2122 my_printf (void *my_object, const char *my_format, ...)
2123 __attribute__ ((format (printf, 2, 3)));
2124 @end smallexample
2125
2126 @noindent
2127 causes the compiler to check the arguments in calls to @code{my_printf}
2128 for consistency with the @code{printf} style format string argument
2129 @code{my_format}.
2130
2131 The parameter @var{archetype} determines how the format string is
2132 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
2133 or @code{strfmon}. (You can also use @code{__printf__},
2134 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
2135 parameter @var{string-index} specifies which argument is the format
2136 string argument (starting from 1), while @var{first-to-check} is the
2137 number of the first argument to check against the format string. For
2138 functions where the arguments are not available to be checked (such as
2139 @code{vprintf}), specify the third parameter as zero. In this case the
2140 compiler only checks the format string for consistency. For
2141 @code{strftime} formats, the third parameter is required to be zero.
2142 Since non-static C++ methods have an implicit @code{this} argument, the
2143 arguments of such methods should be counted from two, not one, when
2144 giving values for @var{string-index} and @var{first-to-check}.
2145
2146 In the example above, the format string (@code{my_format}) is the second
2147 argument of the function @code{my_print}, and the arguments to check
2148 start with the third argument, so the correct parameters for the format
2149 attribute are 2 and 3.
2150
2151 @opindex ffreestanding
2152 The @code{format} attribute allows you to identify your own functions
2153 which take format strings as arguments, so that GCC can check the
2154 calls to these functions for errors. The compiler always (unless
2155 @option{-ffreestanding} is used) checks formats
2156 for the standard library functions @code{printf}, @code{fprintf},
2157 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2158 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2159 warnings are requested (using @option{-Wformat}), so there is no need to
2160 modify the header file @file{stdio.h}. In C99 mode, the functions
2161 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2162 @code{vsscanf} are also checked. Except in strictly conforming C
2163 standard modes, the X/Open function @code{strfmon} is also checked as
2164 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2165 @xref{C Dialect Options,,Options Controlling C Dialect}.
2166
2167 @item format_arg (@var{string-index})
2168 @cindex @code{format_arg} function attribute
2169 @opindex Wformat-nonliteral
2170 The @code{format_arg} attribute specifies that a function takes a format
2171 string for a @code{printf}, @code{scanf}, @code{strftime} or
2172 @code{strfmon} style function and modifies it (for example, to translate
2173 it into another language), so the result can be passed to a
2174 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2175 function (with the remaining arguments to the format function the same
2176 as they would have been for the unmodified string). For example, the
2177 declaration:
2178
2179 @smallexample
2180 extern char *
2181 my_dgettext (char *my_domain, const char *my_format)
2182 __attribute__ ((format_arg (2)));
2183 @end smallexample
2184
2185 @noindent
2186 causes the compiler to check the arguments in calls to a @code{printf},
2187 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2188 format string argument is a call to the @code{my_dgettext} function, for
2189 consistency with the format string argument @code{my_format}. If the
2190 @code{format_arg} attribute had not been specified, all the compiler
2191 could tell in such calls to format functions would be that the format
2192 string argument is not constant; this would generate a warning when
2193 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2194 without the attribute.
2195
2196 The parameter @var{string-index} specifies which argument is the format
2197 string argument (starting from one). Since non-static C++ methods have
2198 an implicit @code{this} argument, the arguments of such methods should
2199 be counted from two.
2200
2201 The @code{format-arg} attribute allows you to identify your own
2202 functions which modify format strings, so that GCC can check the
2203 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2204 type function whose operands are a call to one of your own function.
2205 The compiler always treats @code{gettext}, @code{dgettext}, and
2206 @code{dcgettext} in this manner except when strict ISO C support is
2207 requested by @option{-ansi} or an appropriate @option{-std} option, or
2208 @option{-ffreestanding} is used. @xref{C Dialect Options,,Options
2209 Controlling C Dialect}.
2210
2211 @item nonnull (@var{arg-index}, @dots{})
2212 @cindex @code{nonnull} function attribute
2213 The @code{nonnull} attribute specifies that some function parameters should
2214 be non-null pointers. For instance, the declaration:
2215
2216 @smallexample
2217 extern void *
2218 my_memcpy (void *dest, const void *src, size_t len)
2219 __attribute__((nonnull (1, 2)));
2220 @end smallexample
2221
2222 @noindent
2223 causes the compiler to check that, in calls to @code{my_memcpy},
2224 arguments @var{dest} and @var{src} are non-null. If the compiler
2225 determines that a null pointer is passed in an argument slot marked
2226 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2227 is issued. The compiler may also choose to make optimizations based
2228 on the knowledge that certain function arguments will not be null.
2229
2230 If no argument index list is given to the @code{nonnull} attribute,
2231 all pointer arguments are marked as non-null. To illustrate, the
2232 following declaration is equivalent to the previous example:
2233
2234 @smallexample
2235 extern void *
2236 my_memcpy (void *dest, const void *src, size_t len)
2237 __attribute__((nonnull));
2238 @end smallexample
2239
2240 @item no_instrument_function
2241 @cindex @code{no_instrument_function} function attribute
2242 @opindex finstrument-functions
2243 If @option{-finstrument-functions} is given, profiling function calls will
2244 be generated at entry and exit of most user-compiled functions.
2245 Functions with this attribute will not be so instrumented.
2246
2247 @item section ("@var{section-name}")
2248 @cindex @code{section} function attribute
2249 Normally, the compiler places the code it generates in the @code{text} section.
2250 Sometimes, however, you need additional sections, or you need certain
2251 particular functions to appear in special sections. The @code{section}
2252 attribute specifies that a function lives in a particular section.
2253 For example, the declaration:
2254
2255 @smallexample
2256 extern void foobar (void) __attribute__ ((section ("bar")));
2257 @end smallexample
2258
2259 @noindent
2260 puts the function @code{foobar} in the @code{bar} section.
2261
2262 Some file formats do not support arbitrary sections so the @code{section}
2263 attribute is not available on all platforms.
2264 If you need to map the entire contents of a module to a particular
2265 section, consider using the facilities of the linker instead.
2266
2267 @item constructor
2268 @itemx destructor
2269 @cindex @code{constructor} function attribute
2270 @cindex @code{destructor} function attribute
2271 The @code{constructor} attribute causes the function to be called
2272 automatically before execution enters @code{main ()}. Similarly, the
2273 @code{destructor} attribute causes the function to be called
2274 automatically after @code{main ()} has completed or @code{exit ()} has
2275 been called. Functions with these attributes are useful for
2276 initializing data that will be used implicitly during the execution of
2277 the program.
2278
2279 These attributes are not currently implemented for Objective-C@.
2280
2281 @cindex @code{unused} attribute.
2282 @item unused
2283 This attribute, attached to a function, means that the function is meant
2284 to be possibly unused. GCC will not produce a warning for this
2285 function.
2286
2287 @cindex @code{used} attribute.
2288 @item used
2289 This attribute, attached to a function, means that code must be emitted
2290 for the function even if it appears that the function is not referenced.
2291 This is useful, for example, when the function is referenced only in
2292 inline assembly.
2293
2294 @cindex @code{deprecated} attribute.
2295 @item deprecated
2296 The @code{deprecated} attribute results in a warning if the function
2297 is used anywhere in the source file. This is useful when identifying
2298 functions that are expected to be removed in a future version of a
2299 program. The warning also includes the location of the declaration
2300 of the deprecated function, to enable users to easily find further
2301 information about why the function is deprecated, or what they should
2302 do instead. Note that the warnings only occurs for uses:
2303
2304 @smallexample
2305 int old_fn () __attribute__ ((deprecated));
2306 int old_fn ();
2307 int (*fn_ptr)() = old_fn;
2308 @end smallexample
2309
2310 results in a warning on line 3 but not line 2.
2311
2312 The @code{deprecated} attribute can also be used for variables and
2313 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2314
2315 @item warn_unused_result
2316 @cindex @code{warn_unused_result} attribute
2317 The @code{warn_unused_result} attribute causes a warning to be emitted
2318 if a caller of the function with this attribute does not use its
2319 return value. This is useful for functions where not checking
2320 the result is either a security problem or always a bug, such as
2321 @code{realloc}.
2322
2323 @smallexample
2324 int fn () __attribute__ ((warn_unused_result));
2325 int foo ()
2326 @{
2327 if (fn () < 0) return -1;
2328 fn ();
2329 return 0;
2330 @}
2331 @end smallexample
2332
2333 results in warning on line 5.
2334
2335 @item weak
2336 @cindex @code{weak} attribute
2337 The @code{weak} attribute causes the declaration to be emitted as a weak
2338 symbol rather than a global. This is primarily useful in defining
2339 library functions which can be overridden in user code, though it can
2340 also be used with non-function declarations. Weak symbols are supported
2341 for ELF targets, and also for a.out targets when using the GNU assembler
2342 and linker.
2343
2344 @item malloc
2345 @cindex @code{malloc} attribute
2346 The @code{malloc} attribute is used to tell the compiler that a function
2347 may be treated as if it were the malloc function. The compiler assumes
2348 that calls to malloc result in pointers that cannot alias anything.
2349 This will often improve optimization.
2350
2351 @item alias ("@var{target}")
2352 @cindex @code{alias} attribute
2353 The @code{alias} attribute causes the declaration to be emitted as an
2354 alias for another symbol, which must be specified. For instance,
2355
2356 @smallexample
2357 void __f () @{ /* @r{Do something.} */; @}
2358 void f () __attribute__ ((weak, alias ("__f")));
2359 @end smallexample
2360
2361 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
2362 mangled name for the target must be used.
2363
2364 Not all target machines support this attribute.
2365
2366 @item visibility ("@var{visibility_type}")
2367 @cindex @code{visibility} attribute
2368 The @code{visibility} attribute on ELF targets causes the declaration
2369 to be emitted with default, hidden, protected or internal visibility.
2370
2371 @smallexample
2372 void __attribute__ ((visibility ("protected")))
2373 f () @{ /* @r{Do something.} */; @}
2374 int i __attribute__ ((visibility ("hidden")));
2375 @end smallexample
2376
2377 See the ELF gABI for complete details, but the short story is:
2378
2379 @table @dfn
2380 @item default
2381 Default visibility is the normal case for ELF. This value is
2382 available for the visibility attribute to override other options
2383 that may change the assumed visibility of symbols.
2384
2385 @item hidden
2386 Hidden visibility indicates that the symbol will not be placed into
2387 the dynamic symbol table, so no other @dfn{module} (executable or
2388 shared library) can reference it directly.
2389
2390 @item protected
2391 Protected visibility indicates that the symbol will be placed in the
2392 dynamic symbol table, but that references within the defining module
2393 will bind to the local symbol. That is, the symbol cannot be overridden
2394 by another module.
2395
2396 @item internal
2397 Internal visibility is like hidden visibility, but with additional
2398 processor specific semantics. Unless otherwise specified by the psABI,
2399 gcc defines internal visibility to mean that the function is @emph{never}
2400 called from another module. Note that hidden symbols, while they cannot
2401 be referenced directly by other modules, can be referenced indirectly via
2402 function pointers. By indicating that a symbol cannot be called from
2403 outside the module, gcc may for instance omit the load of a PIC register
2404 since it is known that the calling function loaded the correct value.
2405 @end table
2406
2407 Not all ELF targets support this attribute.
2408
2409 @item regparm (@var{number})
2410 @cindex @code{regparm} attribute
2411 @cindex functions that are passed arguments in registers on the 386
2412 On the Intel 386, the @code{regparm} attribute causes the compiler to
2413 pass up to @var{number} integer arguments in registers EAX,
2414 EDX, and ECX instead of on the stack. Functions that take a
2415 variable number of arguments will continue to be passed all of their
2416 arguments on the stack.
2417
2418 Beware that on some ELF systems this attribute is unsuitable for
2419 global functions in shared libraries with lazy binding (which is the
2420 default). Lazy binding will send the first call via resolving code in
2421 the loader, which might assume EAX, EDX and ECX can be clobbered, as
2422 per the standard calling conventions. Solaris 8 is affected by this.
2423 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
2424 safe since the loaders there save all registers. (Lazy binding can be
2425 disabled with the linker or the loader if desired, to avoid the
2426 problem.)
2427
2428 @item stdcall
2429 @cindex functions that pop the argument stack on the 386
2430 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2431 assume that the called function will pop off the stack space used to
2432 pass arguments, unless it takes a variable number of arguments.
2433
2434 @item fastcall
2435 @cindex functions that pop the argument stack on the 386
2436 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2437 pass the first two arguments in the registers ECX and EDX. Subsequent
2438 arguments are passed on the stack. The called function will pop the
2439 arguments off the stack. If the number of arguments is variable all
2440 arguments are pushed on the stack.
2441
2442 @item cdecl
2443 @cindex functions that do pop the argument stack on the 386
2444 @opindex mrtd
2445 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2446 assume that the calling function will pop off the stack space used to
2447 pass arguments. This is
2448 useful to override the effects of the @option{-mrtd} switch.
2449
2450 @item longcall/shortcall
2451 @cindex functions called via pointer on the RS/6000 and PowerPC
2452 On the RS/6000 and PowerPC, the @code{longcall} attribute causes the
2453 compiler to always call this function via a pointer, just as it would if
2454 the @option{-mlongcall} option had been specified. The @code{shortcall}
2455 attribute causes the compiler not to do this. These attributes override
2456 both the @option{-mlongcall} switch and the @code{#pragma longcall}
2457 setting.
2458
2459 @xref{RS/6000 and PowerPC Options}, for more information on whether long
2460 calls are necessary.
2461
2462 @item long_call/short_call
2463 @cindex indirect calls on ARM
2464 This attribute specifies how a particular function is called on
2465 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2466 command line switch and @code{#pragma long_calls} settings. The
2467 @code{long_call} attribute causes the compiler to always call the
2468 function by first loading its address into a register and then using the
2469 contents of that register. The @code{short_call} attribute always places
2470 the offset to the function from the call site into the @samp{BL}
2471 instruction directly.
2472
2473 @item function_vector
2474 @cindex calling functions through the function vector on the H8/300 processors
2475 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2476 function should be called through the function vector. Calling a
2477 function through the function vector will reduce code size, however;
2478 the function vector has a limited size (maximum 128 entries on the H8/300
2479 and 64 entries on the H8/300H) and shares space with the interrupt vector.
2480
2481 You must use GAS and GLD from GNU binutils version 2.7 or later for
2482 this attribute to work correctly.
2483
2484 @item interrupt
2485 @cindex interrupt handler functions
2486 Use this attribute on the ARM, AVR, C4x, M32R/D and Xstormy16 ports to indicate
2487 that the specified function is an interrupt handler. The compiler will
2488 generate function entry and exit sequences suitable for use in an
2489 interrupt handler when this attribute is present.
2490
2491 Note, interrupt handlers for the H8/300, H8/300H and SH processors can
2492 be specified via the @code{interrupt_handler} attribute.
2493
2494 Note, on the AVR, interrupts will be enabled inside the function.
2495
2496 Note, for the ARM, you can specify the kind of interrupt to be handled by
2497 adding an optional parameter to the interrupt attribute like this:
2498
2499 @smallexample
2500 void f () __attribute__ ((interrupt ("IRQ")));
2501 @end smallexample
2502
2503 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2504
2505 @item interrupt_handler
2506 @cindex interrupt handler functions on the H8/300 and SH processors
2507 Use this attribute on the H8/300, H8/300H and SH to indicate that the
2508 specified function is an interrupt handler. The compiler will generate
2509 function entry and exit sequences suitable for use in an interrupt
2510 handler when this attribute is present.
2511
2512 @item sp_switch
2513 Use this attribute on the SH to indicate an @code{interrupt_handler}
2514 function should switch to an alternate stack. It expects a string
2515 argument that names a global variable holding the address of the
2516 alternate stack.
2517
2518 @smallexample
2519 void *alt_stack;
2520 void f () __attribute__ ((interrupt_handler,
2521 sp_switch ("alt_stack")));
2522 @end smallexample
2523
2524 @item trap_exit
2525 Use this attribute on the SH for an @code{interrupt_handle} to return using
2526 @code{trapa} instead of @code{rte}. This attribute expects an integer
2527 argument specifying the trap number to be used.
2528
2529 @item eightbit_data
2530 @cindex eight bit data on the H8/300 and H8/300H
2531 Use this attribute on the H8/300 and H8/300H to indicate that the specified
2532 variable should be placed into the eight bit data section.
2533 The compiler will generate more efficient code for certain operations
2534 on data in the eight bit data area. Note the eight bit data area is limited to
2535 256 bytes of data.
2536
2537 You must use GAS and GLD from GNU binutils version 2.7 or later for
2538 this attribute to work correctly.
2539
2540 @item tiny_data
2541 @cindex tiny data section on the H8/300H
2542 Use this attribute on the H8/300H to indicate that the specified
2543 variable should be placed into the tiny data section.
2544 The compiler will generate more efficient code for loads and stores
2545 on data in the tiny data section. Note the tiny data area is limited to
2546 slightly under 32kbytes of data.
2547
2548 @item signal
2549 @cindex signal handler functions on the AVR processors
2550 Use this attribute on the AVR to indicate that the specified
2551 function is a signal handler. The compiler will generate function
2552 entry and exit sequences suitable for use in a signal handler when this
2553 attribute is present. Interrupts will be disabled inside the function.
2554
2555 @item naked
2556 @cindex function without a prologue/epilogue code
2557 Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate that the
2558 specified function does not need prologue/epilogue sequences generated by
2559 the compiler. It is up to the programmer to provide these sequences.
2560
2561 @item model (@var{model-name})
2562 @cindex function addressability on the M32R/D
2563 @cindex variable addressability on the IA-64
2564
2565 On the M32R/D, use this attribute to set the addressability of an
2566 object, and of the code generated for a function. The identifier
2567 @var{model-name} is one of @code{small}, @code{medium}, or
2568 @code{large}, representing each of the code models.
2569
2570 Small model objects live in the lower 16MB of memory (so that their
2571 addresses can be loaded with the @code{ld24} instruction), and are
2572 callable with the @code{bl} instruction.
2573
2574 Medium model objects may live anywhere in the 32-bit address space (the
2575 compiler will generate @code{seth/add3} instructions to load their addresses),
2576 and are callable with the @code{bl} instruction.
2577
2578 Large model objects may live anywhere in the 32-bit address space (the
2579 compiler will generate @code{seth/add3} instructions to load their addresses),
2580 and may not be reachable with the @code{bl} instruction (the compiler will
2581 generate the much slower @code{seth/add3/jl} instruction sequence).
2582
2583 On IA-64, use this attribute to set the addressability of an object.
2584 At present, the only supported identifier for @var{model-name} is
2585 @code{small}, indicating addressability via ``small'' (22-bit)
2586 addresses (so that their addresses can be loaded with the @code{addl}
2587 instruction). Caveat: such addressing is by definition not position
2588 independent and hence this attribute must not be used for objects
2589 defined by shared libraries.
2590
2591 @item far
2592 @cindex functions which handle memory bank switching
2593 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
2594 use a calling convention that takes care of switching memory banks when
2595 entering and leaving a function. This calling convention is also the
2596 default when using the @option{-mlong-calls} option.
2597
2598 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
2599 to call and return from a function.
2600
2601 On 68HC11 the compiler will generate a sequence of instructions
2602 to invoke a board-specific routine to switch the memory bank and call the
2603 real function. The board-specific routine simulates a @code{call}.
2604 At the end of a function, it will jump to a board-specific routine
2605 instead of using @code{rts}. The board-specific return routine simulates
2606 the @code{rtc}.
2607
2608 @item near
2609 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2610 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2611 use the normal calling convention based on @code{jsr} and @code{rts}.
2612 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2613 option.
2614
2615 @item dllimport
2616 @cindex @code{__declspec(dllimport)}
2617 On Windows targets, the @code{dllimport} attribute causes the compiler
2618 to reference a function or variable via a global pointer to a pointer
2619 that is set up by the Windows dll library. The pointer name is formed by
2620 combining @code{_imp__} and the function or variable name. The attribute
2621 implies @code{extern} storage.
2622
2623 Currently, the attribute is ignored for inlined functions. If the
2624 attribute is applied to a symbol @emph{definition}, an error is reported.
2625 If a symbol previously declared @code{dllimport} is later defined, the
2626 attribute is ignored in subsequent references, and a warning is emitted.
2627 The attribute is also overriden by a subsequent declaration as
2628 @code{dllexport}.
2629
2630 When applied to C++ classes, the attribute marks non-inlined
2631 member functions and static data members as imports. However, the
2632 attribute is ignored for virtual methods to allow creation of vtables
2633 using thunks.
2634
2635 On cygwin, mingw and arm-pe targets, @code{__declspec(dllimport)} is
2636 recognized as a synonym for @code{__attribute__ ((dllimport))} for
2637 compatibility with other Windows compilers.
2638
2639 The use of the @code{dllimport} attribute on functions is not necessary,
2640 but provides a small performance benefit by eliminating a thunk in the
2641 dll. The use of the @code{dllimport} attribute on imported variables was
2642 required on older versions of GNU ld, but can now be avoided by passing
2643 the @option{--enable-auto-import} switch to ld. As with functions, using
2644 the attribute for a variable eliminates a thunk in the dll.
2645
2646 One drawback to using this attribute is that a pointer to a function or
2647 variable marked as dllimport cannot be used as a constant address. The
2648 attribute can be disabled for functions by setting the
2649 @option{-mnop-fun-dllimport} flag.
2650
2651 @item dllexport
2652 @cindex @code{__declspec(dllexport)}
2653 On Windows targets the @code{dllexport} attribute causes the compiler to
2654 provide a global pointer to a pointer in a dll, so that it can be
2655 referenced with the @code{dllimport} attribute. The pointer name is
2656 formed by combining @code{_imp__} and the function or variable name.
2657
2658 Currently, the @code{dllexport}attribute is ignored for inlined
2659 functions, but export can be forced by using the
2660 @option{-fkeep-inline-functions} flag. The attribute is also ignored for
2661 undefined symbols.
2662
2663 When applied to C++ classes. the attribute marks defined non-inlined
2664 member functions and static data members as exports. Static consts
2665 initialized in-class are not marked unless they are also defined
2666 out-of-class.
2667
2668 On cygwin, mingw and arm-pe targets, @code{__declspec(dllexport)} is
2669 recognized as a synonym for @code{__attribute__ ((dllexport))} for
2670 compatibility with other Windows compilers.
2671
2672 Alternative methods for including the symbol in the dll's export table
2673 are to use a .def file with an @code{EXPORTS} section or, with GNU ld,
2674 using the @option{--export-all} linker flag.
2675
2676 @end table
2677
2678 You can specify multiple attributes in a declaration by separating them
2679 by commas within the double parentheses or by immediately following an
2680 attribute declaration with another attribute declaration.
2681
2682 @cindex @code{#pragma}, reason for not using
2683 @cindex pragma, reason for not using
2684 Some people object to the @code{__attribute__} feature, suggesting that
2685 ISO C's @code{#pragma} should be used instead. At the time
2686 @code{__attribute__} was designed, there were two reasons for not doing
2687 this.
2688
2689 @enumerate
2690 @item
2691 It is impossible to generate @code{#pragma} commands from a macro.
2692
2693 @item
2694 There is no telling what the same @code{#pragma} might mean in another
2695 compiler.
2696 @end enumerate
2697
2698 These two reasons applied to almost any application that might have been
2699 proposed for @code{#pragma}. It was basically a mistake to use
2700 @code{#pragma} for @emph{anything}.
2701
2702 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2703 to be generated from macros. In addition, a @code{#pragma GCC}
2704 namespace is now in use for GCC-specific pragmas. However, it has been
2705 found convenient to use @code{__attribute__} to achieve a natural
2706 attachment of attributes to their corresponding declarations, whereas
2707 @code{#pragma GCC} is of use for constructs that do not naturally form
2708 part of the grammar. @xref{Other Directives,,Miscellaneous
2709 Preprocessing Directives, cpp, The C Preprocessor}.
2710
2711 @node Attribute Syntax
2712 @section Attribute Syntax
2713 @cindex attribute syntax
2714
2715 This section describes the syntax with which @code{__attribute__} may be
2716 used, and the constructs to which attribute specifiers bind, for the C
2717 language. Some details may vary for C++ and Objective-C@. Because of
2718 infelicities in the grammar for attributes, some forms described here
2719 may not be successfully parsed in all cases.
2720
2721 There are some problems with the semantics of attributes in C++. For
2722 example, there are no manglings for attributes, although they may affect
2723 code generation, so problems may arise when attributed types are used in
2724 conjunction with templates or overloading. Similarly, @code{typeid}
2725 does not distinguish between types with different attributes. Support
2726 for attributes in C++ may be restricted in future to attributes on
2727 declarations only, but not on nested declarators.
2728
2729 @xref{Function Attributes}, for details of the semantics of attributes
2730 applying to functions. @xref{Variable Attributes}, for details of the
2731 semantics of attributes applying to variables. @xref{Type Attributes},
2732 for details of the semantics of attributes applying to structure, union
2733 and enumerated types.
2734
2735 An @dfn{attribute specifier} is of the form
2736 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2737 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2738 each attribute is one of the following:
2739
2740 @itemize @bullet
2741 @item
2742 Empty. Empty attributes are ignored.
2743
2744 @item
2745 A word (which may be an identifier such as @code{unused}, or a reserved
2746 word such as @code{const}).
2747
2748 @item
2749 A word, followed by, in parentheses, parameters for the attribute.
2750 These parameters take one of the following forms:
2751
2752 @itemize @bullet
2753 @item
2754 An identifier. For example, @code{mode} attributes use this form.
2755
2756 @item
2757 An identifier followed by a comma and a non-empty comma-separated list
2758 of expressions. For example, @code{format} attributes use this form.
2759
2760 @item
2761 A possibly empty comma-separated list of expressions. For example,
2762 @code{format_arg} attributes use this form with the list being a single
2763 integer constant expression, and @code{alias} attributes use this form
2764 with the list being a single string constant.
2765 @end itemize
2766 @end itemize
2767
2768 An @dfn{attribute specifier list} is a sequence of one or more attribute
2769 specifiers, not separated by any other tokens.
2770
2771 In GNU C, an attribute specifier list may appear after the colon following a
2772 label, other than a @code{case} or @code{default} label. The only
2773 attribute it makes sense to use after a label is @code{unused}. This
2774 feature is intended for code generated by programs which contains labels
2775 that may be unused but which is compiled with @option{-Wall}. It would
2776 not normally be appropriate to use in it human-written code, though it
2777 could be useful in cases where the code that jumps to the label is
2778 contained within an @code{#ifdef} conditional. GNU C++ does not permit
2779 such placement of attribute lists, as it is permissible for a
2780 declaration, which could begin with an attribute list, to be labelled in
2781 C++. Declarations cannot be labelled in C90 or C99, so the ambiguity
2782 does not arise there.
2783
2784 An attribute specifier list may appear as part of a @code{struct},
2785 @code{union} or @code{enum} specifier. It may go either immediately
2786 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2787 the closing brace. It is ignored if the content of the structure, union
2788 or enumerated type is not defined in the specifier in which the
2789 attribute specifier list is used---that is, in usages such as
2790 @code{struct __attribute__((foo)) bar} with no following opening brace.
2791 Where attribute specifiers follow the closing brace, they are considered
2792 to relate to the structure, union or enumerated type defined, not to any
2793 enclosing declaration the type specifier appears in, and the type
2794 defined is not complete until after the attribute specifiers.
2795 @c Otherwise, there would be the following problems: a shift/reduce
2796 @c conflict between attributes binding the struct/union/enum and
2797 @c binding to the list of specifiers/qualifiers; and "aligned"
2798 @c attributes could use sizeof for the structure, but the size could be
2799 @c changed later by "packed" attributes.
2800
2801 Otherwise, an attribute specifier appears as part of a declaration,
2802 counting declarations of unnamed parameters and type names, and relates
2803 to that declaration (which may be nested in another declaration, for
2804 example in the case of a parameter declaration), or to a particular declarator
2805 within a declaration. Where an
2806 attribute specifier is applied to a parameter declared as a function or
2807 an array, it should apply to the function or array rather than the
2808 pointer to which the parameter is implicitly converted, but this is not
2809 yet correctly implemented.
2810
2811 Any list of specifiers and qualifiers at the start of a declaration may
2812 contain attribute specifiers, whether or not such a list may in that
2813 context contain storage class specifiers. (Some attributes, however,
2814 are essentially in the nature of storage class specifiers, and only make
2815 sense where storage class specifiers may be used; for example,
2816 @code{section}.) There is one necessary limitation to this syntax: the
2817 first old-style parameter declaration in a function definition cannot
2818 begin with an attribute specifier, because such an attribute applies to
2819 the function instead by syntax described below (which, however, is not
2820 yet implemented in this case). In some other cases, attribute
2821 specifiers are permitted by this grammar but not yet supported by the
2822 compiler. All attribute specifiers in this place relate to the
2823 declaration as a whole. In the obsolescent usage where a type of
2824 @code{int} is implied by the absence of type specifiers, such a list of
2825 specifiers and qualifiers may be an attribute specifier list with no
2826 other specifiers or qualifiers.
2827
2828 An attribute specifier list may appear immediately before a declarator
2829 (other than the first) in a comma-separated list of declarators in a
2830 declaration of more than one identifier using a single list of
2831 specifiers and qualifiers. Such attribute specifiers apply
2832 only to the identifier before whose declarator they appear. For
2833 example, in
2834
2835 @smallexample
2836 __attribute__((noreturn)) void d0 (void),
2837 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2838 d2 (void)
2839 @end smallexample
2840
2841 @noindent
2842 the @code{noreturn} attribute applies to all the functions
2843 declared; the @code{format} attribute only applies to @code{d1}.
2844
2845 An attribute specifier list may appear immediately before the comma,
2846 @code{=} or semicolon terminating the declaration of an identifier other
2847 than a function definition. At present, such attribute specifiers apply
2848 to the declared object or function, but in future they may attach to the
2849 outermost adjacent declarator. In simple cases there is no difference,
2850 but, for example, in
2851
2852 @smallexample
2853 void (****f)(void) __attribute__((noreturn));
2854 @end smallexample
2855
2856 @noindent
2857 at present the @code{noreturn} attribute applies to @code{f}, which
2858 causes a warning since @code{f} is not a function, but in future it may
2859 apply to the function @code{****f}. The precise semantics of what
2860 attributes in such cases will apply to are not yet specified. Where an
2861 assembler name for an object or function is specified (@pxref{Asm
2862 Labels}), at present the attribute must follow the @code{asm}
2863 specification; in future, attributes before the @code{asm} specification
2864 may apply to the adjacent declarator, and those after it to the declared
2865 object or function.
2866
2867 An attribute specifier list may, in future, be permitted to appear after
2868 the declarator in a function definition (before any old-style parameter
2869 declarations or the function body).
2870
2871 Attribute specifiers may be mixed with type qualifiers appearing inside
2872 the @code{[]} of a parameter array declarator, in the C99 construct by
2873 which such qualifiers are applied to the pointer to which the array is
2874 implicitly converted. Such attribute specifiers apply to the pointer,
2875 not to the array, but at present this is not implemented and they are
2876 ignored.
2877
2878 An attribute specifier list may appear at the start of a nested
2879 declarator. At present, there are some limitations in this usage: the
2880 attributes correctly apply to the declarator, but for most individual
2881 attributes the semantics this implies are not implemented.
2882 When attribute specifiers follow the @code{*} of a pointer
2883 declarator, they may be mixed with any type qualifiers present.
2884 The following describes the formal semantics of this syntax. It will make the
2885 most sense if you are familiar with the formal specification of
2886 declarators in the ISO C standard.
2887
2888 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2889 D1}, where @code{T} contains declaration specifiers that specify a type
2890 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2891 contains an identifier @var{ident}. The type specified for @var{ident}
2892 for derived declarators whose type does not include an attribute
2893 specifier is as in the ISO C standard.
2894
2895 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2896 and the declaration @code{T D} specifies the type
2897 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2898 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2899 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2900
2901 If @code{D1} has the form @code{*
2902 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2903 declaration @code{T D} specifies the type
2904 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2905 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2906 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2907 @var{ident}.
2908
2909 For example,
2910
2911 @smallexample
2912 void (__attribute__((noreturn)) ****f) (void);
2913 @end smallexample
2914
2915 @noindent
2916 specifies the type ``pointer to pointer to pointer to pointer to
2917 non-returning function returning @code{void}''. As another example,
2918
2919 @smallexample
2920 char *__attribute__((aligned(8))) *f;
2921 @end smallexample
2922
2923 @noindent
2924 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2925 Note again that this does not work with most attributes; for example,
2926 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2927 is not yet supported.
2928
2929 For compatibility with existing code written for compiler versions that
2930 did not implement attributes on nested declarators, some laxity is
2931 allowed in the placing of attributes. If an attribute that only applies
2932 to types is applied to a declaration, it will be treated as applying to
2933 the type of that declaration. If an attribute that only applies to
2934 declarations is applied to the type of a declaration, it will be treated
2935 as applying to that declaration; and, for compatibility with code
2936 placing the attributes immediately before the identifier declared, such
2937 an attribute applied to a function return type will be treated as
2938 applying to the function type, and such an attribute applied to an array
2939 element type will be treated as applying to the array type. If an
2940 attribute that only applies to function types is applied to a
2941 pointer-to-function type, it will be treated as applying to the pointer
2942 target type; if such an attribute is applied to a function return type
2943 that is not a pointer-to-function type, it will be treated as applying
2944 to the function type.
2945
2946 @node Function Prototypes
2947 @section Prototypes and Old-Style Function Definitions
2948 @cindex function prototype declarations
2949 @cindex old-style function definitions
2950 @cindex promotion of formal parameters
2951
2952 GNU C extends ISO C to allow a function prototype to override a later
2953 old-style non-prototype definition. Consider the following example:
2954
2955 @example
2956 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2957 #ifdef __STDC__
2958 #define P(x) x
2959 #else
2960 #define P(x) ()
2961 #endif
2962
2963 /* @r{Prototype function declaration.} */
2964 int isroot P((uid_t));
2965
2966 /* @r{Old-style function definition.} */
2967 int
2968 isroot (x) /* ??? lossage here ??? */
2969 uid_t x;
2970 @{
2971 return x == 0;
2972 @}
2973 @end example
2974
2975 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2976 not allow this example, because subword arguments in old-style
2977 non-prototype definitions are promoted. Therefore in this example the
2978 function definition's argument is really an @code{int}, which does not
2979 match the prototype argument type of @code{short}.
2980
2981 This restriction of ISO C makes it hard to write code that is portable
2982 to traditional C compilers, because the programmer does not know
2983 whether the @code{uid_t} type is @code{short}, @code{int}, or
2984 @code{long}. Therefore, in cases like these GNU C allows a prototype
2985 to override a later old-style definition. More precisely, in GNU C, a
2986 function prototype argument type overrides the argument type specified
2987 by a later old-style definition if the former type is the same as the
2988 latter type before promotion. Thus in GNU C the above example is
2989 equivalent to the following:
2990
2991 @example
2992 int isroot (uid_t);
2993
2994 int
2995 isroot (uid_t x)
2996 @{
2997 return x == 0;
2998 @}
2999 @end example
3000
3001 @noindent
3002 GNU C++ does not support old-style function definitions, so this
3003 extension is irrelevant.
3004
3005 @node C++ Comments
3006 @section C++ Style Comments
3007 @cindex //
3008 @cindex C++ comments
3009 @cindex comments, C++ style
3010
3011 In GNU C, you may use C++ style comments, which start with @samp{//} and
3012 continue until the end of the line. Many other C implementations allow
3013 such comments, and they are included in the 1999 C standard. However,
3014 C++ style comments are not recognized if you specify an @option{-std}
3015 option specifying a version of ISO C before C99, or @option{-ansi}
3016 (equivalent to @option{-std=c89}).
3017
3018 @node Dollar Signs
3019 @section Dollar Signs in Identifier Names
3020 @cindex $
3021 @cindex dollar signs in identifier names
3022 @cindex identifier names, dollar signs in
3023
3024 In GNU C, you may normally use dollar signs in identifier names.
3025 This is because many traditional C implementations allow such identifiers.
3026 However, dollar signs in identifiers are not supported on a few target
3027 machines, typically because the target assembler does not allow them.
3028
3029 @node Character Escapes
3030 @section The Character @key{ESC} in Constants
3031
3032 You can use the sequence @samp{\e} in a string or character constant to
3033 stand for the ASCII character @key{ESC}.
3034
3035 @node Alignment
3036 @section Inquiring on Alignment of Types or Variables
3037 @cindex alignment
3038 @cindex type alignment
3039 @cindex variable alignment
3040
3041 The keyword @code{__alignof__} allows you to inquire about how an object
3042 is aligned, or the minimum alignment usually required by a type. Its
3043 syntax is just like @code{sizeof}.
3044
3045 For example, if the target machine requires a @code{double} value to be
3046 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
3047 This is true on many RISC machines. On more traditional machine
3048 designs, @code{__alignof__ (double)} is 4 or even 2.
3049
3050 Some machines never actually require alignment; they allow reference to any
3051 data type even at an odd address. For these machines, @code{__alignof__}
3052 reports the @emph{recommended} alignment of a type.
3053
3054 If the operand of @code{__alignof__} is an lvalue rather than a type,
3055 its value is the required alignment for its type, taking into account
3056 any minimum alignment specified with GCC's @code{__attribute__}
3057 extension (@pxref{Variable Attributes}). For example, after this
3058 declaration:
3059
3060 @example
3061 struct foo @{ int x; char y; @} foo1;
3062 @end example
3063
3064 @noindent
3065 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
3066 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
3067
3068 It is an error to ask for the alignment of an incomplete type.
3069
3070 @node Variable Attributes
3071 @section Specifying Attributes of Variables
3072 @cindex attribute of variables
3073 @cindex variable attributes
3074
3075 The keyword @code{__attribute__} allows you to specify special
3076 attributes of variables or structure fields. This keyword is followed
3077 by an attribute specification inside double parentheses. Some
3078 attributes are currently defined generically for variables.
3079 Other attributes are defined for variables on particular target
3080 systems. Other attributes are available for functions
3081 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
3082 Other front ends might define more attributes
3083 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
3084
3085 You may also specify attributes with @samp{__} preceding and following
3086 each keyword. This allows you to use them in header files without
3087 being concerned about a possible macro of the same name. For example,
3088 you may use @code{__aligned__} instead of @code{aligned}.
3089
3090 @xref{Attribute Syntax}, for details of the exact syntax for using
3091 attributes.
3092
3093 @table @code
3094 @cindex @code{aligned} attribute
3095 @item aligned (@var{alignment})
3096 This attribute specifies a minimum alignment for the variable or
3097 structure field, measured in bytes. For example, the declaration:
3098
3099 @smallexample
3100 int x __attribute__ ((aligned (16))) = 0;
3101 @end smallexample
3102
3103 @noindent
3104 causes the compiler to allocate the global variable @code{x} on a
3105 16-byte boundary. On a 68040, this could be used in conjunction with
3106 an @code{asm} expression to access the @code{move16} instruction which
3107 requires 16-byte aligned operands.
3108
3109 You can also specify the alignment of structure fields. For example, to
3110 create a double-word aligned @code{int} pair, you could write:
3111
3112 @smallexample
3113 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
3114 @end smallexample
3115
3116 @noindent
3117 This is an alternative to creating a union with a @code{double} member
3118 that forces the union to be double-word aligned.
3119
3120 As in the preceding examples, you can explicitly specify the alignment
3121 (in bytes) that you wish the compiler to use for a given variable or
3122 structure field. Alternatively, you can leave out the alignment factor
3123 and just ask the compiler to align a variable or field to the maximum
3124 useful alignment for the target machine you are compiling for. For
3125 example, you could write:
3126
3127 @smallexample
3128 short array[3] __attribute__ ((aligned));
3129 @end smallexample
3130
3131 Whenever you leave out the alignment factor in an @code{aligned} attribute
3132 specification, the compiler automatically sets the alignment for the declared
3133 variable or field to the largest alignment which is ever used for any data
3134 type on the target machine you are compiling for. Doing this can often make
3135 copy operations more efficient, because the compiler can use whatever
3136 instructions copy the biggest chunks of memory when performing copies to
3137 or from the variables or fields that you have aligned this way.
3138
3139 The @code{aligned} attribute can only increase the alignment; but you
3140 can decrease it by specifying @code{packed} as well. See below.
3141
3142 Note that the effectiveness of @code{aligned} attributes may be limited
3143 by inherent limitations in your linker. On many systems, the linker is
3144 only able to arrange for variables to be aligned up to a certain maximum
3145 alignment. (For some linkers, the maximum supported alignment may
3146 be very very small.) If your linker is only able to align variables
3147 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3148 in an @code{__attribute__} will still only provide you with 8 byte
3149 alignment. See your linker documentation for further information.
3150
3151 @item cleanup (@var{cleanup_function})
3152 @cindex @code{cleanup} attribute
3153 The @code{cleanup} attribute runs a function when the variable goes
3154 out of scope. This attribute can only be applied to auto function
3155 scope variables; it may not be applied to parameters or variables
3156 with static storage duration. The function must take one parameter,
3157 a pointer to a type compatible with the variable. The return value
3158 of the function (if any) is ignored.
3159
3160 If @option{-fexceptions} is enabled, then @var{cleanup_function}
3161 will be run during the stack unwinding that happens during the
3162 processing of the exception. Note that the @code{cleanup} attribute
3163 does not allow the exception to be caught, only to perform an action.
3164 It is undefined what happens if @var{cleanup_function} does not
3165 return normally.
3166
3167 @item common
3168 @itemx nocommon
3169 @cindex @code{common} attribute
3170 @cindex @code{nocommon} attribute
3171 @opindex fcommon
3172 @opindex fno-common
3173 The @code{common} attribute requests GCC to place a variable in
3174 ``common'' storage. The @code{nocommon} attribute requests the
3175 opposite -- to allocate space for it directly.
3176
3177 These attributes override the default chosen by the
3178 @option{-fno-common} and @option{-fcommon} flags respectively.
3179
3180 @item deprecated
3181 @cindex @code{deprecated} attribute
3182 The @code{deprecated} attribute results in a warning if the variable
3183 is used anywhere in the source file. This is useful when identifying
3184 variables that are expected to be removed in a future version of a
3185 program. The warning also includes the location of the declaration
3186 of the deprecated variable, to enable users to easily find further
3187 information about why the variable is deprecated, or what they should
3188 do instead. Note that the warning only occurs for uses:
3189
3190 @smallexample
3191 extern int old_var __attribute__ ((deprecated));
3192 extern int old_var;
3193 int new_fn () @{ return old_var; @}
3194 @end smallexample
3195
3196 results in a warning on line 3 but not line 2.
3197
3198 The @code{deprecated} attribute can also be used for functions and
3199 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
3200
3201 @item mode (@var{mode})
3202 @cindex @code{mode} attribute
3203 This attribute specifies the data type for the declaration---whichever
3204 type corresponds to the mode @var{mode}. This in effect lets you
3205 request an integer or floating point type according to its width.
3206
3207 You may also specify a mode of @samp{byte} or @samp{__byte__} to
3208 indicate the mode corresponding to a one-byte integer, @samp{word} or
3209 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
3210 or @samp{__pointer__} for the mode used to represent pointers.
3211
3212 @item packed
3213 @cindex @code{packed} attribute
3214 The @code{packed} attribute specifies that a variable or structure field
3215 should have the smallest possible alignment---one byte for a variable,
3216 and one bit for a field, unless you specify a larger value with the
3217 @code{aligned} attribute.
3218
3219 Here is a structure in which the field @code{x} is packed, so that it
3220 immediately follows @code{a}:
3221
3222 @example
3223 struct foo
3224 @{
3225 char a;
3226 int x[2] __attribute__ ((packed));
3227 @};
3228 @end example
3229
3230 @item section ("@var{section-name}")
3231 @cindex @code{section} variable attribute
3232 Normally, the compiler places the objects it generates in sections like
3233 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
3234 or you need certain particular variables to appear in special sections,
3235 for example to map to special hardware. The @code{section}
3236 attribute specifies that a variable (or function) lives in a particular
3237 section. For example, this small program uses several specific section names:
3238
3239 @smallexample
3240 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
3241 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
3242 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
3243 int init_data __attribute__ ((section ("INITDATA"))) = 0;
3244
3245 main()
3246 @{
3247 /* Initialize stack pointer */
3248 init_sp (stack + sizeof (stack));
3249
3250 /* Initialize initialized data */
3251 memcpy (&init_data, &data, &edata - &data);
3252
3253 /* Turn on the serial ports */
3254 init_duart (&a);
3255 init_duart (&b);
3256 @}
3257 @end smallexample
3258
3259 @noindent
3260 Use the @code{section} attribute with an @emph{initialized} definition
3261 of a @emph{global} variable, as shown in the example. GCC issues
3262 a warning and otherwise ignores the @code{section} attribute in
3263 uninitialized variable declarations.
3264
3265 You may only use the @code{section} attribute with a fully initialized
3266 global definition because of the way linkers work. The linker requires
3267 each object be defined once, with the exception that uninitialized
3268 variables tentatively go in the @code{common} (or @code{bss}) section
3269 and can be multiply ``defined''. You can force a variable to be
3270 initialized with the @option{-fno-common} flag or the @code{nocommon}
3271 attribute.
3272
3273 Some file formats do not support arbitrary sections so the @code{section}
3274 attribute is not available on all platforms.
3275 If you need to map the entire contents of a module to a particular
3276 section, consider using the facilities of the linker instead.
3277
3278 @item shared
3279 @cindex @code{shared} variable attribute
3280 On Windows, in addition to putting variable definitions in a named
3281 section, the section can also be shared among all running copies of an
3282 executable or DLL@. For example, this small program defines shared data
3283 by putting it in a named section @code{shared} and marking the section
3284 shareable:
3285
3286 @smallexample
3287 int foo __attribute__((section ("shared"), shared)) = 0;
3288
3289 int
3290 main()
3291 @{
3292 /* Read and write foo. All running
3293 copies see the same value. */
3294 return 0;
3295 @}
3296 @end smallexample
3297
3298 @noindent
3299 You may only use the @code{shared} attribute along with @code{section}
3300 attribute with a fully initialized global definition because of the way
3301 linkers work. See @code{section} attribute for more information.
3302
3303 The @code{shared} attribute is only available on Windows@.
3304
3305 @item tls_model ("@var{tls_model}")
3306 @cindex @code{tls_model} attribute
3307 The @code{tls_model} attribute sets thread-local storage model
3308 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
3309 overriding @code{-ftls-model=} command line switch on a per-variable
3310 basis.
3311 The @var{tls_model} argument should be one of @code{global-dynamic},
3312 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
3313
3314 Not all targets support this attribute.
3315
3316 @item transparent_union
3317 This attribute, attached to a function parameter which is a union, means
3318 that the corresponding argument may have the type of any union member,
3319 but the argument is passed as if its type were that of the first union
3320 member. For more details see @xref{Type Attributes}. You can also use
3321 this attribute on a @code{typedef} for a union data type; then it
3322 applies to all function parameters with that type.
3323
3324 @item unused
3325 This attribute, attached to a variable, means that the variable is meant
3326 to be possibly unused. GCC will not produce a warning for this
3327 variable.
3328
3329 @item vector_size (@var{bytes})
3330 This attribute specifies the vector size for the variable, measured in
3331 bytes. For example, the declaration:
3332
3333 @smallexample
3334 int foo __attribute__ ((vector_size (16)));
3335 @end smallexample
3336
3337 @noindent
3338 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3339 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3340 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3341
3342 This attribute is only applicable to integral and float scalars,
3343 although arrays, pointers, and function return values are allowed in
3344 conjunction with this construct.
3345
3346 Aggregates with this attribute are invalid, even if they are of the same
3347 size as a corresponding scalar. For example, the declaration:
3348
3349 @smallexample
3350 struct S @{ int a; @};
3351 struct S __attribute__ ((vector_size (16))) foo;
3352 @end smallexample
3353
3354 @noindent
3355 is invalid even if the size of the structure is the same as the size of
3356 the @code{int}.
3357
3358 @item weak
3359 The @code{weak} attribute is described in @xref{Function Attributes}.
3360
3361 @item dllimport
3362 The @code{dllimport} attribute is described in @xref{Function Attributes}.
3363
3364 @item dlexport
3365 The @code{dllexport} attribute is described in @xref{Function Attributes}.
3366
3367 @end table
3368
3369 @subsection M32R/D Variable Attributes
3370
3371 One attribute is currently defined for the M32R/D.
3372
3373 @table @code
3374 @item model (@var{model-name})
3375 @cindex variable addressability on the M32R/D
3376 Use this attribute on the M32R/D to set the addressability of an object.
3377 The identifier @var{model-name} is one of @code{small}, @code{medium},
3378 or @code{large}, representing each of the code models.
3379
3380 Small model objects live in the lower 16MB of memory (so that their
3381 addresses can be loaded with the @code{ld24} instruction).
3382
3383 Medium and large model objects may live anywhere in the 32-bit address space
3384 (the compiler will generate @code{seth/add3} instructions to load their
3385 addresses).
3386 @end table
3387
3388 @subsection i386 Variable Attributes
3389
3390 Two attributes are currently defined for i386 configurations:
3391 @code{ms_struct} and @code{gcc_struct}
3392
3393 @table @code
3394 @item ms_struct
3395 @itemx gcc_struct
3396 @cindex @code{ms_struct} attribute
3397 @cindex @code{gcc_struct} attribute
3398
3399 If @code{packed} is used on a structure, or if bit-fields are used
3400 it may be that the Microsoft ABI packs them differently
3401 than GCC would normally pack them. Particularly when moving packed
3402 data between functions compiled with GCC and the native Microsoft compiler
3403 (either via function call or as data in a file), it may be necessary to access
3404 either format.
3405
3406 Currently @option{-m[no-]ms-bitfields} is provided for the Windows X86
3407 compilers to match the native Microsoft compiler.
3408 @end table
3409
3410 @node Type Attributes
3411 @section Specifying Attributes of Types
3412 @cindex attribute of types
3413 @cindex type attributes
3414
3415 The keyword @code{__attribute__} allows you to specify special
3416 attributes of @code{struct} and @code{union} types when you define such
3417 types. This keyword is followed by an attribute specification inside
3418 double parentheses. Six attributes are currently defined for types:
3419 @code{aligned}, @code{packed}, @code{transparent_union}, @code{unused},
3420 @code{deprecated} and @code{may_alias}. Other attributes are defined for
3421 functions (@pxref{Function Attributes}) and for variables
3422 (@pxref{Variable Attributes}).
3423
3424 You may also specify any one of these attributes with @samp{__}
3425 preceding and following its keyword. This allows you to use these
3426 attributes in header files without being concerned about a possible
3427 macro of the same name. For example, you may use @code{__aligned__}
3428 instead of @code{aligned}.
3429
3430 You may specify the @code{aligned} and @code{transparent_union}
3431 attributes either in a @code{typedef} declaration or just past the
3432 closing curly brace of a complete enum, struct or union type
3433 @emph{definition} and the @code{packed} attribute only past the closing
3434 brace of a definition.
3435
3436 You may also specify attributes between the enum, struct or union
3437 tag and the name of the type rather than after the closing brace.
3438
3439 @xref{Attribute Syntax}, for details of the exact syntax for using
3440 attributes.
3441
3442 @table @code
3443 @cindex @code{aligned} attribute
3444 @item aligned (@var{alignment})
3445 This attribute specifies a minimum alignment (in bytes) for variables
3446 of the specified type. For example, the declarations:
3447
3448 @smallexample
3449 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3450 typedef int more_aligned_int __attribute__ ((aligned (8)));
3451 @end smallexample
3452
3453 @noindent
3454 force the compiler to insure (as far as it can) that each variable whose
3455 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3456 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
3457 variables of type @code{struct S} aligned to 8-byte boundaries allows
3458 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3459 store) instructions when copying one variable of type @code{struct S} to
3460 another, thus improving run-time efficiency.
3461
3462 Note that the alignment of any given @code{struct} or @code{union} type
3463 is required by the ISO C standard to be at least a perfect multiple of
3464 the lowest common multiple of the alignments of all of the members of
3465 the @code{struct} or @code{union} in question. This means that you @emph{can}
3466 effectively adjust the alignment of a @code{struct} or @code{union}
3467 type by attaching an @code{aligned} attribute to any one of the members
3468 of such a type, but the notation illustrated in the example above is a
3469 more obvious, intuitive, and readable way to request the compiler to
3470 adjust the alignment of an entire @code{struct} or @code{union} type.
3471
3472 As in the preceding example, you can explicitly specify the alignment
3473 (in bytes) that you wish the compiler to use for a given @code{struct}
3474 or @code{union} type. Alternatively, you can leave out the alignment factor
3475 and just ask the compiler to align a type to the maximum
3476 useful alignment for the target machine you are compiling for. For
3477 example, you could write:
3478
3479 @smallexample
3480 struct S @{ short f[3]; @} __attribute__ ((aligned));
3481 @end smallexample
3482
3483 Whenever you leave out the alignment factor in an @code{aligned}
3484 attribute specification, the compiler automatically sets the alignment
3485 for the type to the largest alignment which is ever used for any data
3486 type on the target machine you are compiling for. Doing this can often
3487 make copy operations more efficient, because the compiler can use
3488 whatever instructions copy the biggest chunks of memory when performing
3489 copies to or from the variables which have types that you have aligned
3490 this way.
3491
3492 In the example above, if the size of each @code{short} is 2 bytes, then
3493 the size of the entire @code{struct S} type is 6 bytes. The smallest
3494 power of two which is greater than or equal to that is 8, so the
3495 compiler sets the alignment for the entire @code{struct S} type to 8
3496 bytes.
3497
3498 Note that although you can ask the compiler to select a time-efficient
3499 alignment for a given type and then declare only individual stand-alone
3500 objects of that type, the compiler's ability to select a time-efficient
3501 alignment is primarily useful only when you plan to create arrays of
3502 variables having the relevant (efficiently aligned) type. If you
3503 declare or use arrays of variables of an efficiently-aligned type, then
3504 it is likely that your program will also be doing pointer arithmetic (or
3505 subscripting, which amounts to the same thing) on pointers to the
3506 relevant type, and the code that the compiler generates for these
3507 pointer arithmetic operations will often be more efficient for
3508 efficiently-aligned types than for other types.
3509
3510 The @code{aligned} attribute can only increase the alignment; but you
3511 can decrease it by specifying @code{packed} as well. See below.
3512
3513 Note that the effectiveness of @code{aligned} attributes may be limited
3514 by inherent limitations in your linker. On many systems, the linker is
3515 only able to arrange for variables to be aligned up to a certain maximum
3516 alignment. (For some linkers, the maximum supported alignment may
3517 be very very small.) If your linker is only able to align variables
3518 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3519 in an @code{__attribute__} will still only provide you with 8 byte
3520 alignment. See your linker documentation for further information.
3521
3522 @item packed
3523 This attribute, attached to @code{struct} or @code{union} type
3524 definition, specifies that each member of the structure or union is
3525 placed to minimize the memory required. When attached to an @code{enum}
3526 definition, it indicates that the smallest integral type should be used.
3527
3528 @opindex fshort-enums
3529 Specifying this attribute for @code{struct} and @code{union} types is
3530 equivalent to specifying the @code{packed} attribute on each of the
3531 structure or union members. Specifying the @option{-fshort-enums}
3532 flag on the line is equivalent to specifying the @code{packed}
3533 attribute on all @code{enum} definitions.
3534
3535 In the following example @code{struct my_packed_struct}'s members are
3536 packed closely together, but the internal layout of its @code{s} member
3537 is not packed -- to do that, @code{struct my_unpacked_struct} would need to
3538 be packed too.
3539
3540 @smallexample
3541 struct my_unpacked_struct
3542 @{
3543 char c;
3544 int i;
3545 @};
3546
3547 struct my_packed_struct __attribute__ ((__packed__))
3548 @{
3549 char c;
3550 int i;
3551 struct my_unpacked_struct s;
3552 @};
3553 @end smallexample
3554
3555 You may only specify this attribute on the definition of a @code{enum},
3556 @code{struct} or @code{union}, not on a @code{typedef} which does not
3557 also define the enumerated type, structure or union.
3558
3559 @item transparent_union
3560 This attribute, attached to a @code{union} type definition, indicates
3561 that any function parameter having that union type causes calls to that
3562 function to be treated in a special way.
3563
3564 First, the argument corresponding to a transparent union type can be of
3565 any type in the union; no cast is required. Also, if the union contains
3566 a pointer type, the corresponding argument can be a null pointer
3567 constant or a void pointer expression; and if the union contains a void
3568 pointer type, the corresponding argument can be any pointer expression.
3569 If the union member type is a pointer, qualifiers like @code{const} on
3570 the referenced type must be respected, just as with normal pointer
3571 conversions.
3572
3573 Second, the argument is passed to the function using the calling
3574 conventions of the first member of the transparent union, not the calling
3575 conventions of the union itself. All members of the union must have the
3576 same machine representation; this is necessary for this argument passing
3577 to work properly.
3578
3579 Transparent unions are designed for library functions that have multiple
3580 interfaces for compatibility reasons. For example, suppose the
3581 @code{wait} function must accept either a value of type @code{int *} to
3582 comply with Posix, or a value of type @code{union wait *} to comply with
3583 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3584 @code{wait} would accept both kinds of arguments, but it would also
3585 accept any other pointer type and this would make argument type checking
3586 less useful. Instead, @code{<sys/wait.h>} might define the interface
3587 as follows:
3588
3589 @smallexample
3590 typedef union
3591 @{
3592 int *__ip;
3593 union wait *__up;
3594 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3595
3596 pid_t wait (wait_status_ptr_t);
3597 @end smallexample
3598
3599 This interface allows either @code{int *} or @code{union wait *}
3600 arguments to be passed, using the @code{int *} calling convention.
3601 The program can call @code{wait} with arguments of either type:
3602
3603 @example
3604 int w1 () @{ int w; return wait (&w); @}
3605 int w2 () @{ union wait w; return wait (&w); @}
3606 @end example
3607
3608 With this interface, @code{wait}'s implementation might look like this:
3609
3610 @example
3611 pid_t wait (wait_status_ptr_t p)
3612 @{
3613 return waitpid (-1, p.__ip, 0);
3614 @}
3615 @end example
3616
3617 @item unused
3618 When attached to a type (including a @code{union} or a @code{struct}),
3619 this attribute means that variables of that type are meant to appear
3620 possibly unused. GCC will not produce a warning for any variables of
3621 that type, even if the variable appears to do nothing. This is often
3622 the case with lock or thread classes, which are usually defined and then
3623 not referenced, but contain constructors and destructors that have
3624 nontrivial bookkeeping functions.
3625
3626 @item deprecated
3627 The @code{deprecated} attribute results in a warning if the type
3628 is used anywhere in the source file. This is useful when identifying
3629 types that are expected to be removed in a future version of a program.
3630 If possible, the warning also includes the location of the declaration
3631 of the deprecated type, to enable users to easily find further
3632 information about why the type is deprecated, or what they should do
3633 instead. Note that the warnings only occur for uses and then only
3634 if the type is being applied to an identifier that itself is not being
3635 declared as deprecated.
3636
3637 @smallexample
3638 typedef int T1 __attribute__ ((deprecated));
3639 T1 x;
3640 typedef T1 T2;
3641 T2 y;
3642 typedef T1 T3 __attribute__ ((deprecated));
3643 T3 z __attribute__ ((deprecated));
3644 @end smallexample
3645
3646 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
3647 warning is issued for line 4 because T2 is not explicitly
3648 deprecated. Line 5 has no warning because T3 is explicitly
3649 deprecated. Similarly for line 6.
3650
3651 The @code{deprecated} attribute can also be used for functions and
3652 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
3653
3654 @item may_alias
3655 Accesses to objects with types with this attribute are not subjected to
3656 type-based alias analysis, but are instead assumed to be able to alias
3657 any other type of objects, just like the @code{char} type. See
3658 @option{-fstrict-aliasing} for more information on aliasing issues.
3659
3660 Example of use:
3661
3662 @smallexample
3663 typedef short __attribute__((__may_alias__)) short_a;
3664
3665 int
3666 main (void)
3667 @{
3668 int a = 0x12345678;
3669 short_a *b = (short_a *) &a;
3670
3671 b[1] = 0;
3672
3673 if (a == 0x12345678)
3674 abort();
3675
3676 exit(0);
3677 @}
3678 @end smallexample
3679
3680 If you replaced @code{short_a} with @code{short} in the variable
3681 declaration, the above program would abort when compiled with
3682 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
3683 above in recent GCC versions.
3684
3685 @subsection i386 Type Attributes
3686
3687 Two attributes are currently defined for i386 configurations:
3688 @code{ms_struct} and @code{gcc_struct}
3689
3690 @item ms_struct
3691 @itemx gcc_struct
3692 @cindex @code{ms_struct}
3693 @cindex @code{gcc_struct}
3694
3695 If @code{packed} is used on a structure, or if bit-fields are used
3696 it may be that the Microsoft ABI packs them differently
3697 than GCC would normally pack them. Particularly when moving packed
3698 data between functions compiled with GCC and the native Microsoft compiler
3699 (either via function call or as data in a file), it may be necessary to access
3700 either format.
3701
3702 Currently @option{-m[no-]ms-bitfields} is provided for the Windows X86
3703 compilers to match the native Microsoft compiler.
3704 @end table
3705
3706 To specify multiple attributes, separate them by commas within the
3707 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3708 packed))}.
3709
3710 @node Inline
3711 @section An Inline Function is As Fast As a Macro
3712 @cindex inline functions
3713 @cindex integrating function code
3714 @cindex open coding
3715 @cindex macros, inline alternative
3716
3717 By declaring a function @code{inline}, you can direct GCC to
3718 integrate that function's code into the code for its callers. This
3719 makes execution faster by eliminating the function-call overhead; in
3720 addition, if any of the actual argument values are constant, their known
3721 values may permit simplifications at compile time so that not all of the
3722 inline function's code needs to be included. The effect on code size is
3723 less predictable; object code may be larger or smaller with function
3724 inlining, depending on the particular case. Inlining of functions is an
3725 optimization and it really ``works'' only in optimizing compilation. If
3726 you don't use @option{-O}, no function is really inline.
3727
3728 Inline functions are included in the ISO C99 standard, but there are
3729 currently substantial differences between what GCC implements and what
3730 the ISO C99 standard requires.
3731
3732 To declare a function inline, use the @code{inline} keyword in its
3733 declaration, like this:
3734
3735 @example
3736 inline int
3737 inc (int *a)
3738 @{
3739 (*a)++;
3740 @}
3741 @end example
3742
3743 (If you are writing a header file to be included in ISO C programs, write
3744 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3745 You can also make all ``simple enough'' functions inline with the option
3746 @option{-finline-functions}.
3747
3748 @opindex Winline
3749 Note that certain usages in a function definition can make it unsuitable
3750 for inline substitution. Among these usages are: use of varargs, use of
3751 alloca, use of variable sized data types (@pxref{Variable Length}),
3752 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3753 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3754 will warn when a function marked @code{inline} could not be substituted,
3755 and will give the reason for the failure.
3756
3757 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3758 does not affect the linkage of the function.
3759
3760 @cindex automatic @code{inline} for C++ member fns
3761 @cindex @code{inline} automatic for C++ member fns
3762 @cindex member fns, automatically @code{inline}
3763 @cindex C++ member fns, automatically @code{inline}
3764 @opindex fno-default-inline
3765 GCC automatically inlines member functions defined within the class
3766 body of C++ programs even if they are not explicitly declared
3767 @code{inline}. (You can override this with @option{-fno-default-inline};
3768 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3769
3770 @cindex inline functions, omission of
3771 @opindex fkeep-inline-functions
3772 When a function is both inline and @code{static}, if all calls to the
3773 function are integrated into the caller, and the function's address is
3774 never used, then the function's own assembler code is never referenced.
3775 In this case, GCC does not actually output assembler code for the
3776 function, unless you specify the option @option{-fkeep-inline-functions}.
3777 Some calls cannot be integrated for various reasons (in particular,
3778 calls that precede the function's definition cannot be integrated, and
3779 neither can recursive calls within the definition). If there is a
3780 nonintegrated call, then the function is compiled to assembler code as
3781 usual. The function must also be compiled as usual if the program
3782 refers to its address, because that can't be inlined.
3783
3784 @cindex non-static inline function
3785 When an inline function is not @code{static}, then the compiler must assume
3786 that there may be calls from other source files; since a global symbol can
3787 be defined only once in any program, the function must not be defined in
3788 the other source files, so the calls therein cannot be integrated.
3789 Therefore, a non-@code{static} inline function is always compiled on its
3790 own in the usual fashion.
3791
3792 If you specify both @code{inline} and @code{extern} in the function
3793 definition, then the definition is used only for inlining. In no case
3794 is the function compiled on its own, not even if you refer to its
3795 address explicitly. Such an address becomes an external reference, as
3796 if you had only declared the function, and had not defined it.
3797
3798 This combination of @code{inline} and @code{extern} has almost the
3799 effect of a macro. The way to use it is to put a function definition in
3800 a header file with these keywords, and put another copy of the
3801 definition (lacking @code{inline} and @code{extern}) in a library file.
3802 The definition in the header file will cause most calls to the function
3803 to be inlined. If any uses of the function remain, they will refer to
3804 the single copy in the library.
3805
3806 Since GCC eventually will implement ISO C99 semantics for
3807 inline functions, it is best to use @code{static inline} only
3808 to guarentee compatibility. (The
3809 existing semantics will remain available when @option{-std=gnu89} is
3810 specified, but eventually the default will be @option{-std=gnu99} and
3811 that will implement the C99 semantics, though it does not do so yet.)
3812
3813 GCC does not inline any functions when not optimizing unless you specify
3814 the @samp{always_inline} attribute for the function, like this:
3815
3816 @example
3817 /* Prototype. */
3818 inline void foo (const char) __attribute__((always_inline));
3819 @end example
3820
3821 @node Extended Asm
3822 @section Assembler Instructions with C Expression Operands
3823 @cindex extended @code{asm}
3824 @cindex @code{asm} expressions
3825 @cindex assembler instructions
3826 @cindex registers
3827
3828 In an assembler instruction using @code{asm}, you can specify the
3829 operands of the instruction using C expressions. This means you need not
3830 guess which registers or memory locations will contain the data you want
3831 to use.
3832
3833 You must specify an assembler instruction template much like what
3834 appears in a machine description, plus an operand constraint string for
3835 each operand.
3836
3837 For example, here is how to use the 68881's @code{fsinx} instruction:
3838
3839 @example
3840 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3841 @end example
3842
3843 @noindent
3844 Here @code{angle} is the C expression for the input operand while
3845 @code{result} is that of the output operand. Each has @samp{"f"} as its
3846 operand constraint, saying that a floating point register is required.
3847 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3848 output operands' constraints must use @samp{=}. The constraints use the
3849 same language used in the machine description (@pxref{Constraints}).
3850
3851 Each operand is described by an operand-constraint string followed by
3852 the C expression in parentheses. A colon separates the assembler
3853 template from the first output operand and another separates the last
3854 output operand from the first input, if any. Commas separate the
3855 operands within each group. The total number of operands is currently
3856 limited to 30; this limitation may be lifted in some future version of
3857 GCC.
3858
3859 If there are no output operands but there are input operands, you must
3860 place two consecutive colons surrounding the place where the output
3861 operands would go.
3862
3863 As of GCC version 3.1, it is also possible to specify input and output
3864 operands using symbolic names which can be referenced within the
3865 assembler code. These names are specified inside square brackets
3866 preceding the constraint string, and can be referenced inside the
3867 assembler code using @code{%[@var{name}]} instead of a percentage sign
3868 followed by the operand number. Using named operands the above example
3869 could look like:
3870
3871 @example
3872 asm ("fsinx %[angle],%[output]"
3873 : [output] "=f" (result)
3874 : [angle] "f" (angle));
3875 @end example
3876
3877 @noindent
3878 Note that the symbolic operand names have no relation whatsoever to
3879 other C identifiers. You may use any name you like, even those of
3880 existing C symbols, but you must ensure that no two operands within the same
3881 assembler construct use the same symbolic name.
3882
3883 Output operand expressions must be lvalues; the compiler can check this.
3884 The input operands need not be lvalues. The compiler cannot check
3885 whether the operands have data types that are reasonable for the
3886 instruction being executed. It does not parse the assembler instruction
3887 template and does not know what it means or even whether it is valid
3888 assembler input. The extended @code{asm} feature is most often used for
3889 machine instructions the compiler itself does not know exist. If
3890 the output expression cannot be directly addressed (for example, it is a
3891 bit-field), your constraint must allow a register. In that case, GCC
3892 will use the register as the output of the @code{asm}, and then store
3893 that register into the output.
3894
3895 The ordinary output operands must be write-only; GCC will assume that
3896 the values in these operands before the instruction are dead and need
3897 not be generated. Extended asm supports input-output or read-write
3898 operands. Use the constraint character @samp{+} to indicate such an
3899 operand and list it with the output operands.
3900
3901 When the constraints for the read-write operand (or the operand in which
3902 only some of the bits are to be changed) allows a register, you may, as
3903 an alternative, logically split its function into two separate operands,
3904 one input operand and one write-only output operand. The connection
3905 between them is expressed by constraints which say they need to be in
3906 the same location when the instruction executes. You can use the same C
3907 expression for both operands, or different expressions. For example,
3908 here we write the (fictitious) @samp{combine} instruction with
3909 @code{bar} as its read-only source operand and @code{foo} as its
3910 read-write destination:
3911
3912 @example
3913 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3914 @end example
3915
3916 @noindent
3917 The constraint @samp{"0"} for operand 1 says that it must occupy the
3918 same location as operand 0. A number in constraint is allowed only in
3919 an input operand and it must refer to an output operand.
3920
3921 Only a number in the constraint can guarantee that one operand will be in
3922 the same place as another. The mere fact that @code{foo} is the value
3923 of both operands is not enough to guarantee that they will be in the
3924 same place in the generated assembler code. The following would not
3925 work reliably:
3926
3927 @example
3928 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3929 @end example
3930
3931 Various optimizations or reloading could cause operands 0 and 1 to be in
3932 different registers; GCC knows no reason not to do so. For example, the
3933 compiler might find a copy of the value of @code{foo} in one register and
3934 use it for operand 1, but generate the output operand 0 in a different
3935 register (copying it afterward to @code{foo}'s own address). Of course,
3936 since the register for operand 1 is not even mentioned in the assembler
3937 code, the result will not work, but GCC can't tell that.
3938
3939 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3940 the operand number for a matching constraint. For example:
3941
3942 @example
3943 asm ("cmoveq %1,%2,%[result]"
3944 : [result] "=r"(result)
3945 : "r" (test), "r"(new), "[result]"(old));
3946 @end example
3947
3948 Some instructions clobber specific hard registers. To describe this,
3949 write a third colon after the input operands, followed by the names of
3950 the clobbered hard registers (given as strings). Here is a realistic
3951 example for the VAX:
3952
3953 @example
3954 asm volatile ("movc3 %0,%1,%2"
3955 : /* no outputs */
3956 : "g" (from), "g" (to), "g" (count)
3957 : "r0", "r1", "r2", "r3", "r4", "r5");
3958 @end example
3959
3960 You may not write a clobber description in a way that overlaps with an
3961 input or output operand. For example, you may not have an operand
3962 describing a register class with one member if you mention that register
3963 in the clobber list. Variables declared to live in specific registers
3964 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
3965 have no part mentioned in the clobber description.
3966 There is no way for you to specify that an input
3967 operand is modified without also specifying it as an output
3968 operand. Note that if all the output operands you specify are for this
3969 purpose (and hence unused), you will then also need to specify
3970 @code{volatile} for the @code{asm} construct, as described below, to
3971 prevent GCC from deleting the @code{asm} statement as unused.
3972
3973 If you refer to a particular hardware register from the assembler code,
3974 you will probably have to list the register after the third colon to
3975 tell the compiler the register's value is modified. In some assemblers,
3976 the register names begin with @samp{%}; to produce one @samp{%} in the
3977 assembler code, you must write @samp{%%} in the input.
3978
3979 If your assembler instruction can alter the condition code register, add
3980 @samp{cc} to the list of clobbered registers. GCC on some machines
3981 represents the condition codes as a specific hardware register;
3982 @samp{cc} serves to name this register. On other machines, the
3983 condition code is handled differently, and specifying @samp{cc} has no
3984 effect. But it is valid no matter what the machine.
3985
3986 If your assembler instruction modifies memory in an unpredictable
3987 fashion, add @samp{memory} to the list of clobbered registers. This
3988 will cause GCC to not keep memory values cached in registers across
3989 the assembler instruction. You will also want to add the
3990 @code{volatile} keyword if the memory affected is not listed in the
3991 inputs or outputs of the @code{asm}, as the @samp{memory} clobber does
3992 not count as a side-effect of the @code{asm}.
3993
3994 You can put multiple assembler instructions together in a single
3995 @code{asm} template, separated by the characters normally used in assembly
3996 code for the system. A combination that works in most places is a newline
3997 to break the line, plus a tab character to move to the instruction field
3998 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3999 assembler allows semicolons as a line-breaking character. Note that some
4000 assembler dialects use semicolons to start a comment.
4001 The input operands are guaranteed not to use any of the clobbered
4002 registers, and neither will the output operands' addresses, so you can
4003 read and write the clobbered registers as many times as you like. Here
4004 is an example of multiple instructions in a template; it assumes the
4005 subroutine @code{_foo} accepts arguments in registers 9 and 10:
4006
4007 @example
4008 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
4009 : /* no outputs */
4010 : "g" (from), "g" (to)
4011 : "r9", "r10");
4012 @end example
4013
4014 Unless an output operand has the @samp{&} constraint modifier, GCC
4015 may allocate it in the same register as an unrelated input operand, on
4016 the assumption the inputs are consumed before the outputs are produced.
4017 This assumption may be false if the assembler code actually consists of
4018 more than one instruction. In such a case, use @samp{&} for each output
4019 operand that may not overlap an input. @xref{Modifiers}.
4020
4021 If you want to test the condition code produced by an assembler
4022 instruction, you must include a branch and a label in the @code{asm}
4023 construct, as follows:
4024
4025 @example
4026 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
4027 : "g" (result)
4028 : "g" (input));
4029 @end example
4030
4031 @noindent
4032 This assumes your assembler supports local labels, as the GNU assembler
4033 and most Unix assemblers do.
4034
4035 Speaking of labels, jumps from one @code{asm} to another are not
4036 supported. The compiler's optimizers do not know about these jumps, and
4037 therefore they cannot take account of them when deciding how to
4038 optimize.
4039
4040 @cindex macros containing @code{asm}
4041 Usually the most convenient way to use these @code{asm} instructions is to
4042 encapsulate them in macros that look like functions. For example,
4043
4044 @example
4045 #define sin(x) \
4046 (@{ double __value, __arg = (x); \
4047 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
4048 __value; @})
4049 @end example
4050
4051 @noindent
4052 Here the variable @code{__arg} is used to make sure that the instruction
4053 operates on a proper @code{double} value, and to accept only those
4054 arguments @code{x} which can convert automatically to a @code{double}.
4055
4056 Another way to make sure the instruction operates on the correct data
4057 type is to use a cast in the @code{asm}. This is different from using a
4058 variable @code{__arg} in that it converts more different types. For
4059 example, if the desired type were @code{int}, casting the argument to
4060 @code{int} would accept a pointer with no complaint, while assigning the
4061 argument to an @code{int} variable named @code{__arg} would warn about
4062 using a pointer unless the caller explicitly casts it.
4063
4064 If an @code{asm} has output operands, GCC assumes for optimization
4065 purposes the instruction has no side effects except to change the output
4066 operands. This does not mean instructions with a side effect cannot be
4067 used, but you must be careful, because the compiler may eliminate them
4068 if the output operands aren't used, or move them out of loops, or
4069 replace two with one if they constitute a common subexpression. Also,
4070 if your instruction does have a side effect on a variable that otherwise
4071 appears not to change, the old value of the variable may be reused later
4072 if it happens to be found in a register.
4073
4074 You can prevent an @code{asm} instruction from being deleted, moved
4075 significantly, or combined, by writing the keyword @code{volatile} after
4076 the @code{asm}. For example:
4077
4078 @example
4079 #define get_and_set_priority(new) \
4080 (@{ int __old; \
4081 asm volatile ("get_and_set_priority %0, %1" \
4082 : "=g" (__old) : "g" (new)); \
4083 __old; @})
4084 @end example
4085
4086 @noindent
4087 If you write an @code{asm} instruction with no outputs, GCC will know
4088 the instruction has side-effects and will not delete the instruction or
4089 move it outside of loops.
4090
4091 The @code{volatile} keyword indicates that the instruction has
4092 important side-effects. GCC will not delete a volatile @code{asm} if
4093 it is reachable. (The instruction can still be deleted if GCC can
4094 prove that control-flow will never reach the location of the
4095 instruction.) In addition, GCC will not reschedule instructions
4096 across a volatile @code{asm} instruction. For example:
4097
4098 @example
4099 *(volatile int *)addr = foo;
4100 asm volatile ("eieio" : : );
4101 @end example
4102
4103 @noindent
4104 Assume @code{addr} contains the address of a memory mapped device
4105 register. The PowerPC @code{eieio} instruction (Enforce In-order
4106 Execution of I/O) tells the CPU to make sure that the store to that
4107 device register happens before it issues any other I/O@.
4108
4109 Note that even a volatile @code{asm} instruction can be moved in ways
4110 that appear insignificant to the compiler, such as across jump
4111 instructions. You can't expect a sequence of volatile @code{asm}
4112 instructions to remain perfectly consecutive. If you want consecutive
4113 output, use a single @code{asm}. Also, GCC will perform some
4114 optimizations across a volatile @code{asm} instruction; GCC does not
4115 ``forget everything'' when it encounters a volatile @code{asm}
4116 instruction the way some other compilers do.
4117
4118 An @code{asm} instruction without any operands or clobbers (an ``old
4119 style'' @code{asm}) will be treated identically to a volatile
4120 @code{asm} instruction.
4121
4122 It is a natural idea to look for a way to give access to the condition
4123 code left by the assembler instruction. However, when we attempted to
4124 implement this, we found no way to make it work reliably. The problem
4125 is that output operands might need reloading, which would result in
4126 additional following ``store'' instructions. On most machines, these
4127 instructions would alter the condition code before there was time to
4128 test it. This problem doesn't arise for ordinary ``test'' and
4129 ``compare'' instructions because they don't have any output operands.
4130
4131 For reasons similar to those described above, it is not possible to give
4132 an assembler instruction access to the condition code left by previous
4133 instructions.
4134
4135 If you are writing a header file that should be includable in ISO C
4136 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
4137 Keywords}.
4138
4139 @subsection Size of an @code{asm}
4140
4141 Some targets require that GCC track the size of each instruction used in
4142 order to generate correct code. Because the final length of an
4143 @code{asm} is only known by the assembler, GCC must make an estimate as
4144 to how big it will be. The estimate is formed by counting the number of
4145 statements in the pattern of the @code{asm} and multiplying that by the
4146 length of the longest instruction on that processor. Statements in the
4147 @code{asm} are identified by newline characters and whatever statement
4148 separator characters are supported by the assembler; on most processors
4149 this is the `@code{;}' character.
4150
4151 Normally, GCC's estimate is perfectly adequate to ensure that correct
4152 code is generated, but it is possible to confuse the compiler if you use
4153 pseudo instructions or assembler macros that expand into multiple real
4154 instructions or if you use assembler directives that expand to more
4155 space in the object file than would be needed for a single instruction.
4156 If this happens then the assembler will produce a diagnostic saying that
4157 a label is unreachable.
4158
4159 @subsection i386 floating point asm operands
4160
4161 There are several rules on the usage of stack-like regs in
4162 asm_operands insns. These rules apply only to the operands that are
4163 stack-like regs:
4164
4165 @enumerate
4166 @item
4167 Given a set of input regs that die in an asm_operands, it is
4168 necessary to know which are implicitly popped by the asm, and
4169 which must be explicitly popped by gcc.
4170
4171 An input reg that is implicitly popped by the asm must be
4172 explicitly clobbered, unless it is constrained to match an
4173 output operand.
4174
4175 @item
4176 For any input reg that is implicitly popped by an asm, it is
4177 necessary to know how to adjust the stack to compensate for the pop.
4178 If any non-popped input is closer to the top of the reg-stack than
4179 the implicitly popped reg, it would not be possible to know what the
4180 stack looked like---it's not clear how the rest of the stack ``slides
4181 up''.
4182
4183 All implicitly popped input regs must be closer to the top of
4184 the reg-stack than any input that is not implicitly popped.
4185
4186 It is possible that if an input dies in an insn, reload might
4187 use the input reg for an output reload. Consider this example:
4188
4189 @example
4190 asm ("foo" : "=t" (a) : "f" (b));
4191 @end example
4192
4193 This asm says that input B is not popped by the asm, and that
4194 the asm pushes a result onto the reg-stack, i.e., the stack is one
4195 deeper after the asm than it was before. But, it is possible that
4196 reload will think that it can use the same reg for both the input and
4197 the output, if input B dies in this insn.
4198
4199 If any input operand uses the @code{f} constraint, all output reg
4200 constraints must use the @code{&} earlyclobber.
4201
4202 The asm above would be written as
4203
4204 @example
4205 asm ("foo" : "=&t" (a) : "f" (b));
4206 @end example
4207
4208 @item
4209 Some operands need to be in particular places on the stack. All
4210 output operands fall in this category---there is no other way to
4211 know which regs the outputs appear in unless the user indicates
4212 this in the constraints.
4213
4214 Output operands must specifically indicate which reg an output
4215 appears in after an asm. @code{=f} is not allowed: the operand
4216 constraints must select a class with a single reg.
4217
4218 @item
4219 Output operands may not be ``inserted'' between existing stack regs.
4220 Since no 387 opcode uses a read/write operand, all output operands
4221 are dead before the asm_operands, and are pushed by the asm_operands.
4222 It makes no sense to push anywhere but the top of the reg-stack.
4223
4224 Output operands must start at the top of the reg-stack: output
4225 operands may not ``skip'' a reg.
4226
4227 @item
4228 Some asm statements may need extra stack space for internal
4229 calculations. This can be guaranteed by clobbering stack registers
4230 unrelated to the inputs and outputs.
4231
4232 @end enumerate
4233
4234 Here are a couple of reasonable asms to want to write. This asm
4235 takes one input, which is internally popped, and produces two outputs.
4236
4237 @example
4238 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
4239 @end example
4240
4241 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
4242 and replaces them with one output. The user must code the @code{st(1)}
4243 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
4244
4245 @example
4246 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
4247 @end example
4248
4249 @include md.texi
4250
4251 @node Asm Labels
4252 @section Controlling Names Used in Assembler Code
4253 @cindex assembler names for identifiers
4254 @cindex names used in assembler code
4255 @cindex identifiers, names in assembler code
4256
4257 You can specify the name to be used in the assembler code for a C
4258 function or variable by writing the @code{asm} (or @code{__asm__})
4259 keyword after the declarator as follows:
4260
4261 @example
4262 int foo asm ("myfoo") = 2;
4263 @end example
4264
4265 @noindent
4266 This specifies that the name to be used for the variable @code{foo} in
4267 the assembler code should be @samp{myfoo} rather than the usual
4268 @samp{_foo}.
4269
4270 On systems where an underscore is normally prepended to the name of a C
4271 function or variable, this feature allows you to define names for the
4272 linker that do not start with an underscore.
4273
4274 It does not make sense to use this feature with a non-static local
4275 variable since such variables do not have assembler names. If you are
4276 trying to put the variable in a particular register, see @ref{Explicit
4277 Reg Vars}. GCC presently accepts such code with a warning, but will
4278 probably be changed to issue an error, rather than a warning, in the
4279 future.
4280
4281 You cannot use @code{asm} in this way in a function @emph{definition}; but
4282 you can get the same effect by writing a declaration for the function
4283 before its definition and putting @code{asm} there, like this:
4284
4285 @example
4286 extern func () asm ("FUNC");
4287
4288 func (x, y)
4289 int x, y;
4290 /* @r{@dots{}} */
4291 @end example
4292
4293 It is up to you to make sure that the assembler names you choose do not
4294 conflict with any other assembler symbols. Also, you must not use a
4295 register name; that would produce completely invalid assembler code. GCC
4296 does not as yet have the ability to store static variables in registers.
4297 Perhaps that will be added.
4298
4299 @node Explicit Reg Vars
4300 @section Variables in Specified Registers
4301 @cindex explicit register variables
4302 @cindex variables in specified registers
4303 @cindex specified registers
4304 @cindex registers, global allocation
4305
4306 GNU C allows you to put a few global variables into specified hardware
4307 registers. You can also specify the register in which an ordinary
4308 register variable should be allocated.
4309
4310 @itemize @bullet
4311 @item
4312 Global register variables reserve registers throughout the program.
4313 This may be useful in programs such as programming language
4314 interpreters which have a couple of global variables that are accessed
4315 very often.
4316
4317 @item
4318 Local register variables in specific registers do not reserve the
4319 registers. The compiler's data flow analysis is capable of determining
4320 where the specified registers contain live values, and where they are
4321 available for other uses. Stores into local register variables may be deleted
4322 when they appear to be dead according to dataflow analysis. References
4323 to local register variables may be deleted or moved or simplified.
4324
4325 These local variables are sometimes convenient for use with the extended
4326 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
4327 output of the assembler instruction directly into a particular register.
4328 (This will work provided the register you specify fits the constraints
4329 specified for that operand in the @code{asm}.)
4330 @end itemize
4331
4332 @menu
4333 * Global Reg Vars::
4334 * Local Reg Vars::
4335 @end menu
4336
4337 @node Global Reg Vars
4338 @subsection Defining Global Register Variables
4339 @cindex global register variables
4340 @cindex registers, global variables in
4341
4342 You can define a global register variable in GNU C like this:
4343
4344 @example
4345 register int *foo asm ("a5");
4346 @end example
4347
4348 @noindent
4349 Here @code{a5} is the name of the register which should be used. Choose a
4350 register which is normally saved and restored by function calls on your
4351 machine, so that library routines will not clobber it.
4352
4353 Naturally the register name is cpu-dependent, so you would need to
4354 conditionalize your program according to cpu type. The register
4355 @code{a5} would be a good choice on a 68000 for a variable of pointer
4356 type. On machines with register windows, be sure to choose a ``global''
4357 register that is not affected magically by the function call mechanism.
4358
4359 In addition, operating systems on one type of cpu may differ in how they
4360 name the registers; then you would need additional conditionals. For
4361 example, some 68000 operating systems call this register @code{%a5}.
4362
4363 Eventually there may be a way of asking the compiler to choose a register
4364 automatically, but first we need to figure out how it should choose and
4365 how to enable you to guide the choice. No solution is evident.
4366
4367 Defining a global register variable in a certain register reserves that
4368 register entirely for this use, at least within the current compilation.
4369 The register will not be allocated for any other purpose in the functions
4370 in the current compilation. The register will not be saved and restored by
4371 these functions. Stores into this register are never deleted even if they
4372 would appear to be dead, but references may be deleted or moved or
4373 simplified.
4374
4375 It is not safe to access the global register variables from signal
4376 handlers, or from more than one thread of control, because the system
4377 library routines may temporarily use the register for other things (unless
4378 you recompile them specially for the task at hand).
4379
4380 @cindex @code{qsort}, and global register variables
4381 It is not safe for one function that uses a global register variable to
4382 call another such function @code{foo} by way of a third function
4383 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
4384 different source file in which the variable wasn't declared). This is
4385 because @code{lose} might save the register and put some other value there.
4386 For example, you can't expect a global register variable to be available in
4387 the comparison-function that you pass to @code{qsort}, since @code{qsort}
4388 might have put something else in that register. (If you are prepared to
4389 recompile @code{qsort} with the same global register variable, you can
4390 solve this problem.)
4391
4392 If you want to recompile @code{qsort} or other source files which do not
4393 actually use your global register variable, so that they will not use that
4394 register for any other purpose, then it suffices to specify the compiler
4395 option @option{-ffixed-@var{reg}}. You need not actually add a global
4396 register declaration to their source code.
4397
4398 A function which can alter the value of a global register variable cannot
4399 safely be called from a function compiled without this variable, because it
4400 could clobber the value the caller expects to find there on return.
4401 Therefore, the function which is the entry point into the part of the
4402 program that uses the global register variable must explicitly save and
4403 restore the value which belongs to its caller.
4404
4405 @cindex register variable after @code{longjmp}
4406 @cindex global register after @code{longjmp}
4407 @cindex value after @code{longjmp}
4408 @findex longjmp
4409 @findex setjmp
4410 On most machines, @code{longjmp} will restore to each global register
4411 variable the value it had at the time of the @code{setjmp}. On some
4412 machines, however, @code{longjmp} will not change the value of global
4413 register variables. To be portable, the function that called @code{setjmp}
4414 should make other arrangements to save the values of the global register
4415 variables, and to restore them in a @code{longjmp}. This way, the same
4416 thing will happen regardless of what @code{longjmp} does.
4417
4418 All global register variable declarations must precede all function
4419 definitions. If such a declaration could appear after function
4420 definitions, the declaration would be too late to prevent the register from
4421 being used for other purposes in the preceding functions.
4422
4423 Global register variables may not have initial values, because an
4424 executable file has no means to supply initial contents for a register.
4425
4426 On the SPARC, there are reports that g3 @dots{} g7 are suitable
4427 registers, but certain library functions, such as @code{getwd}, as well
4428 as the subroutines for division and remainder, modify g3 and g4. g1 and
4429 g2 are local temporaries.
4430
4431 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
4432 Of course, it will not do to use more than a few of those.
4433
4434 @node Local Reg Vars
4435 @subsection Specifying Registers for Local Variables
4436 @cindex local variables, specifying registers
4437 @cindex specifying registers for local variables
4438 @cindex registers for local variables
4439
4440 You can define a local register variable with a specified register
4441 like this:
4442
4443 @example
4444 register int *foo asm ("a5");
4445 @end example
4446
4447 @noindent
4448 Here @code{a5} is the name of the register which should be used. Note
4449 that this is the same syntax used for defining global register
4450 variables, but for a local variable it would appear within a function.
4451
4452 Naturally the register name is cpu-dependent, but this is not a
4453 problem, since specific registers are most often useful with explicit
4454 assembler instructions (@pxref{Extended Asm}). Both of these things
4455 generally require that you conditionalize your program according to
4456 cpu type.
4457
4458 In addition, operating systems on one type of cpu may differ in how they
4459 name the registers; then you would need additional conditionals. For
4460 example, some 68000 operating systems call this register @code{%a5}.
4461
4462 Defining such a register variable does not reserve the register; it
4463 remains available for other uses in places where flow control determines
4464 the variable's value is not live. However, these registers are made
4465 unavailable for use in the reload pass; excessive use of this feature
4466 leaves the compiler too few available registers to compile certain
4467 functions.
4468
4469 This option does not guarantee that GCC will generate code that has
4470 this variable in the register you specify at all times. You may not
4471 code an explicit reference to this register in an @code{asm} statement
4472 and assume it will always refer to this variable.
4473
4474 Stores into local register variables may be deleted when they appear to be dead
4475 according to dataflow analysis. References to local register variables may
4476 be deleted or moved or simplified.
4477
4478 @node Alternate Keywords
4479 @section Alternate Keywords
4480 @cindex alternate keywords
4481 @cindex keywords, alternate
4482
4483 @option{-ansi} and the various @option{-std} options disable certain
4484 keywords. This causes trouble when you want to use GNU C extensions, or
4485 a general-purpose header file that should be usable by all programs,
4486 including ISO C programs. The keywords @code{asm}, @code{typeof} and
4487 @code{inline} are not available in programs compiled with
4488 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
4489 program compiled with @option{-std=c99}). The ISO C99 keyword
4490 @code{restrict} is only available when @option{-std=gnu99} (which will
4491 eventually be the default) or @option{-std=c99} (or the equivalent
4492 @option{-std=iso9899:1999}) is used.
4493
4494 The way to solve these problems is to put @samp{__} at the beginning and
4495 end of each problematical keyword. For example, use @code{__asm__}
4496 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
4497
4498 Other C compilers won't accept these alternative keywords; if you want to
4499 compile with another compiler, you can define the alternate keywords as
4500 macros to replace them with the customary keywords. It looks like this:
4501
4502 @example
4503 #ifndef __GNUC__
4504 #define __asm__ asm
4505 #endif
4506 @end example
4507
4508 @findex __extension__
4509 @opindex pedantic
4510 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4511 You can
4512 prevent such warnings within one expression by writing
4513 @code{__extension__} before the expression. @code{__extension__} has no
4514 effect aside from this.
4515
4516 @node Incomplete Enums
4517 @section Incomplete @code{enum} Types
4518
4519 You can define an @code{enum} tag without specifying its possible values.
4520 This results in an incomplete type, much like what you get if you write
4521 @code{struct foo} without describing the elements. A later declaration
4522 which does specify the possible values completes the type.
4523
4524 You can't allocate variables or storage using the type while it is
4525 incomplete. However, you can work with pointers to that type.
4526
4527 This extension may not be very useful, but it makes the handling of
4528 @code{enum} more consistent with the way @code{struct} and @code{union}
4529 are handled.
4530
4531 This extension is not supported by GNU C++.
4532
4533 @node Function Names
4534 @section Function Names as Strings
4535 @cindex @code{__func__} identifier
4536 @cindex @code{__FUNCTION__} identifier
4537 @cindex @code{__PRETTY_FUNCTION__} identifier
4538
4539 GCC provides three magic variables which hold the name of the current
4540 function, as a string. The first of these is @code{__func__}, which
4541 is part of the C99 standard:
4542
4543 @display
4544 The identifier @code{__func__} is implicitly declared by the translator
4545 as if, immediately following the opening brace of each function
4546 definition, the declaration
4547
4548 @smallexample
4549 static const char __func__[] = "function-name";
4550 @end smallexample
4551
4552 appeared, where function-name is the name of the lexically-enclosing
4553 function. This name is the unadorned name of the function.
4554 @end display
4555
4556 @code{__FUNCTION__} is another name for @code{__func__}. Older
4557 versions of GCC recognize only this name. However, it is not
4558 standardized. For maximum portability, we recommend you use
4559 @code{__func__}, but provide a fallback definition with the
4560 preprocessor:
4561
4562 @smallexample
4563 #if __STDC_VERSION__ < 199901L
4564 # if __GNUC__ >= 2
4565 # define __func__ __FUNCTION__
4566 # else
4567 # define __func__ "<unknown>"
4568 # endif
4569 #endif
4570 @end smallexample
4571
4572 In C, @code{__PRETTY_FUNCTION__} is yet another name for
4573 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
4574 the type signature of the function as well as its bare name. For
4575 example, this program:
4576
4577 @smallexample
4578 extern "C" @{
4579 extern int printf (char *, ...);
4580 @}
4581
4582 class a @{
4583 public:
4584 void sub (int i)
4585 @{
4586 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4587 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4588 @}
4589 @};
4590
4591 int
4592 main (void)
4593 @{
4594 a ax;
4595 ax.sub (0);
4596 return 0;
4597 @}
4598 @end smallexample
4599
4600 @noindent
4601 gives this output:
4602
4603 @smallexample
4604 __FUNCTION__ = sub
4605 __PRETTY_FUNCTION__ = void a::sub(int)
4606 @end smallexample
4607
4608 These identifiers are not preprocessor macros. In GCC 3.3 and
4609 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
4610 were treated as string literals; they could be used to initialize
4611 @code{char} arrays, and they could be concatenated with other string
4612 literals. GCC 3.4 and later treat them as variables, like
4613 @code{__func__}. In C++, @code{__FUNCTION__} and
4614 @code{__PRETTY_FUNCTION__} have always been variables.
4615
4616 @node Return Address
4617 @section Getting the Return or Frame Address of a Function
4618
4619 These functions may be used to get information about the callers of a
4620 function.
4621
4622 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4623 This function returns the return address of the current function, or of
4624 one of its callers. The @var{level} argument is number of frames to
4625 scan up the call stack. A value of @code{0} yields the return address
4626 of the current function, a value of @code{1} yields the return address
4627 of the caller of the current function, and so forth. When inlining
4628 the expected behavior is that the function will return the address of
4629 the function that will be returned to. To work around this behavior use
4630 the @code{noinline} function attribute.
4631
4632 The @var{level} argument must be a constant integer.
4633
4634 On some machines it may be impossible to determine the return address of
4635 any function other than the current one; in such cases, or when the top
4636 of the stack has been reached, this function will return @code{0} or a
4637 random value. In addition, @code{__builtin_frame_address} may be used
4638 to determine if the top of the stack has been reached.
4639
4640 This function should only be used with a nonzero argument for debugging
4641 purposes.
4642 @end deftypefn
4643
4644 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4645 This function is similar to @code{__builtin_return_address}, but it
4646 returns the address of the function frame rather than the return address
4647 of the function. Calling @code{__builtin_frame_address} with a value of
4648 @code{0} yields the frame address of the current function, a value of
4649 @code{1} yields the frame address of the caller of the current function,
4650 and so forth.
4651
4652 The frame is the area on the stack which holds local variables and saved
4653 registers. The frame address is normally the address of the first word
4654 pushed on to the stack by the function. However, the exact definition
4655 depends upon the processor and the calling convention. If the processor
4656 has a dedicated frame pointer register, and the function has a frame,
4657 then @code{__builtin_frame_address} will return the value of the frame
4658 pointer register.
4659
4660 On some machines it may be impossible to determine the frame address of
4661 any function other than the current one; in such cases, or when the top
4662 of the stack has been reached, this function will return @code{0} if
4663 the first frame pointer is properly initialized by the startup code.
4664
4665 This function should only be used with a nonzero argument for debugging
4666 purposes.
4667 @end deftypefn
4668
4669 @node Vector Extensions
4670 @section Using vector instructions through built-in functions
4671
4672 On some targets, the instruction set contains SIMD vector instructions that
4673 operate on multiple values contained in one large register at the same time.
4674 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4675 this way.
4676
4677 The first step in using these extensions is to provide the necessary data
4678 types. This should be done using an appropriate @code{typedef}:
4679
4680 @example
4681 typedef int v4si __attribute__ ((mode(V4SI)));
4682 @end example
4683
4684 The base type @code{int} is effectively ignored by the compiler, the
4685 actual properties of the new type @code{v4si} are defined by the
4686 @code{__attribute__}. It defines the machine mode to be used; for vector
4687 types these have the form @code{V@var{n}@var{B}}; @var{n} should be the
4688 number of elements in the vector, and @var{B} should be the base mode of the
4689 individual elements. The following can be used as base modes:
4690
4691 @table @code
4692 @item QI
4693 An integer that is as wide as the smallest addressable unit, usually 8 bits.
4694 @item HI
4695 An integer, twice as wide as a QI mode integer, usually 16 bits.
4696 @item SI
4697 An integer, four times as wide as a QI mode integer, usually 32 bits.
4698 @item DI
4699 An integer, eight times as wide as a QI mode integer, usually 64 bits.
4700 @item SF
4701 A floating point value, as wide as a SI mode integer, usually 32 bits.
4702 @item DF
4703 A floating point value, as wide as a DI mode integer, usually 64 bits.
4704 @end table
4705
4706 Specifying a combination that is not valid for the current architecture
4707 will cause gcc to synthesize the instructions using a narrower mode.
4708 For example, if you specify a variable of type @code{V4SI} and your
4709 architecture does not allow for this specific SIMD type, gcc will
4710 produce code that uses 4 @code{SIs}.
4711
4712 The types defined in this manner can be used with a subset of normal C
4713 operations. Currently, gcc will allow using the following operators on
4714 these types: @code{+, -, *, /, unary minus}@.
4715
4716 The operations behave like C++ @code{valarrays}. Addition is defined as
4717 the addition of the corresponding elements of the operands. For
4718 example, in the code below, each of the 4 elements in @var{a} will be
4719 added to the corresponding 4 elements in @var{b} and the resulting
4720 vector will be stored in @var{c}.
4721
4722 @example
4723 typedef int v4si __attribute__ ((mode(V4SI)));
4724
4725 v4si a, b, c;
4726
4727 c = a + b;
4728 @end example
4729
4730 Subtraction, multiplication, and division operate in a similar manner.
4731 Likewise, the result of using the unary minus operator on a vector type
4732 is a vector whose elements are the negative value of the corresponding
4733 elements in the operand.
4734
4735 You can declare variables and use them in function calls and returns, as
4736 well as in assignments and some casts. You can specify a vector type as
4737 a return type for a function. Vector types can also be used as function
4738 arguments. It is possible to cast from one vector type to another,
4739 provided they are of the same size (in fact, you can also cast vectors
4740 to and from other datatypes of the same size).
4741
4742 You cannot operate between vectors of different lengths or different
4743 signedness without a cast.
4744
4745 A port that supports hardware vector operations, usually provides a set
4746 of built-in functions that can be used to operate on vectors. For
4747 example, a function to add two vectors and multiply the result by a
4748 third could look like this:
4749
4750 @example
4751 v4si f (v4si a, v4si b, v4si c)
4752 @{
4753 v4si tmp = __builtin_addv4si (a, b);
4754 return __builtin_mulv4si (tmp, c);
4755 @}
4756
4757 @end example
4758
4759 @node Other Builtins
4760 @section Other built-in functions provided by GCC
4761 @cindex built-in functions
4762 @findex __builtin_isgreater
4763 @findex __builtin_isgreaterequal
4764 @findex __builtin_isless
4765 @findex __builtin_islessequal
4766 @findex __builtin_islessgreater
4767 @findex __builtin_isunordered
4768 @findex _Exit
4769 @findex _exit
4770 @findex abort
4771 @findex abs
4772 @findex acos
4773 @findex acosf
4774 @findex acosh
4775 @findex acoshf
4776 @findex acoshl
4777 @findex acosl
4778 @findex alloca
4779 @findex asin
4780 @findex asinf
4781 @findex asinh
4782 @findex asinhf
4783 @findex asinhl
4784 @findex asinl
4785 @findex atan
4786 @findex atan2
4787 @findex atan2f
4788 @findex atan2l
4789 @findex atanf
4790 @findex atanh
4791 @findex atanhf
4792 @findex atanhl
4793 @findex atanl
4794 @findex bcmp
4795 @findex bzero
4796 @findex cabs
4797 @findex cabsf
4798 @findex cabsl
4799 @findex cacos
4800 @findex cacosf
4801 @findex cacosh
4802 @findex cacoshf
4803 @findex cacoshl
4804 @findex cacosl
4805 @findex calloc
4806 @findex carg
4807 @findex cargf
4808 @findex cargl
4809 @findex casin
4810 @findex casinf
4811 @findex casinh
4812 @findex casinhf
4813 @findex casinhl
4814 @findex casinl
4815 @findex catan
4816 @findex catanf
4817 @findex catanh
4818 @findex catanhf
4819 @findex catanhl
4820 @findex catanl
4821 @findex cbrt
4822 @findex cbrtf
4823 @findex cbrtl
4824 @findex ccos
4825 @findex ccosf
4826 @findex ccosh
4827 @findex ccoshf
4828 @findex ccoshl
4829 @findex ccosl
4830 @findex ceil
4831 @findex ceilf
4832 @findex ceill
4833 @findex cexp
4834 @findex cexpf
4835 @findex cexpl
4836 @findex cimag
4837 @findex cimagf
4838 @findex cimagl
4839 @findex conj
4840 @findex conjf
4841 @findex conjl
4842 @findex copysign
4843 @findex copysignf
4844 @findex copysignl
4845 @findex cos
4846 @findex cosf
4847 @findex cosh
4848 @findex coshf
4849 @findex coshl
4850 @findex cosl
4851 @findex cpow
4852 @findex cpowf
4853 @findex cpowl
4854 @findex cproj
4855 @findex cprojf
4856 @findex cprojl
4857 @findex creal
4858 @findex crealf
4859 @findex creall
4860 @findex csin
4861 @findex csinf
4862 @findex csinh
4863 @findex csinhf
4864 @findex csinhl
4865 @findex csinl
4866 @findex csqrt
4867 @findex csqrtf
4868 @findex csqrtl
4869 @findex ctan
4870 @findex ctanf
4871 @findex ctanh
4872 @findex ctanhf
4873 @findex ctanhl
4874 @findex ctanl
4875 @findex dcgettext
4876 @findex dgettext
4877 @findex drem
4878 @findex dremf
4879 @findex dreml
4880 @findex erf
4881 @findex erfc
4882 @findex erfcf
4883 @findex erfcl
4884 @findex erff
4885 @findex erfl
4886 @findex exit
4887 @findex exp
4888 @findex exp10
4889 @findex exp10f
4890 @findex exp10l
4891 @findex exp2
4892 @findex exp2f
4893 @findex exp2l
4894 @findex expf
4895 @findex expl
4896 @findex expm1
4897 @findex expm1f
4898 @findex expm1l
4899 @findex fabs
4900 @findex fabsf
4901 @findex fabsl
4902 @findex fdim
4903 @findex fdimf
4904 @findex fdiml
4905 @findex ffs
4906 @findex floor
4907 @findex floorf
4908 @findex floorl
4909 @findex fma
4910 @findex fmaf
4911 @findex fmal
4912 @findex fmax
4913 @findex fmaxf
4914 @findex fmaxl
4915 @findex fmin
4916 @findex fminf
4917 @findex fminl
4918 @findex fmod
4919 @findex fmodf
4920 @findex fmodl
4921 @findex fprintf
4922 @findex fprintf_unlocked
4923 @findex fputs
4924 @findex fputs_unlocked
4925 @findex frexp
4926 @findex frexpf
4927 @findex frexpl
4928 @findex fscanf
4929 @findex gamma
4930 @findex gammaf
4931 @findex gammal
4932 @findex gettext
4933 @findex hypot
4934 @findex hypotf
4935 @findex hypotl
4936 @findex ilogb
4937 @findex ilogbf
4938 @findex ilogbl
4939 @findex imaxabs
4940 @findex index
4941 @findex j0
4942 @findex j0f
4943 @findex j0l
4944 @findex j1
4945 @findex j1f
4946 @findex j1l
4947 @findex jn
4948 @findex jnf
4949 @findex jnl
4950 @findex labs
4951 @findex ldexp
4952 @findex ldexpf
4953 @findex ldexpl
4954 @findex lgamma
4955 @findex lgammaf
4956 @findex lgammal
4957 @findex llabs
4958 @findex llrint
4959 @findex llrintf
4960 @findex llrintl
4961 @findex llround
4962 @findex llroundf
4963 @findex llroundl
4964 @findex log
4965 @findex log10
4966 @findex log10f
4967 @findex log10l
4968 @findex log1p
4969 @findex log1pf
4970 @findex log1pl
4971 @findex log2
4972 @findex log2f
4973 @findex log2l
4974 @findex logb
4975 @findex logbf
4976 @findex logbl
4977 @findex logf
4978 @findex logl
4979 @findex lrint
4980 @findex lrintf
4981 @findex lrintl
4982 @findex lround
4983 @findex lroundf
4984 @findex lroundl
4985 @findex malloc
4986 @findex memcmp
4987 @findex memcpy
4988 @findex mempcpy
4989 @findex memset
4990 @findex modf
4991 @findex modff
4992 @findex modfl
4993 @findex nearbyint
4994 @findex nearbyintf
4995 @findex nearbyintl
4996 @findex nextafter
4997 @findex nextafterf
4998 @findex nextafterl
4999 @findex nexttoward
5000 @findex nexttowardf
5001 @findex nexttowardl
5002 @findex pow
5003 @findex pow10
5004 @findex pow10f
5005 @findex pow10l
5006 @findex powf
5007 @findex powl
5008 @findex printf
5009 @findex printf_unlocked
5010 @findex putchar
5011 @findex puts
5012 @findex remainder
5013 @findex remainderf
5014 @findex remainderl
5015 @findex remquo
5016 @findex remquof
5017 @findex remquol
5018 @findex rindex
5019 @findex rint
5020 @findex rintf
5021 @findex rintl
5022 @findex round
5023 @findex roundf
5024 @findex roundl
5025 @findex scalb
5026 @findex scalbf
5027 @findex scalbl
5028 @findex scalbln
5029 @findex scalblnf
5030 @findex scalblnf
5031 @findex scalbn
5032 @findex scalbnf
5033 @findex scanfnl
5034 @findex significand
5035 @findex significandf
5036 @findex significandl
5037 @findex sin
5038 @findex sincos
5039 @findex sincosf
5040 @findex sincosl
5041 @findex sinf
5042 @findex sinh
5043 @findex sinhf
5044 @findex sinhl
5045 @findex sinl
5046 @findex snprintf
5047 @findex sprintf
5048 @findex sqrt
5049 @findex sqrtf
5050 @findex sqrtl
5051 @findex sscanf
5052 @findex stpcpy
5053 @findex strcat
5054 @findex strchr
5055 @findex strcmp
5056 @findex strcpy
5057 @findex strcspn
5058 @findex strdup
5059 @findex strfmon
5060 @findex strftime
5061 @findex strlen
5062 @findex strncat
5063 @findex strncmp
5064 @findex strncpy
5065 @findex strpbrk
5066 @findex strrchr
5067 @findex strspn
5068 @findex strstr
5069 @findex tan
5070 @findex tanf
5071 @findex tanh
5072 @findex tanhf
5073 @findex tanhl
5074 @findex tanl
5075 @findex tgamma
5076 @findex tgammaf
5077 @findex tgammal
5078 @findex trunc
5079 @findex truncf
5080 @findex truncl
5081 @findex vfprintf
5082 @findex vfscanf
5083 @findex vprintf
5084 @findex vscanf
5085 @findex vsnprintf
5086 @findex vsprintf
5087 @findex vsscanf
5088 @findex y0
5089 @findex y0f
5090 @findex y0l
5091 @findex y1
5092 @findex y1f
5093 @findex y1l
5094 @findex yn
5095 @findex ynf
5096 @findex ynl
5097
5098 GCC provides a large number of built-in functions other than the ones
5099 mentioned above. Some of these are for internal use in the processing
5100 of exceptions or variable-length argument lists and will not be
5101 documented here because they may change from time to time; we do not
5102 recommend general use of these functions.
5103
5104 The remaining functions are provided for optimization purposes.
5105
5106 @opindex fno-builtin
5107 GCC includes built-in versions of many of the functions in the standard
5108 C library. The versions prefixed with @code{__builtin_} will always be
5109 treated as having the same meaning as the C library function even if you
5110 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
5111 Many of these functions are only optimized in certain cases; if they are
5112 not optimized in a particular case, a call to the library function will
5113 be emitted.
5114
5115 @opindex ansi
5116 @opindex std
5117 Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or
5118 @option{-std=c99}), the functions
5119 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
5120 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
5121 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
5122 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked}, @code{fputs_unlocked},
5123 @code{gammaf}, @code{gammal}, @code{gamma}, @code{gettext},
5124 @code{index}, @code{j0f}, @code{j0l}, @code{j0}, @code{j1f}, @code{j1l},
5125 @code{j1}, @code{jnf}, @code{jnl}, @code{jn}, @code{mempcpy},
5126 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
5127 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
5128 @code{significandf}, @code{significandl}, @code{significand},
5129 @code{sincosf}, @code{sincosl}, @code{sincos}, @code{stpcpy},
5130 @code{strdup}, @code{strfmon}, @code{y0f}, @code{y0l}, @code{y0},
5131 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and @code{yn}
5132 may be handled as built-in functions.
5133 All these functions have corresponding versions
5134 prefixed with @code{__builtin_}, which may be used even in strict C89
5135 mode.
5136
5137 The ISO C99 functions
5138 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
5139 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
5140 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
5141 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
5142 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
5143 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
5144 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
5145 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
5146 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
5147 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
5148 @code{cimagl}, @code{cimag},
5149 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf},
5150 @code{copysignl}, @code{copysign}, @code{cpowf}, @code{cpowl},
5151 @code{cpow}, @code{cprojf}, @code{cprojl}, @code{cproj}, @code{crealf},
5152 @code{creall}, @code{creal}, @code{csinf}, @code{csinhf}, @code{csinhl},
5153 @code{csinh}, @code{csinl}, @code{csin}, @code{csqrtf}, @code{csqrtl},
5154 @code{csqrt}, @code{ctanf}, @code{ctanhf}, @code{ctanhl}, @code{ctanh},
5155 @code{ctanl}, @code{ctan}, @code{erfcf}, @code{erfcl}, @code{erfc},
5156 @code{erff}, @code{erfl}, @code{erf}, @code{exp2f}, @code{exp2l},
5157 @code{exp2}, @code{expm1f}, @code{expm1l}, @code{expm1}, @code{fdimf},
5158 @code{fdiml}, @code{fdim}, @code{fmaf}, @code{fmal}, @code{fmaxf},
5159 @code{fmaxl}, @code{fmax}, @code{fma}, @code{fminf}, @code{fminl},
5160 @code{fmin}, @code{hypotf}, @code{hypotl}, @code{hypot}, @code{ilogbf},
5161 @code{ilogbl}, @code{ilogb}, @code{imaxabs}, @code{lgammaf},
5162 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf},
5163 @code{llrintl}, @code{llrint}, @code{llroundf}, @code{llroundl},
5164 @code{llround}, @code{log1pf}, @code{log1pl}, @code{log1p},
5165 @code{log2f}, @code{log2l}, @code{log2}, @code{logbf}, @code{logbl},
5166 @code{logb}, @code{lrintf}, @code{lrintl}, @code{lrint}, @code{lroundf},
5167 @code{lroundl}, @code{lround}, @code{nearbyintf}, @code{nearbyintl},
5168 @code{nearbyint}, @code{nextafterf}, @code{nextafterl},
5169 @code{nextafter}, @code{nexttowardf}, @code{nexttowardl},
5170 @code{nexttoward}, @code{remainderf}, @code{remainderl},
5171 @code{remainder}, @code{remquof}, @code{remquol}, @code{remquo},
5172 @code{rintf}, @code{rintl}, @code{rint}, @code{roundf}, @code{roundl},
5173 @code{round}, @code{scalblnf}, @code{scalblnl}, @code{scalbln},
5174 @code{scalbnf}, @code{scalbnl}, @code{scalbn}, @code{snprintf},
5175 @code{tgammaf}, @code{tgammal}, @code{tgamma}, @code{truncf},
5176 @code{truncl}, @code{trunc}, @code{vfscanf}, @code{vscanf},
5177 @code{vsnprintf} and @code{vsscanf}
5178 are handled as built-in functions
5179 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
5180
5181 There are also built-in versions of the ISO C99 functions
5182 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
5183 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
5184 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
5185 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
5186 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
5187 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
5188 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
5189 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
5190 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
5191 that are recognized in any mode since ISO C90 reserves these names for
5192 the purpose to which ISO C99 puts them. All these functions have
5193 corresponding versions prefixed with @code{__builtin_}.
5194
5195 The ISO C90 functions
5196 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
5197 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
5198 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
5199 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf}, @code{labs},
5200 @code{ldexp}, @code{log10}, @code{log}, @code{malloc}, @code{memcmp},
5201 @code{memcpy}, @code{memset}, @code{modf}, @code{pow}, @code{printf},
5202 @code{putchar}, @code{puts}, @code{scanf}, @code{sinh}, @code{sin},
5203 @code{snprintf}, @code{sprintf}, @code{sqrt}, @code{sscanf},
5204 @code{strcat}, @code{strchr}, @code{strcmp}, @code{strcpy},
5205 @code{strcspn}, @code{strlen}, @code{strncat}, @code{strncmp},
5206 @code{strncpy}, @code{strpbrk}, @code{strrchr}, @code{strspn},
5207 @code{strstr}, @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf}
5208 and @code{vsprintf}
5209 are all recognized as built-in functions unless
5210 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
5211 is specified for an individual function). All of these functions have
5212 corresponding versions prefixed with @code{__builtin_}.
5213
5214 GCC provides built-in versions of the ISO C99 floating point comparison
5215 macros that avoid raising exceptions for unordered operands. They have
5216 the same names as the standard macros ( @code{isgreater},
5217 @code{isgreaterequal}, @code{isless}, @code{islessequal},
5218 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
5219 prefixed. We intend for a library implementor to be able to simply
5220 @code{#define} each standard macro to its built-in equivalent.
5221
5222 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
5223
5224 You can use the built-in function @code{__builtin_types_compatible_p} to
5225 determine whether two types are the same.
5226
5227 This built-in function returns 1 if the unqualified versions of the
5228 types @var{type1} and @var{type2} (which are types, not expressions) are
5229 compatible, 0 otherwise. The result of this built-in function can be
5230 used in integer constant expressions.
5231
5232 This built-in function ignores top level qualifiers (e.g., @code{const},
5233 @code{volatile}). For example, @code{int} is equivalent to @code{const
5234 int}.
5235
5236 The type @code{int[]} and @code{int[5]} are compatible. On the other
5237 hand, @code{int} and @code{char *} are not compatible, even if the size
5238 of their types, on the particular architecture are the same. Also, the
5239 amount of pointer indirection is taken into account when determining
5240 similarity. Consequently, @code{short *} is not similar to
5241 @code{short **}. Furthermore, two types that are typedefed are
5242 considered compatible if their underlying types are compatible.
5243
5244 An @code{enum} type is considered to be compatible with another
5245 @code{enum} type. For example, @code{enum @{foo, bar@}} is similar to
5246 @code{enum @{hot, dog@}}.
5247
5248 You would typically use this function in code whose execution varies
5249 depending on the arguments' types. For example:
5250
5251 @smallexample
5252 #define foo(x) \
5253 (@{ \
5254 typeof (x) tmp; \
5255 if (__builtin_types_compatible_p (typeof (x), long double)) \
5256 tmp = foo_long_double (tmp); \
5257 else if (__builtin_types_compatible_p (typeof (x), double)) \
5258 tmp = foo_double (tmp); \
5259 else if (__builtin_types_compatible_p (typeof (x), float)) \
5260 tmp = foo_float (tmp); \
5261 else \
5262 abort (); \
5263 tmp; \
5264 @})
5265 @end smallexample
5266
5267 @emph{Note:} This construct is only available for C.
5268
5269 @end deftypefn
5270
5271 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
5272
5273 You can use the built-in function @code{__builtin_choose_expr} to
5274 evaluate code depending on the value of a constant expression. This
5275 built-in function returns @var{exp1} if @var{const_exp}, which is a
5276 constant expression that must be able to be determined at compile time,
5277 is nonzero. Otherwise it returns 0.
5278
5279 This built-in function is analogous to the @samp{? :} operator in C,
5280 except that the expression returned has its type unaltered by promotion
5281 rules. Also, the built-in function does not evaluate the expression
5282 that was not chosen. For example, if @var{const_exp} evaluates to true,
5283 @var{exp2} is not evaluated even if it has side-effects.
5284
5285 This built-in function can return an lvalue if the chosen argument is an
5286 lvalue.
5287
5288 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
5289 type. Similarly, if @var{exp2} is returned, its return type is the same
5290 as @var{exp2}.
5291
5292 Example:
5293
5294 @smallexample
5295 #define foo(x) \
5296 __builtin_choose_expr ( \
5297 __builtin_types_compatible_p (typeof (x), double), \
5298 foo_double (x), \
5299 __builtin_choose_expr ( \
5300 __builtin_types_compatible_p (typeof (x), float), \
5301 foo_float (x), \
5302 /* @r{The void expression results in a compile-time error} \
5303 @r{when assigning the result to something.} */ \
5304 (void)0))
5305 @end smallexample
5306
5307 @emph{Note:} This construct is only available for C. Furthermore, the
5308 unused expression (@var{exp1} or @var{exp2} depending on the value of
5309 @var{const_exp}) may still generate syntax errors. This may change in
5310 future revisions.
5311
5312 @end deftypefn
5313
5314 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
5315 You can use the built-in function @code{__builtin_constant_p} to
5316 determine if a value is known to be constant at compile-time and hence
5317 that GCC can perform constant-folding on expressions involving that
5318 value. The argument of the function is the value to test. The function
5319 returns the integer 1 if the argument is known to be a compile-time
5320 constant and 0 if it is not known to be a compile-time constant. A
5321 return of 0 does not indicate that the value is @emph{not} a constant,
5322 but merely that GCC cannot prove it is a constant with the specified
5323 value of the @option{-O} option.
5324
5325 You would typically use this function in an embedded application where
5326 memory was a critical resource. If you have some complex calculation,
5327 you may want it to be folded if it involves constants, but need to call
5328 a function if it does not. For example:
5329
5330 @smallexample
5331 #define Scale_Value(X) \
5332 (__builtin_constant_p (X) \
5333 ? ((X) * SCALE + OFFSET) : Scale (X))
5334 @end smallexample
5335
5336 You may use this built-in function in either a macro or an inline
5337 function. However, if you use it in an inlined function and pass an
5338 argument of the function as the argument to the built-in, GCC will
5339 never return 1 when you call the inline function with a string constant
5340 or compound literal (@pxref{Compound Literals}) and will not return 1
5341 when you pass a constant numeric value to the inline function unless you
5342 specify the @option{-O} option.
5343
5344 You may also use @code{__builtin_constant_p} in initializers for static
5345 data. For instance, you can write
5346
5347 @smallexample
5348 static const int table[] = @{
5349 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
5350 /* @r{@dots{}} */
5351 @};
5352 @end smallexample
5353
5354 @noindent
5355 This is an acceptable initializer even if @var{EXPRESSION} is not a
5356 constant expression. GCC must be more conservative about evaluating the
5357 built-in in this case, because it has no opportunity to perform
5358 optimization.
5359
5360 Previous versions of GCC did not accept this built-in in data
5361 initializers. The earliest version where it is completely safe is
5362 3.0.1.
5363 @end deftypefn
5364
5365 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
5366 @opindex fprofile-arcs
5367 You may use @code{__builtin_expect} to provide the compiler with
5368 branch prediction information. In general, you should prefer to
5369 use actual profile feedback for this (@option{-fprofile-arcs}), as
5370 programmers are notoriously bad at predicting how their programs
5371 actually perform. However, there are applications in which this
5372 data is hard to collect.
5373
5374 The return value is the value of @var{exp}, which should be an
5375 integral expression. The value of @var{c} must be a compile-time
5376 constant. The semantics of the built-in are that it is expected
5377 that @var{exp} == @var{c}. For example:
5378
5379 @smallexample
5380 if (__builtin_expect (x, 0))
5381 foo ();
5382 @end smallexample
5383
5384 @noindent
5385 would indicate that we do not expect to call @code{foo}, since
5386 we expect @code{x} to be zero. Since you are limited to integral
5387 expressions for @var{exp}, you should use constructions such as
5388
5389 @smallexample
5390 if (__builtin_expect (ptr != NULL, 1))
5391 error ();
5392 @end smallexample
5393
5394 @noindent
5395 when testing pointer or floating-point values.
5396 @end deftypefn
5397
5398 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
5399 This function is used to minimize cache-miss latency by moving data into
5400 a cache before it is accessed.
5401 You can insert calls to @code{__builtin_prefetch} into code for which
5402 you know addresses of data in memory that is likely to be accessed soon.
5403 If the target supports them, data prefetch instructions will be generated.
5404 If the prefetch is done early enough before the access then the data will
5405 be in the cache by the time it is accessed.
5406
5407 The value of @var{addr} is the address of the memory to prefetch.
5408 There are two optional arguments, @var{rw} and @var{locality}.
5409 The value of @var{rw} is a compile-time constant one or zero; one
5410 means that the prefetch is preparing for a write to the memory address
5411 and zero, the default, means that the prefetch is preparing for a read.
5412 The value @var{locality} must be a compile-time constant integer between
5413 zero and three. A value of zero means that the data has no temporal
5414 locality, so it need not be left in the cache after the access. A value
5415 of three means that the data has a high degree of temporal locality and
5416 should be left in all levels of cache possible. Values of one and two
5417 mean, respectively, a low or moderate degree of temporal locality. The
5418 default is three.
5419
5420 @smallexample
5421 for (i = 0; i < n; i++)
5422 @{
5423 a[i] = a[i] + b[i];
5424 __builtin_prefetch (&a[i+j], 1, 1);
5425 __builtin_prefetch (&b[i+j], 0, 1);
5426 /* @r{@dots{}} */
5427 @}
5428 @end smallexample
5429
5430 Data prefetch does not generate faults if @var{addr} is invalid, but
5431 the address expression itself must be valid. For example, a prefetch
5432 of @code{p->next} will not fault if @code{p->next} is not a valid
5433 address, but evaluation will fault if @code{p} is not a valid address.
5434
5435 If the target does not support data prefetch, the address expression
5436 is evaluated if it includes side effects but no other code is generated
5437 and GCC does not issue a warning.
5438 @end deftypefn
5439
5440 @deftypefn {Built-in Function} double __builtin_huge_val (void)
5441 Returns a positive infinity, if supported by the floating-point format,
5442 else @code{DBL_MAX}. This function is suitable for implementing the
5443 ISO C macro @code{HUGE_VAL}.
5444 @end deftypefn
5445
5446 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
5447 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
5448 @end deftypefn
5449
5450 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
5451 Similar to @code{__builtin_huge_val}, except the return
5452 type is @code{long double}.
5453 @end deftypefn
5454
5455 @deftypefn {Built-in Function} double __builtin_inf (void)
5456 Similar to @code{__builtin_huge_val}, except a warning is generated
5457 if the target floating-point format does not support infinities.
5458 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
5459 @end deftypefn
5460
5461 @deftypefn {Built-in Function} float __builtin_inff (void)
5462 Similar to @code{__builtin_inf}, except the return type is @code{float}.
5463 @end deftypefn
5464
5465 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
5466 Similar to @code{__builtin_inf}, except the return
5467 type is @code{long double}.
5468 @end deftypefn
5469
5470 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
5471 This is an implementation of the ISO C99 function @code{nan}.
5472
5473 Since ISO C99 defines this function in terms of @code{strtod}, which we
5474 do not implement, a description of the parsing is in order. The string
5475 is parsed as by @code{strtol}; that is, the base is recognized by
5476 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
5477 in the significand such that the least significant bit of the number
5478 is at the least significant bit of the significand. The number is
5479 truncated to fit the significand field provided. The significand is
5480 forced to be a quiet NaN.
5481
5482 This function, if given a string literal, is evaluated early enough
5483 that it is considered a compile-time constant.
5484 @end deftypefn
5485
5486 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
5487 Similar to @code{__builtin_nan}, except the return type is @code{float}.
5488 @end deftypefn
5489
5490 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
5491 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
5492 @end deftypefn
5493
5494 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
5495 Similar to @code{__builtin_nan}, except the significand is forced
5496 to be a signaling NaN. The @code{nans} function is proposed by
5497 @uref{http://std.dkuug.dk/JTC1/SC22/WG14/www/docs/n965.htm,,WG14 N965}.
5498 @end deftypefn
5499
5500 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
5501 Similar to @code{__builtin_nans}, except the return type is @code{float}.
5502 @end deftypefn
5503
5504 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
5505 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
5506 @end deftypefn
5507
5508 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
5509 Returns one plus the index of the least significant 1-bit of @var{x}, or
5510 if @var{x} is zero, returns zero.
5511 @end deftypefn
5512
5513 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
5514 Returns the number of leading 0-bits in @var{x}, starting at the most
5515 significant bit position. If @var{x} is 0, the result is undefined.
5516 @end deftypefn
5517
5518 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
5519 Returns the number of trailing 0-bits in @var{x}, starting at the least
5520 significant bit position. If @var{x} is 0, the result is undefined.
5521 @end deftypefn
5522
5523 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
5524 Returns the number of 1-bits in @var{x}.
5525 @end deftypefn
5526
5527 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
5528 Returns the parity of @var{x}, i.@:e. the number of 1-bits in @var{x}
5529 modulo 2.
5530 @end deftypefn
5531
5532 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
5533 Similar to @code{__builtin_ffs}, except the argument type is
5534 @code{unsigned long}.
5535 @end deftypefn
5536
5537 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
5538 Similar to @code{__builtin_clz}, except the argument type is
5539 @code{unsigned long}.
5540 @end deftypefn
5541
5542 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
5543 Similar to @code{__builtin_ctz}, except the argument type is
5544 @code{unsigned long}.
5545 @end deftypefn
5546
5547 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
5548 Similar to @code{__builtin_popcount}, except the argument type is
5549 @code{unsigned long}.
5550 @end deftypefn
5551
5552 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
5553 Similar to @code{__builtin_parity}, except the argument type is
5554 @code{unsigned long}.
5555 @end deftypefn
5556
5557 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
5558 Similar to @code{__builtin_ffs}, except the argument type is
5559 @code{unsigned long long}.
5560 @end deftypefn
5561
5562 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
5563 Similar to @code{__builtin_clz}, except the argument type is
5564 @code{unsigned long long}.
5565 @end deftypefn
5566
5567 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
5568 Similar to @code{__builtin_ctz}, except the argument type is
5569 @code{unsigned long long}.
5570 @end deftypefn
5571
5572 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
5573 Similar to @code{__builtin_popcount}, except the argument type is
5574 @code{unsigned long long}.
5575 @end deftypefn
5576
5577 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
5578 Similar to @code{__builtin_parity}, except the argument type is
5579 @code{unsigned long long}.
5580 @end deftypefn
5581
5582
5583 @node Target Builtins
5584 @section Built-in Functions Specific to Particular Target Machines
5585
5586 On some target machines, GCC supports many built-in functions specific
5587 to those machines. Generally these generate calls to specific machine
5588 instructions, but allow the compiler to schedule those calls.
5589
5590 @menu
5591 * Alpha Built-in Functions::
5592 * ARM Built-in Functions::
5593 * X86 Built-in Functions::
5594 * PowerPC AltiVec Built-in Functions::
5595 @end menu
5596
5597 @node Alpha Built-in Functions
5598 @subsection Alpha Built-in Functions
5599
5600 These built-in functions are available for the Alpha family of
5601 processors, depending on the command-line switches used.
5602
5603 The following built-in functions are always available. They
5604 all generate the machine instruction that is part of the name.
5605
5606 @example
5607 long __builtin_alpha_implver (void)
5608 long __builtin_alpha_rpcc (void)
5609 long __builtin_alpha_amask (long)
5610 long __builtin_alpha_cmpbge (long, long)
5611 long __builtin_alpha_extbl (long, long)
5612 long __builtin_alpha_extwl (long, long)
5613 long __builtin_alpha_extll (long, long)
5614 long __builtin_alpha_extql (long, long)
5615 long __builtin_alpha_extwh (long, long)
5616 long __builtin_alpha_extlh (long, long)
5617 long __builtin_alpha_extqh (long, long)
5618 long __builtin_alpha_insbl (long, long)
5619 long __builtin_alpha_inswl (long, long)
5620 long __builtin_alpha_insll (long, long)
5621 long __builtin_alpha_insql (long, long)
5622 long __builtin_alpha_inswh (long, long)
5623 long __builtin_alpha_inslh (long, long)
5624 long __builtin_alpha_insqh (long, long)
5625 long __builtin_alpha_mskbl (long, long)
5626 long __builtin_alpha_mskwl (long, long)
5627 long __builtin_alpha_mskll (long, long)
5628 long __builtin_alpha_mskql (long, long)
5629 long __builtin_alpha_mskwh (long, long)
5630 long __builtin_alpha_msklh (long, long)
5631 long __builtin_alpha_mskqh (long, long)
5632 long __builtin_alpha_umulh (long, long)
5633 long __builtin_alpha_zap (long, long)
5634 long __builtin_alpha_zapnot (long, long)
5635 @end example
5636
5637 The following built-in functions are always with @option{-mmax}
5638 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
5639 later. They all generate the machine instruction that is part
5640 of the name.
5641
5642 @example
5643 long __builtin_alpha_pklb (long)
5644 long __builtin_alpha_pkwb (long)
5645 long __builtin_alpha_unpkbl (long)
5646 long __builtin_alpha_unpkbw (long)
5647 long __builtin_alpha_minub8 (long, long)
5648 long __builtin_alpha_minsb8 (long, long)
5649 long __builtin_alpha_minuw4 (long, long)
5650 long __builtin_alpha_minsw4 (long, long)
5651 long __builtin_alpha_maxub8 (long, long)
5652 long __builtin_alpha_maxsb8 (long, long)
5653 long __builtin_alpha_maxuw4 (long, long)
5654 long __builtin_alpha_maxsw4 (long, long)
5655 long __builtin_alpha_perr (long, long)
5656 @end example
5657
5658 The following built-in functions are always with @option{-mcix}
5659 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
5660 later. They all generate the machine instruction that is part
5661 of the name.
5662
5663 @example
5664 long __builtin_alpha_cttz (long)
5665 long __builtin_alpha_ctlz (long)
5666 long __builtin_alpha_ctpop (long)
5667 @end example
5668
5669 The following builtins are available on systems that use the OSF/1
5670 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
5671 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
5672 @code{rdval} and @code{wrval}.
5673
5674 @example
5675 void *__builtin_thread_pointer (void)
5676 void __builtin_set_thread_pointer (void *)
5677 @end example
5678
5679 @node ARM Built-in Functions
5680 @subsection ARM Built-in Functions
5681
5682 These built-in functions are available for the ARM family of
5683 processors, when the @option{-mcpu=iwmmxt} switch is used:
5684
5685 @example
5686 typedef int __v2si __attribute__ ((__mode__ (__V2SI__)))
5687
5688 v2si __builtin_arm_waddw (v2si, v2si)
5689 v2si __builtin_arm_waddw (v2si, v2si)
5690 v2si __builtin_arm_wsubw (v2si, v2si)
5691 v2si __builtin_arm_wsubw (v2si, v2si)
5692 v2si __builtin_arm_waddwss (v2si, v2si)
5693 v2si __builtin_arm_wsubwss (v2si, v2si)
5694 v2si __builtin_arm_wsubwss (v2si, v2si)
5695 v2si __builtin_arm_wsubwss (v2si, v2si)
5696 v2si __builtin_arm_wsubwss (v2si, v2si)
5697 v2si __builtin_arm_waddwus (v2si, v2si)
5698 v2si __builtin_arm_wsubwus (v2si, v2si)
5699 v2si __builtin_arm_wsubwus (v2si, v2si)
5700 v2si __builtin_arm_wmaxuw (v2si, v2si)
5701 v2si __builtin_arm_wmaxsw (v2si, v2si)
5702 v2si __builtin_arm_wavg2br (v2si, v2si)
5703 v2si __builtin_arm_wavg2hr (v2si, v2si)
5704 v2si __builtin_arm_wavg2b (v2si, v2si)
5705 v2si __builtin_arm_wavg2h (v2si, v2si)
5706 v2si __builtin_arm_waccb (v2si)
5707 v2si __builtin_arm_wacch (v2si)
5708 v2si __builtin_arm_waccw (v2si)
5709 v2si __builtin_arm_wmacs (v2si, v2si, v2si)
5710 v2si __builtin_arm_wmacsz (v2si, v2si, v2si)
5711 v2si __builtin_arm_wmacu (v2si, v2si, v2si)
5712 v2si __builtin_arm_wmacuz (v2si, v2si)
5713 v2si __builtin_arm_wsadb (v2si, v2si)
5714 v2si __builtin_arm_wsadbz (v2si, v2si)
5715 v2si __builtin_arm_wsadh (v2si, v2si)
5716 v2si __builtin_arm_wsadhz (v2si, v2si)
5717 v2si __builtin_arm_walign (v2si, v2si)
5718 v2si __builtin_arm_tmia (v2si, int, int)
5719 v2si __builtin_arm_tmiaph (v2si, int, int)
5720 v2si __builtin_arm_tmiabb (v2si, int, int)
5721 v2si __builtin_arm_tmiabt (v2si, int, int)
5722 v2si __builtin_arm_tmiatb (v2si, int, int)
5723 v2si __builtin_arm_tmiatt (v2si, int, int)
5724 int __builtin_arm_tmovmskb (v2si)
5725 int __builtin_arm_tmovmskh (v2si)
5726 int __builtin_arm_tmovmskw (v2si)
5727 v2si __builtin_arm_wmadds (v2si, v2si)
5728 v2si __builtin_arm_wmaddu (v2si, v2si)
5729 v2si __builtin_arm_wpackhss (v2si, v2si)
5730 v2si __builtin_arm_wpackwss (v2si, v2si)
5731 v2si __builtin_arm_wpackdss (v2si, v2si)
5732 v2si __builtin_arm_wpackhus (v2si, v2si)
5733 v2si __builtin_arm_wpackwus (v2si, v2si)
5734 v2si __builtin_arm_wpackdus (v2si, v2si)
5735 v2si __builtin_arm_waddb (v2si, v2si)
5736 v2si __builtin_arm_waddh (v2si, v2si)
5737 v2si __builtin_arm_waddw (v2si, v2si)
5738 v2si __builtin_arm_waddbss (v2si, v2si)
5739 v2si __builtin_arm_waddhss (v2si, v2si)
5740 v2si __builtin_arm_waddwss (v2si, v2si)
5741 v2si __builtin_arm_waddbus (v2si, v2si)
5742 v2si __builtin_arm_waddhus (v2si, v2si)
5743 v2si __builtin_arm_waddwus (v2si, v2si)
5744 v2si __builtin_arm_wsubb (v2si, v2si)
5745 v2si __builtin_arm_wsubh (v2si, v2si)
5746 v2si __builtin_arm_wsubw (v2si, v2si)
5747 v2si __builtin_arm_wsubbss (v2si, v2si)
5748 v2si __builtin_arm_wsubhss (v2si, v2si)
5749 v2si __builtin_arm_wsubwss (v2si, v2si)
5750 v2si __builtin_arm_wsubbus (v2si, v2si)
5751 v2si __builtin_arm_wsubhus (v2si, v2si)
5752 v2si __builtin_arm_wsubwus (v2si, v2si)
5753 v2si __builtin_arm_wand (v2si, v2si)
5754 v2si __builtin_arm_wandn (v2si, v2si)
5755 v2si __builtin_arm_wor (v2si, v2si)
5756 v2si __builtin_arm_wxor (v2si, v2si)
5757 v2si __builtin_arm_wcmpeqb (v2si, v2si)
5758 v2si __builtin_arm_wcmpeqh (v2si, v2si)
5759 v2si __builtin_arm_wcmpeqw (v2si, v2si)
5760 v2si __builtin_arm_wcmpgtub (v2si, v2si)
5761 v2si __builtin_arm_wcmpgtuh (v2si, v2si)
5762 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
5763 v2si __builtin_arm_wcmpgtsb (v2si, v2si)
5764 v2si __builtin_arm_wcmpgtsh (v2si, v2si)
5765 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
5766 int __builtin_arm_textrmsb (v2si, int)
5767 int __builtin_arm_textrmsh (v2si, int)
5768 int __builtin_arm_textrmsw (v2si, int)
5769 int __builtin_arm_textrmub (v2si, int)
5770 int __builtin_arm_textrmuh (v2si, int)
5771 int __builtin_arm_textrmuw (v2si, int)
5772 v2si __builtin_arm_tinsrb (v2si, int, int)
5773 v2si __builtin_arm_tinsrh (v2si, int, int)
5774 v2si __builtin_arm_tinsrw (v2si, int, int)
5775 v2si __builtin_arm_wmaxsw (v2si, v2si)
5776 v2si __builtin_arm_wmaxsh (v2si, v2si)
5777 v2si __builtin_arm_wmaxsb (v2si, v2si)
5778 v2si __builtin_arm_wmaxuw (v2si, v2si)
5779 v2si __builtin_arm_wmaxuh (v2si, v2si)
5780 v2si __builtin_arm_wmaxub (v2si, v2si)
5781 v2si __builtin_arm_wminsw (v2si, v2si)
5782 v2si __builtin_arm_wminsh (v2si, v2si)
5783 v2si __builtin_arm_wminsb (v2si, v2si)
5784 v2si __builtin_arm_wminuw (v2si, v2si)
5785 v2si __builtin_arm_wminuh (v2si, v2si)
5786 v2si __builtin_arm_wminub (v2si, v2si)
5787 v2si __builtin_arm_wmuluh (v2si, v2si)
5788 v2si __builtin_arm_wmulsh (v2si, v2si)
5789 v2si __builtin_arm_wmulul (v2si, v2si)
5790 v2si __builtin_arm_wshufh (v2si, int)
5791 v2si __builtin_arm_wsllh (v2si, v2si)
5792 v2si __builtin_arm_wsllw (v2si, v2si)
5793 v2si __builtin_arm_wslld (v2si, v2si)
5794 v2si __builtin_arm_wsrah (v2si, v2si)
5795 v2si __builtin_arm_wsraw (v2si, v2si)
5796 v2si __builtin_arm_wsrad (v2si, v2si)
5797 v2si __builtin_arm_wsrlh (v2si, v2si)
5798 v2si __builtin_arm_wsrlw (v2si, v2si)
5799 v2si __builtin_arm_wsrld (v2si, v2si)
5800 v2si __builtin_arm_wrorh (v2si, v2si)
5801 v2si __builtin_arm_wrorw (v2si, v2si)
5802 v2si __builtin_arm_wrord (v2si, v2si)
5803 v2si __builtin_arm_wsllhi (v2si, int)
5804 v2si __builtin_arm_wsllwi (v2si, int)
5805 v2si __builtin_arm_wslldi (v2si, v2si)
5806 v2si __builtin_arm_wsrahi (v2si, int)
5807 v2si __builtin_arm_wsrawi (v2si, int)
5808 v2si __builtin_arm_wsradi (v2si, v2si)
5809 v2si __builtin_arm_wsrlwi (v2si, int)
5810 v2si __builtin_arm_wsrldi (v2si, int)
5811 v2si __builtin_arm_wrorhi (v2si, int)
5812 v2si __builtin_arm_wrorwi (v2si, int)
5813 v2si __builtin_arm_wrordi (v2si, int)
5814 v2si __builtin_arm_wunpckihb (v2si, v2si)
5815 v2si __builtin_arm_wunpckihh (v2si, v2si)
5816 v2si __builtin_arm_wunpckihw (v2si, v2si)
5817 v2si __builtin_arm_wunpckilb (v2si, v2si)
5818 v2si __builtin_arm_wunpckilh (v2si, v2si)
5819 v2si __builtin_arm_wunpckilw (v2si, v2si)
5820 v2si __builtin_arm_wunpckehsb (v2si)
5821 v2si __builtin_arm_wunpckehsh (v2si)
5822 v2si __builtin_arm_wunpckehsw (v2si)
5823 v2si __builtin_arm_wunpckehub (v2si)
5824 v2si __builtin_arm_wunpckehuh (v2si)
5825 v2si __builtin_arm_wunpckehuw (v2si)
5826 v2si __builtin_arm_wunpckelsb (v2si)
5827 v2si __builtin_arm_wunpckelsh (v2si)
5828 v2si __builtin_arm_wunpckelsw (v2si)
5829 v2si __builtin_arm_wunpckelub (v2si)
5830 v2si __builtin_arm_wunpckeluh (v2si)
5831 v2si __builtin_arm_wunpckeluw (v2si)
5832 v2si __builtin_arm_wsubwss (v2si, v2si)
5833 v2si __builtin_arm_wsraw (v2si, v2si)
5834 v2si __builtin_arm_wsrad (v2si, v2si)
5835 @end example
5836
5837 @node X86 Built-in Functions
5838 @subsection X86 Built-in Functions
5839
5840 These built-in functions are available for the i386 and x86-64 family
5841 of computers, depending on the command-line switches used.
5842
5843 The following machine modes are available for use with MMX built-in functions
5844 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
5845 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
5846 vector of eight 8-bit integers. Some of the built-in functions operate on
5847 MMX registers as a whole 64-bit entity, these use @code{DI} as their mode.
5848
5849 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
5850 of two 32-bit floating point values.
5851
5852 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
5853 floating point values. Some instructions use a vector of four 32-bit
5854 integers, these use @code{V4SI}. Finally, some instructions operate on an
5855 entire vector register, interpreting it as a 128-bit integer, these use mode
5856 @code{TI}.
5857
5858 The following built-in functions are made available by @option{-mmmx}.
5859 All of them generate the machine instruction that is part of the name.
5860
5861 @example
5862 v8qi __builtin_ia32_paddb (v8qi, v8qi)
5863 v4hi __builtin_ia32_paddw (v4hi, v4hi)
5864 v2si __builtin_ia32_paddd (v2si, v2si)
5865 v8qi __builtin_ia32_psubb (v8qi, v8qi)
5866 v4hi __builtin_ia32_psubw (v4hi, v4hi)
5867 v2si __builtin_ia32_psubd (v2si, v2si)
5868 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
5869 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
5870 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
5871 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
5872 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
5873 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
5874 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
5875 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
5876 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
5877 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
5878 di __builtin_ia32_pand (di, di)
5879 di __builtin_ia32_pandn (di,di)
5880 di __builtin_ia32_por (di, di)
5881 di __builtin_ia32_pxor (di, di)
5882 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
5883 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
5884 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
5885 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
5886 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
5887 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
5888 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
5889 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
5890 v2si __builtin_ia32_punpckhdq (v2si, v2si)
5891 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
5892 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
5893 v2si __builtin_ia32_punpckldq (v2si, v2si)
5894 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
5895 v4hi __builtin_ia32_packssdw (v2si, v2si)
5896 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
5897 @end example
5898
5899 The following built-in functions are made available either with
5900 @option{-msse}, or with a combination of @option{-m3dnow} and
5901 @option{-march=athlon}. All of them generate the machine
5902 instruction that is part of the name.
5903
5904 @example
5905 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
5906 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
5907 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
5908 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
5909 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
5910 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
5911 v8qi __builtin_ia32_pminub (v8qi, v8qi)
5912 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
5913 int __builtin_ia32_pextrw (v4hi, int)
5914 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
5915 int __builtin_ia32_pmovmskb (v8qi)
5916 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
5917 void __builtin_ia32_movntq (di *, di)
5918 void __builtin_ia32_sfence (void)
5919 @end example
5920
5921 The following built-in functions are available when @option{-msse} is used.
5922 All of them generate the machine instruction that is part of the name.
5923
5924 @example
5925 int __builtin_ia32_comieq (v4sf, v4sf)
5926 int __builtin_ia32_comineq (v4sf, v4sf)
5927 int __builtin_ia32_comilt (v4sf, v4sf)
5928 int __builtin_ia32_comile (v4sf, v4sf)
5929 int __builtin_ia32_comigt (v4sf, v4sf)
5930 int __builtin_ia32_comige (v4sf, v4sf)
5931 int __builtin_ia32_ucomieq (v4sf, v4sf)
5932 int __builtin_ia32_ucomineq (v4sf, v4sf)
5933 int __builtin_ia32_ucomilt (v4sf, v4sf)
5934 int __builtin_ia32_ucomile (v4sf, v4sf)
5935 int __builtin_ia32_ucomigt (v4sf, v4sf)
5936 int __builtin_ia32_ucomige (v4sf, v4sf)
5937 v4sf __builtin_ia32_addps (v4sf, v4sf)
5938 v4sf __builtin_ia32_subps (v4sf, v4sf)
5939 v4sf __builtin_ia32_mulps (v4sf, v4sf)
5940 v4sf __builtin_ia32_divps (v4sf, v4sf)
5941 v4sf __builtin_ia32_addss (v4sf, v4sf)
5942 v4sf __builtin_ia32_subss (v4sf, v4sf)
5943 v4sf __builtin_ia32_mulss (v4sf, v4sf)
5944 v4sf __builtin_ia32_divss (v4sf, v4sf)
5945 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
5946 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
5947 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
5948 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
5949 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
5950 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
5951 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
5952 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
5953 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
5954 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
5955 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
5956 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
5957 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
5958 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
5959 v4si __builtin_ia32_cmpless (v4sf, v4sf)
5960 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
5961 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
5962 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
5963 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
5964 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
5965 v4sf __builtin_ia32_maxps (v4sf, v4sf)
5966 v4sf __builtin_ia32_maxss (v4sf, v4sf)
5967 v4sf __builtin_ia32_minps (v4sf, v4sf)
5968 v4sf __builtin_ia32_minss (v4sf, v4sf)
5969 v4sf __builtin_ia32_andps (v4sf, v4sf)
5970 v4sf __builtin_ia32_andnps (v4sf, v4sf)
5971 v4sf __builtin_ia32_orps (v4sf, v4sf)
5972 v4sf __builtin_ia32_xorps (v4sf, v4sf)
5973 v4sf __builtin_ia32_movss (v4sf, v4sf)
5974 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
5975 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
5976 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
5977 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
5978 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
5979 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
5980 v2si __builtin_ia32_cvtps2pi (v4sf)
5981 int __builtin_ia32_cvtss2si (v4sf)
5982 v2si __builtin_ia32_cvttps2pi (v4sf)
5983 int __builtin_ia32_cvttss2si (v4sf)
5984 v4sf __builtin_ia32_rcpps (v4sf)
5985 v4sf __builtin_ia32_rsqrtps (v4sf)
5986 v4sf __builtin_ia32_sqrtps (v4sf)
5987 v4sf __builtin_ia32_rcpss (v4sf)
5988 v4sf __builtin_ia32_rsqrtss (v4sf)
5989 v4sf __builtin_ia32_sqrtss (v4sf)
5990 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
5991 void __builtin_ia32_movntps (float *, v4sf)
5992 int __builtin_ia32_movmskps (v4sf)
5993 @end example
5994
5995 The following built-in functions are available when @option{-msse} is used.
5996
5997 @table @code
5998 @item v4sf __builtin_ia32_loadaps (float *)
5999 Generates the @code{movaps} machine instruction as a load from memory.
6000 @item void __builtin_ia32_storeaps (float *, v4sf)
6001 Generates the @code{movaps} machine instruction as a store to memory.
6002 @item v4sf __builtin_ia32_loadups (float *)
6003 Generates the @code{movups} machine instruction as a load from memory.
6004 @item void __builtin_ia32_storeups (float *, v4sf)
6005 Generates the @code{movups} machine instruction as a store to memory.
6006 @item v4sf __builtin_ia32_loadsss (float *)
6007 Generates the @code{movss} machine instruction as a load from memory.
6008 @item void __builtin_ia32_storess (float *, v4sf)
6009 Generates the @code{movss} machine instruction as a store to memory.
6010 @item v4sf __builtin_ia32_loadhps (v4sf, v2si *)
6011 Generates the @code{movhps} machine instruction as a load from memory.
6012 @item v4sf __builtin_ia32_loadlps (v4sf, v2si *)
6013 Generates the @code{movlps} machine instruction as a load from memory
6014 @item void __builtin_ia32_storehps (v4sf, v2si *)
6015 Generates the @code{movhps} machine instruction as a store to memory.
6016 @item void __builtin_ia32_storelps (v4sf, v2si *)
6017 Generates the @code{movlps} machine instruction as a store to memory.
6018 @end table
6019
6020 The following built-in functions are available when @option{-mpni} is used.
6021 All of them generate the machine instruction that is part of the name.
6022
6023 @example
6024 v2df __builtin_ia32_addsubpd (v2df, v2df)
6025 v2df __builtin_ia32_addsubps (v2df, v2df)
6026 v2df __builtin_ia32_haddpd (v2df, v2df)
6027 v2df __builtin_ia32_haddps (v2df, v2df)
6028 v2df __builtin_ia32_hsubpd (v2df, v2df)
6029 v2df __builtin_ia32_hsubps (v2df, v2df)
6030 v16qi __builtin_ia32_lddqu (char const *)
6031 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
6032 v2df __builtin_ia32_movddup (v2df)
6033 v4sf __builtin_ia32_movshdup (v4sf)
6034 v4sf __builtin_ia32_movsldup (v4sf)
6035 void __builtin_ia32_mwait (unsigned int, unsigned int)
6036 @end example
6037
6038 The following built-in functions are available when @option{-mpni} is used.
6039
6040 @table @code
6041 @item v2df __builtin_ia32_loadddup (double const *)
6042 Generates the @code{movddup} machine instruction as a load from memory.
6043 @end table
6044
6045 The following built-in functions are available when @option{-m3dnow} is used.
6046 All of them generate the machine instruction that is part of the name.
6047
6048 @example
6049 void __builtin_ia32_femms (void)
6050 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
6051 v2si __builtin_ia32_pf2id (v2sf)
6052 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
6053 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
6054 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
6055 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
6056 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
6057 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
6058 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
6059 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
6060 v2sf __builtin_ia32_pfrcp (v2sf)
6061 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
6062 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
6063 v2sf __builtin_ia32_pfrsqrt (v2sf)
6064 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
6065 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
6066 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
6067 v2sf __builtin_ia32_pi2fd (v2si)
6068 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
6069 @end example
6070
6071 The following built-in functions are available when both @option{-m3dnow}
6072 and @option{-march=athlon} are used. All of them generate the machine
6073 instruction that is part of the name.
6074
6075 @example
6076 v2si __builtin_ia32_pf2iw (v2sf)
6077 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
6078 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
6079 v2sf __builtin_ia32_pi2fw (v2si)
6080 v2sf __builtin_ia32_pswapdsf (v2sf)
6081 v2si __builtin_ia32_pswapdsi (v2si)
6082 @end example
6083
6084 @node PowerPC AltiVec Built-in Functions
6085 @subsection PowerPC AltiVec Built-in Functions
6086
6087 These built-in functions are available for the PowerPC family
6088 of computers, depending on the command-line switches used.
6089
6090 The following machine modes are available for use with AltiVec built-in
6091 functions (@pxref{Vector Extensions}): @code{V4SI} for a vector of four
6092 32-bit integers, @code{V4SF} for a vector of four 32-bit floating point
6093 numbers, @code{V8HI} for a vector of eight 16-bit integers, and
6094 @code{V16QI} for a vector of sixteen 8-bit integers.
6095
6096 The following functions are made available by including
6097 @code{<altivec.h>} and using @option{-maltivec} and
6098 @option{-mabi=altivec}. The functions implement the functionality
6099 described in Motorola's AltiVec Programming Interface Manual.
6100
6101 There are a few differences from Motorola's documentation and GCC's
6102 implementation. Vector constants are done with curly braces (not
6103 parentheses). Vector initializers require no casts if the vector
6104 constant is of the same type as the variable it is initializing. The
6105 @code{vector bool} type is deprecated and will be discontinued in
6106 further revisions. Use @code{vector signed} instead. If @code{signed}
6107 or @code{unsigned} is omitted, the vector type will default to
6108 @code{signed}. Lastly, all overloaded functions are implemented with macros
6109 for the C implementation. So code the following example will not work:
6110
6111 @smallexample
6112 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
6113 @end smallexample
6114
6115 Since vec_add is a macro, the vector constant in the above example will
6116 be treated as four different arguments. Wrap the entire argument in
6117 parentheses for this to work. The C++ implementation does not use
6118 macros.
6119
6120 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
6121 Internally, GCC uses built-in functions to achieve the functionality in
6122 the aforementioned header file, but they are not supported and are
6123 subject to change without notice.
6124
6125 @smallexample
6126 vector signed char vec_abs (vector signed char, vector signed char);
6127 vector signed short vec_abs (vector signed short, vector signed short);
6128 vector signed int vec_abs (vector signed int, vector signed int);
6129 vector signed float vec_abs (vector signed float, vector signed float);
6130
6131 vector signed char vec_abss (vector signed char, vector signed char);
6132 vector signed short vec_abss (vector signed short, vector signed short);
6133
6134 vector signed char vec_add (vector signed char, vector signed char);
6135 vector unsigned char vec_add (vector signed char, vector unsigned char);
6136
6137 vector unsigned char vec_add (vector unsigned char, vector signed char);
6138
6139 vector unsigned char vec_add (vector unsigned char,
6140 vector unsigned char);
6141 vector signed short vec_add (vector signed short, vector signed short);
6142 vector unsigned short vec_add (vector signed short,
6143 vector unsigned short);
6144 vector unsigned short vec_add (vector unsigned short,
6145 vector signed short);
6146 vector unsigned short vec_add (vector unsigned short,
6147 vector unsigned short);
6148 vector signed int vec_add (vector signed int, vector signed int);
6149 vector unsigned int vec_add (vector signed int, vector unsigned int);
6150 vector unsigned int vec_add (vector unsigned int, vector signed int);
6151 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
6152 vector float vec_add (vector float, vector float);
6153
6154 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
6155
6156 vector unsigned char vec_adds (vector signed char,
6157 vector unsigned char);
6158 vector unsigned char vec_adds (vector unsigned char,
6159 vector signed char);
6160 vector unsigned char vec_adds (vector unsigned char,
6161 vector unsigned char);
6162 vector signed char vec_adds (vector signed char, vector signed char);
6163 vector unsigned short vec_adds (vector signed short,
6164 vector unsigned short);
6165 vector unsigned short vec_adds (vector unsigned short,
6166 vector signed short);
6167 vector unsigned short vec_adds (vector unsigned short,
6168 vector unsigned short);
6169 vector signed short vec_adds (vector signed short, vector signed short);
6170
6171 vector unsigned int vec_adds (vector signed int, vector unsigned int);
6172 vector unsigned int vec_adds (vector unsigned int, vector signed int);
6173 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
6174
6175 vector signed int vec_adds (vector signed int, vector signed int);
6176
6177 vector float vec_and (vector float, vector float);
6178 vector float vec_and (vector float, vector signed int);
6179 vector float vec_and (vector signed int, vector float);
6180 vector signed int vec_and (vector signed int, vector signed int);
6181 vector unsigned int vec_and (vector signed int, vector unsigned int);
6182 vector unsigned int vec_and (vector unsigned int, vector signed int);
6183 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
6184 vector signed short vec_and (vector signed short, vector signed short);
6185 vector unsigned short vec_and (vector signed short,
6186 vector unsigned short);
6187 vector unsigned short vec_and (vector unsigned short,
6188 vector signed short);
6189 vector unsigned short vec_and (vector unsigned short,
6190 vector unsigned short);
6191 vector signed char vec_and (vector signed char, vector signed char);
6192 vector unsigned char vec_and (vector signed char, vector unsigned char);
6193
6194 vector unsigned char vec_and (vector unsigned char, vector signed char);
6195
6196 vector unsigned char vec_and (vector unsigned char,
6197 vector unsigned char);
6198
6199 vector float vec_andc (vector float, vector float);
6200 vector float vec_andc (vector float, vector signed int);
6201 vector float vec_andc (vector signed int, vector float);
6202 vector signed int vec_andc (vector signed int, vector signed int);
6203 vector unsigned int vec_andc (vector signed int, vector unsigned int);
6204 vector unsigned int vec_andc (vector unsigned int, vector signed int);
6205 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
6206
6207 vector signed short vec_andc (vector signed short, vector signed short);
6208
6209 vector unsigned short vec_andc (vector signed short,
6210 vector unsigned short);
6211 vector unsigned short vec_andc (vector unsigned short,
6212 vector signed short);
6213 vector unsigned short vec_andc (vector unsigned short,
6214 vector unsigned short);
6215 vector signed char vec_andc (vector signed char, vector signed char);
6216 vector unsigned char vec_andc (vector signed char,
6217 vector unsigned char);
6218 vector unsigned char vec_andc (vector unsigned char,
6219 vector signed char);
6220 vector unsigned char vec_andc (vector unsigned char,
6221 vector unsigned char);
6222
6223 vector unsigned char vec_avg (vector unsigned char,
6224 vector unsigned char);
6225 vector signed char vec_avg (vector signed char, vector signed char);
6226 vector unsigned short vec_avg (vector unsigned short,
6227 vector unsigned short);
6228 vector signed short vec_avg (vector signed short, vector signed short);
6229 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
6230 vector signed int vec_avg (vector signed int, vector signed int);
6231
6232 vector float vec_ceil (vector float);
6233
6234 vector signed int vec_cmpb (vector float, vector float);
6235
6236 vector signed char vec_cmpeq (vector signed char, vector signed char);
6237 vector signed char vec_cmpeq (vector unsigned char,
6238 vector unsigned char);
6239 vector signed short vec_cmpeq (vector signed short,
6240 vector signed short);
6241 vector signed short vec_cmpeq (vector unsigned short,
6242 vector unsigned short);
6243 vector signed int vec_cmpeq (vector signed int, vector signed int);
6244 vector signed int vec_cmpeq (vector unsigned int, vector unsigned int);
6245 vector signed int vec_cmpeq (vector float, vector float);
6246
6247 vector signed int vec_cmpge (vector float, vector float);
6248
6249 vector signed char vec_cmpgt (vector unsigned char,
6250 vector unsigned char);
6251 vector signed char vec_cmpgt (vector signed char, vector signed char);
6252 vector signed short vec_cmpgt (vector unsigned short,
6253 vector unsigned short);
6254 vector signed short vec_cmpgt (vector signed short,
6255 vector signed short);
6256 vector signed int vec_cmpgt (vector unsigned int, vector unsigned int);
6257 vector signed int vec_cmpgt (vector signed int, vector signed int);
6258 vector signed int vec_cmpgt (vector float, vector float);
6259
6260 vector signed int vec_cmple (vector float, vector float);
6261
6262 vector signed char vec_cmplt (vector unsigned char,
6263 vector unsigned char);
6264 vector signed char vec_cmplt (vector signed char, vector signed char);
6265 vector signed short vec_cmplt (vector unsigned short,
6266 vector unsigned short);
6267 vector signed short vec_cmplt (vector signed short,
6268 vector signed short);
6269 vector signed int vec_cmplt (vector unsigned int, vector unsigned int);
6270 vector signed int vec_cmplt (vector signed int, vector signed int);
6271 vector signed int vec_cmplt (vector float, vector float);
6272
6273 vector float vec_ctf (vector unsigned int, const char);
6274 vector float vec_ctf (vector signed int, const char);
6275
6276 vector signed int vec_cts (vector float, const char);
6277
6278 vector unsigned int vec_ctu (vector float, const char);
6279
6280 void vec_dss (const char);
6281
6282 void vec_dssall (void);
6283
6284 void vec_dst (void *, int, const char);
6285
6286 void vec_dstst (void *, int, const char);
6287
6288 void vec_dststt (void *, int, const char);
6289
6290 void vec_dstt (void *, int, const char);
6291
6292 vector float vec_expte (vector float, vector float);
6293
6294 vector float vec_floor (vector float, vector float);
6295
6296 vector float vec_ld (int, vector float *);
6297 vector float vec_ld (int, float *):
6298 vector signed int vec_ld (int, int *);
6299 vector signed int vec_ld (int, vector signed int *);
6300 vector unsigned int vec_ld (int, vector unsigned int *);
6301 vector unsigned int vec_ld (int, unsigned int *);
6302 vector signed short vec_ld (int, short *, vector signed short *);
6303 vector unsigned short vec_ld (int, unsigned short *,
6304 vector unsigned short *);
6305 vector signed char vec_ld (int, signed char *);
6306 vector signed char vec_ld (int, vector signed char *);
6307 vector unsigned char vec_ld (int, unsigned char *);
6308 vector unsigned char vec_ld (int, vector unsigned char *);
6309
6310 vector signed char vec_lde (int, signed char *);
6311 vector unsigned char vec_lde (int, unsigned char *);
6312 vector signed short vec_lde (int, short *);
6313 vector unsigned short vec_lde (int, unsigned short *);
6314 vector float vec_lde (int, float *);
6315 vector signed int vec_lde (int, int *);
6316 vector unsigned int vec_lde (int, unsigned int *);
6317
6318 void float vec_ldl (int, float *);
6319 void float vec_ldl (int, vector float *);
6320 void signed int vec_ldl (int, vector signed int *);
6321 void signed int vec_ldl (int, int *);
6322 void unsigned int vec_ldl (int, unsigned int *);
6323 void unsigned int vec_ldl (int, vector unsigned int *);
6324 void signed short vec_ldl (int, vector signed short *);
6325 void signed short vec_ldl (int, short *);
6326 void unsigned short vec_ldl (int, vector unsigned short *);
6327 void unsigned short vec_ldl (int, unsigned short *);
6328 void signed char vec_ldl (int, vector signed char *);
6329 void signed char vec_ldl (int, signed char *);
6330 void unsigned char vec_ldl (int, vector unsigned char *);
6331 void unsigned char vec_ldl (int, unsigned char *);
6332
6333 vector float vec_loge (vector float);
6334
6335 vector unsigned char vec_lvsl (int, void *, int *);
6336
6337 vector unsigned char vec_lvsr (int, void *, int *);
6338
6339 vector float vec_madd (vector float, vector float, vector float);
6340
6341 vector signed short vec_madds (vector signed short, vector signed short,
6342 vector signed short);
6343
6344 vector unsigned char vec_max (vector signed char, vector unsigned char);
6345
6346 vector unsigned char vec_max (vector unsigned char, vector signed char);
6347
6348 vector unsigned char vec_max (vector unsigned char,
6349 vector unsigned char);
6350 vector signed char vec_max (vector signed char, vector signed char);
6351 vector unsigned short vec_max (vector signed short,
6352 vector unsigned short);
6353 vector unsigned short vec_max (vector unsigned short,
6354 vector signed short);
6355 vector unsigned short vec_max (vector unsigned short,
6356 vector unsigned short);
6357 vector signed short vec_max (vector signed short, vector signed short);
6358 vector unsigned int vec_max (vector signed int, vector unsigned int);
6359 vector unsigned int vec_max (vector unsigned int, vector signed int);
6360 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
6361 vector signed int vec_max (vector signed int, vector signed int);
6362 vector float vec_max (vector float, vector float);
6363
6364 vector signed char vec_mergeh (vector signed char, vector signed char);
6365 vector unsigned char vec_mergeh (vector unsigned char,
6366 vector unsigned char);
6367 vector signed short vec_mergeh (vector signed short,
6368 vector signed short);
6369 vector unsigned short vec_mergeh (vector unsigned short,
6370 vector unsigned short);
6371 vector float vec_mergeh (vector float, vector float);
6372 vector signed int vec_mergeh (vector signed int, vector signed int);
6373 vector unsigned int vec_mergeh (vector unsigned int,
6374 vector unsigned int);
6375
6376 vector signed char vec_mergel (vector signed char, vector signed char);
6377 vector unsigned char vec_mergel (vector unsigned char,
6378 vector unsigned char);
6379 vector signed short vec_mergel (vector signed short,
6380 vector signed short);
6381 vector unsigned short vec_mergel (vector unsigned short,
6382 vector unsigned short);
6383 vector float vec_mergel (vector float, vector float);
6384 vector signed int vec_mergel (vector signed int, vector signed int);
6385 vector unsigned int vec_mergel (vector unsigned int,
6386 vector unsigned int);
6387
6388 vector unsigned short vec_mfvscr (void);
6389
6390 vector unsigned char vec_min (vector signed char, vector unsigned char);
6391
6392 vector unsigned char vec_min (vector unsigned char, vector signed char);
6393
6394 vector unsigned char vec_min (vector unsigned char,
6395 vector unsigned char);
6396 vector signed char vec_min (vector signed char, vector signed char);
6397 vector unsigned short vec_min (vector signed short,
6398 vector unsigned short);
6399 vector unsigned short vec_min (vector unsigned short,
6400 vector signed short);
6401 vector unsigned short vec_min (vector unsigned short,
6402 vector unsigned short);
6403 vector signed short vec_min (vector signed short, vector signed short);
6404 vector unsigned int vec_min (vector signed int, vector unsigned int);
6405 vector unsigned int vec_min (vector unsigned int, vector signed int);
6406 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
6407 vector signed int vec_min (vector signed int, vector signed int);
6408 vector float vec_min (vector float, vector float);
6409
6410 vector signed short vec_mladd (vector signed short, vector signed short,
6411 vector signed short);
6412 vector signed short vec_mladd (vector signed short,
6413 vector unsigned short,
6414 vector unsigned short);
6415 vector signed short vec_mladd (vector unsigned short,
6416 vector signed short,
6417 vector signed short);
6418 vector unsigned short vec_mladd (vector unsigned short,
6419 vector unsigned short,
6420 vector unsigned short);
6421
6422 vector signed short vec_mradds (vector signed short,
6423 vector signed short,
6424 vector signed short);
6425
6426 vector unsigned int vec_msum (vector unsigned char,
6427 vector unsigned char,
6428 vector unsigned int);
6429 vector signed int vec_msum (vector signed char, vector unsigned char,
6430 vector signed int);
6431 vector unsigned int vec_msum (vector unsigned short,
6432 vector unsigned short,
6433 vector unsigned int);
6434 vector signed int vec_msum (vector signed short, vector signed short,
6435 vector signed int);
6436
6437 vector unsigned int vec_msums (vector unsigned short,
6438 vector unsigned short,
6439 vector unsigned int);
6440 vector signed int vec_msums (vector signed short, vector signed short,
6441 vector signed int);
6442
6443 void vec_mtvscr (vector signed int);
6444 void vec_mtvscr (vector unsigned int);
6445 void vec_mtvscr (vector signed short);
6446 void vec_mtvscr (vector unsigned short);
6447 void vec_mtvscr (vector signed char);
6448 void vec_mtvscr (vector unsigned char);
6449
6450 vector unsigned short vec_mule (vector unsigned char,
6451 vector unsigned char);
6452 vector signed short vec_mule (vector signed char, vector signed char);
6453 vector unsigned int vec_mule (vector unsigned short,
6454 vector unsigned short);
6455 vector signed int vec_mule (vector signed short, vector signed short);
6456
6457 vector unsigned short vec_mulo (vector unsigned char,
6458 vector unsigned char);
6459 vector signed short vec_mulo (vector signed char, vector signed char);
6460 vector unsigned int vec_mulo (vector unsigned short,
6461 vector unsigned short);
6462 vector signed int vec_mulo (vector signed short, vector signed short);
6463
6464 vector float vec_nmsub (vector float, vector float, vector float);
6465
6466 vector float vec_nor (vector float, vector float);
6467 vector signed int vec_nor (vector signed int, vector signed int);
6468 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
6469 vector signed short vec_nor (vector signed short, vector signed short);
6470 vector unsigned short vec_nor (vector unsigned short,
6471 vector unsigned short);
6472 vector signed char vec_nor (vector signed char, vector signed char);
6473 vector unsigned char vec_nor (vector unsigned char,
6474 vector unsigned char);
6475
6476 vector float vec_or (vector float, vector float);
6477 vector float vec_or (vector float, vector signed int);
6478 vector float vec_or (vector signed int, vector float);
6479 vector signed int vec_or (vector signed int, vector signed int);
6480 vector unsigned int vec_or (vector signed int, vector unsigned int);
6481 vector unsigned int vec_or (vector unsigned int, vector signed int);
6482 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
6483 vector signed short vec_or (vector signed short, vector signed short);
6484 vector unsigned short vec_or (vector signed short,
6485 vector unsigned short);
6486 vector unsigned short vec_or (vector unsigned short,
6487 vector signed short);
6488 vector unsigned short vec_or (vector unsigned short,
6489 vector unsigned short);
6490 vector signed char vec_or (vector signed char, vector signed char);
6491 vector unsigned char vec_or (vector signed char, vector unsigned char);
6492 vector unsigned char vec_or (vector unsigned char, vector signed char);
6493 vector unsigned char vec_or (vector unsigned char,
6494 vector unsigned char);
6495
6496 vector signed char vec_pack (vector signed short, vector signed short);
6497 vector unsigned char vec_pack (vector unsigned short,
6498 vector unsigned short);
6499 vector signed short vec_pack (vector signed int, vector signed int);
6500 vector unsigned short vec_pack (vector unsigned int,
6501 vector unsigned int);
6502
6503 vector signed short vec_packpx (vector unsigned int,
6504 vector unsigned int);
6505
6506 vector unsigned char vec_packs (vector unsigned short,
6507 vector unsigned short);
6508 vector signed char vec_packs (vector signed short, vector signed short);
6509
6510 vector unsigned short vec_packs (vector unsigned int,
6511 vector unsigned int);
6512 vector signed short vec_packs (vector signed int, vector signed int);
6513
6514 vector unsigned char vec_packsu (vector unsigned short,
6515 vector unsigned short);
6516 vector unsigned char vec_packsu (vector signed short,
6517 vector signed short);
6518 vector unsigned short vec_packsu (vector unsigned int,
6519 vector unsigned int);
6520 vector unsigned short vec_packsu (vector signed int, vector signed int);
6521
6522 vector float vec_perm (vector float, vector float,
6523 vector unsigned char);
6524 vector signed int vec_perm (vector signed int, vector signed int,
6525 vector unsigned char);
6526 vector unsigned int vec_perm (vector unsigned int, vector unsigned int,
6527 vector unsigned char);
6528 vector signed short vec_perm (vector signed short, vector signed short,
6529 vector unsigned char);
6530 vector unsigned short vec_perm (vector unsigned short,
6531 vector unsigned short,
6532 vector unsigned char);
6533 vector signed char vec_perm (vector signed char, vector signed char,
6534 vector unsigned char);
6535 vector unsigned char vec_perm (vector unsigned char,
6536 vector unsigned char,
6537 vector unsigned char);
6538
6539 vector float vec_re (vector float);
6540
6541 vector signed char vec_rl (vector signed char, vector unsigned char);
6542 vector unsigned char vec_rl (vector unsigned char,
6543 vector unsigned char);
6544 vector signed short vec_rl (vector signed short, vector unsigned short);
6545
6546 vector unsigned short vec_rl (vector unsigned short,
6547 vector unsigned short);
6548 vector signed int vec_rl (vector signed int, vector unsigned int);
6549 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
6550
6551 vector float vec_round (vector float);
6552
6553 vector float vec_rsqrte (vector float);
6554
6555 vector float vec_sel (vector float, vector float, vector signed int);
6556 vector float vec_sel (vector float, vector float, vector unsigned int);
6557 vector signed int vec_sel (vector signed int, vector signed int,
6558 vector signed int);
6559 vector signed int vec_sel (vector signed int, vector signed int,
6560 vector unsigned int);
6561 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
6562 vector signed int);
6563 vector unsigned int vec_sel (vector unsigned int, vector unsigned int,
6564 vector unsigned int);
6565 vector signed short vec_sel (vector signed short, vector signed short,
6566 vector signed short);
6567 vector signed short vec_sel (vector signed short, vector signed short,
6568 vector unsigned short);
6569 vector unsigned short vec_sel (vector unsigned short,
6570 vector unsigned short,
6571 vector signed short);
6572 vector unsigned short vec_sel (vector unsigned short,
6573 vector unsigned short,
6574 vector unsigned short);
6575 vector signed char vec_sel (vector signed char, vector signed char,
6576 vector signed char);
6577 vector signed char vec_sel (vector signed char, vector signed char,
6578 vector unsigned char);
6579 vector unsigned char vec_sel (vector unsigned char,
6580 vector unsigned char,
6581 vector signed char);
6582 vector unsigned char vec_sel (vector unsigned char,
6583 vector unsigned char,
6584 vector unsigned char);
6585
6586 vector signed char vec_sl (vector signed char, vector unsigned char);
6587 vector unsigned char vec_sl (vector unsigned char,
6588 vector unsigned char);
6589 vector signed short vec_sl (vector signed short, vector unsigned short);
6590
6591 vector unsigned short vec_sl (vector unsigned short,
6592 vector unsigned short);
6593 vector signed int vec_sl (vector signed int, vector unsigned int);
6594 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
6595
6596 vector float vec_sld (vector float, vector float, const char);
6597 vector signed int vec_sld (vector signed int, vector signed int,
6598 const char);
6599 vector unsigned int vec_sld (vector unsigned int, vector unsigned int,
6600 const char);
6601 vector signed short vec_sld (vector signed short, vector signed short,
6602 const char);
6603 vector unsigned short vec_sld (vector unsigned short,
6604 vector unsigned short, const char);
6605 vector signed char vec_sld (vector signed char, vector signed char,
6606 const char);
6607 vector unsigned char vec_sld (vector unsigned char,
6608 vector unsigned char,
6609 const char);
6610
6611 vector signed int vec_sll (vector signed int, vector unsigned int);
6612 vector signed int vec_sll (vector signed int, vector unsigned short);
6613 vector signed int vec_sll (vector signed int, vector unsigned char);
6614 vector unsigned int vec_sll (vector unsigned int, vector unsigned int);
6615 vector unsigned int vec_sll (vector unsigned int,
6616 vector unsigned short);
6617 vector unsigned int vec_sll (vector unsigned int, vector unsigned char);
6618
6619 vector signed short vec_sll (vector signed short, vector unsigned int);
6620 vector signed short vec_sll (vector signed short,
6621 vector unsigned short);
6622 vector signed short vec_sll (vector signed short, vector unsigned char);
6623
6624 vector unsigned short vec_sll (vector unsigned short,
6625 vector unsigned int);
6626 vector unsigned short vec_sll (vector unsigned short,
6627 vector unsigned short);
6628 vector unsigned short vec_sll (vector unsigned short,
6629 vector unsigned char);
6630 vector signed char vec_sll (vector signed char, vector unsigned int);
6631 vector signed char vec_sll (vector signed char, vector unsigned short);
6632 vector signed char vec_sll (vector signed char, vector unsigned char);
6633 vector unsigned char vec_sll (vector unsigned char,
6634 vector unsigned int);
6635 vector unsigned char vec_sll (vector unsigned char,
6636 vector unsigned short);
6637 vector unsigned char vec_sll (vector unsigned char,
6638 vector unsigned char);
6639
6640 vector float vec_slo (vector float, vector signed char);
6641 vector float vec_slo (vector float, vector unsigned char);
6642 vector signed int vec_slo (vector signed int, vector signed char);
6643 vector signed int vec_slo (vector signed int, vector unsigned char);
6644 vector unsigned int vec_slo (vector unsigned int, vector signed char);
6645 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
6646
6647 vector signed short vec_slo (vector signed short, vector signed char);
6648 vector signed short vec_slo (vector signed short, vector unsigned char);
6649
6650 vector unsigned short vec_slo (vector unsigned short,
6651 vector signed char);
6652 vector unsigned short vec_slo (vector unsigned short,
6653 vector unsigned char);
6654 vector signed char vec_slo (vector signed char, vector signed char);
6655 vector signed char vec_slo (vector signed char, vector unsigned char);
6656 vector unsigned char vec_slo (vector unsigned char, vector signed char);
6657
6658 vector unsigned char vec_slo (vector unsigned char,
6659 vector unsigned char);
6660
6661 vector signed char vec_splat (vector signed char, const char);
6662 vector unsigned char vec_splat (vector unsigned char, const char);
6663 vector signed short vec_splat (vector signed short, const char);
6664 vector unsigned short vec_splat (vector unsigned short, const char);
6665 vector float vec_splat (vector float, const char);
6666 vector signed int vec_splat (vector signed int, const char);
6667 vector unsigned int vec_splat (vector unsigned int, const char);
6668
6669 vector signed char vec_splat_s8 (const char);
6670
6671 vector signed short vec_splat_s16 (const char);
6672
6673 vector signed int vec_splat_s32 (const char);
6674
6675 vector unsigned char vec_splat_u8 (const char);
6676
6677 vector unsigned short vec_splat_u16 (const char);
6678
6679 vector unsigned int vec_splat_u32 (const char);
6680
6681 vector signed char vec_sr (vector signed char, vector unsigned char);
6682 vector unsigned char vec_sr (vector unsigned char,
6683 vector unsigned char);
6684 vector signed short vec_sr (vector signed short, vector unsigned short);
6685
6686 vector unsigned short vec_sr (vector unsigned short,
6687 vector unsigned short);
6688 vector signed int vec_sr (vector signed int, vector unsigned int);
6689 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
6690
6691 vector signed char vec_sra (vector signed char, vector unsigned char);
6692 vector unsigned char vec_sra (vector unsigned char,
6693 vector unsigned char);
6694 vector signed short vec_sra (vector signed short,
6695 vector unsigned short);
6696 vector unsigned short vec_sra (vector unsigned short,
6697 vector unsigned short);
6698 vector signed int vec_sra (vector signed int, vector unsigned int);
6699 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
6700
6701 vector signed int vec_srl (vector signed int, vector unsigned int);
6702 vector signed int vec_srl (vector signed int, vector unsigned short);
6703 vector signed int vec_srl (vector signed int, vector unsigned char);
6704 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
6705 vector unsigned int vec_srl (vector unsigned int,
6706 vector unsigned short);
6707 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
6708
6709 vector signed short vec_srl (vector signed short, vector unsigned int);
6710 vector signed short vec_srl (vector signed short,
6711 vector unsigned short);
6712 vector signed short vec_srl (vector signed short, vector unsigned char);
6713
6714 vector unsigned short vec_srl (vector unsigned short,
6715 vector unsigned int);
6716 vector unsigned short vec_srl (vector unsigned short,
6717 vector unsigned short);
6718 vector unsigned short vec_srl (vector unsigned short,
6719 vector unsigned char);
6720 vector signed char vec_srl (vector signed char, vector unsigned int);
6721 vector signed char vec_srl (vector signed char, vector unsigned short);
6722 vector signed char vec_srl (vector signed char, vector unsigned char);
6723 vector unsigned char vec_srl (vector unsigned char,
6724 vector unsigned int);
6725 vector unsigned char vec_srl (vector unsigned char,
6726 vector unsigned short);
6727 vector unsigned char vec_srl (vector unsigned char,
6728 vector unsigned char);
6729
6730 vector float vec_sro (vector float, vector signed char);
6731 vector float vec_sro (vector float, vector unsigned char);
6732 vector signed int vec_sro (vector signed int, vector signed char);
6733 vector signed int vec_sro (vector signed int, vector unsigned char);
6734 vector unsigned int vec_sro (vector unsigned int, vector signed char);
6735 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
6736
6737 vector signed short vec_sro (vector signed short, vector signed char);
6738 vector signed short vec_sro (vector signed short, vector unsigned char);
6739
6740 vector unsigned short vec_sro (vector unsigned short,
6741 vector signed char);
6742 vector unsigned short vec_sro (vector unsigned short,
6743 vector unsigned char);
6744 vector signed char vec_sro (vector signed char, vector signed char);
6745 vector signed char vec_sro (vector signed char, vector unsigned char);
6746 vector unsigned char vec_sro (vector unsigned char, vector signed char);
6747
6748 vector unsigned char vec_sro (vector unsigned char,
6749 vector unsigned char);
6750
6751 void vec_st (vector float, int, float *);
6752 void vec_st (vector float, int, vector float *);
6753 void vec_st (vector signed int, int, int *);
6754 void vec_st (vector signed int, int, unsigned int *);
6755 void vec_st (vector unsigned int, int, unsigned int *);
6756 void vec_st (vector unsigned int, int, vector unsigned int *);
6757 void vec_st (vector signed short, int, short *);
6758 void vec_st (vector signed short, int, vector unsigned short *);
6759 void vec_st (vector signed short, int, vector signed short *);
6760 void vec_st (vector unsigned short, int, unsigned short *);
6761 void vec_st (vector unsigned short, int, vector unsigned short *);
6762 void vec_st (vector signed char, int, signed char *);
6763 void vec_st (vector signed char, int, unsigned char *);
6764 void vec_st (vector signed char, int, vector signed char *);
6765 void vec_st (vector unsigned char, int, unsigned char *);
6766 void vec_st (vector unsigned char, int, vector unsigned char *);
6767
6768 void vec_ste (vector signed char, int, unsigned char *);
6769 void vec_ste (vector signed char, int, signed char *);
6770 void vec_ste (vector unsigned char, int, unsigned char *);
6771 void vec_ste (vector signed short, int, short *);
6772 void vec_ste (vector signed short, int, unsigned short *);
6773 void vec_ste (vector unsigned short, int, void *);
6774 void vec_ste (vector signed int, int, unsigned int *);
6775 void vec_ste (vector signed int, int, int *);
6776 void vec_ste (vector unsigned int, int, unsigned int *);
6777 void vec_ste (vector float, int, float *);
6778
6779 void vec_stl (vector float, int, vector float *);
6780 void vec_stl (vector float, int, float *);
6781 void vec_stl (vector signed int, int, vector signed int *);
6782 void vec_stl (vector signed int, int, int *);
6783 void vec_stl (vector signed int, int, unsigned int *);
6784 void vec_stl (vector unsigned int, int, vector unsigned int *);
6785 void vec_stl (vector unsigned int, int, unsigned int *);
6786 void vec_stl (vector signed short, int, short *);
6787 void vec_stl (vector signed short, int, unsigned short *);
6788 void vec_stl (vector signed short, int, vector signed short *);
6789 void vec_stl (vector unsigned short, int, unsigned short *);
6790 void vec_stl (vector unsigned short, int, vector signed short *);
6791 void vec_stl (vector signed char, int, signed char *);
6792 void vec_stl (vector signed char, int, unsigned char *);
6793 void vec_stl (vector signed char, int, vector signed char *);
6794 void vec_stl (vector unsigned char, int, unsigned char *);
6795 void vec_stl (vector unsigned char, int, vector unsigned char *);
6796
6797 vector signed char vec_sub (vector signed char, vector signed char);
6798 vector unsigned char vec_sub (vector signed char, vector unsigned char);
6799
6800 vector unsigned char vec_sub (vector unsigned char, vector signed char);
6801
6802 vector unsigned char vec_sub (vector unsigned char,
6803 vector unsigned char);
6804 vector signed short vec_sub (vector signed short, vector signed short);
6805 vector unsigned short vec_sub (vector signed short,
6806 vector unsigned short);
6807 vector unsigned short vec_sub (vector unsigned short,
6808 vector signed short);
6809 vector unsigned short vec_sub (vector unsigned short,
6810 vector unsigned short);
6811 vector signed int vec_sub (vector signed int, vector signed int);
6812 vector unsigned int vec_sub (vector signed int, vector unsigned int);
6813 vector unsigned int vec_sub (vector unsigned int, vector signed int);
6814 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
6815 vector float vec_sub (vector float, vector float);
6816
6817 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
6818
6819 vector unsigned char vec_subs (vector signed char,
6820 vector unsigned char);
6821 vector unsigned char vec_subs (vector unsigned char,
6822 vector signed char);
6823 vector unsigned char vec_subs (vector unsigned char,
6824 vector unsigned char);
6825 vector signed char vec_subs (vector signed char, vector signed char);
6826 vector unsigned short vec_subs (vector signed short,
6827 vector unsigned short);
6828 vector unsigned short vec_subs (vector unsigned short,
6829 vector signed short);
6830 vector unsigned short vec_subs (vector unsigned short,
6831 vector unsigned short);
6832 vector signed short vec_subs (vector signed short, vector signed short);
6833
6834 vector unsigned int vec_subs (vector signed int, vector unsigned int);
6835 vector unsigned int vec_subs (vector unsigned int, vector signed int);
6836 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
6837
6838 vector signed int vec_subs (vector signed int, vector signed int);
6839
6840 vector unsigned int vec_sum4s (vector unsigned char,
6841 vector unsigned int);
6842 vector signed int vec_sum4s (vector signed char, vector signed int);
6843 vector signed int vec_sum4s (vector signed short, vector signed int);
6844
6845 vector signed int vec_sum2s (vector signed int, vector signed int);
6846
6847 vector signed int vec_sums (vector signed int, vector signed int);
6848
6849 vector float vec_trunc (vector float);
6850
6851 vector signed short vec_unpackh (vector signed char);
6852 vector unsigned int vec_unpackh (vector signed short);
6853 vector signed int vec_unpackh (vector signed short);
6854
6855 vector signed short vec_unpackl (vector signed char);
6856 vector unsigned int vec_unpackl (vector signed short);
6857 vector signed int vec_unpackl (vector signed short);
6858
6859 vector float vec_xor (vector float, vector float);
6860 vector float vec_xor (vector float, vector signed int);
6861 vector float vec_xor (vector signed int, vector float);
6862 vector signed int vec_xor (vector signed int, vector signed int);
6863 vector unsigned int vec_xor (vector signed int, vector unsigned int);
6864 vector unsigned int vec_xor (vector unsigned int, vector signed int);
6865 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
6866 vector signed short vec_xor (vector signed short, vector signed short);
6867 vector unsigned short vec_xor (vector signed short,
6868 vector unsigned short);
6869 vector unsigned short vec_xor (vector unsigned short,
6870 vector signed short);
6871 vector unsigned short vec_xor (vector unsigned short,
6872 vector unsigned short);
6873 vector signed char vec_xor (vector signed char, vector signed char);
6874 vector unsigned char vec_xor (vector signed char, vector unsigned char);
6875
6876 vector unsigned char vec_xor (vector unsigned char, vector signed char);
6877
6878 vector unsigned char vec_xor (vector unsigned char,
6879 vector unsigned char);
6880
6881 vector signed int vec_all_eq (vector signed char, vector unsigned char);
6882
6883 vector signed int vec_all_eq (vector signed char, vector signed char);
6884 vector signed int vec_all_eq (vector unsigned char, vector signed char);
6885
6886 vector signed int vec_all_eq (vector unsigned char,
6887 vector unsigned char);
6888 vector signed int vec_all_eq (vector signed short,
6889 vector unsigned short);
6890 vector signed int vec_all_eq (vector signed short, vector signed short);
6891
6892 vector signed int vec_all_eq (vector unsigned short,
6893 vector signed short);
6894 vector signed int vec_all_eq (vector unsigned short,
6895 vector unsigned short);
6896 vector signed int vec_all_eq (vector signed int, vector unsigned int);
6897 vector signed int vec_all_eq (vector signed int, vector signed int);
6898 vector signed int vec_all_eq (vector unsigned int, vector signed int);
6899 vector signed int vec_all_eq (vector unsigned int, vector unsigned int);
6900
6901 vector signed int vec_all_eq (vector float, vector float);
6902
6903 vector signed int vec_all_ge (vector signed char, vector unsigned char);
6904
6905 vector signed int vec_all_ge (vector unsigned char, vector signed char);
6906
6907 vector signed int vec_all_ge (vector unsigned char,
6908 vector unsigned char);
6909 vector signed int vec_all_ge (vector signed char, vector signed char);
6910 vector signed int vec_all_ge (vector signed short,
6911 vector unsigned short);
6912 vector signed int vec_all_ge (vector unsigned short,
6913 vector signed short);
6914 vector signed int vec_all_ge (vector unsigned short,
6915 vector unsigned short);
6916 vector signed int vec_all_ge (vector signed short, vector signed short);
6917
6918 vector signed int vec_all_ge (vector signed int, vector unsigned int);
6919 vector signed int vec_all_ge (vector unsigned int, vector signed int);
6920 vector signed int vec_all_ge (vector unsigned int, vector unsigned int);
6921
6922 vector signed int vec_all_ge (vector signed int, vector signed int);
6923 vector signed int vec_all_ge (vector float, vector float);
6924
6925 vector signed int vec_all_gt (vector signed char, vector unsigned char);
6926
6927 vector signed int vec_all_gt (vector unsigned char, vector signed char);
6928
6929 vector signed int vec_all_gt (vector unsigned char,
6930 vector unsigned char);
6931 vector signed int vec_all_gt (vector signed char, vector signed char);
6932 vector signed int vec_all_gt (vector signed short,
6933 vector unsigned short);
6934 vector signed int vec_all_gt (vector unsigned short,
6935 vector signed short);
6936 vector signed int vec_all_gt (vector unsigned short,
6937 vector unsigned short);
6938 vector signed int vec_all_gt (vector signed short, vector signed short);
6939
6940 vector signed int vec_all_gt (vector signed int, vector unsigned int);
6941 vector signed int vec_all_gt (vector unsigned int, vector signed int);
6942 vector signed int vec_all_gt (vector unsigned int, vector unsigned int);
6943
6944 vector signed int vec_all_gt (vector signed int, vector signed int);
6945 vector signed int vec_all_gt (vector float, vector float);
6946
6947 vector signed int vec_all_in (vector float, vector float);
6948
6949 vector signed int vec_all_le (vector signed char, vector unsigned char);
6950
6951 vector signed int vec_all_le (vector unsigned char, vector signed char);
6952
6953 vector signed int vec_all_le (vector unsigned char,
6954 vector unsigned char);
6955 vector signed int vec_all_le (vector signed char, vector signed char);
6956 vector signed int vec_all_le (vector signed short,
6957 vector unsigned short);
6958 vector signed int vec_all_le (vector unsigned short,
6959 vector signed short);
6960 vector signed int vec_all_le (vector unsigned short,
6961 vector unsigned short);
6962 vector signed int vec_all_le (vector signed short, vector signed short);
6963
6964 vector signed int vec_all_le (vector signed int, vector unsigned int);
6965 vector signed int vec_all_le (vector unsigned int, vector signed int);
6966 vector signed int vec_all_le (vector unsigned int, vector unsigned int);
6967
6968 vector signed int vec_all_le (vector signed int, vector signed int);
6969 vector signed int vec_all_le (vector float, vector float);
6970
6971 vector signed int vec_all_lt (vector signed char, vector unsigned char);
6972
6973 vector signed int vec_all_lt (vector unsigned char, vector signed char);
6974
6975 vector signed int vec_all_lt (vector unsigned char,
6976 vector unsigned char);
6977 vector signed int vec_all_lt (vector signed char, vector signed char);
6978 vector signed int vec_all_lt (vector signed short,
6979 vector unsigned short);
6980 vector signed int vec_all_lt (vector unsigned short,
6981 vector signed short);
6982 vector signed int vec_all_lt (vector unsigned short,
6983 vector unsigned short);
6984 vector signed int vec_all_lt (vector signed short, vector signed short);
6985
6986 vector signed int vec_all_lt (vector signed int, vector unsigned int);
6987 vector signed int vec_all_lt (vector unsigned int, vector signed int);
6988 vector signed int vec_all_lt (vector unsigned int, vector unsigned int);
6989
6990 vector signed int vec_all_lt (vector signed int, vector signed int);
6991 vector signed int vec_all_lt (vector float, vector float);
6992
6993 vector signed int vec_all_nan (vector float);
6994
6995 vector signed int vec_all_ne (vector signed char, vector unsigned char);
6996
6997 vector signed int vec_all_ne (vector signed char, vector signed char);
6998 vector signed int vec_all_ne (vector unsigned char, vector signed char);
6999
7000 vector signed int vec_all_ne (vector unsigned char,
7001 vector unsigned char);
7002 vector signed int vec_all_ne (vector signed short,
7003 vector unsigned short);
7004 vector signed int vec_all_ne (vector signed short, vector signed short);
7005
7006 vector signed int vec_all_ne (vector unsigned short,
7007 vector signed short);
7008 vector signed int vec_all_ne (vector unsigned short,
7009 vector unsigned short);
7010 vector signed int vec_all_ne (vector signed int, vector unsigned int);
7011 vector signed int vec_all_ne (vector signed int, vector signed int);
7012 vector signed int vec_all_ne (vector unsigned int, vector signed int);
7013 vector signed int vec_all_ne (vector unsigned int, vector unsigned int);
7014
7015 vector signed int vec_all_ne (vector float, vector float);
7016
7017 vector signed int vec_all_nge (vector float, vector float);
7018
7019 vector signed int vec_all_ngt (vector float, vector float);
7020
7021 vector signed int vec_all_nle (vector float, vector float);
7022
7023 vector signed int vec_all_nlt (vector float, vector float);
7024
7025 vector signed int vec_all_numeric (vector float);
7026
7027 vector signed int vec_any_eq (vector signed char, vector unsigned char);
7028
7029 vector signed int vec_any_eq (vector signed char, vector signed char);
7030 vector signed int vec_any_eq (vector unsigned char, vector signed char);
7031
7032 vector signed int vec_any_eq (vector unsigned char,
7033 vector unsigned char);
7034 vector signed int vec_any_eq (vector signed short,
7035 vector unsigned short);
7036 vector signed int vec_any_eq (vector signed short, vector signed short);
7037
7038 vector signed int vec_any_eq (vector unsigned short,
7039 vector signed short);
7040 vector signed int vec_any_eq (vector unsigned short,
7041 vector unsigned short);
7042 vector signed int vec_any_eq (vector signed int, vector unsigned int);
7043 vector signed int vec_any_eq (vector signed int, vector signed int);
7044 vector signed int vec_any_eq (vector unsigned int, vector signed int);
7045 vector signed int vec_any_eq (vector unsigned int, vector unsigned int);
7046
7047 vector signed int vec_any_eq (vector float, vector float);
7048
7049 vector signed int vec_any_ge (vector signed char, vector unsigned char);
7050
7051 vector signed int vec_any_ge (vector unsigned char, vector signed char);
7052
7053 vector signed int vec_any_ge (vector unsigned char,
7054 vector unsigned char);
7055 vector signed int vec_any_ge (vector signed char, vector signed char);
7056 vector signed int vec_any_ge (vector signed short,
7057 vector unsigned short);
7058 vector signed int vec_any_ge (vector unsigned short,
7059 vector signed short);
7060 vector signed int vec_any_ge (vector unsigned short,
7061 vector unsigned short);
7062 vector signed int vec_any_ge (vector signed short, vector signed short);
7063
7064 vector signed int vec_any_ge (vector signed int, vector unsigned int);
7065 vector signed int vec_any_ge (vector unsigned int, vector signed int);
7066 vector signed int vec_any_ge (vector unsigned int, vector unsigned int);
7067
7068 vector signed int vec_any_ge (vector signed int, vector signed int);
7069 vector signed int vec_any_ge (vector float, vector float);
7070
7071 vector signed int vec_any_gt (vector signed char, vector unsigned char);
7072
7073 vector signed int vec_any_gt (vector unsigned char, vector signed char);
7074
7075 vector signed int vec_any_gt (vector unsigned char,
7076 vector unsigned char);
7077 vector signed int vec_any_gt (vector signed char, vector signed char);
7078 vector signed int vec_any_gt (vector signed short,
7079 vector unsigned short);
7080 vector signed int vec_any_gt (vector unsigned short,
7081 vector signed short);
7082 vector signed int vec_any_gt (vector unsigned short,
7083 vector unsigned short);
7084 vector signed int vec_any_gt (vector signed short, vector signed short);
7085
7086 vector signed int vec_any_gt (vector signed int, vector unsigned int);
7087 vector signed int vec_any_gt (vector unsigned int, vector signed int);
7088 vector signed int vec_any_gt (vector unsigned int, vector unsigned int);
7089
7090 vector signed int vec_any_gt (vector signed int, vector signed int);
7091 vector signed int vec_any_gt (vector float, vector float);
7092
7093 vector signed int vec_any_le (vector signed char, vector unsigned char);
7094
7095 vector signed int vec_any_le (vector unsigned char, vector signed char);
7096
7097 vector signed int vec_any_le (vector unsigned char,
7098 vector unsigned char);
7099 vector signed int vec_any_le (vector signed char, vector signed char);
7100 vector signed int vec_any_le (vector signed short,
7101 vector unsigned short);
7102 vector signed int vec_any_le (vector unsigned short,
7103 vector signed short);
7104 vector signed int vec_any_le (vector unsigned short,
7105 vector unsigned short);
7106 vector signed int vec_any_le (vector signed short, vector signed short);
7107
7108 vector signed int vec_any_le (vector signed int, vector unsigned int);
7109 vector signed int vec_any_le (vector unsigned int, vector signed int);
7110 vector signed int vec_any_le (vector unsigned int, vector unsigned int);
7111
7112 vector signed int vec_any_le (vector signed int, vector signed int);
7113 vector signed int vec_any_le (vector float, vector float);
7114
7115 vector signed int vec_any_lt (vector signed char, vector unsigned char);
7116
7117 vector signed int vec_any_lt (vector unsigned char, vector signed char);
7118
7119 vector signed int vec_any_lt (vector unsigned char,
7120 vector unsigned char);
7121 vector signed int vec_any_lt (vector signed char, vector signed char);
7122 vector signed int vec_any_lt (vector signed short,
7123 vector unsigned short);
7124 vector signed int vec_any_lt (vector unsigned short,
7125 vector signed short);
7126 vector signed int vec_any_lt (vector unsigned short,
7127 vector unsigned short);
7128 vector signed int vec_any_lt (vector signed short, vector signed short);
7129
7130 vector signed int vec_any_lt (vector signed int, vector unsigned int);
7131 vector signed int vec_any_lt (vector unsigned int, vector signed int);
7132 vector signed int vec_any_lt (vector unsigned int, vector unsigned int);
7133
7134 vector signed int vec_any_lt (vector signed int, vector signed int);
7135 vector signed int vec_any_lt (vector float, vector float);
7136
7137 vector signed int vec_any_nan (vector float);
7138
7139 vector signed int vec_any_ne (vector signed char, vector unsigned char);
7140
7141 vector signed int vec_any_ne (vector signed char, vector signed char);
7142 vector signed int vec_any_ne (vector unsigned char, vector signed char);
7143
7144 vector signed int vec_any_ne (vector unsigned char,
7145 vector unsigned char);
7146 vector signed int vec_any_ne (vector signed short,
7147 vector unsigned short);
7148 vector signed int vec_any_ne (vector signed short, vector signed short);
7149
7150 vector signed int vec_any_ne (vector unsigned short,
7151 vector signed short);
7152 vector signed int vec_any_ne (vector unsigned short,
7153 vector unsigned short);
7154 vector signed int vec_any_ne (vector signed int, vector unsigned int);
7155 vector signed int vec_any_ne (vector signed int, vector signed int);
7156 vector signed int vec_any_ne (vector unsigned int, vector signed int);
7157 vector signed int vec_any_ne (vector unsigned int, vector unsigned int);
7158
7159 vector signed int vec_any_ne (vector float, vector float);
7160
7161 vector signed int vec_any_nge (vector float, vector float);
7162
7163 vector signed int vec_any_ngt (vector float, vector float);
7164
7165 vector signed int vec_any_nle (vector float, vector float);
7166
7167 vector signed int vec_any_nlt (vector float, vector float);
7168
7169 vector signed int vec_any_numeric (vector float);
7170
7171 vector signed int vec_any_out (vector float, vector float);
7172 @end smallexample
7173
7174 @node Pragmas
7175 @section Pragmas Accepted by GCC
7176 @cindex pragmas
7177 @cindex #pragma
7178
7179 GCC supports several types of pragmas, primarily in order to compile
7180 code originally written for other compilers. Note that in general
7181 we do not recommend the use of pragmas; @xref{Function Attributes},
7182 for further explanation.
7183
7184 @menu
7185 * ARM Pragmas::
7186 * RS/6000 and PowerPC Pragmas::
7187 * Darwin Pragmas::
7188 * Solaris Pragmas::
7189 * Tru64 Pragmas::
7190 @end menu
7191
7192 @node ARM Pragmas
7193 @subsection ARM Pragmas
7194
7195 The ARM target defines pragmas for controlling the default addition of
7196 @code{long_call} and @code{short_call} attributes to functions.
7197 @xref{Function Attributes}, for information about the effects of these
7198 attributes.
7199
7200 @table @code
7201 @item long_calls
7202 @cindex pragma, long_calls
7203 Set all subsequent functions to have the @code{long_call} attribute.
7204
7205 @item no_long_calls
7206 @cindex pragma, no_long_calls
7207 Set all subsequent functions to have the @code{short_call} attribute.
7208
7209 @item long_calls_off
7210 @cindex pragma, long_calls_off
7211 Do not affect the @code{long_call} or @code{short_call} attributes of
7212 subsequent functions.
7213 @end table
7214
7215 @node RS/6000 and PowerPC Pragmas
7216 @subsection RS/6000 and PowerPC Pragmas
7217
7218 The RS/6000 and PowerPC targets define one pragma for controlling
7219 whether or not the @code{longcall} attribute is added to function
7220 declarations by default. This pragma overrides the @option{-mlongcall}
7221 option, but not the @code{longcall} and @code{shortcall} attributes.
7222 @xref{RS/6000 and PowerPC Options}, for more information about when long
7223 calls are and are not necessary.
7224
7225 @table @code
7226 @item longcall (1)
7227 @cindex pragma, longcall
7228 Apply the @code{longcall} attribute to all subsequent function
7229 declarations.
7230
7231 @item longcall (0)
7232 Do not apply the @code{longcall} attribute to subsequent function
7233 declarations.
7234 @end table
7235
7236 @c Describe c4x pragmas here.
7237 @c Describe h8300 pragmas here.
7238 @c Describe i370 pragmas here.
7239 @c Describe i960 pragmas here.
7240 @c Describe sh pragmas here.
7241 @c Describe v850 pragmas here.
7242
7243 @node Darwin Pragmas
7244 @subsection Darwin Pragmas
7245
7246 The following pragmas are available for all architectures running the
7247 Darwin operating system. These are useful for compatibility with other
7248 Mac OS compilers.
7249
7250 @table @code
7251 @item mark @var{tokens}@dots{}
7252 @cindex pragma, mark
7253 This pragma is accepted, but has no effect.
7254
7255 @item options align=@var{alignment}
7256 @cindex pragma, options align
7257 This pragma sets the alignment of fields in structures. The values of
7258 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
7259 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
7260 properly; to restore the previous setting, use @code{reset} for the
7261 @var{alignment}.
7262
7263 @item segment @var{tokens}@dots{}
7264 @cindex pragma, segment
7265 This pragma is accepted, but has no effect.
7266
7267 @item unused (@var{var} [, @var{var}]@dots{})
7268 @cindex pragma, unused
7269 This pragma declares variables to be possibly unused. GCC will not
7270 produce warnings for the listed variables. The effect is similar to
7271 that of the @code{unused} attribute, except that this pragma may appear
7272 anywhere within the variables' scopes.
7273 @end table
7274
7275 @node Solaris Pragmas
7276 @subsection Solaris Pragmas
7277
7278 For compatibility with the SunPRO compiler, the following pragma
7279 is supported.
7280
7281 @table @code
7282 @item redefine_extname @var{oldname} @var{newname}
7283 @cindex pragma, redefine_extname
7284
7285 This pragma gives the C function @var{oldname} the assembler label
7286 @var{newname}. The pragma must appear before the function declaration.
7287 This pragma is equivalent to the asm labels extension (@pxref{Asm
7288 Labels}). The preprocessor defines @code{__PRAGMA_REDEFINE_EXTNAME}
7289 if the pragma is available.
7290 @end table
7291
7292 @node Tru64 Pragmas
7293 @subsection Tru64 Pragmas
7294
7295 For compatibility with the Compaq C compiler, the following pragma
7296 is supported.
7297
7298 @table @code
7299 @item extern_prefix @var{string}
7300 @cindex pragma, extern_prefix
7301
7302 This pragma renames all subsequent function and variable declarations
7303 such that @var{string} is prepended to the name. This effect may be
7304 terminated by using another @code{extern_prefix} pragma with the
7305 empty string.
7306
7307 This pragma is similar in intent to to the asm labels extension
7308 (@pxref{Asm Labels}) in that the system programmer wants to change
7309 the assembly-level ABI without changing the source-level API. The
7310 preprocessor defines @code{__PRAGMA_EXTERN_PREFIX} if the pragma is
7311 available.
7312 @end table
7313
7314 @node Unnamed Fields
7315 @section Unnamed struct/union fields within structs/unions.
7316 @cindex struct
7317 @cindex union
7318
7319 For compatibility with other compilers, GCC allows you to define
7320 a structure or union that contains, as fields, structures and unions
7321 without names. For example:
7322
7323 @example
7324 struct @{
7325 int a;
7326 union @{
7327 int b;
7328 float c;
7329 @};
7330 int d;
7331 @} foo;
7332 @end example
7333
7334 In this example, the user would be able to access members of the unnamed
7335 union with code like @samp{foo.b}. Note that only unnamed structs and
7336 unions are allowed, you may not have, for example, an unnamed
7337 @code{int}.
7338
7339 You must never create such structures that cause ambiguous field definitions.
7340 For example, this structure:
7341
7342 @example
7343 struct @{
7344 int a;
7345 struct @{
7346 int a;
7347 @};
7348 @} foo;
7349 @end example
7350
7351 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
7352 Such constructs are not supported and must be avoided. In the future,
7353 such constructs may be detected and treated as compilation errors.
7354
7355 @node Thread-Local
7356 @section Thread-Local Storage
7357 @cindex Thread-Local Storage
7358 @cindex @acronym{TLS}
7359 @cindex __thread
7360
7361 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
7362 are allocated such that there is one instance of the variable per extant
7363 thread. The run-time model GCC uses to implement this originates
7364 in the IA-64 processor-specific ABI, but has since been migrated
7365 to other processors as well. It requires significant support from
7366 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
7367 system libraries (@file{libc.so} and @file{libpthread.so}), so it
7368 is not available everywhere.
7369
7370 At the user level, the extension is visible with a new storage
7371 class keyword: @code{__thread}. For example:
7372
7373 @example
7374 __thread int i;
7375 extern __thread struct state s;
7376 static __thread char *p;
7377 @end example
7378
7379 The @code{__thread} specifier may be used alone, with the @code{extern}
7380 or @code{static} specifiers, but with no other storage class specifier.
7381 When used with @code{extern} or @code{static}, @code{__thread} must appear
7382 immediately after the other storage class specifier.
7383
7384 The @code{__thread} specifier may be applied to any global, file-scoped
7385 static, function-scoped static, or static data member of a class. It may
7386 not be applied to block-scoped automatic or non-static data member.
7387
7388 When the address-of operator is applied to a thread-local variable, it is
7389 evaluated at run-time and returns the address of the current thread's
7390 instance of that variable. An address so obtained may be used by any
7391 thread. When a thread terminates, any pointers to thread-local variables
7392 in that thread become invalid.
7393
7394 No static initialization may refer to the address of a thread-local variable.
7395
7396 In C++, if an initializer is present for a thread-local variable, it must
7397 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
7398 standard.
7399
7400 See @uref{http://people.redhat.com/drepper/tls.pdf,
7401 ELF Handling For Thread-Local Storage} for a detailed explanation of
7402 the four thread-local storage addressing models, and how the run-time
7403 is expected to function.
7404
7405 @menu
7406 * C99 Thread-Local Edits::
7407 * C++98 Thread-Local Edits::
7408 @end menu
7409
7410 @node C99 Thread-Local Edits
7411 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
7412
7413 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
7414 that document the exact semantics of the language extension.
7415
7416 @itemize @bullet
7417 @item
7418 @cite{5.1.2 Execution environments}
7419
7420 Add new text after paragraph 1
7421
7422 @quotation
7423 Within either execution environment, a @dfn{thread} is a flow of
7424 control within a program. It is implementation defined whether
7425 or not there may be more than one thread associated with a program.
7426 It is implementation defined how threads beyond the first are
7427 created, the name and type of the function called at thread
7428 startup, and how threads may be terminated. However, objects
7429 with thread storage duration shall be initialized before thread
7430 startup.
7431 @end quotation
7432
7433 @item
7434 @cite{6.2.4 Storage durations of objects}
7435
7436 Add new text before paragraph 3
7437
7438 @quotation
7439 An object whose identifier is declared with the storage-class
7440 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
7441 Its lifetime is the entire execution of the thread, and its
7442 stored value is initialized only once, prior to thread startup.
7443 @end quotation
7444
7445 @item
7446 @cite{6.4.1 Keywords}
7447
7448 Add @code{__thread}.
7449
7450 @item
7451 @cite{6.7.1 Storage-class specifiers}
7452
7453 Add @code{__thread} to the list of storage class specifiers in
7454 paragraph 1.
7455
7456 Change paragraph 2 to
7457
7458 @quotation
7459 With the exception of @code{__thread}, at most one storage-class
7460 specifier may be given [@dots{}]. The @code{__thread} specifier may
7461 be used alone, or immediately following @code{extern} or
7462 @code{static}.
7463 @end quotation
7464
7465 Add new text after paragraph 6
7466
7467 @quotation
7468 The declaration of an identifier for a variable that has
7469 block scope that specifies @code{__thread} shall also
7470 specify either @code{extern} or @code{static}.
7471
7472 The @code{__thread} specifier shall be used only with
7473 variables.
7474 @end quotation
7475 @end itemize
7476
7477 @node C++98 Thread-Local Edits
7478 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
7479
7480 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
7481 that document the exact semantics of the language extension.
7482
7483 @itemize @bullet
7484 @item
7485 @b{[intro.execution]}
7486
7487 New text after paragraph 4
7488
7489 @quotation
7490 A @dfn{thread} is a flow of control within the abstract machine.
7491 It is implementation defined whether or not there may be more than
7492 one thread.
7493 @end quotation
7494
7495 New text after paragraph 7
7496
7497 @quotation
7498 It is unspecified whether additional action must be taken to
7499 ensure when and whether side effects are visible to other threads.
7500 @end quotation
7501
7502 @item
7503 @b{[lex.key]}
7504
7505 Add @code{__thread}.
7506
7507 @item
7508 @b{[basic.start.main]}
7509
7510 Add after paragraph 5
7511
7512 @quotation
7513 The thread that begins execution at the @code{main} function is called
7514 the @dfn{main thread}. It is implementation defined how functions
7515 beginning threads other than the main thread are designated or typed.
7516 A function so designated, as well as the @code{main} function, is called
7517 a @dfn{thread startup function}. It is implementation defined what
7518 happens if a thread startup function returns. It is implementation
7519 defined what happens to other threads when any thread calls @code{exit}.
7520 @end quotation
7521
7522 @item
7523 @b{[basic.start.init]}
7524
7525 Add after paragraph 4
7526
7527 @quotation
7528 The storage for an object of thread storage duration shall be
7529 statically initialized before the first statement of the thread startup
7530 function. An object of thread storage duration shall not require
7531 dynamic initialization.
7532 @end quotation
7533
7534 @item
7535 @b{[basic.start.term]}
7536
7537 Add after paragraph 3
7538
7539 @quotation
7540 The type of an object with thread storage duration shall not have a
7541 non-trivial destructor, nor shall it be an array type whose elements
7542 (directly or indirectly) have non-trivial destructors.
7543 @end quotation
7544
7545 @item
7546 @b{[basic.stc]}
7547
7548 Add ``thread storage duration'' to the list in paragraph 1.
7549
7550 Change paragraph 2
7551
7552 @quotation
7553 Thread, static, and automatic storage durations are associated with
7554 objects introduced by declarations [@dots{}].
7555 @end quotation
7556
7557 Add @code{__thread} to the list of specifiers in paragraph 3.
7558
7559 @item
7560 @b{[basic.stc.thread]}
7561
7562 New section before @b{[basic.stc.static]}
7563
7564 @quotation
7565 The keyword @code{__thread} applied to a non-local object gives the
7566 object thread storage duration.
7567
7568 A local variable or class data member declared both @code{static}
7569 and @code{__thread} gives the variable or member thread storage
7570 duration.
7571 @end quotation
7572
7573 @item
7574 @b{[basic.stc.static]}
7575
7576 Change paragraph 1
7577
7578 @quotation
7579 All objects which have neither thread storage duration, dynamic
7580 storage duration nor are local [@dots{}].
7581 @end quotation
7582
7583 @item
7584 @b{[dcl.stc]}
7585
7586 Add @code{__thread} to the list in paragraph 1.
7587
7588 Change paragraph 1
7589
7590 @quotation
7591 With the exception of @code{__thread}, at most one
7592 @var{storage-class-specifier} shall appear in a given
7593 @var{decl-specifier-seq}. The @code{__thread} specifier may
7594 be used alone, or immediately following the @code{extern} or
7595 @code{static} specifiers. [@dots{}]
7596 @end quotation
7597
7598 Add after paragraph 5
7599
7600 @quotation
7601 The @code{__thread} specifier can be applied only to the names of objects
7602 and to anonymous unions.
7603 @end quotation
7604
7605 @item
7606 @b{[class.mem]}
7607
7608 Add after paragraph 6
7609
7610 @quotation
7611 Non-@code{static} members shall not be @code{__thread}.
7612 @end quotation
7613 @end itemize
7614
7615 @node C++ Extensions
7616 @chapter Extensions to the C++ Language
7617 @cindex extensions, C++ language
7618 @cindex C++ language extensions
7619
7620 The GNU compiler provides these extensions to the C++ language (and you
7621 can also use most of the C language extensions in your C++ programs). If you
7622 want to write code that checks whether these features are available, you can
7623 test for the GNU compiler the same way as for C programs: check for a
7624 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
7625 test specifically for GNU C++ (@pxref{Standard Predefined,,Standard
7626 Predefined Macros,cpp.info,The C Preprocessor}).
7627
7628 @menu
7629 * Min and Max:: C++ Minimum and maximum operators.
7630 * Volatiles:: What constitutes an access to a volatile object.
7631 * Restricted Pointers:: C99 restricted pointers and references.
7632 * Vague Linkage:: Where G++ puts inlines, vtables and such.
7633 * C++ Interface:: You can use a single C++ header file for both
7634 declarations and definitions.
7635 * Template Instantiation:: Methods for ensuring that exactly one copy of
7636 each needed template instantiation is emitted.
7637 * Bound member functions:: You can extract a function pointer to the
7638 method denoted by a @samp{->*} or @samp{.*} expression.
7639 * C++ Attributes:: Variable, function, and type attributes for C++ only.
7640 * Java Exceptions:: Tweaking exception handling to work with Java.
7641 * Deprecated Features:: Things will disappear from g++.
7642 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
7643 @end menu
7644
7645 @node Min and Max
7646 @section Minimum and Maximum Operators in C++
7647
7648 It is very convenient to have operators which return the ``minimum'' or the
7649 ``maximum'' of two arguments. In GNU C++ (but not in GNU C),
7650
7651 @table @code
7652 @item @var{a} <? @var{b}
7653 @findex <?
7654 @cindex minimum operator
7655 is the @dfn{minimum}, returning the smaller of the numeric values
7656 @var{a} and @var{b};
7657
7658 @item @var{a} >? @var{b}
7659 @findex >?
7660 @cindex maximum operator
7661 is the @dfn{maximum}, returning the larger of the numeric values @var{a}
7662 and @var{b}.
7663 @end table
7664
7665 These operations are not primitive in ordinary C++, since you can
7666 use a macro to return the minimum of two things in C++, as in the
7667 following example.
7668
7669 @example
7670 #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y))
7671 @end example
7672
7673 @noindent
7674 You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to
7675 the minimum value of variables @var{i} and @var{j}.
7676
7677 However, side effects in @code{X} or @code{Y} may cause unintended
7678 behavior. For example, @code{MIN (i++, j++)} will fail, incrementing
7679 the smaller counter twice. The GNU C @code{typeof} extension allows you
7680 to write safe macros that avoid this kind of problem (@pxref{Typeof}).
7681 However, writing @code{MIN} and @code{MAX} as macros also forces you to
7682 use function-call notation for a fundamental arithmetic operation.
7683 Using GNU C++ extensions, you can write @w{@samp{int min = i <? j;}}
7684 instead.
7685
7686 Since @code{<?} and @code{>?} are built into the compiler, they properly
7687 handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}}
7688 works correctly.
7689
7690 @node Volatiles
7691 @section When is a Volatile Object Accessed?
7692 @cindex accessing volatiles
7693 @cindex volatile read
7694 @cindex volatile write
7695 @cindex volatile access
7696
7697 Both the C and C++ standard have the concept of volatile objects. These
7698 are normally accessed by pointers and used for accessing hardware. The
7699 standards encourage compilers to refrain from optimizations
7700 concerning accesses to volatile objects that it might perform on
7701 non-volatile objects. The C standard leaves it implementation defined
7702 as to what constitutes a volatile access. The C++ standard omits to
7703 specify this, except to say that C++ should behave in a similar manner
7704 to C with respect to volatiles, where possible. The minimum either
7705 standard specifies is that at a sequence point all previous accesses to
7706 volatile objects have stabilized and no subsequent accesses have
7707 occurred. Thus an implementation is free to reorder and combine
7708 volatile accesses which occur between sequence points, but cannot do so
7709 for accesses across a sequence point. The use of volatiles does not
7710 allow you to violate the restriction on updating objects multiple times
7711 within a sequence point.
7712
7713 In most expressions, it is intuitively obvious what is a read and what is
7714 a write. For instance
7715
7716 @example
7717 volatile int *dst = @var{somevalue};
7718 volatile int *src = @var{someothervalue};
7719 *dst = *src;
7720 @end example
7721
7722 @noindent
7723 will cause a read of the volatile object pointed to by @var{src} and stores the
7724 value into the volatile object pointed to by @var{dst}. There is no
7725 guarantee that these reads and writes are atomic, especially for objects
7726 larger than @code{int}.
7727
7728 Less obvious expressions are where something which looks like an access
7729 is used in a void context. An example would be,
7730
7731 @example
7732 volatile int *src = @var{somevalue};
7733 *src;
7734 @end example
7735
7736 With C, such expressions are rvalues, and as rvalues cause a read of
7737 the object, GCC interprets this as a read of the volatile being pointed
7738 to. The C++ standard specifies that such expressions do not undergo
7739 lvalue to rvalue conversion, and that the type of the dereferenced
7740 object may be incomplete. The C++ standard does not specify explicitly
7741 that it is this lvalue to rvalue conversion which is responsible for
7742 causing an access. However, there is reason to believe that it is,
7743 because otherwise certain simple expressions become undefined. However,
7744 because it would surprise most programmers, G++ treats dereferencing a
7745 pointer to volatile object of complete type in a void context as a read
7746 of the object. When the object has incomplete type, G++ issues a
7747 warning.
7748
7749 @example
7750 struct S;
7751 struct T @{int m;@};
7752 volatile S *ptr1 = @var{somevalue};
7753 volatile T *ptr2 = @var{somevalue};
7754 *ptr1;
7755 *ptr2;
7756 @end example
7757
7758 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
7759 causes a read of the object pointed to. If you wish to force an error on
7760 the first case, you must force a conversion to rvalue with, for instance
7761 a static cast, @code{static_cast<S>(*ptr1)}.
7762
7763 When using a reference to volatile, G++ does not treat equivalent
7764 expressions as accesses to volatiles, but instead issues a warning that
7765 no volatile is accessed. The rationale for this is that otherwise it
7766 becomes difficult to determine where volatile access occur, and not
7767 possible to ignore the return value from functions returning volatile
7768 references. Again, if you wish to force a read, cast the reference to
7769 an rvalue.
7770
7771 @node Restricted Pointers
7772 @section Restricting Pointer Aliasing
7773 @cindex restricted pointers
7774 @cindex restricted references
7775 @cindex restricted this pointer
7776
7777 As with gcc, g++ understands the C99 feature of restricted pointers,
7778 specified with the @code{__restrict__}, or @code{__restrict} type
7779 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
7780 language flag, @code{restrict} is not a keyword in C++.
7781
7782 In addition to allowing restricted pointers, you can specify restricted
7783 references, which indicate that the reference is not aliased in the local
7784 context.
7785
7786 @example
7787 void fn (int *__restrict__ rptr, int &__restrict__ rref)
7788 @{
7789 /* @r{@dots{}} */
7790 @}
7791 @end example
7792
7793 @noindent
7794 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
7795 @var{rref} refers to a (different) unaliased integer.
7796
7797 You may also specify whether a member function's @var{this} pointer is
7798 unaliased by using @code{__restrict__} as a member function qualifier.
7799
7800 @example
7801 void T::fn () __restrict__
7802 @{
7803 /* @r{@dots{}} */
7804 @}
7805 @end example
7806
7807 @noindent
7808 Within the body of @code{T::fn}, @var{this} will have the effective
7809 definition @code{T *__restrict__ const this}. Notice that the
7810 interpretation of a @code{__restrict__} member function qualifier is
7811 different to that of @code{const} or @code{volatile} qualifier, in that it
7812 is applied to the pointer rather than the object. This is consistent with
7813 other compilers which implement restricted pointers.
7814
7815 As with all outermost parameter qualifiers, @code{__restrict__} is
7816 ignored in function definition matching. This means you only need to
7817 specify @code{__restrict__} in a function definition, rather than
7818 in a function prototype as well.
7819
7820 @node Vague Linkage
7821 @section Vague Linkage
7822 @cindex vague linkage
7823
7824 There are several constructs in C++ which require space in the object
7825 file but are not clearly tied to a single translation unit. We say that
7826 these constructs have ``vague linkage''. Typically such constructs are
7827 emitted wherever they are needed, though sometimes we can be more
7828 clever.
7829
7830 @table @asis
7831 @item Inline Functions
7832 Inline functions are typically defined in a header file which can be
7833 included in many different compilations. Hopefully they can usually be
7834 inlined, but sometimes an out-of-line copy is necessary, if the address
7835 of the function is taken or if inlining fails. In general, we emit an
7836 out-of-line copy in all translation units where one is needed. As an
7837 exception, we only emit inline virtual functions with the vtable, since
7838 it will always require a copy.
7839
7840 Local static variables and string constants used in an inline function
7841 are also considered to have vague linkage, since they must be shared
7842 between all inlined and out-of-line instances of the function.
7843
7844 @item VTables
7845 @cindex vtable
7846 C++ virtual functions are implemented in most compilers using a lookup
7847 table, known as a vtable. The vtable contains pointers to the virtual
7848 functions provided by a class, and each object of the class contains a
7849 pointer to its vtable (or vtables, in some multiple-inheritance
7850 situations). If the class declares any non-inline, non-pure virtual
7851 functions, the first one is chosen as the ``key method'' for the class,
7852 and the vtable is only emitted in the translation unit where the key
7853 method is defined.
7854
7855 @emph{Note:} If the chosen key method is later defined as inline, the
7856 vtable will still be emitted in every translation unit which defines it.
7857 Make sure that any inline virtuals are declared inline in the class
7858 body, even if they are not defined there.
7859
7860 @item type_info objects
7861 @cindex type_info
7862 @cindex RTTI
7863 C++ requires information about types to be written out in order to
7864 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
7865 For polymorphic classes (classes with virtual functions), the type_info
7866 object is written out along with the vtable so that @samp{dynamic_cast}
7867 can determine the dynamic type of a class object at runtime. For all
7868 other types, we write out the type_info object when it is used: when
7869 applying @samp{typeid} to an expression, throwing an object, or
7870 referring to a type in a catch clause or exception specification.
7871
7872 @item Template Instantiations
7873 Most everything in this section also applies to template instantiations,
7874 but there are other options as well.
7875 @xref{Template Instantiation,,Where's the Template?}.
7876
7877 @end table
7878
7879 When used with GNU ld version 2.8 or later on an ELF system such as
7880 Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of
7881 these constructs will be discarded at link time. This is known as
7882 COMDAT support.
7883
7884 On targets that don't support COMDAT, but do support weak symbols, GCC
7885 will use them. This way one copy will override all the others, but
7886 the unused copies will still take up space in the executable.
7887
7888 For targets which do not support either COMDAT or weak symbols,
7889 most entities with vague linkage will be emitted as local symbols to
7890 avoid duplicate definition errors from the linker. This will not happen
7891 for local statics in inlines, however, as having multiple copies will
7892 almost certainly break things.
7893
7894 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
7895 another way to control placement of these constructs.
7896
7897 @node C++ Interface
7898 @section Declarations and Definitions in One Header
7899
7900 @cindex interface and implementation headers, C++
7901 @cindex C++ interface and implementation headers
7902 C++ object definitions can be quite complex. In principle, your source
7903 code will need two kinds of things for each object that you use across
7904 more than one source file. First, you need an @dfn{interface}
7905 specification, describing its structure with type declarations and
7906 function prototypes. Second, you need the @dfn{implementation} itself.
7907 It can be tedious to maintain a separate interface description in a
7908 header file, in parallel to the actual implementation. It is also
7909 dangerous, since separate interface and implementation definitions may
7910 not remain parallel.
7911
7912 @cindex pragmas, interface and implementation
7913 With GNU C++, you can use a single header file for both purposes.
7914
7915 @quotation
7916 @emph{Warning:} The mechanism to specify this is in transition. For the
7917 nonce, you must use one of two @code{#pragma} commands; in a future
7918 release of GNU C++, an alternative mechanism will make these
7919 @code{#pragma} commands unnecessary.
7920 @end quotation
7921
7922 The header file contains the full definitions, but is marked with
7923 @samp{#pragma interface} in the source code. This allows the compiler
7924 to use the header file only as an interface specification when ordinary
7925 source files incorporate it with @code{#include}. In the single source
7926 file where the full implementation belongs, you can use either a naming
7927 convention or @samp{#pragma implementation} to indicate this alternate
7928 use of the header file.
7929
7930 @table @code
7931 @item #pragma interface
7932 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
7933 @kindex #pragma interface
7934 Use this directive in @emph{header files} that define object classes, to save
7935 space in most of the object files that use those classes. Normally,
7936 local copies of certain information (backup copies of inline member
7937 functions, debugging information, and the internal tables that implement
7938 virtual functions) must be kept in each object file that includes class
7939 definitions. You can use this pragma to avoid such duplication. When a
7940 header file containing @samp{#pragma interface} is included in a
7941 compilation, this auxiliary information will not be generated (unless
7942 the main input source file itself uses @samp{#pragma implementation}).
7943 Instead, the object files will contain references to be resolved at link
7944 time.
7945
7946 The second form of this directive is useful for the case where you have
7947 multiple headers with the same name in different directories. If you
7948 use this form, you must specify the same string to @samp{#pragma
7949 implementation}.
7950
7951 @item #pragma implementation
7952 @itemx #pragma implementation "@var{objects}.h"
7953 @kindex #pragma implementation
7954 Use this pragma in a @emph{main input file}, when you want full output from
7955 included header files to be generated (and made globally visible). The
7956 included header file, in turn, should use @samp{#pragma interface}.
7957 Backup copies of inline member functions, debugging information, and the
7958 internal tables used to implement virtual functions are all generated in
7959 implementation files.
7960
7961 @cindex implied @code{#pragma implementation}
7962 @cindex @code{#pragma implementation}, implied
7963 @cindex naming convention, implementation headers
7964 If you use @samp{#pragma implementation} with no argument, it applies to
7965 an include file with the same basename@footnote{A file's @dfn{basename}
7966 was the name stripped of all leading path information and of trailing
7967 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
7968 file. For example, in @file{allclass.cc}, giving just
7969 @samp{#pragma implementation}
7970 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
7971
7972 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
7973 an implementation file whenever you would include it from
7974 @file{allclass.cc} even if you never specified @samp{#pragma
7975 implementation}. This was deemed to be more trouble than it was worth,
7976 however, and disabled.
7977
7978 If you use an explicit @samp{#pragma implementation}, it must appear in
7979 your source file @emph{before} you include the affected header files.
7980
7981 Use the string argument if you want a single implementation file to
7982 include code from multiple header files. (You must also use
7983 @samp{#include} to include the header file; @samp{#pragma
7984 implementation} only specifies how to use the file---it doesn't actually
7985 include it.)
7986
7987 There is no way to split up the contents of a single header file into
7988 multiple implementation files.
7989 @end table
7990
7991 @cindex inlining and C++ pragmas
7992 @cindex C++ pragmas, effect on inlining
7993 @cindex pragmas in C++, effect on inlining
7994 @samp{#pragma implementation} and @samp{#pragma interface} also have an
7995 effect on function inlining.
7996
7997 If you define a class in a header file marked with @samp{#pragma
7998 interface}, the effect on a function defined in that class is similar to
7999 an explicit @code{extern} declaration---the compiler emits no code at
8000 all to define an independent version of the function. Its definition
8001 is used only for inlining with its callers.
8002
8003 @opindex fno-implement-inlines
8004 Conversely, when you include the same header file in a main source file
8005 that declares it as @samp{#pragma implementation}, the compiler emits
8006 code for the function itself; this defines a version of the function
8007 that can be found via pointers (or by callers compiled without
8008 inlining). If all calls to the function can be inlined, you can avoid
8009 emitting the function by compiling with @option{-fno-implement-inlines}.
8010 If any calls were not inlined, you will get linker errors.
8011
8012 @node Template Instantiation
8013 @section Where's the Template?
8014 @cindex template instantiation
8015
8016 C++ templates are the first language feature to require more
8017 intelligence from the environment than one usually finds on a UNIX
8018 system. Somehow the compiler and linker have to make sure that each
8019 template instance occurs exactly once in the executable if it is needed,
8020 and not at all otherwise. There are two basic approaches to this
8021 problem, which I will refer to as the Borland model and the Cfront model.
8022
8023 @table @asis
8024 @item Borland model
8025 Borland C++ solved the template instantiation problem by adding the code
8026 equivalent of common blocks to their linker; the compiler emits template
8027 instances in each translation unit that uses them, and the linker
8028 collapses them together. The advantage of this model is that the linker
8029 only has to consider the object files themselves; there is no external
8030 complexity to worry about. This disadvantage is that compilation time
8031 is increased because the template code is being compiled repeatedly.
8032 Code written for this model tends to include definitions of all
8033 templates in the header file, since they must be seen to be
8034 instantiated.
8035
8036 @item Cfront model
8037 The AT&T C++ translator, Cfront, solved the template instantiation
8038 problem by creating the notion of a template repository, an
8039 automatically maintained place where template instances are stored. A
8040 more modern version of the repository works as follows: As individual
8041 object files are built, the compiler places any template definitions and
8042 instantiations encountered in the repository. At link time, the link
8043 wrapper adds in the objects in the repository and compiles any needed
8044 instances that were not previously emitted. The advantages of this
8045 model are more optimal compilation speed and the ability to use the
8046 system linker; to implement the Borland model a compiler vendor also
8047 needs to replace the linker. The disadvantages are vastly increased
8048 complexity, and thus potential for error; for some code this can be
8049 just as transparent, but in practice it can been very difficult to build
8050 multiple programs in one directory and one program in multiple
8051 directories. Code written for this model tends to separate definitions
8052 of non-inline member templates into a separate file, which should be
8053 compiled separately.
8054 @end table
8055
8056 When used with GNU ld version 2.8 or later on an ELF system such as
8057 Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the
8058 Borland model. On other systems, g++ implements neither automatic
8059 model.
8060
8061 A future version of g++ will support a hybrid model whereby the compiler
8062 will emit any instantiations for which the template definition is
8063 included in the compile, and store template definitions and
8064 instantiation context information into the object file for the rest.
8065 The link wrapper will extract that information as necessary and invoke
8066 the compiler to produce the remaining instantiations. The linker will
8067 then combine duplicate instantiations.
8068
8069 In the mean time, you have the following options for dealing with
8070 template instantiations:
8071
8072 @enumerate
8073 @item
8074 @opindex frepo
8075 Compile your template-using code with @option{-frepo}. The compiler will
8076 generate files with the extension @samp{.rpo} listing all of the
8077 template instantiations used in the corresponding object files which
8078 could be instantiated there; the link wrapper, @samp{collect2}, will
8079 then update the @samp{.rpo} files to tell the compiler where to place
8080 those instantiations and rebuild any affected object files. The
8081 link-time overhead is negligible after the first pass, as the compiler
8082 will continue to place the instantiations in the same files.
8083
8084 This is your best option for application code written for the Borland
8085 model, as it will just work. Code written for the Cfront model will
8086 need to be modified so that the template definitions are available at
8087 one or more points of instantiation; usually this is as simple as adding
8088 @code{#include <tmethods.cc>} to the end of each template header.
8089
8090 For library code, if you want the library to provide all of the template
8091 instantiations it needs, just try to link all of its object files
8092 together; the link will fail, but cause the instantiations to be
8093 generated as a side effect. Be warned, however, that this may cause
8094 conflicts if multiple libraries try to provide the same instantiations.
8095 For greater control, use explicit instantiation as described in the next
8096 option.
8097
8098 @item
8099 @opindex fno-implicit-templates
8100 Compile your code with @option{-fno-implicit-templates} to disable the
8101 implicit generation of template instances, and explicitly instantiate
8102 all the ones you use. This approach requires more knowledge of exactly
8103 which instances you need than do the others, but it's less
8104 mysterious and allows greater control. You can scatter the explicit
8105 instantiations throughout your program, perhaps putting them in the
8106 translation units where the instances are used or the translation units
8107 that define the templates themselves; you can put all of the explicit
8108 instantiations you need into one big file; or you can create small files
8109 like
8110
8111 @example
8112 #include "Foo.h"
8113 #include "Foo.cc"
8114
8115 template class Foo<int>;
8116 template ostream& operator <<
8117 (ostream&, const Foo<int>&);
8118 @end example
8119
8120 for each of the instances you need, and create a template instantiation
8121 library from those.
8122
8123 If you are using Cfront-model code, you can probably get away with not
8124 using @option{-fno-implicit-templates} when compiling files that don't
8125 @samp{#include} the member template definitions.
8126
8127 If you use one big file to do the instantiations, you may want to
8128 compile it without @option{-fno-implicit-templates} so you get all of the
8129 instances required by your explicit instantiations (but not by any
8130 other files) without having to specify them as well.
8131
8132 g++ has extended the template instantiation syntax given in the ISO
8133 standard to allow forward declaration of explicit instantiations
8134 (with @code{extern}), instantiation of the compiler support data for a
8135 template class (i.e.@: the vtable) without instantiating any of its
8136 members (with @code{inline}), and instantiation of only the static data
8137 members of a template class, without the support data or member
8138 functions (with (@code{static}):
8139
8140 @example
8141 extern template int max (int, int);
8142 inline template class Foo<int>;
8143 static template class Foo<int>;
8144 @end example
8145
8146 @item
8147 Do nothing. Pretend g++ does implement automatic instantiation
8148 management. Code written for the Borland model will work fine, but
8149 each translation unit will contain instances of each of the templates it
8150 uses. In a large program, this can lead to an unacceptable amount of code
8151 duplication.
8152
8153 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
8154 more discussion of these pragmas.
8155 @end enumerate
8156
8157 @node Bound member functions
8158 @section Extracting the function pointer from a bound pointer to member function
8159 @cindex pmf
8160 @cindex pointer to member function
8161 @cindex bound pointer to member function
8162
8163 In C++, pointer to member functions (PMFs) are implemented using a wide
8164 pointer of sorts to handle all the possible call mechanisms; the PMF
8165 needs to store information about how to adjust the @samp{this} pointer,
8166 and if the function pointed to is virtual, where to find the vtable, and
8167 where in the vtable to look for the member function. If you are using
8168 PMFs in an inner loop, you should really reconsider that decision. If
8169 that is not an option, you can extract the pointer to the function that
8170 would be called for a given object/PMF pair and call it directly inside
8171 the inner loop, to save a bit of time.
8172
8173 Note that you will still be paying the penalty for the call through a
8174 function pointer; on most modern architectures, such a call defeats the
8175 branch prediction features of the CPU@. This is also true of normal
8176 virtual function calls.
8177
8178 The syntax for this extension is
8179
8180 @example
8181 extern A a;
8182 extern int (A::*fp)();
8183 typedef int (*fptr)(A *);
8184
8185 fptr p = (fptr)(a.*fp);
8186 @end example
8187
8188 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
8189 no object is needed to obtain the address of the function. They can be
8190 converted to function pointers directly:
8191
8192 @example
8193 fptr p1 = (fptr)(&A::foo);
8194 @end example
8195
8196 @opindex Wno-pmf-conversions
8197 You must specify @option{-Wno-pmf-conversions} to use this extension.
8198
8199 @node C++ Attributes
8200 @section C++-Specific Variable, Function, and Type Attributes
8201
8202 Some attributes only make sense for C++ programs.
8203
8204 @table @code
8205 @item init_priority (@var{priority})
8206 @cindex init_priority attribute
8207
8208
8209 In Standard C++, objects defined at namespace scope are guaranteed to be
8210 initialized in an order in strict accordance with that of their definitions
8211 @emph{in a given translation unit}. No guarantee is made for initializations
8212 across translation units. However, GNU C++ allows users to control the
8213 order of initialization of objects defined at namespace scope with the
8214 @code{init_priority} attribute by specifying a relative @var{priority},
8215 a constant integral expression currently bounded between 101 and 65535
8216 inclusive. Lower numbers indicate a higher priority.
8217
8218 In the following example, @code{A} would normally be created before
8219 @code{B}, but the @code{init_priority} attribute has reversed that order:
8220
8221 @smallexample
8222 Some_Class A __attribute__ ((init_priority (2000)));
8223 Some_Class B __attribute__ ((init_priority (543)));
8224 @end smallexample
8225
8226 @noindent
8227 Note that the particular values of @var{priority} do not matter; only their
8228 relative ordering.
8229
8230 @item java_interface
8231 @cindex java_interface attribute
8232
8233 This type attribute informs C++ that the class is a Java interface. It may
8234 only be applied to classes declared within an @code{extern "Java"} block.
8235 Calls to methods declared in this interface will be dispatched using GCJ's
8236 interface table mechanism, instead of regular virtual table dispatch.
8237
8238 @end table
8239
8240 @node Java Exceptions
8241 @section Java Exceptions
8242
8243 The Java language uses a slightly different exception handling model
8244 from C++. Normally, GNU C++ will automatically detect when you are
8245 writing C++ code that uses Java exceptions, and handle them
8246 appropriately. However, if C++ code only needs to execute destructors
8247 when Java exceptions are thrown through it, GCC will guess incorrectly.
8248 Sample problematic code is:
8249
8250 @smallexample
8251 struct S @{ ~S(); @};
8252 extern void bar(); // is written in Java, and may throw exceptions
8253 void foo()
8254 @{
8255 S s;
8256 bar();
8257 @}
8258 @end smallexample
8259
8260 @noindent
8261 The usual effect of an incorrect guess is a link failure, complaining of
8262 a missing routine called @samp{__gxx_personality_v0}.
8263
8264 You can inform the compiler that Java exceptions are to be used in a
8265 translation unit, irrespective of what it might think, by writing
8266 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
8267 @samp{#pragma} must appear before any functions that throw or catch
8268 exceptions, or run destructors when exceptions are thrown through them.
8269
8270 You cannot mix Java and C++ exceptions in the same translation unit. It
8271 is believed to be safe to throw a C++ exception from one file through
8272 another file compiled for the Java exception model, or vice versa, but
8273 there may be bugs in this area.
8274
8275 @node Deprecated Features
8276 @section Deprecated Features
8277
8278 In the past, the GNU C++ compiler was extended to experiment with new
8279 features, at a time when the C++ language was still evolving. Now that
8280 the C++ standard is complete, some of those features are superseded by
8281 superior alternatives. Using the old features might cause a warning in
8282 some cases that the feature will be dropped in the future. In other
8283 cases, the feature might be gone already.
8284
8285 While the list below is not exhaustive, it documents some of the options
8286 that are now deprecated:
8287
8288 @table @code
8289 @item -fexternal-templates
8290 @itemx -falt-external-templates
8291 These are two of the many ways for g++ to implement template
8292 instantiation. @xref{Template Instantiation}. The C++ standard clearly
8293 defines how template definitions have to be organized across
8294 implementation units. g++ has an implicit instantiation mechanism that
8295 should work just fine for standard-conforming code.
8296
8297 @item -fstrict-prototype
8298 @itemx -fno-strict-prototype
8299 Previously it was possible to use an empty prototype parameter list to
8300 indicate an unspecified number of parameters (like C), rather than no
8301 parameters, as C++ demands. This feature has been removed, except where
8302 it is required for backwards compatibility @xref{Backwards Compatibility}.
8303 @end table
8304
8305 The named return value extension has been deprecated, and is now
8306 removed from g++.
8307
8308 The use of initializer lists with new expressions has been deprecated,
8309 and is now removed from g++.
8310
8311 Floating and complex non-type template parameters have been deprecated,
8312 and are now removed from g++.
8313
8314 The implicit typename extension has been deprecated and is now
8315 removed from g++.
8316
8317 The use of default arguments in function pointers, function typedefs and
8318 and other places where they are not permitted by the standard is
8319 deprecated and will be removed from a future version of g++.
8320
8321 @node Backwards Compatibility
8322 @section Backwards Compatibility
8323 @cindex Backwards Compatibility
8324 @cindex ARM [Annotated C++ Reference Manual]
8325
8326 Now that there is a definitive ISO standard C++, G++ has a specification
8327 to adhere to. The C++ language evolved over time, and features that
8328 used to be acceptable in previous drafts of the standard, such as the ARM
8329 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
8330 compilation of C++ written to such drafts, G++ contains some backwards
8331 compatibilities. @emph{All such backwards compatibility features are
8332 liable to disappear in future versions of G++.} They should be considered
8333 deprecated @xref{Deprecated Features}.
8334
8335 @table @code
8336 @item For scope
8337 If a variable is declared at for scope, it used to remain in scope until
8338 the end of the scope which contained the for statement (rather than just
8339 within the for scope). G++ retains this, but issues a warning, if such a
8340 variable is accessed outside the for scope.
8341
8342 @item Implicit C language
8343 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
8344 scope to set the language. On such systems, all header files are
8345 implicitly scoped inside a C language scope. Also, an empty prototype
8346 @code{()} will be treated as an unspecified number of arguments, rather
8347 than no arguments, as C++ demands.
8348 @end table