1 @c Copyright (C) 1988, 1989, 1992, 1993, 1994, 1996, 1998, 1999, 2000, 2001,
2 @c 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
3 @c Free Software Foundation, Inc.
5 @c This is part of the GCC manual.
6 @c For copying conditions, see the file gcc.texi.
9 @chapter Extensions to the C Language Family
10 @cindex extensions, C language
11 @cindex C language extensions
14 GNU C provides several language features not found in ISO standard C@.
15 (The @option{-pedantic} option directs GCC to print a warning message if
16 any of these features is used.) To test for the availability of these
17 features in conditional compilation, check for a predefined macro
18 @code{__GNUC__}, which is always defined under GCC@.
20 These extensions are available in C and Objective-C@. Most of them are
21 also available in C++. @xref{C++ Extensions,,Extensions to the
22 C++ Language}, for extensions that apply @emph{only} to C++.
24 Some features that are in ISO C99 but not C90 or C++ are also, as
25 extensions, accepted by GCC in C90 mode and in C++.
28 * Statement Exprs:: Putting statements and declarations inside expressions.
29 * Local Labels:: Labels local to a block.
30 * Labels as Values:: Getting pointers to labels, and computed gotos.
31 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
32 * Constructing Calls:: Dispatching a call to another function.
33 * Typeof:: @code{typeof}: referring to the type of an expression.
34 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
35 * Long Long:: Double-word integers---@code{long long int}.
36 * __int128:: 128-bit integers---@code{__int128}.
37 * Complex:: Data types for complex numbers.
38 * Floating Types:: Additional Floating Types.
39 * Half-Precision:: Half-Precision Floating Point.
40 * Decimal Float:: Decimal Floating Types.
41 * Hex Floats:: Hexadecimal floating-point constants.
42 * Fixed-Point:: Fixed-Point Types.
43 * Named Address Spaces::Named address spaces.
44 * Zero Length:: Zero-length arrays.
45 * Variable Length:: Arrays whose length is computed at run time.
46 * Empty Structures:: Structures with no members.
47 * Variadic Macros:: Macros with a variable number of arguments.
48 * Escaped Newlines:: Slightly looser rules for escaped newlines.
49 * Subscripting:: Any array can be subscripted, even if not an lvalue.
50 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
51 * Initializers:: Non-constant initializers.
52 * Compound Literals:: Compound literals give structures, unions
54 * Designated Inits:: Labeling elements of initializers.
55 * Cast to Union:: Casting to union type from any member of the union.
56 * Case Ranges:: `case 1 ... 9' and such.
57 * Mixed Declarations:: Mixing declarations and code.
58 * Function Attributes:: Declaring that functions have no side effects,
59 or that they can never return.
60 * Attribute Syntax:: Formal syntax for attributes.
61 * Function Prototypes:: Prototype declarations and old-style definitions.
62 * C++ Comments:: C++ comments are recognized.
63 * Dollar Signs:: Dollar sign is allowed in identifiers.
64 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
65 * Variable Attributes:: Specifying attributes of variables.
66 * Type Attributes:: Specifying attributes of types.
67 * Alignment:: Inquiring about the alignment of a type or variable.
68 * Inline:: Defining inline functions (as fast as macros).
69 * Volatiles:: What constitutes an access to a volatile object.
70 * Extended Asm:: Assembler instructions with C expressions as operands.
71 (With them you can define ``built-in'' functions.)
72 * Constraints:: Constraints for asm operands
73 * Asm Labels:: Specifying the assembler name to use for a C symbol.
74 * Explicit Reg Vars:: Defining variables residing in specified registers.
75 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
76 * Incomplete Enums:: @code{enum foo;}, with details to follow.
77 * Function Names:: Printable strings which are the name of the current
79 * Return Address:: Getting the return or frame address of a function.
80 * Vector Extensions:: Using vector instructions through built-in functions.
81 * Offsetof:: Special syntax for implementing @code{offsetof}.
82 * Atomic Builtins:: Built-in functions for atomic memory access.
83 * Object Size Checking:: Built-in functions for limited buffer overflow
85 * Other Builtins:: Other built-in functions.
86 * Target Builtins:: Built-in functions specific to particular targets.
87 * Target Format Checks:: Format checks specific to particular targets.
88 * Pragmas:: Pragmas accepted by GCC.
89 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
90 * Thread-Local:: Per-thread variables.
91 * Binary constants:: Binary constants using the @samp{0b} prefix.
95 @section Statements and Declarations in Expressions
96 @cindex statements inside expressions
97 @cindex declarations inside expressions
98 @cindex expressions containing statements
99 @cindex macros, statements in expressions
101 @c the above section title wrapped and causes an underfull hbox.. i
102 @c changed it from "within" to "in". --mew 4feb93
103 A compound statement enclosed in parentheses may appear as an expression
104 in GNU C@. This allows you to use loops, switches, and local variables
105 within an expression.
107 Recall that a compound statement is a sequence of statements surrounded
108 by braces; in this construct, parentheses go around the braces. For
112 (@{ int y = foo (); int z;
119 is a valid (though slightly more complex than necessary) expression
120 for the absolute value of @code{foo ()}.
122 The last thing in the compound statement should be an expression
123 followed by a semicolon; the value of this subexpression serves as the
124 value of the entire construct. (If you use some other kind of statement
125 last within the braces, the construct has type @code{void}, and thus
126 effectively no value.)
128 This feature is especially useful in making macro definitions ``safe'' (so
129 that they evaluate each operand exactly once). For example, the
130 ``maximum'' function is commonly defined as a macro in standard C as
134 #define max(a,b) ((a) > (b) ? (a) : (b))
138 @cindex side effects, macro argument
139 But this definition computes either @var{a} or @var{b} twice, with bad
140 results if the operand has side effects. In GNU C, if you know the
141 type of the operands (here taken as @code{int}), you can define
142 the macro safely as follows:
145 #define maxint(a,b) \
146 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
149 Embedded statements are not allowed in constant expressions, such as
150 the value of an enumeration constant, the width of a bit-field, or
151 the initial value of a static variable.
153 If you don't know the type of the operand, you can still do this, but you
154 must use @code{typeof} (@pxref{Typeof}).
156 In G++, the result value of a statement expression undergoes array and
157 function pointer decay, and is returned by value to the enclosing
158 expression. For instance, if @code{A} is a class, then
167 will construct a temporary @code{A} object to hold the result of the
168 statement expression, and that will be used to invoke @code{Foo}.
169 Therefore the @code{this} pointer observed by @code{Foo} will not be the
172 Any temporaries created within a statement within a statement expression
173 will be destroyed at the statement's end. This makes statement
174 expressions inside macros slightly different from function calls. In
175 the latter case temporaries introduced during argument evaluation will
176 be destroyed at the end of the statement that includes the function
177 call. In the statement expression case they will be destroyed during
178 the statement expression. For instance,
181 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
182 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
192 will have different places where temporaries are destroyed. For the
193 @code{macro} case, the temporary @code{X} will be destroyed just after
194 the initialization of @code{b}. In the @code{function} case that
195 temporary will be destroyed when the function returns.
197 These considerations mean that it is probably a bad idea to use
198 statement-expressions of this form in header files that are designed to
199 work with C++. (Note that some versions of the GNU C Library contained
200 header files using statement-expression that lead to precisely this
203 Jumping into a statement expression with @code{goto} or using a
204 @code{switch} statement outside the statement expression with a
205 @code{case} or @code{default} label inside the statement expression is
206 not permitted. Jumping into a statement expression with a computed
207 @code{goto} (@pxref{Labels as Values}) yields undefined behavior.
208 Jumping out of a statement expression is permitted, but if the
209 statement expression is part of a larger expression then it is
210 unspecified which other subexpressions of that expression have been
211 evaluated except where the language definition requires certain
212 subexpressions to be evaluated before or after the statement
213 expression. In any case, as with a function call the evaluation of a
214 statement expression is not interleaved with the evaluation of other
215 parts of the containing expression. For example,
218 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
222 will call @code{foo} and @code{bar1} and will not call @code{baz} but
223 may or may not call @code{bar2}. If @code{bar2} is called, it will be
224 called after @code{foo} and before @code{bar1}
227 @section Locally Declared Labels
229 @cindex macros, local labels
231 GCC allows you to declare @dfn{local labels} in any nested block
232 scope. A local label is just like an ordinary label, but you can
233 only reference it (with a @code{goto} statement, or by taking its
234 address) within the block in which it was declared.
236 A local label declaration looks like this:
239 __label__ @var{label};
246 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
249 Local label declarations must come at the beginning of the block,
250 before any ordinary declarations or statements.
252 The label declaration defines the label @emph{name}, but does not define
253 the label itself. You must do this in the usual way, with
254 @code{@var{label}:}, within the statements of the statement expression.
256 The local label feature is useful for complex macros. If a macro
257 contains nested loops, a @code{goto} can be useful for breaking out of
258 them. However, an ordinary label whose scope is the whole function
259 cannot be used: if the macro can be expanded several times in one
260 function, the label will be multiply defined in that function. A
261 local label avoids this problem. For example:
264 #define SEARCH(value, array, target) \
267 typeof (target) _SEARCH_target = (target); \
268 typeof (*(array)) *_SEARCH_array = (array); \
271 for (i = 0; i < max; i++) \
272 for (j = 0; j < max; j++) \
273 if (_SEARCH_array[i][j] == _SEARCH_target) \
274 @{ (value) = i; goto found; @} \
280 This could also be written using a statement-expression:
283 #define SEARCH(array, target) \
286 typeof (target) _SEARCH_target = (target); \
287 typeof (*(array)) *_SEARCH_array = (array); \
290 for (i = 0; i < max; i++) \
291 for (j = 0; j < max; j++) \
292 if (_SEARCH_array[i][j] == _SEARCH_target) \
293 @{ value = i; goto found; @} \
300 Local label declarations also make the labels they declare visible to
301 nested functions, if there are any. @xref{Nested Functions}, for details.
303 @node Labels as Values
304 @section Labels as Values
305 @cindex labels as values
306 @cindex computed gotos
307 @cindex goto with computed label
308 @cindex address of a label
310 You can get the address of a label defined in the current function
311 (or a containing function) with the unary operator @samp{&&}. The
312 value has type @code{void *}. This value is a constant and can be used
313 wherever a constant of that type is valid. For example:
321 To use these values, you need to be able to jump to one. This is done
322 with the computed goto statement@footnote{The analogous feature in
323 Fortran is called an assigned goto, but that name seems inappropriate in
324 C, where one can do more than simply store label addresses in label
325 variables.}, @code{goto *@var{exp};}. For example,
332 Any expression of type @code{void *} is allowed.
334 One way of using these constants is in initializing a static array that
335 will serve as a jump table:
338 static void *array[] = @{ &&foo, &&bar, &&hack @};
341 Then you can select a label with indexing, like this:
348 Note that this does not check whether the subscript is in bounds---array
349 indexing in C never does that.
351 Such an array of label values serves a purpose much like that of the
352 @code{switch} statement. The @code{switch} statement is cleaner, so
353 use that rather than an array unless the problem does not fit a
354 @code{switch} statement very well.
356 Another use of label values is in an interpreter for threaded code.
357 The labels within the interpreter function can be stored in the
358 threaded code for super-fast dispatching.
360 You may not use this mechanism to jump to code in a different function.
361 If you do that, totally unpredictable things will happen. The best way to
362 avoid this is to store the label address only in automatic variables and
363 never pass it as an argument.
365 An alternate way to write the above example is
368 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
370 goto *(&&foo + array[i]);
374 This is more friendly to code living in shared libraries, as it reduces
375 the number of dynamic relocations that are needed, and by consequence,
376 allows the data to be read-only.
378 The @code{&&foo} expressions for the same label might have different
379 values if the containing function is inlined or cloned. If a program
380 relies on them being always the same,
381 @code{__attribute__((__noinline__,__noclone__))} should be used to
382 prevent inlining and cloning. If @code{&&foo} is used in a static
383 variable initializer, inlining and cloning is forbidden.
385 @node Nested Functions
386 @section Nested Functions
387 @cindex nested functions
388 @cindex downward funargs
391 A @dfn{nested function} is a function defined inside another function.
392 (Nested functions are not supported for GNU C++.) The nested function's
393 name is local to the block where it is defined. For example, here we
394 define a nested function named @code{square}, and call it twice:
398 foo (double a, double b)
400 double square (double z) @{ return z * z; @}
402 return square (a) + square (b);
407 The nested function can access all the variables of the containing
408 function that are visible at the point of its definition. This is
409 called @dfn{lexical scoping}. For example, here we show a nested
410 function which uses an inherited variable named @code{offset}:
414 bar (int *array, int offset, int size)
416 int access (int *array, int index)
417 @{ return array[index + offset]; @}
420 for (i = 0; i < size; i++)
421 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
426 Nested function definitions are permitted within functions in the places
427 where variable definitions are allowed; that is, in any block, mixed
428 with the other declarations and statements in the block.
430 It is possible to call the nested function from outside the scope of its
431 name by storing its address or passing the address to another function:
434 hack (int *array, int size)
436 void store (int index, int value)
437 @{ array[index] = value; @}
439 intermediate (store, size);
443 Here, the function @code{intermediate} receives the address of
444 @code{store} as an argument. If @code{intermediate} calls @code{store},
445 the arguments given to @code{store} are used to store into @code{array}.
446 But this technique works only so long as the containing function
447 (@code{hack}, in this example) does not exit.
449 If you try to call the nested function through its address after the
450 containing function has exited, all hell will break loose. If you try
451 to call it after a containing scope level has exited, and if it refers
452 to some of the variables that are no longer in scope, you may be lucky,
453 but it's not wise to take the risk. If, however, the nested function
454 does not refer to anything that has gone out of scope, you should be
457 GCC implements taking the address of a nested function using a technique
458 called @dfn{trampolines}. This technique was described in
459 @cite{Lexical Closures for C++} (Thomas M. Breuel, USENIX
460 C++ Conference Proceedings, October 17-21, 1988).
462 A nested function can jump to a label inherited from a containing
463 function, provided the label was explicitly declared in the containing
464 function (@pxref{Local Labels}). Such a jump returns instantly to the
465 containing function, exiting the nested function which did the
466 @code{goto} and any intermediate functions as well. Here is an example:
470 bar (int *array, int offset, int size)
473 int access (int *array, int index)
477 return array[index + offset];
481 for (i = 0; i < size; i++)
482 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
486 /* @r{Control comes here from @code{access}
487 if it detects an error.} */
494 A nested function always has no linkage. Declaring one with
495 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
496 before its definition, use @code{auto} (which is otherwise meaningless
497 for function declarations).
500 bar (int *array, int offset, int size)
503 auto int access (int *, int);
505 int access (int *array, int index)
509 return array[index + offset];
515 @node Constructing Calls
516 @section Constructing Function Calls
517 @cindex constructing calls
518 @cindex forwarding calls
520 Using the built-in functions described below, you can record
521 the arguments a function received, and call another function
522 with the same arguments, without knowing the number or types
525 You can also record the return value of that function call,
526 and later return that value, without knowing what data type
527 the function tried to return (as long as your caller expects
530 However, these built-in functions may interact badly with some
531 sophisticated features or other extensions of the language. It
532 is, therefore, not recommended to use them outside very simple
533 functions acting as mere forwarders for their arguments.
535 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
536 This built-in function returns a pointer to data
537 describing how to perform a call with the same arguments as were passed
538 to the current function.
540 The function saves the arg pointer register, structure value address,
541 and all registers that might be used to pass arguments to a function
542 into a block of memory allocated on the stack. Then it returns the
543 address of that block.
546 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
547 This built-in function invokes @var{function}
548 with a copy of the parameters described by @var{arguments}
551 The value of @var{arguments} should be the value returned by
552 @code{__builtin_apply_args}. The argument @var{size} specifies the size
553 of the stack argument data, in bytes.
555 This function returns a pointer to data describing
556 how to return whatever value was returned by @var{function}. The data
557 is saved in a block of memory allocated on the stack.
559 It is not always simple to compute the proper value for @var{size}. The
560 value is used by @code{__builtin_apply} to compute the amount of data
561 that should be pushed on the stack and copied from the incoming argument
565 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
566 This built-in function returns the value described by @var{result} from
567 the containing function. You should specify, for @var{result}, a value
568 returned by @code{__builtin_apply}.
571 @deftypefn {Built-in Function} {} __builtin_va_arg_pack ()
572 This built-in function represents all anonymous arguments of an inline
573 function. It can be used only in inline functions which will be always
574 inlined, never compiled as a separate function, such as those using
575 @code{__attribute__ ((__always_inline__))} or
576 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
577 It must be only passed as last argument to some other function
578 with variable arguments. This is useful for writing small wrapper
579 inlines for variable argument functions, when using preprocessor
580 macros is undesirable. For example:
582 extern int myprintf (FILE *f, const char *format, ...);
583 extern inline __attribute__ ((__gnu_inline__)) int
584 myprintf (FILE *f, const char *format, ...)
586 int r = fprintf (f, "myprintf: ");
589 int s = fprintf (f, format, __builtin_va_arg_pack ());
597 @deftypefn {Built-in Function} {size_t} __builtin_va_arg_pack_len ()
598 This built-in function returns the number of anonymous arguments of
599 an inline function. It can be used only in inline functions which
600 will be always inlined, never compiled as a separate function, such
601 as those using @code{__attribute__ ((__always_inline__))} or
602 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
603 For example following will do link or runtime checking of open
604 arguments for optimized code:
607 extern inline __attribute__((__gnu_inline__)) int
608 myopen (const char *path, int oflag, ...)
610 if (__builtin_va_arg_pack_len () > 1)
611 warn_open_too_many_arguments ();
613 if (__builtin_constant_p (oflag))
615 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1)
617 warn_open_missing_mode ();
618 return __open_2 (path, oflag);
620 return open (path, oflag, __builtin_va_arg_pack ());
623 if (__builtin_va_arg_pack_len () < 1)
624 return __open_2 (path, oflag);
626 return open (path, oflag, __builtin_va_arg_pack ());
633 @section Referring to a Type with @code{typeof}
636 @cindex macros, types of arguments
638 Another way to refer to the type of an expression is with @code{typeof}.
639 The syntax of using of this keyword looks like @code{sizeof}, but the
640 construct acts semantically like a type name defined with @code{typedef}.
642 There are two ways of writing the argument to @code{typeof}: with an
643 expression or with a type. Here is an example with an expression:
650 This assumes that @code{x} is an array of pointers to functions;
651 the type described is that of the values of the functions.
653 Here is an example with a typename as the argument:
660 Here the type described is that of pointers to @code{int}.
662 If you are writing a header file that must work when included in ISO C
663 programs, write @code{__typeof__} instead of @code{typeof}.
664 @xref{Alternate Keywords}.
666 A @code{typeof}-construct can be used anywhere a typedef name could be
667 used. For example, you can use it in a declaration, in a cast, or inside
668 of @code{sizeof} or @code{typeof}.
670 The operand of @code{typeof} is evaluated for its side effects if and
671 only if it is an expression of variably modified type or the name of
674 @code{typeof} is often useful in conjunction with the
675 statements-within-expressions feature. Here is how the two together can
676 be used to define a safe ``maximum'' macro that operates on any
677 arithmetic type and evaluates each of its arguments exactly once:
681 (@{ typeof (a) _a = (a); \
682 typeof (b) _b = (b); \
683 _a > _b ? _a : _b; @})
686 @cindex underscores in variables in macros
687 @cindex @samp{_} in variables in macros
688 @cindex local variables in macros
689 @cindex variables, local, in macros
690 @cindex macros, local variables in
692 The reason for using names that start with underscores for the local
693 variables is to avoid conflicts with variable names that occur within the
694 expressions that are substituted for @code{a} and @code{b}. Eventually we
695 hope to design a new form of declaration syntax that allows you to declare
696 variables whose scopes start only after their initializers; this will be a
697 more reliable way to prevent such conflicts.
700 Some more examples of the use of @code{typeof}:
704 This declares @code{y} with the type of what @code{x} points to.
711 This declares @code{y} as an array of such values.
718 This declares @code{y} as an array of pointers to characters:
721 typeof (typeof (char *)[4]) y;
725 It is equivalent to the following traditional C declaration:
731 To see the meaning of the declaration using @code{typeof}, and why it
732 might be a useful way to write, rewrite it with these macros:
735 #define pointer(T) typeof(T *)
736 #define array(T, N) typeof(T [N])
740 Now the declaration can be rewritten this way:
743 array (pointer (char), 4) y;
747 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
748 pointers to @code{char}.
751 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
752 a more limited extension which permitted one to write
755 typedef @var{T} = @var{expr};
759 with the effect of declaring @var{T} to have the type of the expression
760 @var{expr}. This extension does not work with GCC 3 (versions between
761 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
762 relies on it should be rewritten to use @code{typeof}:
765 typedef typeof(@var{expr}) @var{T};
769 This will work with all versions of GCC@.
772 @section Conditionals with Omitted Operands
773 @cindex conditional expressions, extensions
774 @cindex omitted middle-operands
775 @cindex middle-operands, omitted
776 @cindex extensions, @code{?:}
777 @cindex @code{?:} extensions
779 The middle operand in a conditional expression may be omitted. Then
780 if the first operand is nonzero, its value is the value of the conditional
783 Therefore, the expression
790 has the value of @code{x} if that is nonzero; otherwise, the value of
793 This example is perfectly equivalent to
799 @cindex side effect in @code{?:}
800 @cindex @code{?:} side effect
802 In this simple case, the ability to omit the middle operand is not
803 especially useful. When it becomes useful is when the first operand does,
804 or may (if it is a macro argument), contain a side effect. Then repeating
805 the operand in the middle would perform the side effect twice. Omitting
806 the middle operand uses the value already computed without the undesirable
807 effects of recomputing it.
810 @section 128-bits integers
811 @cindex @code{__int128} data types
813 As an extension the integer scalar type @code{__int128} is supported for
814 targets having an integer mode wide enough to hold 128-bit.
815 Simply write @code{__int128} for a signed 128-bit integer, or
816 @code{unsigned __int128} for an unsigned 128-bit integer. There is no
817 support in GCC to express an integer constant of type @code{__int128}
818 for targets having @code{long long} integer with less then 128 bit width.
821 @section Double-Word Integers
822 @cindex @code{long long} data types
823 @cindex double-word arithmetic
824 @cindex multiprecision arithmetic
825 @cindex @code{LL} integer suffix
826 @cindex @code{ULL} integer suffix
828 ISO C99 supports data types for integers that are at least 64 bits wide,
829 and as an extension GCC supports them in C90 mode and in C++.
830 Simply write @code{long long int} for a signed integer, or
831 @code{unsigned long long int} for an unsigned integer. To make an
832 integer constant of type @code{long long int}, add the suffix @samp{LL}
833 to the integer. To make an integer constant of type @code{unsigned long
834 long int}, add the suffix @samp{ULL} to the integer.
836 You can use these types in arithmetic like any other integer types.
837 Addition, subtraction, and bitwise boolean operations on these types
838 are open-coded on all types of machines. Multiplication is open-coded
839 if the machine supports fullword-to-doubleword a widening multiply
840 instruction. Division and shifts are open-coded only on machines that
841 provide special support. The operations that are not open-coded use
842 special library routines that come with GCC@.
844 There may be pitfalls when you use @code{long long} types for function
845 arguments, unless you declare function prototypes. If a function
846 expects type @code{int} for its argument, and you pass a value of type
847 @code{long long int}, confusion will result because the caller and the
848 subroutine will disagree about the number of bytes for the argument.
849 Likewise, if the function expects @code{long long int} and you pass
850 @code{int}. The best way to avoid such problems is to use prototypes.
853 @section Complex Numbers
854 @cindex complex numbers
855 @cindex @code{_Complex} keyword
856 @cindex @code{__complex__} keyword
858 ISO C99 supports complex floating data types, and as an extension GCC
859 supports them in C90 mode and in C++, and supports complex integer data
860 types which are not part of ISO C99. You can declare complex types
861 using the keyword @code{_Complex}. As an extension, the older GNU
862 keyword @code{__complex__} is also supported.
864 For example, @samp{_Complex double x;} declares @code{x} as a
865 variable whose real part and imaginary part are both of type
866 @code{double}. @samp{_Complex short int y;} declares @code{y} to
867 have real and imaginary parts of type @code{short int}; this is not
868 likely to be useful, but it shows that the set of complex types is
871 To write a constant with a complex data type, use the suffix @samp{i} or
872 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
873 has type @code{_Complex float} and @code{3i} has type
874 @code{_Complex int}. Such a constant always has a pure imaginary
875 value, but you can form any complex value you like by adding one to a
876 real constant. This is a GNU extension; if you have an ISO C99
877 conforming C library (such as GNU libc), and want to construct complex
878 constants of floating type, you should include @code{<complex.h>} and
879 use the macros @code{I} or @code{_Complex_I} instead.
881 @cindex @code{__real__} keyword
882 @cindex @code{__imag__} keyword
883 To extract the real part of a complex-valued expression @var{exp}, write
884 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
885 extract the imaginary part. This is a GNU extension; for values of
886 floating type, you should use the ISO C99 functions @code{crealf},
887 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
888 @code{cimagl}, declared in @code{<complex.h>} and also provided as
889 built-in functions by GCC@.
891 @cindex complex conjugation
892 The operator @samp{~} performs complex conjugation when used on a value
893 with a complex type. This is a GNU extension; for values of
894 floating type, you should use the ISO C99 functions @code{conjf},
895 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
896 provided as built-in functions by GCC@.
898 GCC can allocate complex automatic variables in a noncontiguous
899 fashion; it's even possible for the real part to be in a register while
900 the imaginary part is on the stack (or vice-versa). Only the DWARF2
901 debug info format can represent this, so use of DWARF2 is recommended.
902 If you are using the stabs debug info format, GCC describes a noncontiguous
903 complex variable as if it were two separate variables of noncomplex type.
904 If the variable's actual name is @code{foo}, the two fictitious
905 variables are named @code{foo$real} and @code{foo$imag}. You can
906 examine and set these two fictitious variables with your debugger.
909 @section Additional Floating Types
910 @cindex additional floating types
911 @cindex @code{__float80} data type
912 @cindex @code{__float128} data type
913 @cindex @code{w} floating point suffix
914 @cindex @code{q} floating point suffix
915 @cindex @code{W} floating point suffix
916 @cindex @code{Q} floating point suffix
918 As an extension, the GNU C compiler supports additional floating
919 types, @code{__float80} and @code{__float128} to support 80bit
920 (@code{XFmode}) and 128 bit (@code{TFmode}) floating types.
921 Support for additional types includes the arithmetic operators:
922 add, subtract, multiply, divide; unary arithmetic operators;
923 relational operators; equality operators; and conversions to and from
924 integer and other floating types. Use a suffix @samp{w} or @samp{W}
925 in a literal constant of type @code{__float80} and @samp{q} or @samp{Q}
926 for @code{_float128}. You can declare complex types using the
927 corresponding internal complex type, @code{XCmode} for @code{__float80}
928 type and @code{TCmode} for @code{__float128} type:
931 typedef _Complex float __attribute__((mode(TC))) _Complex128;
932 typedef _Complex float __attribute__((mode(XC))) _Complex80;
935 Not all targets support additional floating point types. @code{__float80}
936 and @code{__float128} types are supported on i386, x86_64 and ia64 targets.
939 @section Half-Precision Floating Point
940 @cindex half-precision floating point
941 @cindex @code{__fp16} data type
943 On ARM targets, GCC supports half-precision (16-bit) floating point via
944 the @code{__fp16} type. You must enable this type explicitly
945 with the @option{-mfp16-format} command-line option in order to use it.
947 ARM supports two incompatible representations for half-precision
948 floating-point values. You must choose one of the representations and
949 use it consistently in your program.
951 Specifying @option{-mfp16-format=ieee} selects the IEEE 754-2008 format.
952 This format can represent normalized values in the range of @math{2^{-14}} to 65504.
953 There are 11 bits of significand precision, approximately 3
956 Specifying @option{-mfp16-format=alternative} selects the ARM
957 alternative format. This representation is similar to the IEEE
958 format, but does not support infinities or NaNs. Instead, the range
959 of exponents is extended, so that this format can represent normalized
960 values in the range of @math{2^{-14}} to 131008.
962 The @code{__fp16} type is a storage format only. For purposes
963 of arithmetic and other operations, @code{__fp16} values in C or C++
964 expressions are automatically promoted to @code{float}. In addition,
965 you cannot declare a function with a return value or parameters
966 of type @code{__fp16}.
968 Note that conversions from @code{double} to @code{__fp16}
969 involve an intermediate conversion to @code{float}. Because
970 of rounding, this can sometimes produce a different result than a
973 ARM provides hardware support for conversions between
974 @code{__fp16} and @code{float} values
975 as an extension to VFP and NEON (Advanced SIMD). GCC generates
976 code using these hardware instructions if you compile with
977 options to select an FPU that provides them;
978 for example, @option{-mfpu=neon-fp16 -mfloat-abi=softfp},
979 in addition to the @option{-mfp16-format} option to select
980 a half-precision format.
982 Language-level support for the @code{__fp16} data type is
983 independent of whether GCC generates code using hardware floating-point
984 instructions. In cases where hardware support is not specified, GCC
985 implements conversions between @code{__fp16} and @code{float} values
989 @section Decimal Floating Types
990 @cindex decimal floating types
991 @cindex @code{_Decimal32} data type
992 @cindex @code{_Decimal64} data type
993 @cindex @code{_Decimal128} data type
994 @cindex @code{df} integer suffix
995 @cindex @code{dd} integer suffix
996 @cindex @code{dl} integer suffix
997 @cindex @code{DF} integer suffix
998 @cindex @code{DD} integer suffix
999 @cindex @code{DL} integer suffix
1001 As an extension, the GNU C compiler supports decimal floating types as
1002 defined in the N1312 draft of ISO/IEC WDTR24732. Support for decimal
1003 floating types in GCC will evolve as the draft technical report changes.
1004 Calling conventions for any target might also change. Not all targets
1005 support decimal floating types.
1007 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
1008 @code{_Decimal128}. They use a radix of ten, unlike the floating types
1009 @code{float}, @code{double}, and @code{long double} whose radix is not
1010 specified by the C standard but is usually two.
1012 Support for decimal floating types includes the arithmetic operators
1013 add, subtract, multiply, divide; unary arithmetic operators;
1014 relational operators; equality operators; and conversions to and from
1015 integer and other floating types. Use a suffix @samp{df} or
1016 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
1017 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
1020 GCC support of decimal float as specified by the draft technical report
1025 When the value of a decimal floating type cannot be represented in the
1026 integer type to which it is being converted, the result is undefined
1027 rather than the result value specified by the draft technical report.
1030 GCC does not provide the C library functionality associated with
1031 @file{math.h}, @file{fenv.h}, @file{stdio.h}, @file{stdlib.h}, and
1032 @file{wchar.h}, which must come from a separate C library implementation.
1033 Because of this the GNU C compiler does not define macro
1034 @code{__STDC_DEC_FP__} to indicate that the implementation conforms to
1035 the technical report.
1038 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
1039 are supported by the DWARF2 debug information format.
1045 ISO C99 supports floating-point numbers written not only in the usual
1046 decimal notation, such as @code{1.55e1}, but also numbers such as
1047 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1048 supports this in C90 mode (except in some cases when strictly
1049 conforming) and in C++. In that format the
1050 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1051 mandatory. The exponent is a decimal number that indicates the power of
1052 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1059 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1060 is the same as @code{1.55e1}.
1062 Unlike for floating-point numbers in the decimal notation the exponent
1063 is always required in the hexadecimal notation. Otherwise the compiler
1064 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1065 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1066 extension for floating-point constants of type @code{float}.
1069 @section Fixed-Point Types
1070 @cindex fixed-point types
1071 @cindex @code{_Fract} data type
1072 @cindex @code{_Accum} data type
1073 @cindex @code{_Sat} data type
1074 @cindex @code{hr} fixed-suffix
1075 @cindex @code{r} fixed-suffix
1076 @cindex @code{lr} fixed-suffix
1077 @cindex @code{llr} fixed-suffix
1078 @cindex @code{uhr} fixed-suffix
1079 @cindex @code{ur} fixed-suffix
1080 @cindex @code{ulr} fixed-suffix
1081 @cindex @code{ullr} fixed-suffix
1082 @cindex @code{hk} fixed-suffix
1083 @cindex @code{k} fixed-suffix
1084 @cindex @code{lk} fixed-suffix
1085 @cindex @code{llk} fixed-suffix
1086 @cindex @code{uhk} fixed-suffix
1087 @cindex @code{uk} fixed-suffix
1088 @cindex @code{ulk} fixed-suffix
1089 @cindex @code{ullk} fixed-suffix
1090 @cindex @code{HR} fixed-suffix
1091 @cindex @code{R} fixed-suffix
1092 @cindex @code{LR} fixed-suffix
1093 @cindex @code{LLR} fixed-suffix
1094 @cindex @code{UHR} fixed-suffix
1095 @cindex @code{UR} fixed-suffix
1096 @cindex @code{ULR} fixed-suffix
1097 @cindex @code{ULLR} fixed-suffix
1098 @cindex @code{HK} fixed-suffix
1099 @cindex @code{K} fixed-suffix
1100 @cindex @code{LK} fixed-suffix
1101 @cindex @code{LLK} fixed-suffix
1102 @cindex @code{UHK} fixed-suffix
1103 @cindex @code{UK} fixed-suffix
1104 @cindex @code{ULK} fixed-suffix
1105 @cindex @code{ULLK} fixed-suffix
1107 As an extension, the GNU C compiler supports fixed-point types as
1108 defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point
1109 types in GCC will evolve as the draft technical report changes.
1110 Calling conventions for any target might also change. Not all targets
1111 support fixed-point types.
1113 The fixed-point types are
1114 @code{short _Fract},
1117 @code{long long _Fract},
1118 @code{unsigned short _Fract},
1119 @code{unsigned _Fract},
1120 @code{unsigned long _Fract},
1121 @code{unsigned long long _Fract},
1122 @code{_Sat short _Fract},
1124 @code{_Sat long _Fract},
1125 @code{_Sat long long _Fract},
1126 @code{_Sat unsigned short _Fract},
1127 @code{_Sat unsigned _Fract},
1128 @code{_Sat unsigned long _Fract},
1129 @code{_Sat unsigned long long _Fract},
1130 @code{short _Accum},
1133 @code{long long _Accum},
1134 @code{unsigned short _Accum},
1135 @code{unsigned _Accum},
1136 @code{unsigned long _Accum},
1137 @code{unsigned long long _Accum},
1138 @code{_Sat short _Accum},
1140 @code{_Sat long _Accum},
1141 @code{_Sat long long _Accum},
1142 @code{_Sat unsigned short _Accum},
1143 @code{_Sat unsigned _Accum},
1144 @code{_Sat unsigned long _Accum},
1145 @code{_Sat unsigned long long _Accum}.
1147 Fixed-point data values contain fractional and optional integral parts.
1148 The format of fixed-point data varies and depends on the target machine.
1150 Support for fixed-point types includes:
1153 prefix and postfix increment and decrement operators (@code{++}, @code{--})
1155 unary arithmetic operators (@code{+}, @code{-}, @code{!})
1157 binary arithmetic operators (@code{+}, @code{-}, @code{*}, @code{/})
1159 binary shift operators (@code{<<}, @code{>>})
1161 relational operators (@code{<}, @code{<=}, @code{>=}, @code{>})
1163 equality operators (@code{==}, @code{!=})
1165 assignment operators (@code{+=}, @code{-=}, @code{*=}, @code{/=},
1166 @code{<<=}, @code{>>=})
1168 conversions to and from integer, floating-point, or fixed-point types
1171 Use a suffix in a fixed-point literal constant:
1173 @item @samp{hr} or @samp{HR} for @code{short _Fract} and
1174 @code{_Sat short _Fract}
1175 @item @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract}
1176 @item @samp{lr} or @samp{LR} for @code{long _Fract} and
1177 @code{_Sat long _Fract}
1178 @item @samp{llr} or @samp{LLR} for @code{long long _Fract} and
1179 @code{_Sat long long _Fract}
1180 @item @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and
1181 @code{_Sat unsigned short _Fract}
1182 @item @samp{ur} or @samp{UR} for @code{unsigned _Fract} and
1183 @code{_Sat unsigned _Fract}
1184 @item @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and
1185 @code{_Sat unsigned long _Fract}
1186 @item @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract}
1187 and @code{_Sat unsigned long long _Fract}
1188 @item @samp{hk} or @samp{HK} for @code{short _Accum} and
1189 @code{_Sat short _Accum}
1190 @item @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum}
1191 @item @samp{lk} or @samp{LK} for @code{long _Accum} and
1192 @code{_Sat long _Accum}
1193 @item @samp{llk} or @samp{LLK} for @code{long long _Accum} and
1194 @code{_Sat long long _Accum}
1195 @item @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and
1196 @code{_Sat unsigned short _Accum}
1197 @item @samp{uk} or @samp{UK} for @code{unsigned _Accum} and
1198 @code{_Sat unsigned _Accum}
1199 @item @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and
1200 @code{_Sat unsigned long _Accum}
1201 @item @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum}
1202 and @code{_Sat unsigned long long _Accum}
1205 GCC support of fixed-point types as specified by the draft technical report
1210 Pragmas to control overflow and rounding behaviors are not implemented.
1213 Fixed-point types are supported by the DWARF2 debug information format.
1215 @node Named Address Spaces
1216 @section Named address spaces
1217 @cindex named address spaces
1219 As an extension, the GNU C compiler supports named address spaces as
1220 defined in the N1275 draft of ISO/IEC DTR 18037. Support for named
1221 address spaces in GCC will evolve as the draft technical report changes.
1222 Calling conventions for any target might also change. At present, only
1223 the SPU and M32C targets support other address spaces. On the SPU target, for
1224 example, variables may be declared as belonging to another address space
1225 by qualifying the type with the @code{__ea} address space identifier:
1231 When the variable @code{i} is accessed, the compiler will generate
1232 special code to access this variable. It may use runtime library
1233 support, or generate special machine instructions to access that address
1236 The @code{__ea} identifier may be used exactly like any other C type
1237 qualifier (e.g., @code{const} or @code{volatile}). See the N1275
1238 document for more details.
1240 On the M32C target, with the R8C and M16C cpu variants, variables
1241 qualified with @code{__far} are accessed using 32-bit addresses in
1242 order to access memory beyond the first 64k bytes. If @code{__far} is
1243 used with the M32CM or M32C cpu variants, it has no effect.
1246 @section Arrays of Length Zero
1247 @cindex arrays of length zero
1248 @cindex zero-length arrays
1249 @cindex length-zero arrays
1250 @cindex flexible array members
1252 Zero-length arrays are allowed in GNU C@. They are very useful as the
1253 last element of a structure which is really a header for a variable-length
1262 struct line *thisline = (struct line *)
1263 malloc (sizeof (struct line) + this_length);
1264 thisline->length = this_length;
1267 In ISO C90, you would have to give @code{contents} a length of 1, which
1268 means either you waste space or complicate the argument to @code{malloc}.
1270 In ISO C99, you would use a @dfn{flexible array member}, which is
1271 slightly different in syntax and semantics:
1275 Flexible array members are written as @code{contents[]} without
1279 Flexible array members have incomplete type, and so the @code{sizeof}
1280 operator may not be applied. As a quirk of the original implementation
1281 of zero-length arrays, @code{sizeof} evaluates to zero.
1284 Flexible array members may only appear as the last member of a
1285 @code{struct} that is otherwise non-empty.
1288 A structure containing a flexible array member, or a union containing
1289 such a structure (possibly recursively), may not be a member of a
1290 structure or an element of an array. (However, these uses are
1291 permitted by GCC as extensions.)
1294 GCC versions before 3.0 allowed zero-length arrays to be statically
1295 initialized, as if they were flexible arrays. In addition to those
1296 cases that were useful, it also allowed initializations in situations
1297 that would corrupt later data. Non-empty initialization of zero-length
1298 arrays is now treated like any case where there are more initializer
1299 elements than the array holds, in that a suitable warning about "excess
1300 elements in array" is given, and the excess elements (all of them, in
1301 this case) are ignored.
1303 Instead GCC allows static initialization of flexible array members.
1304 This is equivalent to defining a new structure containing the original
1305 structure followed by an array of sufficient size to contain the data.
1306 I.e.@: in the following, @code{f1} is constructed as if it were declared
1312 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1315 struct f1 f1; int data[3];
1316 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1320 The convenience of this extension is that @code{f1} has the desired
1321 type, eliminating the need to consistently refer to @code{f2.f1}.
1323 This has symmetry with normal static arrays, in that an array of
1324 unknown size is also written with @code{[]}.
1326 Of course, this extension only makes sense if the extra data comes at
1327 the end of a top-level object, as otherwise we would be overwriting
1328 data at subsequent offsets. To avoid undue complication and confusion
1329 with initialization of deeply nested arrays, we simply disallow any
1330 non-empty initialization except when the structure is the top-level
1331 object. For example:
1334 struct foo @{ int x; int y[]; @};
1335 struct bar @{ struct foo z; @};
1337 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1338 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1339 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1340 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1343 @node Empty Structures
1344 @section Structures With No Members
1345 @cindex empty structures
1346 @cindex zero-size structures
1348 GCC permits a C structure to have no members:
1355 The structure will have size zero. In C++, empty structures are part
1356 of the language. G++ treats empty structures as if they had a single
1357 member of type @code{char}.
1359 @node Variable Length
1360 @section Arrays of Variable Length
1361 @cindex variable-length arrays
1362 @cindex arrays of variable length
1365 Variable-length automatic arrays are allowed in ISO C99, and as an
1366 extension GCC accepts them in C90 mode and in C++. These arrays are
1367 declared like any other automatic arrays, but with a length that is not
1368 a constant expression. The storage is allocated at the point of
1369 declaration and deallocated when the brace-level is exited. For
1374 concat_fopen (char *s1, char *s2, char *mode)
1376 char str[strlen (s1) + strlen (s2) + 1];
1379 return fopen (str, mode);
1383 @cindex scope of a variable length array
1384 @cindex variable-length array scope
1385 @cindex deallocating variable length arrays
1386 Jumping or breaking out of the scope of the array name deallocates the
1387 storage. Jumping into the scope is not allowed; you get an error
1390 @cindex @code{alloca} vs variable-length arrays
1391 You can use the function @code{alloca} to get an effect much like
1392 variable-length arrays. The function @code{alloca} is available in
1393 many other C implementations (but not in all). On the other hand,
1394 variable-length arrays are more elegant.
1396 There are other differences between these two methods. Space allocated
1397 with @code{alloca} exists until the containing @emph{function} returns.
1398 The space for a variable-length array is deallocated as soon as the array
1399 name's scope ends. (If you use both variable-length arrays and
1400 @code{alloca} in the same function, deallocation of a variable-length array
1401 will also deallocate anything more recently allocated with @code{alloca}.)
1403 You can also use variable-length arrays as arguments to functions:
1407 tester (int len, char data[len][len])
1413 The length of an array is computed once when the storage is allocated
1414 and is remembered for the scope of the array in case you access it with
1417 If you want to pass the array first and the length afterward, you can
1418 use a forward declaration in the parameter list---another GNU extension.
1422 tester (int len; char data[len][len], int len)
1428 @cindex parameter forward declaration
1429 The @samp{int len} before the semicolon is a @dfn{parameter forward
1430 declaration}, and it serves the purpose of making the name @code{len}
1431 known when the declaration of @code{data} is parsed.
1433 You can write any number of such parameter forward declarations in the
1434 parameter list. They can be separated by commas or semicolons, but the
1435 last one must end with a semicolon, which is followed by the ``real''
1436 parameter declarations. Each forward declaration must match a ``real''
1437 declaration in parameter name and data type. ISO C99 does not support
1438 parameter forward declarations.
1440 @node Variadic Macros
1441 @section Macros with a Variable Number of Arguments.
1442 @cindex variable number of arguments
1443 @cindex macro with variable arguments
1444 @cindex rest argument (in macro)
1445 @cindex variadic macros
1447 In the ISO C standard of 1999, a macro can be declared to accept a
1448 variable number of arguments much as a function can. The syntax for
1449 defining the macro is similar to that of a function. Here is an
1453 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1456 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1457 such a macro, it represents the zero or more tokens until the closing
1458 parenthesis that ends the invocation, including any commas. This set of
1459 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1460 wherever it appears. See the CPP manual for more information.
1462 GCC has long supported variadic macros, and used a different syntax that
1463 allowed you to give a name to the variable arguments just like any other
1464 argument. Here is an example:
1467 #define debug(format, args...) fprintf (stderr, format, args)
1470 This is in all ways equivalent to the ISO C example above, but arguably
1471 more readable and descriptive.
1473 GNU CPP has two further variadic macro extensions, and permits them to
1474 be used with either of the above forms of macro definition.
1476 In standard C, you are not allowed to leave the variable argument out
1477 entirely; but you are allowed to pass an empty argument. For example,
1478 this invocation is invalid in ISO C, because there is no comma after
1485 GNU CPP permits you to completely omit the variable arguments in this
1486 way. In the above examples, the compiler would complain, though since
1487 the expansion of the macro still has the extra comma after the format
1490 To help solve this problem, CPP behaves specially for variable arguments
1491 used with the token paste operator, @samp{##}. If instead you write
1494 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1497 and if the variable arguments are omitted or empty, the @samp{##}
1498 operator causes the preprocessor to remove the comma before it. If you
1499 do provide some variable arguments in your macro invocation, GNU CPP
1500 does not complain about the paste operation and instead places the
1501 variable arguments after the comma. Just like any other pasted macro
1502 argument, these arguments are not macro expanded.
1504 @node Escaped Newlines
1505 @section Slightly Looser Rules for Escaped Newlines
1506 @cindex escaped newlines
1507 @cindex newlines (escaped)
1509 Recently, the preprocessor has relaxed its treatment of escaped
1510 newlines. Previously, the newline had to immediately follow a
1511 backslash. The current implementation allows whitespace in the form
1512 of spaces, horizontal and vertical tabs, and form feeds between the
1513 backslash and the subsequent newline. The preprocessor issues a
1514 warning, but treats it as a valid escaped newline and combines the two
1515 lines to form a single logical line. This works within comments and
1516 tokens, as well as between tokens. Comments are @emph{not} treated as
1517 whitespace for the purposes of this relaxation, since they have not
1518 yet been replaced with spaces.
1521 @section Non-Lvalue Arrays May Have Subscripts
1522 @cindex subscripting
1523 @cindex arrays, non-lvalue
1525 @cindex subscripting and function values
1526 In ISO C99, arrays that are not lvalues still decay to pointers, and
1527 may be subscripted, although they may not be modified or used after
1528 the next sequence point and the unary @samp{&} operator may not be
1529 applied to them. As an extension, GCC allows such arrays to be
1530 subscripted in C90 mode, though otherwise they do not decay to
1531 pointers outside C99 mode. For example,
1532 this is valid in GNU C though not valid in C90:
1536 struct foo @{int a[4];@};
1542 return f().a[index];
1548 @section Arithmetic on @code{void}- and Function-Pointers
1549 @cindex void pointers, arithmetic
1550 @cindex void, size of pointer to
1551 @cindex function pointers, arithmetic
1552 @cindex function, size of pointer to
1554 In GNU C, addition and subtraction operations are supported on pointers to
1555 @code{void} and on pointers to functions. This is done by treating the
1556 size of a @code{void} or of a function as 1.
1558 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1559 and on function types, and returns 1.
1561 @opindex Wpointer-arith
1562 The option @option{-Wpointer-arith} requests a warning if these extensions
1566 @section Non-Constant Initializers
1567 @cindex initializers, non-constant
1568 @cindex non-constant initializers
1570 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1571 automatic variable are not required to be constant expressions in GNU C@.
1572 Here is an example of an initializer with run-time varying elements:
1575 foo (float f, float g)
1577 float beat_freqs[2] = @{ f-g, f+g @};
1582 @node Compound Literals
1583 @section Compound Literals
1584 @cindex constructor expressions
1585 @cindex initializations in expressions
1586 @cindex structures, constructor expression
1587 @cindex expressions, constructor
1588 @cindex compound literals
1589 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1591 ISO C99 supports compound literals. A compound literal looks like
1592 a cast containing an initializer. Its value is an object of the
1593 type specified in the cast, containing the elements specified in
1594 the initializer; it is an lvalue. As an extension, GCC supports
1595 compound literals in C90 mode and in C++.
1597 Usually, the specified type is a structure. Assume that
1598 @code{struct foo} and @code{structure} are declared as shown:
1601 struct foo @{int a; char b[2];@} structure;
1605 Here is an example of constructing a @code{struct foo} with a compound literal:
1608 structure = ((struct foo) @{x + y, 'a', 0@});
1612 This is equivalent to writing the following:
1616 struct foo temp = @{x + y, 'a', 0@};
1621 You can also construct an array. If all the elements of the compound literal
1622 are (made up of) simple constant expressions, suitable for use in
1623 initializers of objects of static storage duration, then the compound
1624 literal can be coerced to a pointer to its first element and used in
1625 such an initializer, as shown here:
1628 char **foo = (char *[]) @{ "x", "y", "z" @};
1631 Compound literals for scalar types and union types are is
1632 also allowed, but then the compound literal is equivalent
1635 As a GNU extension, GCC allows initialization of objects with static storage
1636 duration by compound literals (which is not possible in ISO C99, because
1637 the initializer is not a constant).
1638 It is handled as if the object was initialized only with the bracket
1639 enclosed list if the types of the compound literal and the object match.
1640 The initializer list of the compound literal must be constant.
1641 If the object being initialized has array type of unknown size, the size is
1642 determined by compound literal size.
1645 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1646 static int y[] = (int []) @{1, 2, 3@};
1647 static int z[] = (int [3]) @{1@};
1651 The above lines are equivalent to the following:
1653 static struct foo x = @{1, 'a', 'b'@};
1654 static int y[] = @{1, 2, 3@};
1655 static int z[] = @{1, 0, 0@};
1658 @node Designated Inits
1659 @section Designated Initializers
1660 @cindex initializers with labeled elements
1661 @cindex labeled elements in initializers
1662 @cindex case labels in initializers
1663 @cindex designated initializers
1665 Standard C90 requires the elements of an initializer to appear in a fixed
1666 order, the same as the order of the elements in the array or structure
1669 In ISO C99 you can give the elements in any order, specifying the array
1670 indices or structure field names they apply to, and GNU C allows this as
1671 an extension in C90 mode as well. This extension is not
1672 implemented in GNU C++.
1674 To specify an array index, write
1675 @samp{[@var{index}] =} before the element value. For example,
1678 int a[6] = @{ [4] = 29, [2] = 15 @};
1685 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1689 The index values must be constant expressions, even if the array being
1690 initialized is automatic.
1692 An alternative syntax for this which has been obsolete since GCC 2.5 but
1693 GCC still accepts is to write @samp{[@var{index}]} before the element
1694 value, with no @samp{=}.
1696 To initialize a range of elements to the same value, write
1697 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1698 extension. For example,
1701 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1705 If the value in it has side-effects, the side-effects will happen only once,
1706 not for each initialized field by the range initializer.
1709 Note that the length of the array is the highest value specified
1712 In a structure initializer, specify the name of a field to initialize
1713 with @samp{.@var{fieldname} =} before the element value. For example,
1714 given the following structure,
1717 struct point @{ int x, y; @};
1721 the following initialization
1724 struct point p = @{ .y = yvalue, .x = xvalue @};
1731 struct point p = @{ xvalue, yvalue @};
1734 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1735 @samp{@var{fieldname}:}, as shown here:
1738 struct point p = @{ y: yvalue, x: xvalue @};
1742 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1743 @dfn{designator}. You can also use a designator (or the obsolete colon
1744 syntax) when initializing a union, to specify which element of the union
1745 should be used. For example,
1748 union foo @{ int i; double d; @};
1750 union foo f = @{ .d = 4 @};
1754 will convert 4 to a @code{double} to store it in the union using
1755 the second element. By contrast, casting 4 to type @code{union foo}
1756 would store it into the union as the integer @code{i}, since it is
1757 an integer. (@xref{Cast to Union}.)
1759 You can combine this technique of naming elements with ordinary C
1760 initialization of successive elements. Each initializer element that
1761 does not have a designator applies to the next consecutive element of the
1762 array or structure. For example,
1765 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1772 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1775 Labeling the elements of an array initializer is especially useful
1776 when the indices are characters or belong to an @code{enum} type.
1781 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1782 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1785 @cindex designator lists
1786 You can also write a series of @samp{.@var{fieldname}} and
1787 @samp{[@var{index}]} designators before an @samp{=} to specify a
1788 nested subobject to initialize; the list is taken relative to the
1789 subobject corresponding to the closest surrounding brace pair. For
1790 example, with the @samp{struct point} declaration above:
1793 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1797 If the same field is initialized multiple times, it will have value from
1798 the last initialization. If any such overridden initialization has
1799 side-effect, it is unspecified whether the side-effect happens or not.
1800 Currently, GCC will discard them and issue a warning.
1803 @section Case Ranges
1805 @cindex ranges in case statements
1807 You can specify a range of consecutive values in a single @code{case} label,
1811 case @var{low} ... @var{high}:
1815 This has the same effect as the proper number of individual @code{case}
1816 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1818 This feature is especially useful for ranges of ASCII character codes:
1824 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1825 it may be parsed wrong when you use it with integer values. For example,
1840 @section Cast to a Union Type
1841 @cindex cast to a union
1842 @cindex union, casting to a
1844 A cast to union type is similar to other casts, except that the type
1845 specified is a union type. You can specify the type either with
1846 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1847 a constructor though, not a cast, and hence does not yield an lvalue like
1848 normal casts. (@xref{Compound Literals}.)
1850 The types that may be cast to the union type are those of the members
1851 of the union. Thus, given the following union and variables:
1854 union foo @{ int i; double d; @};
1860 both @code{x} and @code{y} can be cast to type @code{union foo}.
1862 Using the cast as the right-hand side of an assignment to a variable of
1863 union type is equivalent to storing in a member of the union:
1868 u = (union foo) x @equiv{} u.i = x
1869 u = (union foo) y @equiv{} u.d = y
1872 You can also use the union cast as a function argument:
1875 void hack (union foo);
1877 hack ((union foo) x);
1880 @node Mixed Declarations
1881 @section Mixed Declarations and Code
1882 @cindex mixed declarations and code
1883 @cindex declarations, mixed with code
1884 @cindex code, mixed with declarations
1886 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1887 within compound statements. As an extension, GCC also allows this in
1888 C90 mode. For example, you could do:
1897 Each identifier is visible from where it is declared until the end of
1898 the enclosing block.
1900 @node Function Attributes
1901 @section Declaring Attributes of Functions
1902 @cindex function attributes
1903 @cindex declaring attributes of functions
1904 @cindex functions that never return
1905 @cindex functions that return more than once
1906 @cindex functions that have no side effects
1907 @cindex functions in arbitrary sections
1908 @cindex functions that behave like malloc
1909 @cindex @code{volatile} applied to function
1910 @cindex @code{const} applied to function
1911 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1912 @cindex functions with non-null pointer arguments
1913 @cindex functions that are passed arguments in registers on the 386
1914 @cindex functions that pop the argument stack on the 386
1915 @cindex functions that do not pop the argument stack on the 386
1916 @cindex functions that have different compilation options on the 386
1917 @cindex functions that have different optimization options
1918 @cindex functions that are dynamically resolved
1920 In GNU C, you declare certain things about functions called in your program
1921 which help the compiler optimize function calls and check your code more
1924 The keyword @code{__attribute__} allows you to specify special
1925 attributes when making a declaration. This keyword is followed by an
1926 attribute specification inside double parentheses. The following
1927 attributes are currently defined for functions on all targets:
1928 @code{aligned}, @code{alloc_size}, @code{noreturn},
1929 @code{returns_twice}, @code{noinline}, @code{noclone},
1930 @code{always_inline}, @code{flatten}, @code{pure}, @code{const},
1931 @code{nothrow}, @code{sentinel}, @code{format}, @code{format_arg},
1932 @code{no_instrument_function}, @code{no_split_stack},
1933 @code{section}, @code{constructor},
1934 @code{destructor}, @code{used}, @code{unused}, @code{deprecated},
1935 @code{weak}, @code{malloc}, @code{alias}, @code{ifunc},
1936 @code{warn_unused_result}, @code{nonnull}, @code{gnu_inline},
1937 @code{externally_visible}, @code{hot}, @code{cold}, @code{artificial},
1938 @code{error} and @code{warning}. Several other attributes are defined
1939 for functions on particular target systems. Other attributes,
1940 including @code{section} are supported for variables declarations
1941 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1943 GCC plugins may provide their own attributes.
1945 You may also specify attributes with @samp{__} preceding and following
1946 each keyword. This allows you to use them in header files without
1947 being concerned about a possible macro of the same name. For example,
1948 you may use @code{__noreturn__} instead of @code{noreturn}.
1950 @xref{Attribute Syntax}, for details of the exact syntax for using
1954 @c Keep this table alphabetized by attribute name. Treat _ as space.
1956 @item alias ("@var{target}")
1957 @cindex @code{alias} attribute
1958 The @code{alias} attribute causes the declaration to be emitted as an
1959 alias for another symbol, which must be specified. For instance,
1962 void __f () @{ /* @r{Do something.} */; @}
1963 void f () __attribute__ ((weak, alias ("__f")));
1966 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
1967 mangled name for the target must be used. It is an error if @samp{__f}
1968 is not defined in the same translation unit.
1970 Not all target machines support this attribute.
1972 @item aligned (@var{alignment})
1973 @cindex @code{aligned} attribute
1974 This attribute specifies a minimum alignment for the function,
1977 You cannot use this attribute to decrease the alignment of a function,
1978 only to increase it. However, when you explicitly specify a function
1979 alignment this will override the effect of the
1980 @option{-falign-functions} (@pxref{Optimize Options}) option for this
1983 Note that the effectiveness of @code{aligned} attributes may be
1984 limited by inherent limitations in your linker. On many systems, the
1985 linker is only able to arrange for functions to be aligned up to a
1986 certain maximum alignment. (For some linkers, the maximum supported
1987 alignment may be very very small.) See your linker documentation for
1988 further information.
1990 The @code{aligned} attribute can also be used for variables and fields
1991 (@pxref{Variable Attributes}.)
1994 @cindex @code{alloc_size} attribute
1995 The @code{alloc_size} attribute is used to tell the compiler that the
1996 function return value points to memory, where the size is given by
1997 one or two of the functions parameters. GCC uses this
1998 information to improve the correctness of @code{__builtin_object_size}.
2000 The function parameter(s) denoting the allocated size are specified by
2001 one or two integer arguments supplied to the attribute. The allocated size
2002 is either the value of the single function argument specified or the product
2003 of the two function arguments specified. Argument numbering starts at
2009 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
2010 void my_realloc(void*, size_t) __attribute__((alloc_size(2)))
2013 declares that my_calloc will return memory of the size given by
2014 the product of parameter 1 and 2 and that my_realloc will return memory
2015 of the size given by parameter 2.
2018 @cindex @code{always_inline} function attribute
2019 Generally, functions are not inlined unless optimization is specified.
2020 For functions declared inline, this attribute inlines the function even
2021 if no optimization level was specified.
2024 @cindex @code{gnu_inline} function attribute
2025 This attribute should be used with a function which is also declared
2026 with the @code{inline} keyword. It directs GCC to treat the function
2027 as if it were defined in gnu90 mode even when compiling in C99 or
2030 If the function is declared @code{extern}, then this definition of the
2031 function is used only for inlining. In no case is the function
2032 compiled as a standalone function, not even if you take its address
2033 explicitly. Such an address becomes an external reference, as if you
2034 had only declared the function, and had not defined it. This has
2035 almost the effect of a macro. The way to use this is to put a
2036 function definition in a header file with this attribute, and put
2037 another copy of the function, without @code{extern}, in a library
2038 file. The definition in the header file will cause most calls to the
2039 function to be inlined. If any uses of the function remain, they will
2040 refer to the single copy in the library. Note that the two
2041 definitions of the functions need not be precisely the same, although
2042 if they do not have the same effect your program may behave oddly.
2044 In C, if the function is neither @code{extern} nor @code{static}, then
2045 the function is compiled as a standalone function, as well as being
2046 inlined where possible.
2048 This is how GCC traditionally handled functions declared
2049 @code{inline}. Since ISO C99 specifies a different semantics for
2050 @code{inline}, this function attribute is provided as a transition
2051 measure and as a useful feature in its own right. This attribute is
2052 available in GCC 4.1.3 and later. It is available if either of the
2053 preprocessor macros @code{__GNUC_GNU_INLINE__} or
2054 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
2055 Function is As Fast As a Macro}.
2057 In C++, this attribute does not depend on @code{extern} in any way,
2058 but it still requires the @code{inline} keyword to enable its special
2062 @cindex @code{artificial} function attribute
2063 This attribute is useful for small inline wrappers which if possible
2064 should appear during debugging as a unit, depending on the debug
2065 info format it will either mean marking the function as artificial
2066 or using the caller location for all instructions within the inlined
2070 @cindex interrupt handler functions
2071 When added to an interrupt handler with the M32C port, causes the
2072 prologue and epilogue to use bank switching to preserve the registers
2073 rather than saving them on the stack.
2076 @cindex @code{flatten} function attribute
2077 Generally, inlining into a function is limited. For a function marked with
2078 this attribute, every call inside this function will be inlined, if possible.
2079 Whether the function itself is considered for inlining depends on its size and
2080 the current inlining parameters.
2082 @item error ("@var{message}")
2083 @cindex @code{error} function attribute
2084 If this attribute is used on a function declaration and a call to such a function
2085 is not eliminated through dead code elimination or other optimizations, an error
2086 which will include @var{message} will be diagnosed. This is useful
2087 for compile time checking, especially together with @code{__builtin_constant_p}
2088 and inline functions where checking the inline function arguments is not
2089 possible through @code{extern char [(condition) ? 1 : -1];} tricks.
2090 While it is possible to leave the function undefined and thus invoke
2091 a link failure, when using this attribute the problem will be diagnosed
2092 earlier and with exact location of the call even in presence of inline
2093 functions or when not emitting debugging information.
2095 @item warning ("@var{message}")
2096 @cindex @code{warning} function attribute
2097 If this attribute is used on a function declaration and a call to such a function
2098 is not eliminated through dead code elimination or other optimizations, a warning
2099 which will include @var{message} will be diagnosed. This is useful
2100 for compile time checking, especially together with @code{__builtin_constant_p}
2101 and inline functions. While it is possible to define the function with
2102 a message in @code{.gnu.warning*} section, when using this attribute the problem
2103 will be diagnosed earlier and with exact location of the call even in presence
2104 of inline functions or when not emitting debugging information.
2107 @cindex functions that do pop the argument stack on the 386
2109 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2110 assume that the calling function will pop off the stack space used to
2111 pass arguments. This is
2112 useful to override the effects of the @option{-mrtd} switch.
2115 @cindex @code{const} function attribute
2116 Many functions do not examine any values except their arguments, and
2117 have no effects except the return value. Basically this is just slightly
2118 more strict class than the @code{pure} attribute below, since function is not
2119 allowed to read global memory.
2121 @cindex pointer arguments
2122 Note that a function that has pointer arguments and examines the data
2123 pointed to must @emph{not} be declared @code{const}. Likewise, a
2124 function that calls a non-@code{const} function usually must not be
2125 @code{const}. It does not make sense for a @code{const} function to
2128 The attribute @code{const} is not implemented in GCC versions earlier
2129 than 2.5. An alternative way to declare that a function has no side
2130 effects, which works in the current version and in some older versions,
2134 typedef int intfn ();
2136 extern const intfn square;
2139 This approach does not work in GNU C++ from 2.6.0 on, since the language
2140 specifies that the @samp{const} must be attached to the return value.
2144 @itemx constructor (@var{priority})
2145 @itemx destructor (@var{priority})
2146 @cindex @code{constructor} function attribute
2147 @cindex @code{destructor} function attribute
2148 The @code{constructor} attribute causes the function to be called
2149 automatically before execution enters @code{main ()}. Similarly, the
2150 @code{destructor} attribute causes the function to be called
2151 automatically after @code{main ()} has completed or @code{exit ()} has
2152 been called. Functions with these attributes are useful for
2153 initializing data that will be used implicitly during the execution of
2156 You may provide an optional integer priority to control the order in
2157 which constructor and destructor functions are run. A constructor
2158 with a smaller priority number runs before a constructor with a larger
2159 priority number; the opposite relationship holds for destructors. So,
2160 if you have a constructor that allocates a resource and a destructor
2161 that deallocates the same resource, both functions typically have the
2162 same priority. The priorities for constructor and destructor
2163 functions are the same as those specified for namespace-scope C++
2164 objects (@pxref{C++ Attributes}).
2166 These attributes are not currently implemented for Objective-C@.
2169 @itemx deprecated (@var{msg})
2170 @cindex @code{deprecated} attribute.
2171 The @code{deprecated} attribute results in a warning if the function
2172 is used anywhere in the source file. This is useful when identifying
2173 functions that are expected to be removed in a future version of a
2174 program. The warning also includes the location of the declaration
2175 of the deprecated function, to enable users to easily find further
2176 information about why the function is deprecated, or what they should
2177 do instead. Note that the warnings only occurs for uses:
2180 int old_fn () __attribute__ ((deprecated));
2182 int (*fn_ptr)() = old_fn;
2185 results in a warning on line 3 but not line 2. The optional msg
2186 argument, which must be a string, will be printed in the warning if
2189 The @code{deprecated} attribute can also be used for variables and
2190 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2193 @cindex @code{disinterrupt} attribute
2194 On MeP targets, this attribute causes the compiler to emit
2195 instructions to disable interrupts for the duration of the given
2199 @cindex @code{__declspec(dllexport)}
2200 On Microsoft Windows targets and Symbian OS targets the
2201 @code{dllexport} attribute causes the compiler to provide a global
2202 pointer to a pointer in a DLL, so that it can be referenced with the
2203 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
2204 name is formed by combining @code{_imp__} and the function or variable
2207 You can use @code{__declspec(dllexport)} as a synonym for
2208 @code{__attribute__ ((dllexport))} for compatibility with other
2211 On systems that support the @code{visibility} attribute, this
2212 attribute also implies ``default'' visibility. It is an error to
2213 explicitly specify any other visibility.
2215 Currently, the @code{dllexport} attribute is ignored for inlined
2216 functions, unless the @option{-fkeep-inline-functions} flag has been
2217 used. The attribute is also ignored for undefined symbols.
2219 When applied to C++ classes, the attribute marks defined non-inlined
2220 member functions and static data members as exports. Static consts
2221 initialized in-class are not marked unless they are also defined
2224 For Microsoft Windows targets there are alternative methods for
2225 including the symbol in the DLL's export table such as using a
2226 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
2227 the @option{--export-all} linker flag.
2230 @cindex @code{__declspec(dllimport)}
2231 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
2232 attribute causes the compiler to reference a function or variable via
2233 a global pointer to a pointer that is set up by the DLL exporting the
2234 symbol. The attribute implies @code{extern}. On Microsoft Windows
2235 targets, the pointer name is formed by combining @code{_imp__} and the
2236 function or variable name.
2238 You can use @code{__declspec(dllimport)} as a synonym for
2239 @code{__attribute__ ((dllimport))} for compatibility with other
2242 On systems that support the @code{visibility} attribute, this
2243 attribute also implies ``default'' visibility. It is an error to
2244 explicitly specify any other visibility.
2246 Currently, the attribute is ignored for inlined functions. If the
2247 attribute is applied to a symbol @emph{definition}, an error is reported.
2248 If a symbol previously declared @code{dllimport} is later defined, the
2249 attribute is ignored in subsequent references, and a warning is emitted.
2250 The attribute is also overridden by a subsequent declaration as
2253 When applied to C++ classes, the attribute marks non-inlined
2254 member functions and static data members as imports. However, the
2255 attribute is ignored for virtual methods to allow creation of vtables
2258 On the SH Symbian OS target the @code{dllimport} attribute also has
2259 another affect---it can cause the vtable and run-time type information
2260 for a class to be exported. This happens when the class has a
2261 dllimport'ed constructor or a non-inline, non-pure virtual function
2262 and, for either of those two conditions, the class also has an inline
2263 constructor or destructor and has a key function that is defined in
2264 the current translation unit.
2266 For Microsoft Windows based targets the use of the @code{dllimport}
2267 attribute on functions is not necessary, but provides a small
2268 performance benefit by eliminating a thunk in the DLL@. The use of the
2269 @code{dllimport} attribute on imported variables was required on older
2270 versions of the GNU linker, but can now be avoided by passing the
2271 @option{--enable-auto-import} switch to the GNU linker. As with
2272 functions, using the attribute for a variable eliminates a thunk in
2275 One drawback to using this attribute is that a pointer to a
2276 @emph{variable} marked as @code{dllimport} cannot be used as a constant
2277 address. However, a pointer to a @emph{function} with the
2278 @code{dllimport} attribute can be used as a constant initializer; in
2279 this case, the address of a stub function in the import lib is
2280 referenced. On Microsoft Windows targets, the attribute can be disabled
2281 for functions by setting the @option{-mnop-fun-dllimport} flag.
2284 @cindex eight bit data on the H8/300, H8/300H, and H8S
2285 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2286 variable should be placed into the eight bit data section.
2287 The compiler will generate more efficient code for certain operations
2288 on data in the eight bit data area. Note the eight bit data area is limited to
2291 You must use GAS and GLD from GNU binutils version 2.7 or later for
2292 this attribute to work correctly.
2294 @item exception_handler
2295 @cindex exception handler functions on the Blackfin processor
2296 Use this attribute on the Blackfin to indicate that the specified function
2297 is an exception handler. The compiler will generate function entry and
2298 exit sequences suitable for use in an exception handler when this
2299 attribute is present.
2301 @item externally_visible
2302 @cindex @code{externally_visible} attribute.
2303 This attribute, attached to a global variable or function, nullifies
2304 the effect of the @option{-fwhole-program} command-line option, so the
2305 object remains visible outside the current compilation unit. If @option{-fwhole-program} is used together with @option{-flto} and @command{gold} is used as the linker plugin, @code{externally_visible} attributes are automatically added to functions (not variable yet due to a current @command{gold} issue) that are accessed outside of LTO objects according to resolution file produced by @command{gold}. For other linkers that cannot generate resolution file, explicit @code{externally_visible} attributes are still necessary.
2308 @cindex functions which handle memory bank switching
2309 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
2310 use a calling convention that takes care of switching memory banks when
2311 entering and leaving a function. This calling convention is also the
2312 default when using the @option{-mlong-calls} option.
2314 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
2315 to call and return from a function.
2317 On 68HC11 the compiler will generate a sequence of instructions
2318 to invoke a board-specific routine to switch the memory bank and call the
2319 real function. The board-specific routine simulates a @code{call}.
2320 At the end of a function, it will jump to a board-specific routine
2321 instead of using @code{rts}. The board-specific return routine simulates
2324 On MeP targets this causes the compiler to use a calling convention
2325 which assumes the called function is too far away for the built-in
2328 @item fast_interrupt
2329 @cindex interrupt handler functions
2330 Use this attribute on the M32C and RX ports to indicate that the specified
2331 function is a fast interrupt handler. This is just like the
2332 @code{interrupt} attribute, except that @code{freit} is used to return
2333 instead of @code{reit}.
2336 @cindex functions that pop the argument stack on the 386
2337 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2338 pass the first argument (if of integral type) in the register ECX and
2339 the second argument (if of integral type) in the register EDX@. Subsequent
2340 and other typed arguments are passed on the stack. The called function will
2341 pop the arguments off the stack. If the number of arguments is variable all
2342 arguments are pushed on the stack.
2345 @cindex functions that pop the argument stack on the 386
2346 On the Intel 386, the @code{thiscall} attribute causes the compiler to
2347 pass the first argument (if of integral type) in the register ECX.
2348 Subsequent and other typed arguments are passed on the stack. The called
2349 function will pop the arguments off the stack.
2350 If the number of arguments is variable all arguments are pushed on the
2352 The @code{thiscall} attribute is intended for C++ non-static member functions.
2353 As gcc extension this calling convention can be used for C-functions
2354 and for static member methods.
2356 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2357 @cindex @code{format} function attribute
2359 The @code{format} attribute specifies that a function takes @code{printf},
2360 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2361 should be type-checked against a format string. For example, the
2366 my_printf (void *my_object, const char *my_format, ...)
2367 __attribute__ ((format (printf, 2, 3)));
2371 causes the compiler to check the arguments in calls to @code{my_printf}
2372 for consistency with the @code{printf} style format string argument
2375 The parameter @var{archetype} determines how the format string is
2376 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime},
2377 @code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or
2378 @code{strfmon}. (You can also use @code{__printf__},
2379 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On
2380 MinGW targets, @code{ms_printf}, @code{ms_scanf}, and
2381 @code{ms_strftime} are also present.
2382 @var{archtype} values such as @code{printf} refer to the formats accepted
2383 by the system's C run-time library, while @code{gnu_} values always refer
2384 to the formats accepted by the GNU C Library. On Microsoft Windows
2385 targets, @code{ms_} values refer to the formats accepted by the
2386 @file{msvcrt.dll} library.
2387 The parameter @var{string-index}
2388 specifies which argument is the format string argument (starting
2389 from 1), while @var{first-to-check} is the number of the first
2390 argument to check against the format string. For functions
2391 where the arguments are not available to be checked (such as
2392 @code{vprintf}), specify the third parameter as zero. In this case the
2393 compiler only checks the format string for consistency. For
2394 @code{strftime} formats, the third parameter is required to be zero.
2395 Since non-static C++ methods have an implicit @code{this} argument, the
2396 arguments of such methods should be counted from two, not one, when
2397 giving values for @var{string-index} and @var{first-to-check}.
2399 In the example above, the format string (@code{my_format}) is the second
2400 argument of the function @code{my_print}, and the arguments to check
2401 start with the third argument, so the correct parameters for the format
2402 attribute are 2 and 3.
2404 @opindex ffreestanding
2405 @opindex fno-builtin
2406 The @code{format} attribute allows you to identify your own functions
2407 which take format strings as arguments, so that GCC can check the
2408 calls to these functions for errors. The compiler always (unless
2409 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2410 for the standard library functions @code{printf}, @code{fprintf},
2411 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2412 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2413 warnings are requested (using @option{-Wformat}), so there is no need to
2414 modify the header file @file{stdio.h}. In C99 mode, the functions
2415 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2416 @code{vsscanf} are also checked. Except in strictly conforming C
2417 standard modes, the X/Open function @code{strfmon} is also checked as
2418 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2419 @xref{C Dialect Options,,Options Controlling C Dialect}.
2421 For Objective-C dialects, @code{NSString} (or @code{__NSString__}) is
2422 recognized in the same context. Declarations including these format attributes
2423 will be parsed for correct syntax, however the result of checking of such format
2424 strings is not yet defined, and will not be carried out by this version of the
2427 The target may also provide additional types of format checks.
2428 @xref{Target Format Checks,,Format Checks Specific to Particular
2431 @item format_arg (@var{string-index})
2432 @cindex @code{format_arg} function attribute
2433 @opindex Wformat-nonliteral
2434 The @code{format_arg} attribute specifies that a function takes a format
2435 string for a @code{printf}, @code{scanf}, @code{strftime} or
2436 @code{strfmon} style function and modifies it (for example, to translate
2437 it into another language), so the result can be passed to a
2438 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2439 function (with the remaining arguments to the format function the same
2440 as they would have been for the unmodified string). For example, the
2445 my_dgettext (char *my_domain, const char *my_format)
2446 __attribute__ ((format_arg (2)));
2450 causes the compiler to check the arguments in calls to a @code{printf},
2451 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2452 format string argument is a call to the @code{my_dgettext} function, for
2453 consistency with the format string argument @code{my_format}. If the
2454 @code{format_arg} attribute had not been specified, all the compiler
2455 could tell in such calls to format functions would be that the format
2456 string argument is not constant; this would generate a warning when
2457 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2458 without the attribute.
2460 The parameter @var{string-index} specifies which argument is the format
2461 string argument (starting from one). Since non-static C++ methods have
2462 an implicit @code{this} argument, the arguments of such methods should
2463 be counted from two.
2465 The @code{format-arg} attribute allows you to identify your own
2466 functions which modify format strings, so that GCC can check the
2467 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2468 type function whose operands are a call to one of your own function.
2469 The compiler always treats @code{gettext}, @code{dgettext}, and
2470 @code{dcgettext} in this manner except when strict ISO C support is
2471 requested by @option{-ansi} or an appropriate @option{-std} option, or
2472 @option{-ffreestanding} or @option{-fno-builtin}
2473 is used. @xref{C Dialect Options,,Options
2474 Controlling C Dialect}.
2476 For Objective-C dialects, the @code{format-arg} attribute may refer to an
2477 @code{NSString} reference for compatibility with the @code{format} attribute
2480 The target may also allow additional types in @code{format-arg} attributes.
2481 @xref{Target Format Checks,,Format Checks Specific to Particular
2484 @item function_vector
2485 @cindex calling functions through the function vector on H8/300, M16C, M32C and SH2A processors
2486 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2487 function should be called through the function vector. Calling a
2488 function through the function vector will reduce code size, however;
2489 the function vector has a limited size (maximum 128 entries on the H8/300
2490 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
2492 In SH2A target, this attribute declares a function to be called using the
2493 TBR relative addressing mode. The argument to this attribute is the entry
2494 number of the same function in a vector table containing all the TBR
2495 relative addressable functions. For the successful jump, register TBR
2496 should contain the start address of this TBR relative vector table.
2497 In the startup routine of the user application, user needs to care of this
2498 TBR register initialization. The TBR relative vector table can have at
2499 max 256 function entries. The jumps to these functions will be generated
2500 using a SH2A specific, non delayed branch instruction JSR/N @@(disp8,TBR).
2501 You must use GAS and GLD from GNU binutils version 2.7 or later for
2502 this attribute to work correctly.
2504 Please refer the example of M16C target, to see the use of this
2505 attribute while declaring a function,
2507 In an application, for a function being called once, this attribute will
2508 save at least 8 bytes of code; and if other successive calls are being
2509 made to the same function, it will save 2 bytes of code per each of these
2512 On M16C/M32C targets, the @code{function_vector} attribute declares a
2513 special page subroutine call function. Use of this attribute reduces
2514 the code size by 2 bytes for each call generated to the
2515 subroutine. The argument to the attribute is the vector number entry
2516 from the special page vector table which contains the 16 low-order
2517 bits of the subroutine's entry address. Each vector table has special
2518 page number (18 to 255) which are used in @code{jsrs} instruction.
2519 Jump addresses of the routines are generated by adding 0x0F0000 (in
2520 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the 2
2521 byte addresses set in the vector table. Therefore you need to ensure
2522 that all the special page vector routines should get mapped within the
2523 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
2526 In the following example 2 bytes will be saved for each call to
2527 function @code{foo}.
2530 void foo (void) __attribute__((function_vector(0x18)));
2541 If functions are defined in one file and are called in another file,
2542 then be sure to write this declaration in both files.
2544 This attribute is ignored for R8C target.
2547 @cindex interrupt handler functions
2548 Use this attribute on the ARM, AVR, CRX, M32C, M32R/D, m68k, MeP, MIPS,
2549 RX and Xstormy16 ports to indicate that the specified function is an
2550 interrupt handler. The compiler will generate function entry and exit
2551 sequences suitable for use in an interrupt handler when this attribute
2554 Note, interrupt handlers for the Blackfin, H8/300, H8/300H, H8S, MicroBlaze,
2555 and SH processors can be specified via the @code{interrupt_handler} attribute.
2557 Note, on the AVR, interrupts will be enabled inside the function.
2559 Note, for the ARM, you can specify the kind of interrupt to be handled by
2560 adding an optional parameter to the interrupt attribute like this:
2563 void f () __attribute__ ((interrupt ("IRQ")));
2566 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2568 On ARMv7-M the interrupt type is ignored, and the attribute means the function
2569 may be called with a word aligned stack pointer.
2571 On MIPS targets, you can use the following attributes to modify the behavior
2572 of an interrupt handler:
2574 @item use_shadow_register_set
2575 @cindex @code{use_shadow_register_set} attribute
2576 Assume that the handler uses a shadow register set, instead of
2577 the main general-purpose registers.
2579 @item keep_interrupts_masked
2580 @cindex @code{keep_interrupts_masked} attribute
2581 Keep interrupts masked for the whole function. Without this attribute,
2582 GCC tries to reenable interrupts for as much of the function as it can.
2584 @item use_debug_exception_return
2585 @cindex @code{use_debug_exception_return} attribute
2586 Return using the @code{deret} instruction. Interrupt handlers that don't
2587 have this attribute return using @code{eret} instead.
2590 You can use any combination of these attributes, as shown below:
2592 void __attribute__ ((interrupt)) v0 ();
2593 void __attribute__ ((interrupt, use_shadow_register_set)) v1 ();
2594 void __attribute__ ((interrupt, keep_interrupts_masked)) v2 ();
2595 void __attribute__ ((interrupt, use_debug_exception_return)) v3 ();
2596 void __attribute__ ((interrupt, use_shadow_register_set,
2597 keep_interrupts_masked)) v4 ();
2598 void __attribute__ ((interrupt, use_shadow_register_set,
2599 use_debug_exception_return)) v5 ();
2600 void __attribute__ ((interrupt, keep_interrupts_masked,
2601 use_debug_exception_return)) v6 ();
2602 void __attribute__ ((interrupt, use_shadow_register_set,
2603 keep_interrupts_masked,
2604 use_debug_exception_return)) v7 ();
2607 @item ifunc ("@var{resolver}")
2608 @cindex @code{ifunc} attribute
2609 The @code{ifunc} attribute is used to mark a function as an indirect
2610 function using the STT_GNU_IFUNC symbol type extension to the ELF
2611 standard. This allows the resolution of the symbol value to be
2612 determined dynamically at load time, and an optimized version of the
2613 routine can be selected for the particular processor or other system
2614 characteristics determined then. To use this attribute, first define
2615 the implementation functions available, and a resolver function that
2616 returns a pointer to the selected implementation function. The
2617 implementation functions' declarations must match the API of the
2618 function being implemented, the resolver's declaration is be a
2619 function returning pointer to void function returning void:
2622 void *my_memcpy (void *dst, const void *src, size_t len)
2627 static void (*resolve_memcpy (void)) (void)
2629 return my_memcpy; // we'll just always select this routine
2633 The exported header file declaring the function the user calls would
2637 extern void *memcpy (void *, const void *, size_t);
2640 allowing the user to call this as a regular function, unaware of the
2641 implementation. Finally, the indirect function needs to be defined in
2642 the same translation unit as the resolver function:
2645 void *memcpy (void *, const void *, size_t)
2646 __attribute__ ((ifunc ("resolve_memcpy")));
2649 Indirect functions cannot be weak, and require a recent binutils (at
2650 least version 2.20.1), and GNU C library (at least version 2.11.1).
2652 @item interrupt_handler
2653 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
2654 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
2655 indicate that the specified function is an interrupt handler. The compiler
2656 will generate function entry and exit sequences suitable for use in an
2657 interrupt handler when this attribute is present.
2659 @item interrupt_thread
2660 @cindex interrupt thread functions on fido
2661 Use this attribute on fido, a subarchitecture of the m68k, to indicate
2662 that the specified function is an interrupt handler that is designed
2663 to run as a thread. The compiler omits generate prologue/epilogue
2664 sequences and replaces the return instruction with a @code{sleep}
2665 instruction. This attribute is available only on fido.
2668 @cindex interrupt service routines on ARM
2669 Use this attribute on ARM to write Interrupt Service Routines. This is an
2670 alias to the @code{interrupt} attribute above.
2673 @cindex User stack pointer in interrupts on the Blackfin
2674 When used together with @code{interrupt_handler}, @code{exception_handler}
2675 or @code{nmi_handler}, code will be generated to load the stack pointer
2676 from the USP register in the function prologue.
2679 @cindex @code{l1_text} function attribute
2680 This attribute specifies a function to be placed into L1 Instruction
2681 SRAM@. The function will be put into a specific section named @code{.l1.text}.
2682 With @option{-mfdpic}, function calls with a such function as the callee
2683 or caller will use inlined PLT.
2686 @cindex @code{l2} function attribute
2687 On the Blackfin, this attribute specifies a function to be placed into L2
2688 SRAM. The function will be put into a specific section named
2689 @code{.l1.text}. With @option{-mfdpic}, callers of such functions will use
2693 @cindex @code{leaf} function attribute
2694 Calls to external functions with this attribute must return to the current
2695 compilation unit only by return or by exception handling. In particular, leaf
2696 functions are not allowed to call callback function passed to it from current
2697 compilation unit or directly call functions exported by the unit or longjmp
2698 into the unit. Still leaf function might call functions from other complation
2699 units and thus they are not neccesarily leaf in the sense that they contains no
2700 function calls at all.
2702 The attribute is intended for library functions to improve dataflow analysis.
2703 Compiler takes the hint that any data not escaping current compilation unit can
2704 not be used or modified by the leaf function. For example, function @code{sin}
2705 is leaf, function @code{qsort} is not.
2707 Note that the leaf functions might invoke signals and signal handlers might be
2708 defined in the current compilation unit and use static variables. Only
2709 compliant way to write such a signal handler is to declare such variables
2712 The attribute has no effect on functions defined within current compilation
2713 unit. This is to allow easy merging of multiple compilation units into one,
2714 for example, by using the link time optimization. For this reason the
2715 attribute is not allowed on types to annotate indirect calls.
2717 @item long_call/short_call
2718 @cindex indirect calls on ARM
2719 This attribute specifies how a particular function is called on
2720 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2721 command-line switch and @code{#pragma long_calls} settings. The
2722 @code{long_call} attribute indicates that the function might be far
2723 away from the call site and require a different (more expensive)
2724 calling sequence. The @code{short_call} attribute always places
2725 the offset to the function from the call site into the @samp{BL}
2726 instruction directly.
2728 @item longcall/shortcall
2729 @cindex functions called via pointer on the RS/6000 and PowerPC
2730 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute
2731 indicates that the function might be far away from the call site and
2732 require a different (more expensive) calling sequence. The
2733 @code{shortcall} attribute indicates that the function is always close
2734 enough for the shorter calling sequence to be used. These attributes
2735 override both the @option{-mlongcall} switch and, on the RS/6000 and
2736 PowerPC, the @code{#pragma longcall} setting.
2738 @xref{RS/6000 and PowerPC Options}, for more information on whether long
2739 calls are necessary.
2741 @item long_call/near/far
2742 @cindex indirect calls on MIPS
2743 These attributes specify how a particular function is called on MIPS@.
2744 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
2745 command-line switch. The @code{long_call} and @code{far} attributes are
2746 synonyms, and cause the compiler to always call
2747 the function by first loading its address into a register, and then using
2748 the contents of that register. The @code{near} attribute has the opposite
2749 effect; it specifies that non-PIC calls should be made using the more
2750 efficient @code{jal} instruction.
2753 @cindex @code{malloc} attribute
2754 The @code{malloc} attribute is used to tell the compiler that a function
2755 may be treated as if any non-@code{NULL} pointer it returns cannot
2756 alias any other pointer valid when the function returns.
2757 This will often improve optimization.
2758 Standard functions with this property include @code{malloc} and
2759 @code{calloc}. @code{realloc}-like functions have this property as
2760 long as the old pointer is never referred to (including comparing it
2761 to the new pointer) after the function returns a non-@code{NULL}
2764 @item mips16/nomips16
2765 @cindex @code{mips16} attribute
2766 @cindex @code{nomips16} attribute
2768 On MIPS targets, you can use the @code{mips16} and @code{nomips16}
2769 function attributes to locally select or turn off MIPS16 code generation.
2770 A function with the @code{mips16} attribute is emitted as MIPS16 code,
2771 while MIPS16 code generation is disabled for functions with the
2772 @code{nomips16} attribute. These attributes override the
2773 @option{-mips16} and @option{-mno-mips16} options on the command line
2774 (@pxref{MIPS Options}).
2776 When compiling files containing mixed MIPS16 and non-MIPS16 code, the
2777 preprocessor symbol @code{__mips16} reflects the setting on the command line,
2778 not that within individual functions. Mixed MIPS16 and non-MIPS16 code
2779 may interact badly with some GCC extensions such as @code{__builtin_apply}
2780 (@pxref{Constructing Calls}).
2782 @item model (@var{model-name})
2783 @cindex function addressability on the M32R/D
2784 @cindex variable addressability on the IA-64
2786 On the M32R/D, use this attribute to set the addressability of an
2787 object, and of the code generated for a function. The identifier
2788 @var{model-name} is one of @code{small}, @code{medium}, or
2789 @code{large}, representing each of the code models.
2791 Small model objects live in the lower 16MB of memory (so that their
2792 addresses can be loaded with the @code{ld24} instruction), and are
2793 callable with the @code{bl} instruction.
2795 Medium model objects may live anywhere in the 32-bit address space (the
2796 compiler will generate @code{seth/add3} instructions to load their addresses),
2797 and are callable with the @code{bl} instruction.
2799 Large model objects may live anywhere in the 32-bit address space (the
2800 compiler will generate @code{seth/add3} instructions to load their addresses),
2801 and may not be reachable with the @code{bl} instruction (the compiler will
2802 generate the much slower @code{seth/add3/jl} instruction sequence).
2804 On IA-64, use this attribute to set the addressability of an object.
2805 At present, the only supported identifier for @var{model-name} is
2806 @code{small}, indicating addressability via ``small'' (22-bit)
2807 addresses (so that their addresses can be loaded with the @code{addl}
2808 instruction). Caveat: such addressing is by definition not position
2809 independent and hence this attribute must not be used for objects
2810 defined by shared libraries.
2812 @item ms_abi/sysv_abi
2813 @cindex @code{ms_abi} attribute
2814 @cindex @code{sysv_abi} attribute
2816 On 64-bit x86_64-*-* targets, you can use an ABI attribute to indicate
2817 which calling convention should be used for a function. The @code{ms_abi}
2818 attribute tells the compiler to use the Microsoft ABI, while the
2819 @code{sysv_abi} attribute tells the compiler to use the ABI used on
2820 GNU/Linux and other systems. The default is to use the Microsoft ABI
2821 when targeting Windows. On all other systems, the default is the AMD ABI.
2823 Note, the @code{ms_abi} attribute for Windows targets currently requires
2824 the @option{-maccumulate-outgoing-args} option.
2826 @item ms_hook_prologue
2827 @cindex @code{ms_hook_prologue} attribute
2829 On 32 bit i[34567]86-*-* targets and 64 bit x86_64-*-* targets, you can use
2830 this function attribute to make gcc generate the "hot-patching" function
2831 prologue used in Win32 API functions in Microsoft Windows XP Service Pack 2
2835 @cindex function without a prologue/epilogue code
2836 Use this attribute on the ARM, AVR, MCORE, RX and SPU ports to indicate that
2837 the specified function does not need prologue/epilogue sequences generated by
2838 the compiler. It is up to the programmer to provide these sequences. The
2839 only statements that can be safely included in naked functions are
2840 @code{asm} statements that do not have operands. All other statements,
2841 including declarations of local variables, @code{if} statements, and so
2842 forth, should be avoided. Naked functions should be used to implement the
2843 body of an assembly function, while allowing the compiler to construct
2844 the requisite function declaration for the assembler.
2847 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2848 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2849 use the normal calling convention based on @code{jsr} and @code{rts}.
2850 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2853 On MeP targets this attribute causes the compiler to assume the called
2854 function is close enough to use the normal calling convention,
2855 overriding the @code{-mtf} command line option.
2858 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
2859 Use this attribute together with @code{interrupt_handler},
2860 @code{exception_handler} or @code{nmi_handler} to indicate that the function
2861 entry code should enable nested interrupts or exceptions.
2864 @cindex NMI handler functions on the Blackfin processor
2865 Use this attribute on the Blackfin to indicate that the specified function
2866 is an NMI handler. The compiler will generate function entry and
2867 exit sequences suitable for use in an NMI handler when this
2868 attribute is present.
2870 @item no_instrument_function
2871 @cindex @code{no_instrument_function} function attribute
2872 @opindex finstrument-functions
2873 If @option{-finstrument-functions} is given, profiling function calls will
2874 be generated at entry and exit of most user-compiled functions.
2875 Functions with this attribute will not be so instrumented.
2877 @item no_split_stack
2878 @cindex @code{no_split_stack} function attribute
2879 @opindex fsplit-stack
2880 If @option{-fsplit-stack} is given, functions will have a small
2881 prologue which decides whether to split the stack. Functions with the
2882 @code{no_split_stack} attribute will not have that prologue, and thus
2883 may run with only a small amount of stack space available.
2886 @cindex @code{noinline} function attribute
2887 This function attribute prevents a function from being considered for
2889 @c Don't enumerate the optimizations by name here; we try to be
2890 @c future-compatible with this mechanism.
2891 If the function does not have side-effects, there are optimizations
2892 other than inlining that causes function calls to be optimized away,
2893 although the function call is live. To keep such calls from being
2898 (@pxref{Extended Asm}) in the called function, to serve as a special
2902 @cindex @code{noclone} function attribute
2903 This function attribute prevents a function from being considered for
2904 cloning - a mechanism which produces specialized copies of functions
2905 and which is (currently) performed by interprocedural constant
2908 @item nonnull (@var{arg-index}, @dots{})
2909 @cindex @code{nonnull} function attribute
2910 The @code{nonnull} attribute specifies that some function parameters should
2911 be non-null pointers. For instance, the declaration:
2915 my_memcpy (void *dest, const void *src, size_t len)
2916 __attribute__((nonnull (1, 2)));
2920 causes the compiler to check that, in calls to @code{my_memcpy},
2921 arguments @var{dest} and @var{src} are non-null. If the compiler
2922 determines that a null pointer is passed in an argument slot marked
2923 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2924 is issued. The compiler may also choose to make optimizations based
2925 on the knowledge that certain function arguments will not be null.
2927 If no argument index list is given to the @code{nonnull} attribute,
2928 all pointer arguments are marked as non-null. To illustrate, the
2929 following declaration is equivalent to the previous example:
2933 my_memcpy (void *dest, const void *src, size_t len)
2934 __attribute__((nonnull));
2938 @cindex @code{noreturn} function attribute
2939 A few standard library functions, such as @code{abort} and @code{exit},
2940 cannot return. GCC knows this automatically. Some programs define
2941 their own functions that never return. You can declare them
2942 @code{noreturn} to tell the compiler this fact. For example,
2946 void fatal () __attribute__ ((noreturn));
2949 fatal (/* @r{@dots{}} */)
2951 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2957 The @code{noreturn} keyword tells the compiler to assume that
2958 @code{fatal} cannot return. It can then optimize without regard to what
2959 would happen if @code{fatal} ever did return. This makes slightly
2960 better code. More importantly, it helps avoid spurious warnings of
2961 uninitialized variables.
2963 The @code{noreturn} keyword does not affect the exceptional path when that
2964 applies: a @code{noreturn}-marked function may still return to the caller
2965 by throwing an exception or calling @code{longjmp}.
2967 Do not assume that registers saved by the calling function are
2968 restored before calling the @code{noreturn} function.
2970 It does not make sense for a @code{noreturn} function to have a return
2971 type other than @code{void}.
2973 The attribute @code{noreturn} is not implemented in GCC versions
2974 earlier than 2.5. An alternative way to declare that a function does
2975 not return, which works in the current version and in some older
2976 versions, is as follows:
2979 typedef void voidfn ();
2981 volatile voidfn fatal;
2984 This approach does not work in GNU C++.
2987 @cindex @code{nothrow} function attribute
2988 The @code{nothrow} attribute is used to inform the compiler that a
2989 function cannot throw an exception. For example, most functions in
2990 the standard C library can be guaranteed not to throw an exception
2991 with the notable exceptions of @code{qsort} and @code{bsearch} that
2992 take function pointer arguments. The @code{nothrow} attribute is not
2993 implemented in GCC versions earlier than 3.3.
2996 @cindex @code{optimize} function attribute
2997 The @code{optimize} attribute is used to specify that a function is to
2998 be compiled with different optimization options than specified on the
2999 command line. Arguments can either be numbers or strings. Numbers
3000 are assumed to be an optimization level. Strings that begin with
3001 @code{O} are assumed to be an optimization option, while other options
3002 are assumed to be used with a @code{-f} prefix. You can also use the
3003 @samp{#pragma GCC optimize} pragma to set the optimization options
3004 that affect more than one function.
3005 @xref{Function Specific Option Pragmas}, for details about the
3006 @samp{#pragma GCC optimize} pragma.
3008 This can be used for instance to have frequently executed functions
3009 compiled with more aggressive optimization options that produce faster
3010 and larger code, while other functions can be called with less
3014 @cindex @code{pcs} function attribute
3016 The @code{pcs} attribute can be used to control the calling convention
3017 used for a function on ARM. The attribute takes an argument that specifies
3018 the calling convention to use.
3020 When compiling using the AAPCS ABI (or a variant of that) then valid
3021 values for the argument are @code{"aapcs"} and @code{"aapcs-vfp"}. In
3022 order to use a variant other than @code{"aapcs"} then the compiler must
3023 be permitted to use the appropriate co-processor registers (i.e., the
3024 VFP registers must be available in order to use @code{"aapcs-vfp"}).
3028 /* Argument passed in r0, and result returned in r0+r1. */
3029 double f2d (float) __attribute__((pcs("aapcs")));
3032 Variadic functions always use the @code{"aapcs"} calling convention and
3033 the compiler will reject attempts to specify an alternative.
3036 @cindex @code{pure} function attribute
3037 Many functions have no effects except the return value and their
3038 return value depends only on the parameters and/or global variables.
3039 Such a function can be subject
3040 to common subexpression elimination and loop optimization just as an
3041 arithmetic operator would be. These functions should be declared
3042 with the attribute @code{pure}. For example,
3045 int square (int) __attribute__ ((pure));
3049 says that the hypothetical function @code{square} is safe to call
3050 fewer times than the program says.
3052 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
3053 Interesting non-pure functions are functions with infinite loops or those
3054 depending on volatile memory or other system resource, that may change between
3055 two consecutive calls (such as @code{feof} in a multithreading environment).
3057 The attribute @code{pure} is not implemented in GCC versions earlier
3061 @cindex @code{hot} function attribute
3062 The @code{hot} attribute is used to inform the compiler that a function is a
3063 hot spot of the compiled program. The function is optimized more aggressively
3064 and on many target it is placed into special subsection of the text section so
3065 all hot functions appears close together improving locality.
3067 When profile feedback is available, via @option{-fprofile-use}, hot functions
3068 are automatically detected and this attribute is ignored.
3070 The @code{hot} attribute is not implemented in GCC versions earlier
3074 @cindex @code{cold} function attribute
3075 The @code{cold} attribute is used to inform the compiler that a function is
3076 unlikely executed. The function is optimized for size rather than speed and on
3077 many targets it is placed into special subsection of the text section so all
3078 cold functions appears close together improving code locality of non-cold parts
3079 of program. The paths leading to call of cold functions within code are marked
3080 as unlikely by the branch prediction mechanism. It is thus useful to mark
3081 functions used to handle unlikely conditions, such as @code{perror}, as cold to
3082 improve optimization of hot functions that do call marked functions in rare
3085 When profile feedback is available, via @option{-fprofile-use}, hot functions
3086 are automatically detected and this attribute is ignored.
3088 The @code{cold} attribute is not implemented in GCC versions earlier than 4.3.
3090 @item regparm (@var{number})
3091 @cindex @code{regparm} attribute
3092 @cindex functions that are passed arguments in registers on the 386
3093 On the Intel 386, the @code{regparm} attribute causes the compiler to
3094 pass arguments number one to @var{number} if they are of integral type
3095 in registers EAX, EDX, and ECX instead of on the stack. Functions that
3096 take a variable number of arguments will continue to be passed all of their
3097 arguments on the stack.
3099 Beware that on some ELF systems this attribute is unsuitable for
3100 global functions in shared libraries with lazy binding (which is the
3101 default). Lazy binding will send the first call via resolving code in
3102 the loader, which might assume EAX, EDX and ECX can be clobbered, as
3103 per the standard calling conventions. Solaris 8 is affected by this.
3104 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
3105 safe since the loaders there save EAX, EDX and ECX. (Lazy binding can be
3106 disabled with the linker or the loader if desired, to avoid the
3110 @cindex @code{sseregparm} attribute
3111 On the Intel 386 with SSE support, the @code{sseregparm} attribute
3112 causes the compiler to pass up to 3 floating point arguments in
3113 SSE registers instead of on the stack. Functions that take a
3114 variable number of arguments will continue to pass all of their
3115 floating point arguments on the stack.
3117 @item force_align_arg_pointer
3118 @cindex @code{force_align_arg_pointer} attribute
3119 On the Intel x86, the @code{force_align_arg_pointer} attribute may be
3120 applied to individual function definitions, generating an alternate
3121 prologue and epilogue that realigns the runtime stack if necessary.
3122 This supports mixing legacy codes that run with a 4-byte aligned stack
3123 with modern codes that keep a 16-byte stack for SSE compatibility.
3126 @cindex @code{resbank} attribute
3127 On the SH2A target, this attribute enables the high-speed register
3128 saving and restoration using a register bank for @code{interrupt_handler}
3129 routines. Saving to the bank is performed automatically after the CPU
3130 accepts an interrupt that uses a register bank.
3132 The nineteen 32-bit registers comprising general register R0 to R14,
3133 control register GBR, and system registers MACH, MACL, and PR and the
3134 vector table address offset are saved into a register bank. Register
3135 banks are stacked in first-in last-out (FILO) sequence. Restoration
3136 from the bank is executed by issuing a RESBANK instruction.
3139 @cindex @code{returns_twice} attribute
3140 The @code{returns_twice} attribute tells the compiler that a function may
3141 return more than one time. The compiler will ensure that all registers
3142 are dead before calling such a function and will emit a warning about
3143 the variables that may be clobbered after the second return from the
3144 function. Examples of such functions are @code{setjmp} and @code{vfork}.
3145 The @code{longjmp}-like counterpart of such function, if any, might need
3146 to be marked with the @code{noreturn} attribute.
3149 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
3150 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
3151 all registers except the stack pointer should be saved in the prologue
3152 regardless of whether they are used or not.
3154 @item save_volatiles
3155 @cindex save volatile registers on the MicroBlaze
3156 Use this attribute on the MicroBlaze to indicate that the function is
3157 an interrupt handler. All volatile registers (in addition to non-volatile
3158 registers) will be saved in the function prologue. If the function is a leaf
3159 function, only volatiles used by the function are saved. A normal function
3160 return is generated instead of a return from interrupt.
3162 @item section ("@var{section-name}")
3163 @cindex @code{section} function attribute
3164 Normally, the compiler places the code it generates in the @code{text} section.
3165 Sometimes, however, you need additional sections, or you need certain
3166 particular functions to appear in special sections. The @code{section}
3167 attribute specifies that a function lives in a particular section.
3168 For example, the declaration:
3171 extern void foobar (void) __attribute__ ((section ("bar")));
3175 puts the function @code{foobar} in the @code{bar} section.
3177 Some file formats do not support arbitrary sections so the @code{section}
3178 attribute is not available on all platforms.
3179 If you need to map the entire contents of a module to a particular
3180 section, consider using the facilities of the linker instead.
3183 @cindex @code{sentinel} function attribute
3184 This function attribute ensures that a parameter in a function call is
3185 an explicit @code{NULL}. The attribute is only valid on variadic
3186 functions. By default, the sentinel is located at position zero, the
3187 last parameter of the function call. If an optional integer position
3188 argument P is supplied to the attribute, the sentinel must be located at
3189 position P counting backwards from the end of the argument list.
3192 __attribute__ ((sentinel))
3194 __attribute__ ((sentinel(0)))
3197 The attribute is automatically set with a position of 0 for the built-in
3198 functions @code{execl} and @code{execlp}. The built-in function
3199 @code{execle} has the attribute set with a position of 1.
3201 A valid @code{NULL} in this context is defined as zero with any pointer
3202 type. If your system defines the @code{NULL} macro with an integer type
3203 then you need to add an explicit cast. GCC replaces @code{stddef.h}
3204 with a copy that redefines NULL appropriately.
3206 The warnings for missing or incorrect sentinels are enabled with
3210 See long_call/short_call.
3213 See longcall/shortcall.
3216 @cindex signal handler functions on the AVR processors
3217 Use this attribute on the AVR to indicate that the specified
3218 function is a signal handler. The compiler will generate function
3219 entry and exit sequences suitable for use in a signal handler when this
3220 attribute is present. Interrupts will be disabled inside the function.
3223 Use this attribute on the SH to indicate an @code{interrupt_handler}
3224 function should switch to an alternate stack. It expects a string
3225 argument that names a global variable holding the address of the
3230 void f () __attribute__ ((interrupt_handler,
3231 sp_switch ("alt_stack")));
3235 @cindex functions that pop the argument stack on the 386
3236 On the Intel 386, the @code{stdcall} attribute causes the compiler to
3237 assume that the called function will pop off the stack space used to
3238 pass arguments, unless it takes a variable number of arguments.
3240 @item syscall_linkage
3241 @cindex @code{syscall_linkage} attribute
3242 This attribute is used to modify the IA64 calling convention by marking
3243 all input registers as live at all function exits. This makes it possible
3244 to restart a system call after an interrupt without having to save/restore
3245 the input registers. This also prevents kernel data from leaking into
3249 @cindex @code{target} function attribute
3250 The @code{target} attribute is used to specify that a function is to
3251 be compiled with different target options than specified on the
3252 command line. This can be used for instance to have functions
3253 compiled with a different ISA (instruction set architecture) than the
3254 default. You can also use the @samp{#pragma GCC target} pragma to set
3255 more than one function to be compiled with specific target options.
3256 @xref{Function Specific Option Pragmas}, for details about the
3257 @samp{#pragma GCC target} pragma.
3259 For instance on a 386, you could compile one function with
3260 @code{target("sse4.1,arch=core2")} and another with
3261 @code{target("sse4a,arch=amdfam10")} that would be equivalent to
3262 compiling the first function with @option{-msse4.1} and
3263 @option{-march=core2} options, and the second function with
3264 @option{-msse4a} and @option{-march=amdfam10} options. It is up to the
3265 user to make sure that a function is only invoked on a machine that
3266 supports the particular ISA it was compiled for (for example by using
3267 @code{cpuid} on 386 to determine what feature bits and architecture
3271 int core2_func (void) __attribute__ ((__target__ ("arch=core2")));
3272 int sse3_func (void) __attribute__ ((__target__ ("sse3")));
3276 @item i386 target attributes
3277 On the 386, the following options are allowed:
3282 @cindex @code{target("abm")} attribute
3283 Enable/disable the generation of the advanced bit instructions.
3287 @cindex @code{target("aes")} attribute
3288 Enable/disable the generation of the AES instructions.
3292 @cindex @code{target("mmx")} attribute
3293 Enable/disable the generation of the MMX instructions.
3297 @cindex @code{target("pclmul")} attribute
3298 Enable/disable the generation of the PCLMUL instructions.
3302 @cindex @code{target("popcnt")} attribute
3303 Enable/disable the generation of the POPCNT instruction.
3307 @cindex @code{target("sse")} attribute
3308 Enable/disable the generation of the SSE instructions.
3312 @cindex @code{target("sse2")} attribute
3313 Enable/disable the generation of the SSE2 instructions.
3317 @cindex @code{target("sse3")} attribute
3318 Enable/disable the generation of the SSE3 instructions.
3322 @cindex @code{target("sse4")} attribute
3323 Enable/disable the generation of the SSE4 instructions (both SSE4.1
3328 @cindex @code{target("sse4.1")} attribute
3329 Enable/disable the generation of the sse4.1 instructions.
3333 @cindex @code{target("sse4.2")} attribute
3334 Enable/disable the generation of the sse4.2 instructions.
3338 @cindex @code{target("sse4a")} attribute
3339 Enable/disable the generation of the SSE4A instructions.
3343 @cindex @code{target("fma4")} attribute
3344 Enable/disable the generation of the FMA4 instructions.
3348 @cindex @code{target("xop")} attribute
3349 Enable/disable the generation of the XOP instructions.
3353 @cindex @code{target("lwp")} attribute
3354 Enable/disable the generation of the LWP instructions.
3358 @cindex @code{target("ssse3")} attribute
3359 Enable/disable the generation of the SSSE3 instructions.
3363 @cindex @code{target("cld")} attribute
3364 Enable/disable the generation of the CLD before string moves.
3366 @item fancy-math-387
3367 @itemx no-fancy-math-387
3368 @cindex @code{target("fancy-math-387")} attribute
3369 Enable/disable the generation of the @code{sin}, @code{cos}, and
3370 @code{sqrt} instructions on the 387 floating point unit.
3373 @itemx no-fused-madd
3374 @cindex @code{target("fused-madd")} attribute
3375 Enable/disable the generation of the fused multiply/add instructions.
3379 @cindex @code{target("ieee-fp")} attribute
3380 Enable/disable the generation of floating point that depends on IEEE arithmetic.
3382 @item inline-all-stringops
3383 @itemx no-inline-all-stringops
3384 @cindex @code{target("inline-all-stringops")} attribute
3385 Enable/disable inlining of string operations.
3387 @item inline-stringops-dynamically
3388 @itemx no-inline-stringops-dynamically
3389 @cindex @code{target("inline-stringops-dynamically")} attribute
3390 Enable/disable the generation of the inline code to do small string
3391 operations and calling the library routines for large operations.
3393 @item align-stringops
3394 @itemx no-align-stringops
3395 @cindex @code{target("align-stringops")} attribute
3396 Do/do not align destination of inlined string operations.
3400 @cindex @code{target("recip")} attribute
3401 Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS
3402 instructions followed an additional Newton-Raphson step instead of
3403 doing a floating point division.
3405 @item arch=@var{ARCH}
3406 @cindex @code{target("arch=@var{ARCH}")} attribute
3407 Specify the architecture to generate code for in compiling the function.
3409 @item tune=@var{TUNE}
3410 @cindex @code{target("tune=@var{TUNE}")} attribute
3411 Specify the architecture to tune for in compiling the function.
3413 @item fpmath=@var{FPMATH}
3414 @cindex @code{target("fpmath=@var{FPMATH}")} attribute
3415 Specify which floating point unit to use. The
3416 @code{target("fpmath=sse,387")} option must be specified as
3417 @code{target("fpmath=sse+387")} because the comma would separate
3420 @item PowerPC target attributes
3421 On the PowerPC, the following options are allowed:
3426 @cindex @code{target("altivec")} attribute
3427 Generate code that uses (does not use) AltiVec instructions. In
3428 32-bit code, you cannot enable Altivec instructions unless
3429 @option{-mabi=altivec} was used on the command line.
3433 @cindex @code{target("cmpb")} attribute
3434 Generate code that uses (does not use) the compare bytes instruction
3435 implemented on the POWER6 processor and other processors that support
3436 the PowerPC V2.05 architecture.
3440 @cindex @code{target("dlmzb")} attribute
3441 Generate code that uses (does not use) the string-search @samp{dlmzb}
3442 instruction on the IBM 405, 440, 464 and 476 processors. This instruction is
3443 generated by default when targetting those processors.
3447 @cindex @code{target("fprnd")} attribute
3448 Generate code that uses (does not use) the FP round to integer
3449 instructions implemented on the POWER5+ processor and other processors
3450 that support the PowerPC V2.03 architecture.
3454 @cindex @code{target("hard-dfp")} attribute
3455 Generate code that uses (does not use) the decimal floating point
3456 instructions implemented on some POWER processors.
3460 @cindex @code{target("isel")} attribute
3461 Generate code that uses (does not use) ISEL instruction.
3465 @cindex @code{target("mfcrf")} attribute
3466 Generate code that uses (does not use) the move from condition
3467 register field instruction implemented on the POWER4 processor and
3468 other processors that support the PowerPC V2.01 architecture.
3472 @cindex @code{target("mfpgpr")} attribute
3473 Generate code that uses (does not use) the FP move to/from general
3474 purpose register instructions implemented on the POWER6X processor and
3475 other processors that support the extended PowerPC V2.05 architecture.
3479 @cindex @code{target("mulhw")} attribute
3480 Generate code that uses (does not use) the half-word multiply and
3481 multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors.
3482 These instructions are generated by default when targetting those
3487 @cindex @code{target("multiple")} attribute
3488 Generate code that uses (does not use) the load multiple word
3489 instructions and the store multiple word instructions.
3493 @cindex @code{target("update")} attribute
3494 Generate code that uses (does not use) the load or store instructions
3495 that update the base register to the address of the calculated memory
3500 @cindex @code{target("popcntb")} attribute
3501 Generate code that uses (does not use) the popcount and double
3502 precision FP reciprocal estimate instruction implemented on the POWER5
3503 processor and other processors that support the PowerPC V2.02
3508 @cindex @code{target("popcntd")} attribute
3509 Generate code that uses (does not use) the popcount instruction
3510 implemented on the POWER7 processor and other processors that support
3511 the PowerPC V2.06 architecture.
3513 @item powerpc-gfxopt
3514 @itemx no-powerpc-gfxopt
3515 @cindex @code{target("powerpc-gfxopt")} attribute
3516 Generate code that uses (does not use) the optional PowerPC
3517 architecture instructions in the Graphics group, including
3518 floating-point select.
3521 @itemx no-powerpc-gpopt
3522 @cindex @code{target("powerpc-gpopt")} attribute
3523 Generate code that uses (does not use) the optional PowerPC
3524 architecture instructions in the General Purpose group, including
3525 floating-point square root.
3527 @item recip-precision
3528 @itemx no-recip-precision
3529 @cindex @code{target("recip-precision")} attribute
3530 Assume (do not assume) that the reciprocal estimate instructions
3531 provide higher precision estimates than is mandated by the powerpc
3536 @cindex @code{target("string")} attribute
3537 Generate code that uses (does not use) the load string instructions
3538 and the store string word instructions to save multiple registers and
3539 do small block moves.
3543 @cindex @code{target("vsx")} attribute
3544 Generate code that uses (does not use) vector/scalar (VSX)
3545 instructions, and also enable the use of built-in functions that allow
3546 more direct access to the VSX instruction set. In 32-bit code, you
3547 cannot enable VSX or Altivec instructions unless
3548 @option{-mabi=altivec} was used on the command line.
3552 @cindex @code{target("friz")} attribute
3553 Generate (do not generate) the @code{friz} instruction when the
3554 @option{-funsafe-math-optimizations} option is used to optimize
3555 rounding a floating point value to 64-bit integer and back to floating
3556 point. The @code{friz} instruction does not return the same value if
3557 the floating point number is too large to fit in an integer.
3559 @item avoid-indexed-addresses
3560 @itemx no-avoid-indexed-addresses
3561 @cindex @code{target("avoid-indexed-addresses")} attribute
3562 Generate code that tries to avoid (not avoid) the use of indexed load
3563 or store instructions.
3567 @cindex @code{target("paired")} attribute
3568 Generate code that uses (does not use) the generation of PAIRED simd
3573 @cindex @code{target("longcall")} attribute
3574 Generate code that assumes (does not assume) that all calls are far
3575 away so that a longer more expensive calling sequence is required.
3578 @cindex @code{target("cpu=@var{CPU}")} attribute
3579 Specify the architecture to generate code for in compiling the
3580 function. If you select @code{"target("cpu=power7)"} attribute when
3581 generating 32-bit code, VSX and Altivec instructions are not generated
3582 unless you use the @option{-mabi=altivec} option on the command line.
3584 @item tune=@var{TUNE}
3585 @cindex @code{target("tune=@var{TUNE}")} attribute
3586 Specify the architecture to tune for in compiling the function. If
3587 you do not specify the @code{target("tune=@var{TUNE}")} attribute and
3588 you do specifiy the @code{target("cpu=@var{CPU}")} attribute,
3589 compilation will tune for the @var{CPU} architecture, and not the
3590 default tuning specified on the command line.
3595 On the 386/x86_64 and PowerPC backends, you can use either multiple
3596 strings to specify multiple options, or you can separate the option
3597 with a comma (@code{,}).
3599 On the 386/x86_64 and PowerPC backends, the inliner will not inline a
3600 function that has different target options than the caller, unless the
3601 callee has a subset of the target options of the caller. For example
3602 a function declared with @code{target("sse3")} can inline a function
3603 with @code{target("sse2")}, since @code{-msse3} implies @code{-msse2}.
3605 The @code{target} attribute is not implemented in GCC versions earlier
3606 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC backends. It is
3607 not currently implemented for other backends.
3610 @cindex tiny data section on the H8/300H and H8S
3611 Use this attribute on the H8/300H and H8S to indicate that the specified
3612 variable should be placed into the tiny data section.
3613 The compiler will generate more efficient code for loads and stores
3614 on data in the tiny data section. Note the tiny data area is limited to
3615 slightly under 32kbytes of data.
3618 Use this attribute on the SH for an @code{interrupt_handler} to return using
3619 @code{trapa} instead of @code{rte}. This attribute expects an integer
3620 argument specifying the trap number to be used.
3623 @cindex @code{unused} attribute.
3624 This attribute, attached to a function, means that the function is meant
3625 to be possibly unused. GCC will not produce a warning for this
3629 @cindex @code{used} attribute.
3630 This attribute, attached to a function, means that code must be emitted
3631 for the function even if it appears that the function is not referenced.
3632 This is useful, for example, when the function is referenced only in
3636 @cindex @code{version_id} attribute
3637 This IA64 HP-UX attribute, attached to a global variable or function, renames a
3638 symbol to contain a version string, thus allowing for function level
3639 versioning. HP-UX system header files may use version level functioning
3640 for some system calls.
3643 extern int foo () __attribute__((version_id ("20040821")));
3646 Calls to @var{foo} will be mapped to calls to @var{foo@{20040821@}}.
3648 @item visibility ("@var{visibility_type}")
3649 @cindex @code{visibility} attribute
3650 This attribute affects the linkage of the declaration to which it is attached.
3651 There are four supported @var{visibility_type} values: default,
3652 hidden, protected or internal visibility.
3655 void __attribute__ ((visibility ("protected")))
3656 f () @{ /* @r{Do something.} */; @}
3657 int i __attribute__ ((visibility ("hidden")));
3660 The possible values of @var{visibility_type} correspond to the
3661 visibility settings in the ELF gABI.
3664 @c keep this list of visibilities in alphabetical order.
3667 Default visibility is the normal case for the object file format.
3668 This value is available for the visibility attribute to override other
3669 options that may change the assumed visibility of entities.
3671 On ELF, default visibility means that the declaration is visible to other
3672 modules and, in shared libraries, means that the declared entity may be
3675 On Darwin, default visibility means that the declaration is visible to
3678 Default visibility corresponds to ``external linkage'' in the language.
3681 Hidden visibility indicates that the entity declared will have a new
3682 form of linkage, which we'll call ``hidden linkage''. Two
3683 declarations of an object with hidden linkage refer to the same object
3684 if they are in the same shared object.
3687 Internal visibility is like hidden visibility, but with additional
3688 processor specific semantics. Unless otherwise specified by the
3689 psABI, GCC defines internal visibility to mean that a function is
3690 @emph{never} called from another module. Compare this with hidden
3691 functions which, while they cannot be referenced directly by other
3692 modules, can be referenced indirectly via function pointers. By
3693 indicating that a function cannot be called from outside the module,
3694 GCC may for instance omit the load of a PIC register since it is known
3695 that the calling function loaded the correct value.
3698 Protected visibility is like default visibility except that it
3699 indicates that references within the defining module will bind to the
3700 definition in that module. That is, the declared entity cannot be
3701 overridden by another module.
3705 All visibilities are supported on many, but not all, ELF targets
3706 (supported when the assembler supports the @samp{.visibility}
3707 pseudo-op). Default visibility is supported everywhere. Hidden
3708 visibility is supported on Darwin targets.
3710 The visibility attribute should be applied only to declarations which
3711 would otherwise have external linkage. The attribute should be applied
3712 consistently, so that the same entity should not be declared with
3713 different settings of the attribute.
3715 In C++, the visibility attribute applies to types as well as functions
3716 and objects, because in C++ types have linkage. A class must not have
3717 greater visibility than its non-static data member types and bases,
3718 and class members default to the visibility of their class. Also, a
3719 declaration without explicit visibility is limited to the visibility
3722 In C++, you can mark member functions and static member variables of a
3723 class with the visibility attribute. This is useful if you know a
3724 particular method or static member variable should only be used from
3725 one shared object; then you can mark it hidden while the rest of the
3726 class has default visibility. Care must be taken to avoid breaking
3727 the One Definition Rule; for example, it is usually not useful to mark
3728 an inline method as hidden without marking the whole class as hidden.
3730 A C++ namespace declaration can also have the visibility attribute.
3731 This attribute applies only to the particular namespace body, not to
3732 other definitions of the same namespace; it is equivalent to using
3733 @samp{#pragma GCC visibility} before and after the namespace
3734 definition (@pxref{Visibility Pragmas}).
3736 In C++, if a template argument has limited visibility, this
3737 restriction is implicitly propagated to the template instantiation.
3738 Otherwise, template instantiations and specializations default to the
3739 visibility of their template.
3741 If both the template and enclosing class have explicit visibility, the
3742 visibility from the template is used.
3745 @cindex @code{vliw} attribute
3746 On MeP, the @code{vliw} attribute tells the compiler to emit
3747 instructions in VLIW mode instead of core mode. Note that this
3748 attribute is not allowed unless a VLIW coprocessor has been configured
3749 and enabled through command line options.
3751 @item warn_unused_result
3752 @cindex @code{warn_unused_result} attribute
3753 The @code{warn_unused_result} attribute causes a warning to be emitted
3754 if a caller of the function with this attribute does not use its
3755 return value. This is useful for functions where not checking
3756 the result is either a security problem or always a bug, such as
3760 int fn () __attribute__ ((warn_unused_result));
3763 if (fn () < 0) return -1;
3769 results in warning on line 5.
3772 @cindex @code{weak} attribute
3773 The @code{weak} attribute causes the declaration to be emitted as a weak
3774 symbol rather than a global. This is primarily useful in defining
3775 library functions which can be overridden in user code, though it can
3776 also be used with non-function declarations. Weak symbols are supported
3777 for ELF targets, and also for a.out targets when using the GNU assembler
3781 @itemx weakref ("@var{target}")
3782 @cindex @code{weakref} attribute
3783 The @code{weakref} attribute marks a declaration as a weak reference.
3784 Without arguments, it should be accompanied by an @code{alias} attribute
3785 naming the target symbol. Optionally, the @var{target} may be given as
3786 an argument to @code{weakref} itself. In either case, @code{weakref}
3787 implicitly marks the declaration as @code{weak}. Without a
3788 @var{target}, given as an argument to @code{weakref} or to @code{alias},
3789 @code{weakref} is equivalent to @code{weak}.
3792 static int x() __attribute__ ((weakref ("y")));
3793 /* is equivalent to... */
3794 static int x() __attribute__ ((weak, weakref, alias ("y")));
3796 static int x() __attribute__ ((weakref));
3797 static int x() __attribute__ ((alias ("y")));
3800 A weak reference is an alias that does not by itself require a
3801 definition to be given for the target symbol. If the target symbol is
3802 only referenced through weak references, then it becomes a @code{weak}
3803 undefined symbol. If it is directly referenced, however, then such
3804 strong references prevail, and a definition will be required for the
3805 symbol, not necessarily in the same translation unit.
3807 The effect is equivalent to moving all references to the alias to a
3808 separate translation unit, renaming the alias to the aliased symbol,
3809 declaring it as weak, compiling the two separate translation units and
3810 performing a reloadable link on them.
3812 At present, a declaration to which @code{weakref} is attached can
3813 only be @code{static}.
3817 You can specify multiple attributes in a declaration by separating them
3818 by commas within the double parentheses or by immediately following an
3819 attribute declaration with another attribute declaration.
3821 @cindex @code{#pragma}, reason for not using
3822 @cindex pragma, reason for not using
3823 Some people object to the @code{__attribute__} feature, suggesting that
3824 ISO C's @code{#pragma} should be used instead. At the time
3825 @code{__attribute__} was designed, there were two reasons for not doing
3830 It is impossible to generate @code{#pragma} commands from a macro.
3833 There is no telling what the same @code{#pragma} might mean in another
3837 These two reasons applied to almost any application that might have been
3838 proposed for @code{#pragma}. It was basically a mistake to use
3839 @code{#pragma} for @emph{anything}.
3841 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
3842 to be generated from macros. In addition, a @code{#pragma GCC}
3843 namespace is now in use for GCC-specific pragmas. However, it has been
3844 found convenient to use @code{__attribute__} to achieve a natural
3845 attachment of attributes to their corresponding declarations, whereas
3846 @code{#pragma GCC} is of use for constructs that do not naturally form
3847 part of the grammar. @xref{Other Directives,,Miscellaneous
3848 Preprocessing Directives, cpp, The GNU C Preprocessor}.
3850 @node Attribute Syntax
3851 @section Attribute Syntax
3852 @cindex attribute syntax
3854 This section describes the syntax with which @code{__attribute__} may be
3855 used, and the constructs to which attribute specifiers bind, for the C
3856 language. Some details may vary for C++ and Objective-C@. Because of
3857 infelicities in the grammar for attributes, some forms described here
3858 may not be successfully parsed in all cases.
3860 There are some problems with the semantics of attributes in C++. For
3861 example, there are no manglings for attributes, although they may affect
3862 code generation, so problems may arise when attributed types are used in
3863 conjunction with templates or overloading. Similarly, @code{typeid}
3864 does not distinguish between types with different attributes. Support
3865 for attributes in C++ may be restricted in future to attributes on
3866 declarations only, but not on nested declarators.
3868 @xref{Function Attributes}, for details of the semantics of attributes
3869 applying to functions. @xref{Variable Attributes}, for details of the
3870 semantics of attributes applying to variables. @xref{Type Attributes},
3871 for details of the semantics of attributes applying to structure, union
3872 and enumerated types.
3874 An @dfn{attribute specifier} is of the form
3875 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
3876 is a possibly empty comma-separated sequence of @dfn{attributes}, where
3877 each attribute is one of the following:
3881 Empty. Empty attributes are ignored.
3884 A word (which may be an identifier such as @code{unused}, or a reserved
3885 word such as @code{const}).
3888 A word, followed by, in parentheses, parameters for the attribute.
3889 These parameters take one of the following forms:
3893 An identifier. For example, @code{mode} attributes use this form.
3896 An identifier followed by a comma and a non-empty comma-separated list
3897 of expressions. For example, @code{format} attributes use this form.
3900 A possibly empty comma-separated list of expressions. For example,
3901 @code{format_arg} attributes use this form with the list being a single
3902 integer constant expression, and @code{alias} attributes use this form
3903 with the list being a single string constant.
3907 An @dfn{attribute specifier list} is a sequence of one or more attribute
3908 specifiers, not separated by any other tokens.
3910 In GNU C, an attribute specifier list may appear after the colon following a
3911 label, other than a @code{case} or @code{default} label. The only
3912 attribute it makes sense to use after a label is @code{unused}. This
3913 feature is intended for code generated by programs which contains labels
3914 that may be unused but which is compiled with @option{-Wall}. It would
3915 not normally be appropriate to use in it human-written code, though it
3916 could be useful in cases where the code that jumps to the label is
3917 contained within an @code{#ifdef} conditional. GNU C++ only permits
3918 attributes on labels if the attribute specifier is immediately
3919 followed by a semicolon (i.e., the label applies to an empty
3920 statement). If the semicolon is missing, C++ label attributes are
3921 ambiguous, as it is permissible for a declaration, which could begin
3922 with an attribute list, to be labelled in C++. Declarations cannot be
3923 labelled in C90 or C99, so the ambiguity does not arise there.
3925 An attribute specifier list may appear as part of a @code{struct},
3926 @code{union} or @code{enum} specifier. It may go either immediately
3927 after the @code{struct}, @code{union} or @code{enum} keyword, or after
3928 the closing brace. The former syntax is preferred.
3929 Where attribute specifiers follow the closing brace, they are considered
3930 to relate to the structure, union or enumerated type defined, not to any
3931 enclosing declaration the type specifier appears in, and the type
3932 defined is not complete until after the attribute specifiers.
3933 @c Otherwise, there would be the following problems: a shift/reduce
3934 @c conflict between attributes binding the struct/union/enum and
3935 @c binding to the list of specifiers/qualifiers; and "aligned"
3936 @c attributes could use sizeof for the structure, but the size could be
3937 @c changed later by "packed" attributes.
3939 Otherwise, an attribute specifier appears as part of a declaration,
3940 counting declarations of unnamed parameters and type names, and relates
3941 to that declaration (which may be nested in another declaration, for
3942 example in the case of a parameter declaration), or to a particular declarator
3943 within a declaration. Where an
3944 attribute specifier is applied to a parameter declared as a function or
3945 an array, it should apply to the function or array rather than the
3946 pointer to which the parameter is implicitly converted, but this is not
3947 yet correctly implemented.
3949 Any list of specifiers and qualifiers at the start of a declaration may
3950 contain attribute specifiers, whether or not such a list may in that
3951 context contain storage class specifiers. (Some attributes, however,
3952 are essentially in the nature of storage class specifiers, and only make
3953 sense where storage class specifiers may be used; for example,
3954 @code{section}.) There is one necessary limitation to this syntax: the
3955 first old-style parameter declaration in a function definition cannot
3956 begin with an attribute specifier, because such an attribute applies to
3957 the function instead by syntax described below (which, however, is not
3958 yet implemented in this case). In some other cases, attribute
3959 specifiers are permitted by this grammar but not yet supported by the
3960 compiler. All attribute specifiers in this place relate to the
3961 declaration as a whole. In the obsolescent usage where a type of
3962 @code{int} is implied by the absence of type specifiers, such a list of
3963 specifiers and qualifiers may be an attribute specifier list with no
3964 other specifiers or qualifiers.
3966 At present, the first parameter in a function prototype must have some
3967 type specifier which is not an attribute specifier; this resolves an
3968 ambiguity in the interpretation of @code{void f(int
3969 (__attribute__((foo)) x))}, but is subject to change. At present, if
3970 the parentheses of a function declarator contain only attributes then
3971 those attributes are ignored, rather than yielding an error or warning
3972 or implying a single parameter of type int, but this is subject to
3975 An attribute specifier list may appear immediately before a declarator
3976 (other than the first) in a comma-separated list of declarators in a
3977 declaration of more than one identifier using a single list of
3978 specifiers and qualifiers. Such attribute specifiers apply
3979 only to the identifier before whose declarator they appear. For
3983 __attribute__((noreturn)) void d0 (void),
3984 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
3989 the @code{noreturn} attribute applies to all the functions
3990 declared; the @code{format} attribute only applies to @code{d1}.
3992 An attribute specifier list may appear immediately before the comma,
3993 @code{=} or semicolon terminating the declaration of an identifier other
3994 than a function definition. Such attribute specifiers apply
3995 to the declared object or function. Where an
3996 assembler name for an object or function is specified (@pxref{Asm
3997 Labels}), the attribute must follow the @code{asm}
4000 An attribute specifier list may, in future, be permitted to appear after
4001 the declarator in a function definition (before any old-style parameter
4002 declarations or the function body).
4004 Attribute specifiers may be mixed with type qualifiers appearing inside
4005 the @code{[]} of a parameter array declarator, in the C99 construct by
4006 which such qualifiers are applied to the pointer to which the array is
4007 implicitly converted. Such attribute specifiers apply to the pointer,
4008 not to the array, but at present this is not implemented and they are
4011 An attribute specifier list may appear at the start of a nested
4012 declarator. At present, there are some limitations in this usage: the
4013 attributes correctly apply to the declarator, but for most individual
4014 attributes the semantics this implies are not implemented.
4015 When attribute specifiers follow the @code{*} of a pointer
4016 declarator, they may be mixed with any type qualifiers present.
4017 The following describes the formal semantics of this syntax. It will make the
4018 most sense if you are familiar with the formal specification of
4019 declarators in the ISO C standard.
4021 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
4022 D1}, where @code{T} contains declaration specifiers that specify a type
4023 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
4024 contains an identifier @var{ident}. The type specified for @var{ident}
4025 for derived declarators whose type does not include an attribute
4026 specifier is as in the ISO C standard.
4028 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
4029 and the declaration @code{T D} specifies the type
4030 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4031 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4032 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
4034 If @code{D1} has the form @code{*
4035 @var{type-qualifier-and-attribute-specifier-list} D}, and the
4036 declaration @code{T D} specifies the type
4037 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4038 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4039 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
4045 void (__attribute__((noreturn)) ****f) (void);
4049 specifies the type ``pointer to pointer to pointer to pointer to
4050 non-returning function returning @code{void}''. As another example,
4053 char *__attribute__((aligned(8))) *f;
4057 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
4058 Note again that this does not work with most attributes; for example,
4059 the usage of @samp{aligned} and @samp{noreturn} attributes given above
4060 is not yet supported.
4062 For compatibility with existing code written for compiler versions that
4063 did not implement attributes on nested declarators, some laxity is
4064 allowed in the placing of attributes. If an attribute that only applies
4065 to types is applied to a declaration, it will be treated as applying to
4066 the type of that declaration. If an attribute that only applies to
4067 declarations is applied to the type of a declaration, it will be treated
4068 as applying to that declaration; and, for compatibility with code
4069 placing the attributes immediately before the identifier declared, such
4070 an attribute applied to a function return type will be treated as
4071 applying to the function type, and such an attribute applied to an array
4072 element type will be treated as applying to the array type. If an
4073 attribute that only applies to function types is applied to a
4074 pointer-to-function type, it will be treated as applying to the pointer
4075 target type; if such an attribute is applied to a function return type
4076 that is not a pointer-to-function type, it will be treated as applying
4077 to the function type.
4079 @node Function Prototypes
4080 @section Prototypes and Old-Style Function Definitions
4081 @cindex function prototype declarations
4082 @cindex old-style function definitions
4083 @cindex promotion of formal parameters
4085 GNU C extends ISO C to allow a function prototype to override a later
4086 old-style non-prototype definition. Consider the following example:
4089 /* @r{Use prototypes unless the compiler is old-fashioned.} */
4096 /* @r{Prototype function declaration.} */
4097 int isroot P((uid_t));
4099 /* @r{Old-style function definition.} */
4101 isroot (x) /* @r{??? lossage here ???} */
4108 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
4109 not allow this example, because subword arguments in old-style
4110 non-prototype definitions are promoted. Therefore in this example the
4111 function definition's argument is really an @code{int}, which does not
4112 match the prototype argument type of @code{short}.
4114 This restriction of ISO C makes it hard to write code that is portable
4115 to traditional C compilers, because the programmer does not know
4116 whether the @code{uid_t} type is @code{short}, @code{int}, or
4117 @code{long}. Therefore, in cases like these GNU C allows a prototype
4118 to override a later old-style definition. More precisely, in GNU C, a
4119 function prototype argument type overrides the argument type specified
4120 by a later old-style definition if the former type is the same as the
4121 latter type before promotion. Thus in GNU C the above example is
4122 equivalent to the following:
4135 GNU C++ does not support old-style function definitions, so this
4136 extension is irrelevant.
4139 @section C++ Style Comments
4141 @cindex C++ comments
4142 @cindex comments, C++ style
4144 In GNU C, you may use C++ style comments, which start with @samp{//} and
4145 continue until the end of the line. Many other C implementations allow
4146 such comments, and they are included in the 1999 C standard. However,
4147 C++ style comments are not recognized if you specify an @option{-std}
4148 option specifying a version of ISO C before C99, or @option{-ansi}
4149 (equivalent to @option{-std=c90}).
4152 @section Dollar Signs in Identifier Names
4154 @cindex dollar signs in identifier names
4155 @cindex identifier names, dollar signs in
4157 In GNU C, you may normally use dollar signs in identifier names.
4158 This is because many traditional C implementations allow such identifiers.
4159 However, dollar signs in identifiers are not supported on a few target
4160 machines, typically because the target assembler does not allow them.
4162 @node Character Escapes
4163 @section The Character @key{ESC} in Constants
4165 You can use the sequence @samp{\e} in a string or character constant to
4166 stand for the ASCII character @key{ESC}.
4169 @section Inquiring on Alignment of Types or Variables
4171 @cindex type alignment
4172 @cindex variable alignment
4174 The keyword @code{__alignof__} allows you to inquire about how an object
4175 is aligned, or the minimum alignment usually required by a type. Its
4176 syntax is just like @code{sizeof}.
4178 For example, if the target machine requires a @code{double} value to be
4179 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
4180 This is true on many RISC machines. On more traditional machine
4181 designs, @code{__alignof__ (double)} is 4 or even 2.
4183 Some machines never actually require alignment; they allow reference to any
4184 data type even at an odd address. For these machines, @code{__alignof__}
4185 reports the smallest alignment that GCC will give the data type, usually as
4186 mandated by the target ABI.
4188 If the operand of @code{__alignof__} is an lvalue rather than a type,
4189 its value is the required alignment for its type, taking into account
4190 any minimum alignment specified with GCC's @code{__attribute__}
4191 extension (@pxref{Variable Attributes}). For example, after this
4195 struct foo @{ int x; char y; @} foo1;
4199 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
4200 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
4202 It is an error to ask for the alignment of an incomplete type.
4204 @node Variable Attributes
4205 @section Specifying Attributes of Variables
4206 @cindex attribute of variables
4207 @cindex variable attributes
4209 The keyword @code{__attribute__} allows you to specify special
4210 attributes of variables or structure fields. This keyword is followed
4211 by an attribute specification inside double parentheses. Some
4212 attributes are currently defined generically for variables.
4213 Other attributes are defined for variables on particular target
4214 systems. Other attributes are available for functions
4215 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
4216 Other front ends might define more attributes
4217 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
4219 You may also specify attributes with @samp{__} preceding and following
4220 each keyword. This allows you to use them in header files without
4221 being concerned about a possible macro of the same name. For example,
4222 you may use @code{__aligned__} instead of @code{aligned}.
4224 @xref{Attribute Syntax}, for details of the exact syntax for using
4228 @cindex @code{aligned} attribute
4229 @item aligned (@var{alignment})
4230 This attribute specifies a minimum alignment for the variable or
4231 structure field, measured in bytes. For example, the declaration:
4234 int x __attribute__ ((aligned (16))) = 0;
4238 causes the compiler to allocate the global variable @code{x} on a
4239 16-byte boundary. On a 68040, this could be used in conjunction with
4240 an @code{asm} expression to access the @code{move16} instruction which
4241 requires 16-byte aligned operands.
4243 You can also specify the alignment of structure fields. For example, to
4244 create a double-word aligned @code{int} pair, you could write:
4247 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
4251 This is an alternative to creating a union with a @code{double} member
4252 that forces the union to be double-word aligned.
4254 As in the preceding examples, you can explicitly specify the alignment
4255 (in bytes) that you wish the compiler to use for a given variable or
4256 structure field. Alternatively, you can leave out the alignment factor
4257 and just ask the compiler to align a variable or field to the
4258 default alignment for the target architecture you are compiling for.
4259 The default alignment is sufficient for all scalar types, but may not be
4260 enough for all vector types on a target which supports vector operations.
4261 The default alignment is fixed for a particular target ABI.
4263 Gcc also provides a target specific macro @code{__BIGGEST_ALIGNMENT__},
4264 which is the largest alignment ever used for any data type on the
4265 target machine you are compiling for. For example, you could write:
4268 short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__)));
4271 The compiler automatically sets the alignment for the declared
4272 variable or field to @code{__BIGGEST_ALIGNMENT__}. Doing this can
4273 often make copy operations more efficient, because the compiler can
4274 use whatever instructions copy the biggest chunks of memory when
4275 performing copies to or from the variables or fields that you have
4276 aligned this way. Note that the value of @code{__BIGGEST_ALIGNMENT__}
4277 may change depending on command line options.
4279 When used on a struct, or struct member, the @code{aligned} attribute can
4280 only increase the alignment; in order to decrease it, the @code{packed}
4281 attribute must be specified as well. When used as part of a typedef, the
4282 @code{aligned} attribute can both increase and decrease alignment, and
4283 specifying the @code{packed} attribute will generate a warning.
4285 Note that the effectiveness of @code{aligned} attributes may be limited
4286 by inherent limitations in your linker. On many systems, the linker is
4287 only able to arrange for variables to be aligned up to a certain maximum
4288 alignment. (For some linkers, the maximum supported alignment may
4289 be very very small.) If your linker is only able to align variables
4290 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
4291 in an @code{__attribute__} will still only provide you with 8 byte
4292 alignment. See your linker documentation for further information.
4294 The @code{aligned} attribute can also be used for functions
4295 (@pxref{Function Attributes}.)
4297 @item cleanup (@var{cleanup_function})
4298 @cindex @code{cleanup} attribute
4299 The @code{cleanup} attribute runs a function when the variable goes
4300 out of scope. This attribute can only be applied to auto function
4301 scope variables; it may not be applied to parameters or variables
4302 with static storage duration. The function must take one parameter,
4303 a pointer to a type compatible with the variable. The return value
4304 of the function (if any) is ignored.
4306 If @option{-fexceptions} is enabled, then @var{cleanup_function}
4307 will be run during the stack unwinding that happens during the
4308 processing of the exception. Note that the @code{cleanup} attribute
4309 does not allow the exception to be caught, only to perform an action.
4310 It is undefined what happens if @var{cleanup_function} does not
4315 @cindex @code{common} attribute
4316 @cindex @code{nocommon} attribute
4319 The @code{common} attribute requests GCC to place a variable in
4320 ``common'' storage. The @code{nocommon} attribute requests the
4321 opposite---to allocate space for it directly.
4323 These attributes override the default chosen by the
4324 @option{-fno-common} and @option{-fcommon} flags respectively.
4327 @itemx deprecated (@var{msg})
4328 @cindex @code{deprecated} attribute
4329 The @code{deprecated} attribute results in a warning if the variable
4330 is used anywhere in the source file. This is useful when identifying
4331 variables that are expected to be removed in a future version of a
4332 program. The warning also includes the location of the declaration
4333 of the deprecated variable, to enable users to easily find further
4334 information about why the variable is deprecated, or what they should
4335 do instead. Note that the warning only occurs for uses:
4338 extern int old_var __attribute__ ((deprecated));
4340 int new_fn () @{ return old_var; @}
4343 results in a warning on line 3 but not line 2. The optional msg
4344 argument, which must be a string, will be printed in the warning if
4347 The @code{deprecated} attribute can also be used for functions and
4348 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
4350 @item mode (@var{mode})
4351 @cindex @code{mode} attribute
4352 This attribute specifies the data type for the declaration---whichever
4353 type corresponds to the mode @var{mode}. This in effect lets you
4354 request an integer or floating point type according to its width.
4356 You may also specify a mode of @samp{byte} or @samp{__byte__} to
4357 indicate the mode corresponding to a one-byte integer, @samp{word} or
4358 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
4359 or @samp{__pointer__} for the mode used to represent pointers.
4362 @cindex @code{packed} attribute
4363 The @code{packed} attribute specifies that a variable or structure field
4364 should have the smallest possible alignment---one byte for a variable,
4365 and one bit for a field, unless you specify a larger value with the
4366 @code{aligned} attribute.
4368 Here is a structure in which the field @code{x} is packed, so that it
4369 immediately follows @code{a}:
4375 int x[2] __attribute__ ((packed));
4379 @emph{Note:} The 4.1, 4.2 and 4.3 series of GCC ignore the
4380 @code{packed} attribute on bit-fields of type @code{char}. This has
4381 been fixed in GCC 4.4 but the change can lead to differences in the
4382 structure layout. See the documentation of
4383 @option{-Wpacked-bitfield-compat} for more information.
4385 @item section ("@var{section-name}")
4386 @cindex @code{section} variable attribute
4387 Normally, the compiler places the objects it generates in sections like
4388 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
4389 or you need certain particular variables to appear in special sections,
4390 for example to map to special hardware. The @code{section}
4391 attribute specifies that a variable (or function) lives in a particular
4392 section. For example, this small program uses several specific section names:
4395 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
4396 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
4397 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
4398 int init_data __attribute__ ((section ("INITDATA")));
4402 /* @r{Initialize stack pointer} */
4403 init_sp (stack + sizeof (stack));
4405 /* @r{Initialize initialized data} */
4406 memcpy (&init_data, &data, &edata - &data);
4408 /* @r{Turn on the serial ports} */
4415 Use the @code{section} attribute with
4416 @emph{global} variables and not @emph{local} variables,
4417 as shown in the example.
4419 You may use the @code{section} attribute with initialized or
4420 uninitialized global variables but the linker requires
4421 each object be defined once, with the exception that uninitialized
4422 variables tentatively go in the @code{common} (or @code{bss}) section
4423 and can be multiply ``defined''. Using the @code{section} attribute
4424 will change what section the variable goes into and may cause the
4425 linker to issue an error if an uninitialized variable has multiple
4426 definitions. You can force a variable to be initialized with the
4427 @option{-fno-common} flag or the @code{nocommon} attribute.
4429 Some file formats do not support arbitrary sections so the @code{section}
4430 attribute is not available on all platforms.
4431 If you need to map the entire contents of a module to a particular
4432 section, consider using the facilities of the linker instead.
4435 @cindex @code{shared} variable attribute
4436 On Microsoft Windows, in addition to putting variable definitions in a named
4437 section, the section can also be shared among all running copies of an
4438 executable or DLL@. For example, this small program defines shared data
4439 by putting it in a named section @code{shared} and marking the section
4443 int foo __attribute__((section ("shared"), shared)) = 0;
4448 /* @r{Read and write foo. All running
4449 copies see the same value.} */
4455 You may only use the @code{shared} attribute along with @code{section}
4456 attribute with a fully initialized global definition because of the way
4457 linkers work. See @code{section} attribute for more information.
4459 The @code{shared} attribute is only available on Microsoft Windows@.
4461 @item tls_model ("@var{tls_model}")
4462 @cindex @code{tls_model} attribute
4463 The @code{tls_model} attribute sets thread-local storage model
4464 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
4465 overriding @option{-ftls-model=} command-line switch on a per-variable
4467 The @var{tls_model} argument should be one of @code{global-dynamic},
4468 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
4470 Not all targets support this attribute.
4473 This attribute, attached to a variable, means that the variable is meant
4474 to be possibly unused. GCC will not produce a warning for this
4478 This attribute, attached to a variable, means that the variable must be
4479 emitted even if it appears that the variable is not referenced.
4481 @item vector_size (@var{bytes})
4482 This attribute specifies the vector size for the variable, measured in
4483 bytes. For example, the declaration:
4486 int foo __attribute__ ((vector_size (16)));
4490 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
4491 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
4492 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
4494 This attribute is only applicable to integral and float scalars,
4495 although arrays, pointers, and function return values are allowed in
4496 conjunction with this construct.
4498 Aggregates with this attribute are invalid, even if they are of the same
4499 size as a corresponding scalar. For example, the declaration:
4502 struct S @{ int a; @};
4503 struct S __attribute__ ((vector_size (16))) foo;
4507 is invalid even if the size of the structure is the same as the size of
4511 The @code{selectany} attribute causes an initialized global variable to
4512 have link-once semantics. When multiple definitions of the variable are
4513 encountered by the linker, the first is selected and the remainder are
4514 discarded. Following usage by the Microsoft compiler, the linker is told
4515 @emph{not} to warn about size or content differences of the multiple
4518 Although the primary usage of this attribute is for POD types, the
4519 attribute can also be applied to global C++ objects that are initialized
4520 by a constructor. In this case, the static initialization and destruction
4521 code for the object is emitted in each translation defining the object,
4522 but the calls to the constructor and destructor are protected by a
4523 link-once guard variable.
4525 The @code{selectany} attribute is only available on Microsoft Windows
4526 targets. You can use @code{__declspec (selectany)} as a synonym for
4527 @code{__attribute__ ((selectany))} for compatibility with other
4531 The @code{weak} attribute is described in @ref{Function Attributes}.
4534 The @code{dllimport} attribute is described in @ref{Function Attributes}.
4537 The @code{dllexport} attribute is described in @ref{Function Attributes}.
4541 @subsection Blackfin Variable Attributes
4543 Three attributes are currently defined for the Blackfin.
4549 @cindex @code{l1_data} variable attribute
4550 @cindex @code{l1_data_A} variable attribute
4551 @cindex @code{l1_data_B} variable attribute
4552 Use these attributes on the Blackfin to place the variable into L1 Data SRAM.
4553 Variables with @code{l1_data} attribute will be put into the specific section
4554 named @code{.l1.data}. Those with @code{l1_data_A} attribute will be put into
4555 the specific section named @code{.l1.data.A}. Those with @code{l1_data_B}
4556 attribute will be put into the specific section named @code{.l1.data.B}.
4559 @cindex @code{l2} variable attribute
4560 Use this attribute on the Blackfin to place the variable into L2 SRAM.
4561 Variables with @code{l2} attribute will be put into the specific section
4562 named @code{.l2.data}.
4565 @subsection M32R/D Variable Attributes
4567 One attribute is currently defined for the M32R/D@.
4570 @item model (@var{model-name})
4571 @cindex variable addressability on the M32R/D
4572 Use this attribute on the M32R/D to set the addressability of an object.
4573 The identifier @var{model-name} is one of @code{small}, @code{medium},
4574 or @code{large}, representing each of the code models.
4576 Small model objects live in the lower 16MB of memory (so that their
4577 addresses can be loaded with the @code{ld24} instruction).
4579 Medium and large model objects may live anywhere in the 32-bit address space
4580 (the compiler will generate @code{seth/add3} instructions to load their
4584 @anchor{MeP Variable Attributes}
4585 @subsection MeP Variable Attributes
4587 The MeP target has a number of addressing modes and busses. The
4588 @code{near} space spans the standard memory space's first 16 megabytes
4589 (24 bits). The @code{far} space spans the entire 32-bit memory space.
4590 The @code{based} space is a 128 byte region in the memory space which
4591 is addressed relative to the @code{$tp} register. The @code{tiny}
4592 space is a 65536 byte region relative to the @code{$gp} register. In
4593 addition to these memory regions, the MeP target has a separate 16-bit
4594 control bus which is specified with @code{cb} attributes.
4599 Any variable with the @code{based} attribute will be assigned to the
4600 @code{.based} section, and will be accessed with relative to the
4601 @code{$tp} register.
4604 Likewise, the @code{tiny} attribute assigned variables to the
4605 @code{.tiny} section, relative to the @code{$gp} register.
4608 Variables with the @code{near} attribute are assumed to have addresses
4609 that fit in a 24-bit addressing mode. This is the default for large
4610 variables (@code{-mtiny=4} is the default) but this attribute can
4611 override @code{-mtiny=} for small variables, or override @code{-ml}.
4614 Variables with the @code{far} attribute are addressed using a full
4615 32-bit address. Since this covers the entire memory space, this
4616 allows modules to make no assumptions about where variables might be
4620 @itemx io (@var{addr})
4621 Variables with the @code{io} attribute are used to address
4622 memory-mapped peripherals. If an address is specified, the variable
4623 is assigned that address, else it is not assigned an address (it is
4624 assumed some other module will assign an address). Example:
4627 int timer_count __attribute__((io(0x123)));
4631 @itemx cb (@var{addr})
4632 Variables with the @code{cb} attribute are used to access the control
4633 bus, using special instructions. @code{addr} indicates the control bus
4637 int cpu_clock __attribute__((cb(0x123)));
4642 @anchor{i386 Variable Attributes}
4643 @subsection i386 Variable Attributes
4645 Two attributes are currently defined for i386 configurations:
4646 @code{ms_struct} and @code{gcc_struct}
4651 @cindex @code{ms_struct} attribute
4652 @cindex @code{gcc_struct} attribute
4654 If @code{packed} is used on a structure, or if bit-fields are used
4655 it may be that the Microsoft ABI packs them differently
4656 than GCC would normally pack them. Particularly when moving packed
4657 data between functions compiled with GCC and the native Microsoft compiler
4658 (either via function call or as data in a file), it may be necessary to access
4661 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
4662 compilers to match the native Microsoft compiler.
4664 The Microsoft structure layout algorithm is fairly simple with the exception
4665 of the bitfield packing:
4667 The padding and alignment of members of structures and whether a bit field
4668 can straddle a storage-unit boundary
4671 @item Structure members are stored sequentially in the order in which they are
4672 declared: the first member has the lowest memory address and the last member
4675 @item Every data object has an alignment-requirement. The alignment-requirement
4676 for all data except structures, unions, and arrays is either the size of the
4677 object or the current packing size (specified with either the aligned attribute
4678 or the pack pragma), whichever is less. For structures, unions, and arrays,
4679 the alignment-requirement is the largest alignment-requirement of its members.
4680 Every object is allocated an offset so that:
4682 offset % alignment-requirement == 0
4684 @item Adjacent bit fields are packed into the same 1-, 2-, or 4-byte allocation
4685 unit if the integral types are the same size and if the next bit field fits
4686 into the current allocation unit without crossing the boundary imposed by the
4687 common alignment requirements of the bit fields.
4690 Handling of zero-length bitfields:
4692 MSVC interprets zero-length bitfields in the following ways:
4695 @item If a zero-length bitfield is inserted between two bitfields that would
4696 normally be coalesced, the bitfields will not be coalesced.
4703 unsigned long bf_1 : 12;
4705 unsigned long bf_2 : 12;
4709 The size of @code{t1} would be 8 bytes with the zero-length bitfield. If the
4710 zero-length bitfield were removed, @code{t1}'s size would be 4 bytes.
4712 @item If a zero-length bitfield is inserted after a bitfield, @code{foo}, and the
4713 alignment of the zero-length bitfield is greater than the member that follows it,
4714 @code{bar}, @code{bar} will be aligned as the type of the zero-length bitfield.
4734 For @code{t2}, @code{bar} will be placed at offset 2, rather than offset 1.
4735 Accordingly, the size of @code{t2} will be 4. For @code{t3}, the zero-length
4736 bitfield will not affect the alignment of @code{bar} or, as a result, the size
4739 Taking this into account, it is important to note the following:
4742 @item If a zero-length bitfield follows a normal bitfield, the type of the
4743 zero-length bitfield may affect the alignment of the structure as whole. For
4744 example, @code{t2} has a size of 4 bytes, since the zero-length bitfield follows a
4745 normal bitfield, and is of type short.
4747 @item Even if a zero-length bitfield is not followed by a normal bitfield, it may
4748 still affect the alignment of the structure:
4758 Here, @code{t4} will take up 4 bytes.
4761 @item Zero-length bitfields following non-bitfield members are ignored:
4772 Here, @code{t5} will take up 2 bytes.
4776 @subsection PowerPC Variable Attributes
4778 Three attributes currently are defined for PowerPC configurations:
4779 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
4781 For full documentation of the struct attributes please see the
4782 documentation in @ref{i386 Variable Attributes}.
4784 For documentation of @code{altivec} attribute please see the
4785 documentation in @ref{PowerPC Type Attributes}.
4787 @subsection SPU Variable Attributes
4789 The SPU supports the @code{spu_vector} attribute for variables. For
4790 documentation of this attribute please see the documentation in
4791 @ref{SPU Type Attributes}.
4793 @subsection Xstormy16 Variable Attributes
4795 One attribute is currently defined for xstormy16 configurations:
4800 @cindex @code{below100} attribute
4802 If a variable has the @code{below100} attribute (@code{BELOW100} is
4803 allowed also), GCC will place the variable in the first 0x100 bytes of
4804 memory and use special opcodes to access it. Such variables will be
4805 placed in either the @code{.bss_below100} section or the
4806 @code{.data_below100} section.
4810 @subsection AVR Variable Attributes
4814 @cindex @code{progmem} variable attribute
4815 The @code{progmem} attribute is used on the AVR to place data in the Program
4816 Memory address space. The AVR is a Harvard Architecture processor and data
4817 normally resides in the Data Memory address space.
4820 @node Type Attributes
4821 @section Specifying Attributes of Types
4822 @cindex attribute of types
4823 @cindex type attributes
4825 The keyword @code{__attribute__} allows you to specify special
4826 attributes of @code{struct} and @code{union} types when you define
4827 such types. This keyword is followed by an attribute specification
4828 inside double parentheses. Seven attributes are currently defined for
4829 types: @code{aligned}, @code{packed}, @code{transparent_union},
4830 @code{unused}, @code{deprecated}, @code{visibility}, and
4831 @code{may_alias}. Other attributes are defined for functions
4832 (@pxref{Function Attributes}) and for variables (@pxref{Variable
4835 You may also specify any one of these attributes with @samp{__}
4836 preceding and following its keyword. This allows you to use these
4837 attributes in header files without being concerned about a possible
4838 macro of the same name. For example, you may use @code{__aligned__}
4839 instead of @code{aligned}.
4841 You may specify type attributes in an enum, struct or union type
4842 declaration or definition, or for other types in a @code{typedef}
4845 For an enum, struct or union type, you may specify attributes either
4846 between the enum, struct or union tag and the name of the type, or
4847 just past the closing curly brace of the @emph{definition}. The
4848 former syntax is preferred.
4850 @xref{Attribute Syntax}, for details of the exact syntax for using
4854 @cindex @code{aligned} attribute
4855 @item aligned (@var{alignment})
4856 This attribute specifies a minimum alignment (in bytes) for variables
4857 of the specified type. For example, the declarations:
4860 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
4861 typedef int more_aligned_int __attribute__ ((aligned (8)));
4865 force the compiler to insure (as far as it can) that each variable whose
4866 type is @code{struct S} or @code{more_aligned_int} will be allocated and
4867 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
4868 variables of type @code{struct S} aligned to 8-byte boundaries allows
4869 the compiler to use the @code{ldd} and @code{std} (doubleword load and
4870 store) instructions when copying one variable of type @code{struct S} to
4871 another, thus improving run-time efficiency.
4873 Note that the alignment of any given @code{struct} or @code{union} type
4874 is required by the ISO C standard to be at least a perfect multiple of
4875 the lowest common multiple of the alignments of all of the members of
4876 the @code{struct} or @code{union} in question. This means that you @emph{can}
4877 effectively adjust the alignment of a @code{struct} or @code{union}
4878 type by attaching an @code{aligned} attribute to any one of the members
4879 of such a type, but the notation illustrated in the example above is a
4880 more obvious, intuitive, and readable way to request the compiler to
4881 adjust the alignment of an entire @code{struct} or @code{union} type.
4883 As in the preceding example, you can explicitly specify the alignment
4884 (in bytes) that you wish the compiler to use for a given @code{struct}
4885 or @code{union} type. Alternatively, you can leave out the alignment factor
4886 and just ask the compiler to align a type to the maximum
4887 useful alignment for the target machine you are compiling for. For
4888 example, you could write:
4891 struct S @{ short f[3]; @} __attribute__ ((aligned));
4894 Whenever you leave out the alignment factor in an @code{aligned}
4895 attribute specification, the compiler automatically sets the alignment
4896 for the type to the largest alignment which is ever used for any data
4897 type on the target machine you are compiling for. Doing this can often
4898 make copy operations more efficient, because the compiler can use
4899 whatever instructions copy the biggest chunks of memory when performing
4900 copies to or from the variables which have types that you have aligned
4903 In the example above, if the size of each @code{short} is 2 bytes, then
4904 the size of the entire @code{struct S} type is 6 bytes. The smallest
4905 power of two which is greater than or equal to that is 8, so the
4906 compiler sets the alignment for the entire @code{struct S} type to 8
4909 Note that although you can ask the compiler to select a time-efficient
4910 alignment for a given type and then declare only individual stand-alone
4911 objects of that type, the compiler's ability to select a time-efficient
4912 alignment is primarily useful only when you plan to create arrays of
4913 variables having the relevant (efficiently aligned) type. If you
4914 declare or use arrays of variables of an efficiently-aligned type, then
4915 it is likely that your program will also be doing pointer arithmetic (or
4916 subscripting, which amounts to the same thing) on pointers to the
4917 relevant type, and the code that the compiler generates for these
4918 pointer arithmetic operations will often be more efficient for
4919 efficiently-aligned types than for other types.
4921 The @code{aligned} attribute can only increase the alignment; but you
4922 can decrease it by specifying @code{packed} as well. See below.
4924 Note that the effectiveness of @code{aligned} attributes may be limited
4925 by inherent limitations in your linker. On many systems, the linker is
4926 only able to arrange for variables to be aligned up to a certain maximum
4927 alignment. (For some linkers, the maximum supported alignment may
4928 be very very small.) If your linker is only able to align variables
4929 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
4930 in an @code{__attribute__} will still only provide you with 8 byte
4931 alignment. See your linker documentation for further information.
4934 This attribute, attached to @code{struct} or @code{union} type
4935 definition, specifies that each member (other than zero-width bitfields)
4936 of the structure or union is placed to minimize the memory required. When
4937 attached to an @code{enum} definition, it indicates that the smallest
4938 integral type should be used.
4940 @opindex fshort-enums
4941 Specifying this attribute for @code{struct} and @code{union} types is
4942 equivalent to specifying the @code{packed} attribute on each of the
4943 structure or union members. Specifying the @option{-fshort-enums}
4944 flag on the line is equivalent to specifying the @code{packed}
4945 attribute on all @code{enum} definitions.
4947 In the following example @code{struct my_packed_struct}'s members are
4948 packed closely together, but the internal layout of its @code{s} member
4949 is not packed---to do that, @code{struct my_unpacked_struct} would need to
4953 struct my_unpacked_struct
4959 struct __attribute__ ((__packed__)) my_packed_struct
4963 struct my_unpacked_struct s;
4967 You may only specify this attribute on the definition of an @code{enum},
4968 @code{struct} or @code{union}, not on a @code{typedef} which does not
4969 also define the enumerated type, structure or union.
4971 @item transparent_union
4972 This attribute, attached to a @code{union} type definition, indicates
4973 that any function parameter having that union type causes calls to that
4974 function to be treated in a special way.
4976 First, the argument corresponding to a transparent union type can be of
4977 any type in the union; no cast is required. Also, if the union contains
4978 a pointer type, the corresponding argument can be a null pointer
4979 constant or a void pointer expression; and if the union contains a void
4980 pointer type, the corresponding argument can be any pointer expression.
4981 If the union member type is a pointer, qualifiers like @code{const} on
4982 the referenced type must be respected, just as with normal pointer
4985 Second, the argument is passed to the function using the calling
4986 conventions of the first member of the transparent union, not the calling
4987 conventions of the union itself. All members of the union must have the
4988 same machine representation; this is necessary for this argument passing
4991 Transparent unions are designed for library functions that have multiple
4992 interfaces for compatibility reasons. For example, suppose the
4993 @code{wait} function must accept either a value of type @code{int *} to
4994 comply with Posix, or a value of type @code{union wait *} to comply with
4995 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
4996 @code{wait} would accept both kinds of arguments, but it would also
4997 accept any other pointer type and this would make argument type checking
4998 less useful. Instead, @code{<sys/wait.h>} might define the interface
5002 typedef union __attribute__ ((__transparent_union__))
5006 @} wait_status_ptr_t;
5008 pid_t wait (wait_status_ptr_t);
5011 This interface allows either @code{int *} or @code{union wait *}
5012 arguments to be passed, using the @code{int *} calling convention.
5013 The program can call @code{wait} with arguments of either type:
5016 int w1 () @{ int w; return wait (&w); @}
5017 int w2 () @{ union wait w; return wait (&w); @}
5020 With this interface, @code{wait}'s implementation might look like this:
5023 pid_t wait (wait_status_ptr_t p)
5025 return waitpid (-1, p.__ip, 0);
5030 When attached to a type (including a @code{union} or a @code{struct}),
5031 this attribute means that variables of that type are meant to appear
5032 possibly unused. GCC will not produce a warning for any variables of
5033 that type, even if the variable appears to do nothing. This is often
5034 the case with lock or thread classes, which are usually defined and then
5035 not referenced, but contain constructors and destructors that have
5036 nontrivial bookkeeping functions.
5039 @itemx deprecated (@var{msg})
5040 The @code{deprecated} attribute results in a warning if the type
5041 is used anywhere in the source file. This is useful when identifying
5042 types that are expected to be removed in a future version of a program.
5043 If possible, the warning also includes the location of the declaration
5044 of the deprecated type, to enable users to easily find further
5045 information about why the type is deprecated, or what they should do
5046 instead. Note that the warnings only occur for uses and then only
5047 if the type is being applied to an identifier that itself is not being
5048 declared as deprecated.
5051 typedef int T1 __attribute__ ((deprecated));
5055 typedef T1 T3 __attribute__ ((deprecated));
5056 T3 z __attribute__ ((deprecated));
5059 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
5060 warning is issued for line 4 because T2 is not explicitly
5061 deprecated. Line 5 has no warning because T3 is explicitly
5062 deprecated. Similarly for line 6. The optional msg
5063 argument, which must be a string, will be printed in the warning if
5066 The @code{deprecated} attribute can also be used for functions and
5067 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
5070 Accesses through pointers to types with this attribute are not subject
5071 to type-based alias analysis, but are instead assumed to be able to alias
5072 any other type of objects. In the context of 6.5/7 an lvalue expression
5073 dereferencing such a pointer is treated like having a character type.
5074 See @option{-fstrict-aliasing} for more information on aliasing issues.
5075 This extension exists to support some vector APIs, in which pointers to
5076 one vector type are permitted to alias pointers to a different vector type.
5078 Note that an object of a type with this attribute does not have any
5084 typedef short __attribute__((__may_alias__)) short_a;
5090 short_a *b = (short_a *) &a;
5094 if (a == 0x12345678)
5101 If you replaced @code{short_a} with @code{short} in the variable
5102 declaration, the above program would abort when compiled with
5103 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
5104 above in recent GCC versions.
5107 In C++, attribute visibility (@pxref{Function Attributes}) can also be
5108 applied to class, struct, union and enum types. Unlike other type
5109 attributes, the attribute must appear between the initial keyword and
5110 the name of the type; it cannot appear after the body of the type.
5112 Note that the type visibility is applied to vague linkage entities
5113 associated with the class (vtable, typeinfo node, etc.). In
5114 particular, if a class is thrown as an exception in one shared object
5115 and caught in another, the class must have default visibility.
5116 Otherwise the two shared objects will be unable to use the same
5117 typeinfo node and exception handling will break.
5121 @subsection ARM Type Attributes
5123 On those ARM targets that support @code{dllimport} (such as Symbian
5124 OS), you can use the @code{notshared} attribute to indicate that the
5125 virtual table and other similar data for a class should not be
5126 exported from a DLL@. For example:
5129 class __declspec(notshared) C @{
5131 __declspec(dllimport) C();
5135 __declspec(dllexport)
5139 In this code, @code{C::C} is exported from the current DLL, but the
5140 virtual table for @code{C} is not exported. (You can use
5141 @code{__attribute__} instead of @code{__declspec} if you prefer, but
5142 most Symbian OS code uses @code{__declspec}.)
5144 @anchor{MeP Type Attributes}
5145 @subsection MeP Type Attributes
5147 Many of the MeP variable attributes may be applied to types as well.
5148 Specifically, the @code{based}, @code{tiny}, @code{near}, and
5149 @code{far} attributes may be applied to either. The @code{io} and
5150 @code{cb} attributes may not be applied to types.
5152 @anchor{i386 Type Attributes}
5153 @subsection i386 Type Attributes
5155 Two attributes are currently defined for i386 configurations:
5156 @code{ms_struct} and @code{gcc_struct}.
5162 @cindex @code{ms_struct}
5163 @cindex @code{gcc_struct}
5165 If @code{packed} is used on a structure, or if bit-fields are used
5166 it may be that the Microsoft ABI packs them differently
5167 than GCC would normally pack them. Particularly when moving packed
5168 data between functions compiled with GCC and the native Microsoft compiler
5169 (either via function call or as data in a file), it may be necessary to access
5172 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
5173 compilers to match the native Microsoft compiler.
5176 To specify multiple attributes, separate them by commas within the
5177 double parentheses: for example, @samp{__attribute__ ((aligned (16),
5180 @anchor{PowerPC Type Attributes}
5181 @subsection PowerPC Type Attributes
5183 Three attributes currently are defined for PowerPC configurations:
5184 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
5186 For full documentation of the @code{ms_struct} and @code{gcc_struct}
5187 attributes please see the documentation in @ref{i386 Type Attributes}.
5189 The @code{altivec} attribute allows one to declare AltiVec vector data
5190 types supported by the AltiVec Programming Interface Manual. The
5191 attribute requires an argument to specify one of three vector types:
5192 @code{vector__}, @code{pixel__} (always followed by unsigned short),
5193 and @code{bool__} (always followed by unsigned).
5196 __attribute__((altivec(vector__)))
5197 __attribute__((altivec(pixel__))) unsigned short
5198 __attribute__((altivec(bool__))) unsigned
5201 These attributes mainly are intended to support the @code{__vector},
5202 @code{__pixel}, and @code{__bool} AltiVec keywords.
5204 @anchor{SPU Type Attributes}
5205 @subsection SPU Type Attributes
5207 The SPU supports the @code{spu_vector} attribute for types. This attribute
5208 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
5209 Language Extensions Specification. It is intended to support the
5210 @code{__vector} keyword.
5214 @section An Inline Function is As Fast As a Macro
5215 @cindex inline functions
5216 @cindex integrating function code
5218 @cindex macros, inline alternative
5220 By declaring a function inline, you can direct GCC to make
5221 calls to that function faster. One way GCC can achieve this is to
5222 integrate that function's code into the code for its callers. This
5223 makes execution faster by eliminating the function-call overhead; in
5224 addition, if any of the actual argument values are constant, their
5225 known values may permit simplifications at compile time so that not
5226 all of the inline function's code needs to be included. The effect on
5227 code size is less predictable; object code may be larger or smaller
5228 with function inlining, depending on the particular case. You can
5229 also direct GCC to try to integrate all ``simple enough'' functions
5230 into their callers with the option @option{-finline-functions}.
5232 GCC implements three different semantics of declaring a function
5233 inline. One is available with @option{-std=gnu89} or
5234 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
5235 on all inline declarations, another when
5236 @option{-std=c99}, @option{-std=c1x},
5237 @option{-std=gnu99} or @option{-std=gnu1x}
5238 (without @option{-fgnu89-inline}), and the third
5239 is used when compiling C++.
5241 To declare a function inline, use the @code{inline} keyword in its
5242 declaration, like this:
5252 If you are writing a header file to be included in ISO C90 programs, write
5253 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
5255 The three types of inlining behave similarly in two important cases:
5256 when the @code{inline} keyword is used on a @code{static} function,
5257 like the example above, and when a function is first declared without
5258 using the @code{inline} keyword and then is defined with
5259 @code{inline}, like this:
5262 extern int inc (int *a);
5270 In both of these common cases, the program behaves the same as if you
5271 had not used the @code{inline} keyword, except for its speed.
5273 @cindex inline functions, omission of
5274 @opindex fkeep-inline-functions
5275 When a function is both inline and @code{static}, if all calls to the
5276 function are integrated into the caller, and the function's address is
5277 never used, then the function's own assembler code is never referenced.
5278 In this case, GCC does not actually output assembler code for the
5279 function, unless you specify the option @option{-fkeep-inline-functions}.
5280 Some calls cannot be integrated for various reasons (in particular,
5281 calls that precede the function's definition cannot be integrated, and
5282 neither can recursive calls within the definition). If there is a
5283 nonintegrated call, then the function is compiled to assembler code as
5284 usual. The function must also be compiled as usual if the program
5285 refers to its address, because that can't be inlined.
5288 Note that certain usages in a function definition can make it unsuitable
5289 for inline substitution. Among these usages are: use of varargs, use of
5290 alloca, use of variable sized data types (@pxref{Variable Length}),
5291 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
5292 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
5293 will warn when a function marked @code{inline} could not be substituted,
5294 and will give the reason for the failure.
5296 @cindex automatic @code{inline} for C++ member fns
5297 @cindex @code{inline} automatic for C++ member fns
5298 @cindex member fns, automatically @code{inline}
5299 @cindex C++ member fns, automatically @code{inline}
5300 @opindex fno-default-inline
5301 As required by ISO C++, GCC considers member functions defined within
5302 the body of a class to be marked inline even if they are
5303 not explicitly declared with the @code{inline} keyword. You can
5304 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
5305 Options,,Options Controlling C++ Dialect}.
5307 GCC does not inline any functions when not optimizing unless you specify
5308 the @samp{always_inline} attribute for the function, like this:
5311 /* @r{Prototype.} */
5312 inline void foo (const char) __attribute__((always_inline));
5315 The remainder of this section is specific to GNU C90 inlining.
5317 @cindex non-static inline function
5318 When an inline function is not @code{static}, then the compiler must assume
5319 that there may be calls from other source files; since a global symbol can
5320 be defined only once in any program, the function must not be defined in
5321 the other source files, so the calls therein cannot be integrated.
5322 Therefore, a non-@code{static} inline function is always compiled on its
5323 own in the usual fashion.
5325 If you specify both @code{inline} and @code{extern} in the function
5326 definition, then the definition is used only for inlining. In no case
5327 is the function compiled on its own, not even if you refer to its
5328 address explicitly. Such an address becomes an external reference, as
5329 if you had only declared the function, and had not defined it.
5331 This combination of @code{inline} and @code{extern} has almost the
5332 effect of a macro. The way to use it is to put a function definition in
5333 a header file with these keywords, and put another copy of the
5334 definition (lacking @code{inline} and @code{extern}) in a library file.
5335 The definition in the header file will cause most calls to the function
5336 to be inlined. If any uses of the function remain, they will refer to
5337 the single copy in the library.
5340 @section When is a Volatile Object Accessed?
5341 @cindex accessing volatiles
5342 @cindex volatile read
5343 @cindex volatile write
5344 @cindex volatile access
5346 C has the concept of volatile objects. These are normally accessed by
5347 pointers and used for accessing hardware or inter-thread
5348 communication. The standard encourage compilers to refrain from
5349 optimizations concerning accesses to volatile objects, but leaves it
5350 implementation defined as to what constitutes a volatile access. The
5351 minimum requirement is that at a sequence point all previous accesses
5352 to volatile objects have stabilized and no subsequent accesses have
5353 occurred. Thus an implementation is free to reorder and combine
5354 volatile accesses which occur between sequence points, but cannot do
5355 so for accesses across a sequence point. The use of volatiles does
5356 not allow you to violate the restriction on updating objects multiple
5357 times between two sequence points.
5359 Accesses to non-volatile objects are not ordered with respect to
5360 volatile accesses. You cannot use a volatile object as a memory
5361 barrier to order a sequence of writes to non-volatile memory. For
5365 int *ptr = @var{something};
5367 *ptr = @var{something};
5371 Unless @var{*ptr} and @var{vobj} can be aliased, it is not guaranteed
5372 that the write to @var{*ptr} will have occurred by the time the update
5373 of @var{vobj} has happened. If you need this guarantee, you must use
5374 a stronger memory barrier such as:
5377 int *ptr = @var{something};
5379 *ptr = @var{something};
5380 asm volatile ("" : : : "memory");
5384 A scalar volatile object is read, when it is accessed in a void context:
5387 volatile int *src = @var{somevalue};
5391 Such expressions are rvalues, and GCC implements this as a
5392 read of the volatile object being pointed to.
5394 Assignments are also expressions and have an rvalue. However when
5395 assigning to a scalar volatile, the volatile object is not reread,
5396 regardless of whether the assignment expression's rvalue is used or
5397 not. If the assignment's rvalue is used, the value is that assigned
5398 to the volatile object. For instance, there is no read of @var{vobj}
5399 in all the following cases:
5404 vobj = @var{something};
5405 obj = vobj = @var{something};
5406 obj ? vobj = @var{onething} : vobj = @var{anotherthing};
5407 obj = (@var{something}, vobj = @var{anotherthing});
5410 If you need to read the volatile object after an assignment has
5411 occurred, you must use a separate expression with an intervening
5414 As bitfields are not individually addressable, volatile bitfields may
5415 be implicitly read when written to, or when adjacent bitfields are
5416 accessed. Bitfield operations may be optimized such that adjacent
5417 bitfields are only partially accessed, if they straddle a storage unit
5418 boundary. For these reasons it is unwise to use volatile bitfields to
5422 @section Assembler Instructions with C Expression Operands
5423 @cindex extended @code{asm}
5424 @cindex @code{asm} expressions
5425 @cindex assembler instructions
5428 In an assembler instruction using @code{asm}, you can specify the
5429 operands of the instruction using C expressions. This means you need not
5430 guess which registers or memory locations will contain the data you want
5433 You must specify an assembler instruction template much like what
5434 appears in a machine description, plus an operand constraint string for
5437 For example, here is how to use the 68881's @code{fsinx} instruction:
5440 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
5444 Here @code{angle} is the C expression for the input operand while
5445 @code{result} is that of the output operand. Each has @samp{"f"} as its
5446 operand constraint, saying that a floating point register is required.
5447 The @samp{=} in @samp{=f} indicates that the operand is an output; all
5448 output operands' constraints must use @samp{=}. The constraints use the
5449 same language used in the machine description (@pxref{Constraints}).
5451 Each operand is described by an operand-constraint string followed by
5452 the C expression in parentheses. A colon separates the assembler
5453 template from the first output operand and another separates the last
5454 output operand from the first input, if any. Commas separate the
5455 operands within each group. The total number of operands is currently
5456 limited to 30; this limitation may be lifted in some future version of
5459 If there are no output operands but there are input operands, you must
5460 place two consecutive colons surrounding the place where the output
5463 As of GCC version 3.1, it is also possible to specify input and output
5464 operands using symbolic names which can be referenced within the
5465 assembler code. These names are specified inside square brackets
5466 preceding the constraint string, and can be referenced inside the
5467 assembler code using @code{%[@var{name}]} instead of a percentage sign
5468 followed by the operand number. Using named operands the above example
5472 asm ("fsinx %[angle],%[output]"
5473 : [output] "=f" (result)
5474 : [angle] "f" (angle));
5478 Note that the symbolic operand names have no relation whatsoever to
5479 other C identifiers. You may use any name you like, even those of
5480 existing C symbols, but you must ensure that no two operands within the same
5481 assembler construct use the same symbolic name.
5483 Output operand expressions must be lvalues; the compiler can check this.
5484 The input operands need not be lvalues. The compiler cannot check
5485 whether the operands have data types that are reasonable for the
5486 instruction being executed. It does not parse the assembler instruction
5487 template and does not know what it means or even whether it is valid
5488 assembler input. The extended @code{asm} feature is most often used for
5489 machine instructions the compiler itself does not know exist. If
5490 the output expression cannot be directly addressed (for example, it is a
5491 bit-field), your constraint must allow a register. In that case, GCC
5492 will use the register as the output of the @code{asm}, and then store
5493 that register into the output.
5495 The ordinary output operands must be write-only; GCC will assume that
5496 the values in these operands before the instruction are dead and need
5497 not be generated. Extended asm supports input-output or read-write
5498 operands. Use the constraint character @samp{+} to indicate such an
5499 operand and list it with the output operands. You should only use
5500 read-write operands when the constraints for the operand (or the
5501 operand in which only some of the bits are to be changed) allow a
5504 You may, as an alternative, logically split its function into two
5505 separate operands, one input operand and one write-only output
5506 operand. The connection between them is expressed by constraints
5507 which say they need to be in the same location when the instruction
5508 executes. You can use the same C expression for both operands, or
5509 different expressions. For example, here we write the (fictitious)
5510 @samp{combine} instruction with @code{bar} as its read-only source
5511 operand and @code{foo} as its read-write destination:
5514 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
5518 The constraint @samp{"0"} for operand 1 says that it must occupy the
5519 same location as operand 0. A number in constraint is allowed only in
5520 an input operand and it must refer to an output operand.
5522 Only a number in the constraint can guarantee that one operand will be in
5523 the same place as another. The mere fact that @code{foo} is the value
5524 of both operands is not enough to guarantee that they will be in the
5525 same place in the generated assembler code. The following would not
5529 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
5532 Various optimizations or reloading could cause operands 0 and 1 to be in
5533 different registers; GCC knows no reason not to do so. For example, the
5534 compiler might find a copy of the value of @code{foo} in one register and
5535 use it for operand 1, but generate the output operand 0 in a different
5536 register (copying it afterward to @code{foo}'s own address). Of course,
5537 since the register for operand 1 is not even mentioned in the assembler
5538 code, the result will not work, but GCC can't tell that.
5540 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
5541 the operand number for a matching constraint. For example:
5544 asm ("cmoveq %1,%2,%[result]"
5545 : [result] "=r"(result)
5546 : "r" (test), "r"(new), "[result]"(old));
5549 Sometimes you need to make an @code{asm} operand be a specific register,
5550 but there's no matching constraint letter for that register @emph{by
5551 itself}. To force the operand into that register, use a local variable
5552 for the operand and specify the register in the variable declaration.
5553 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
5554 register constraint letter that matches the register:
5557 register int *p1 asm ("r0") = @dots{};
5558 register int *p2 asm ("r1") = @dots{};
5559 register int *result asm ("r0");
5560 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
5563 @anchor{Example of asm with clobbered asm reg}
5564 In the above example, beware that a register that is call-clobbered by
5565 the target ABI will be overwritten by any function call in the
5566 assignment, including library calls for arithmetic operators.
5567 Also a register may be clobbered when generating some operations,
5568 like variable shift, memory copy or memory move on x86.
5569 Assuming it is a call-clobbered register, this may happen to @code{r0}
5570 above by the assignment to @code{p2}. If you have to use such a
5571 register, use temporary variables for expressions between the register
5576 register int *p1 asm ("r0") = @dots{};
5577 register int *p2 asm ("r1") = t1;
5578 register int *result asm ("r0");
5579 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
5582 Some instructions clobber specific hard registers. To describe this,
5583 write a third colon after the input operands, followed by the names of
5584 the clobbered hard registers (given as strings). Here is a realistic
5585 example for the VAX:
5588 asm volatile ("movc3 %0,%1,%2"
5589 : /* @r{no outputs} */
5590 : "g" (from), "g" (to), "g" (count)
5591 : "r0", "r1", "r2", "r3", "r4", "r5");
5594 You may not write a clobber description in a way that overlaps with an
5595 input or output operand. For example, you may not have an operand
5596 describing a register class with one member if you mention that register
5597 in the clobber list. Variables declared to live in specific registers
5598 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
5599 have no part mentioned in the clobber description.
5600 There is no way for you to specify that an input
5601 operand is modified without also specifying it as an output
5602 operand. Note that if all the output operands you specify are for this
5603 purpose (and hence unused), you will then also need to specify
5604 @code{volatile} for the @code{asm} construct, as described below, to
5605 prevent GCC from deleting the @code{asm} statement as unused.
5607 If you refer to a particular hardware register from the assembler code,
5608 you will probably have to list the register after the third colon to
5609 tell the compiler the register's value is modified. In some assemblers,
5610 the register names begin with @samp{%}; to produce one @samp{%} in the
5611 assembler code, you must write @samp{%%} in the input.
5613 If your assembler instruction can alter the condition code register, add
5614 @samp{cc} to the list of clobbered registers. GCC on some machines
5615 represents the condition codes as a specific hardware register;
5616 @samp{cc} serves to name this register. On other machines, the
5617 condition code is handled differently, and specifying @samp{cc} has no
5618 effect. But it is valid no matter what the machine.
5620 If your assembler instructions access memory in an unpredictable
5621 fashion, add @samp{memory} to the list of clobbered registers. This
5622 will cause GCC to not keep memory values cached in registers across the
5623 assembler instruction and not optimize stores or loads to that memory.
5624 You will also want to add the @code{volatile} keyword if the memory
5625 affected is not listed in the inputs or outputs of the @code{asm}, as
5626 the @samp{memory} clobber does not count as a side-effect of the
5627 @code{asm}. If you know how large the accessed memory is, you can add
5628 it as input or output but if this is not known, you should add
5629 @samp{memory}. As an example, if you access ten bytes of a string, you
5630 can use a memory input like:
5633 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
5636 Note that in the following example the memory input is necessary,
5637 otherwise GCC might optimize the store to @code{x} away:
5644 asm ("magic stuff accessing an 'int' pointed to by '%1'"
5645 "=&d" (r) : "a" (y), "m" (*y));
5650 You can put multiple assembler instructions together in a single
5651 @code{asm} template, separated by the characters normally used in assembly
5652 code for the system. A combination that works in most places is a newline
5653 to break the line, plus a tab character to move to the instruction field
5654 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
5655 assembler allows semicolons as a line-breaking character. Note that some
5656 assembler dialects use semicolons to start a comment.
5657 The input operands are guaranteed not to use any of the clobbered
5658 registers, and neither will the output operands' addresses, so you can
5659 read and write the clobbered registers as many times as you like. Here
5660 is an example of multiple instructions in a template; it assumes the
5661 subroutine @code{_foo} accepts arguments in registers 9 and 10:
5664 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
5666 : "g" (from), "g" (to)
5670 Unless an output operand has the @samp{&} constraint modifier, GCC
5671 may allocate it in the same register as an unrelated input operand, on
5672 the assumption the inputs are consumed before the outputs are produced.
5673 This assumption may be false if the assembler code actually consists of
5674 more than one instruction. In such a case, use @samp{&} for each output
5675 operand that may not overlap an input. @xref{Modifiers}.
5677 If you want to test the condition code produced by an assembler
5678 instruction, you must include a branch and a label in the @code{asm}
5679 construct, as follows:
5682 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
5688 This assumes your assembler supports local labels, as the GNU assembler
5689 and most Unix assemblers do.
5691 Speaking of labels, jumps from one @code{asm} to another are not
5692 supported. The compiler's optimizers do not know about these jumps, and
5693 therefore they cannot take account of them when deciding how to
5694 optimize. @xref{Extended asm with goto}.
5696 @cindex macros containing @code{asm}
5697 Usually the most convenient way to use these @code{asm} instructions is to
5698 encapsulate them in macros that look like functions. For example,
5702 (@{ double __value, __arg = (x); \
5703 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
5708 Here the variable @code{__arg} is used to make sure that the instruction
5709 operates on a proper @code{double} value, and to accept only those
5710 arguments @code{x} which can convert automatically to a @code{double}.
5712 Another way to make sure the instruction operates on the correct data
5713 type is to use a cast in the @code{asm}. This is different from using a
5714 variable @code{__arg} in that it converts more different types. For
5715 example, if the desired type were @code{int}, casting the argument to
5716 @code{int} would accept a pointer with no complaint, while assigning the
5717 argument to an @code{int} variable named @code{__arg} would warn about
5718 using a pointer unless the caller explicitly casts it.
5720 If an @code{asm} has output operands, GCC assumes for optimization
5721 purposes the instruction has no side effects except to change the output
5722 operands. This does not mean instructions with a side effect cannot be
5723 used, but you must be careful, because the compiler may eliminate them
5724 if the output operands aren't used, or move them out of loops, or
5725 replace two with one if they constitute a common subexpression. Also,
5726 if your instruction does have a side effect on a variable that otherwise
5727 appears not to change, the old value of the variable may be reused later
5728 if it happens to be found in a register.
5730 You can prevent an @code{asm} instruction from being deleted
5731 by writing the keyword @code{volatile} after
5732 the @code{asm}. For example:
5735 #define get_and_set_priority(new) \
5737 asm volatile ("get_and_set_priority %0, %1" \
5738 : "=g" (__old) : "g" (new)); \
5743 The @code{volatile} keyword indicates that the instruction has
5744 important side-effects. GCC will not delete a volatile @code{asm} if
5745 it is reachable. (The instruction can still be deleted if GCC can
5746 prove that control-flow will never reach the location of the
5747 instruction.) Note that even a volatile @code{asm} instruction
5748 can be moved relative to other code, including across jump
5749 instructions. For example, on many targets there is a system
5750 register which can be set to control the rounding mode of
5751 floating point operations. You might try
5752 setting it with a volatile @code{asm}, like this PowerPC example:
5755 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
5760 This will not work reliably, as the compiler may move the addition back
5761 before the volatile @code{asm}. To make it work you need to add an
5762 artificial dependency to the @code{asm} referencing a variable in the code
5763 you don't want moved, for example:
5766 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
5770 Similarly, you can't expect a
5771 sequence of volatile @code{asm} instructions to remain perfectly
5772 consecutive. If you want consecutive output, use a single @code{asm}.
5773 Also, GCC will perform some optimizations across a volatile @code{asm}
5774 instruction; GCC does not ``forget everything'' when it encounters
5775 a volatile @code{asm} instruction the way some other compilers do.
5777 An @code{asm} instruction without any output operands will be treated
5778 identically to a volatile @code{asm} instruction.
5780 It is a natural idea to look for a way to give access to the condition
5781 code left by the assembler instruction. However, when we attempted to
5782 implement this, we found no way to make it work reliably. The problem
5783 is that output operands might need reloading, which would result in
5784 additional following ``store'' instructions. On most machines, these
5785 instructions would alter the condition code before there was time to
5786 test it. This problem doesn't arise for ordinary ``test'' and
5787 ``compare'' instructions because they don't have any output operands.
5789 For reasons similar to those described above, it is not possible to give
5790 an assembler instruction access to the condition code left by previous
5793 @anchor{Extended asm with goto}
5794 As of GCC version 4.5, @code{asm goto} may be used to have the assembly
5795 jump to one or more C labels. In this form, a fifth section after the
5796 clobber list contains a list of all C labels to which the assembly may jump.
5797 Each label operand is implicitly self-named. The @code{asm} is also assumed
5798 to fall through to the next statement.
5800 This form of @code{asm} is restricted to not have outputs. This is due
5801 to a internal restriction in the compiler that control transfer instructions
5802 cannot have outputs. This restriction on @code{asm goto} may be lifted
5803 in some future version of the compiler. In the mean time, @code{asm goto}
5804 may include a memory clobber, and so leave outputs in memory.
5810 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
5811 : : "r"(x), "r"(&y) : "r5", "memory" : error);
5818 In this (inefficient) example, the @code{frob} instruction sets the
5819 carry bit to indicate an error. The @code{jc} instruction detects
5820 this and branches to the @code{error} label. Finally, the output
5821 of the @code{frob} instruction (@code{%r5}) is stored into the memory
5822 for variable @code{y}, which is later read by the @code{return} statement.
5828 asm goto ("mfsr %%r1, 123; jmp %%r1;"
5829 ".pushsection doit_table;"
5830 ".long %l0, %l1, %l2, %l3;"
5832 : : : "r1" : label1, label2, label3, label4);
5833 __builtin_unreachable ();
5848 In this (also inefficient) example, the @code{mfsr} instruction reads
5849 an address from some out-of-band machine register, and the following
5850 @code{jmp} instruction branches to that address. The address read by
5851 the @code{mfsr} instruction is assumed to have been previously set via
5852 some application-specific mechanism to be one of the four values stored
5853 in the @code{doit_table} section. Finally, the @code{asm} is followed
5854 by a call to @code{__builtin_unreachable} to indicate that the @code{asm}
5855 does not in fact fall through.
5858 #define TRACE1(NUM) \
5860 asm goto ("0: nop;" \
5861 ".pushsection trace_table;" \
5864 : : : : trace#NUM); \
5865 if (0) @{ trace#NUM: trace(); @} \
5867 #define TRACE TRACE1(__COUNTER__)
5870 In this example (which in fact inspired the @code{asm goto} feature)
5871 we want on rare occasions to call the @code{trace} function; on other
5872 occasions we'd like to keep the overhead to the absolute minimum.
5873 The normal code path consists of a single @code{nop} instruction.
5874 However, we record the address of this @code{nop} together with the
5875 address of a label that calls the @code{trace} function. This allows
5876 the @code{nop} instruction to be patched at runtime to be an
5877 unconditional branch to the stored label. It is assumed that an
5878 optimizing compiler will move the labeled block out of line, to
5879 optimize the fall through path from the @code{asm}.
5881 If you are writing a header file that should be includable in ISO C
5882 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
5885 @subsection Size of an @code{asm}
5887 Some targets require that GCC track the size of each instruction used in
5888 order to generate correct code. Because the final length of an
5889 @code{asm} is only known by the assembler, GCC must make an estimate as
5890 to how big it will be. The estimate is formed by counting the number of
5891 statements in the pattern of the @code{asm} and multiplying that by the
5892 length of the longest instruction on that processor. Statements in the
5893 @code{asm} are identified by newline characters and whatever statement
5894 separator characters are supported by the assembler; on most processors
5895 this is the `@code{;}' character.
5897 Normally, GCC's estimate is perfectly adequate to ensure that correct
5898 code is generated, but it is possible to confuse the compiler if you use
5899 pseudo instructions or assembler macros that expand into multiple real
5900 instructions or if you use assembler directives that expand to more
5901 space in the object file than would be needed for a single instruction.
5902 If this happens then the assembler will produce a diagnostic saying that
5903 a label is unreachable.
5905 @subsection i386 floating point asm operands
5907 There are several rules on the usage of stack-like regs in
5908 asm_operands insns. These rules apply only to the operands that are
5913 Given a set of input regs that die in an asm_operands, it is
5914 necessary to know which are implicitly popped by the asm, and
5915 which must be explicitly popped by gcc.
5917 An input reg that is implicitly popped by the asm must be
5918 explicitly clobbered, unless it is constrained to match an
5922 For any input reg that is implicitly popped by an asm, it is
5923 necessary to know how to adjust the stack to compensate for the pop.
5924 If any non-popped input is closer to the top of the reg-stack than
5925 the implicitly popped reg, it would not be possible to know what the
5926 stack looked like---it's not clear how the rest of the stack ``slides
5929 All implicitly popped input regs must be closer to the top of
5930 the reg-stack than any input that is not implicitly popped.
5932 It is possible that if an input dies in an insn, reload might
5933 use the input reg for an output reload. Consider this example:
5936 asm ("foo" : "=t" (a) : "f" (b));
5939 This asm says that input B is not popped by the asm, and that
5940 the asm pushes a result onto the reg-stack, i.e., the stack is one
5941 deeper after the asm than it was before. But, it is possible that
5942 reload will think that it can use the same reg for both the input and
5943 the output, if input B dies in this insn.
5945 If any input operand uses the @code{f} constraint, all output reg
5946 constraints must use the @code{&} earlyclobber.
5948 The asm above would be written as
5951 asm ("foo" : "=&t" (a) : "f" (b));
5955 Some operands need to be in particular places on the stack. All
5956 output operands fall in this category---there is no other way to
5957 know which regs the outputs appear in unless the user indicates
5958 this in the constraints.
5960 Output operands must specifically indicate which reg an output
5961 appears in after an asm. @code{=f} is not allowed: the operand
5962 constraints must select a class with a single reg.
5965 Output operands may not be ``inserted'' between existing stack regs.
5966 Since no 387 opcode uses a read/write operand, all output operands
5967 are dead before the asm_operands, and are pushed by the asm_operands.
5968 It makes no sense to push anywhere but the top of the reg-stack.
5970 Output operands must start at the top of the reg-stack: output
5971 operands may not ``skip'' a reg.
5974 Some asm statements may need extra stack space for internal
5975 calculations. This can be guaranteed by clobbering stack registers
5976 unrelated to the inputs and outputs.
5980 Here are a couple of reasonable asms to want to write. This asm
5981 takes one input, which is internally popped, and produces two outputs.
5984 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
5987 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
5988 and replaces them with one output. The user must code the @code{st(1)}
5989 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
5992 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
5998 @section Controlling Names Used in Assembler Code
5999 @cindex assembler names for identifiers
6000 @cindex names used in assembler code
6001 @cindex identifiers, names in assembler code
6003 You can specify the name to be used in the assembler code for a C
6004 function or variable by writing the @code{asm} (or @code{__asm__})
6005 keyword after the declarator as follows:
6008 int foo asm ("myfoo") = 2;
6012 This specifies that the name to be used for the variable @code{foo} in
6013 the assembler code should be @samp{myfoo} rather than the usual
6016 On systems where an underscore is normally prepended to the name of a C
6017 function or variable, this feature allows you to define names for the
6018 linker that do not start with an underscore.
6020 It does not make sense to use this feature with a non-static local
6021 variable since such variables do not have assembler names. If you are
6022 trying to put the variable in a particular register, see @ref{Explicit
6023 Reg Vars}. GCC presently accepts such code with a warning, but will
6024 probably be changed to issue an error, rather than a warning, in the
6027 You cannot use @code{asm} in this way in a function @emph{definition}; but
6028 you can get the same effect by writing a declaration for the function
6029 before its definition and putting @code{asm} there, like this:
6032 extern func () asm ("FUNC");
6039 It is up to you to make sure that the assembler names you choose do not
6040 conflict with any other assembler symbols. Also, you must not use a
6041 register name; that would produce completely invalid assembler code. GCC
6042 does not as yet have the ability to store static variables in registers.
6043 Perhaps that will be added.
6045 @node Explicit Reg Vars
6046 @section Variables in Specified Registers
6047 @cindex explicit register variables
6048 @cindex variables in specified registers
6049 @cindex specified registers
6050 @cindex registers, global allocation
6052 GNU C allows you to put a few global variables into specified hardware
6053 registers. You can also specify the register in which an ordinary
6054 register variable should be allocated.
6058 Global register variables reserve registers throughout the program.
6059 This may be useful in programs such as programming language
6060 interpreters which have a couple of global variables that are accessed
6064 Local register variables in specific registers do not reserve the
6065 registers, except at the point where they are used as input or output
6066 operands in an @code{asm} statement and the @code{asm} statement itself is
6067 not deleted. The compiler's data flow analysis is capable of determining
6068 where the specified registers contain live values, and where they are
6069 available for other uses. Stores into local register variables may be deleted
6070 when they appear to be dead according to dataflow analysis. References
6071 to local register variables may be deleted or moved or simplified.
6073 These local variables are sometimes convenient for use with the extended
6074 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
6075 output of the assembler instruction directly into a particular register.
6076 (This will work provided the register you specify fits the constraints
6077 specified for that operand in the @code{asm}.)
6085 @node Global Reg Vars
6086 @subsection Defining Global Register Variables
6087 @cindex global register variables
6088 @cindex registers, global variables in
6090 You can define a global register variable in GNU C like this:
6093 register int *foo asm ("a5");
6097 Here @code{a5} is the name of the register which should be used. Choose a
6098 register which is normally saved and restored by function calls on your
6099 machine, so that library routines will not clobber it.
6101 Naturally the register name is cpu-dependent, so you would need to
6102 conditionalize your program according to cpu type. The register
6103 @code{a5} would be a good choice on a 68000 for a variable of pointer
6104 type. On machines with register windows, be sure to choose a ``global''
6105 register that is not affected magically by the function call mechanism.
6107 In addition, operating systems on one type of cpu may differ in how they
6108 name the registers; then you would need additional conditionals. For
6109 example, some 68000 operating systems call this register @code{%a5}.
6111 Eventually there may be a way of asking the compiler to choose a register
6112 automatically, but first we need to figure out how it should choose and
6113 how to enable you to guide the choice. No solution is evident.
6115 Defining a global register variable in a certain register reserves that
6116 register entirely for this use, at least within the current compilation.
6117 The register will not be allocated for any other purpose in the functions
6118 in the current compilation. The register will not be saved and restored by
6119 these functions. Stores into this register are never deleted even if they
6120 would appear to be dead, but references may be deleted or moved or
6123 It is not safe to access the global register variables from signal
6124 handlers, or from more than one thread of control, because the system
6125 library routines may temporarily use the register for other things (unless
6126 you recompile them specially for the task at hand).
6128 @cindex @code{qsort}, and global register variables
6129 It is not safe for one function that uses a global register variable to
6130 call another such function @code{foo} by way of a third function
6131 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
6132 different source file in which the variable wasn't declared). This is
6133 because @code{lose} might save the register and put some other value there.
6134 For example, you can't expect a global register variable to be available in
6135 the comparison-function that you pass to @code{qsort}, since @code{qsort}
6136 might have put something else in that register. (If you are prepared to
6137 recompile @code{qsort} with the same global register variable, you can
6138 solve this problem.)
6140 If you want to recompile @code{qsort} or other source files which do not
6141 actually use your global register variable, so that they will not use that
6142 register for any other purpose, then it suffices to specify the compiler
6143 option @option{-ffixed-@var{reg}}. You need not actually add a global
6144 register declaration to their source code.
6146 A function which can alter the value of a global register variable cannot
6147 safely be called from a function compiled without this variable, because it
6148 could clobber the value the caller expects to find there on return.
6149 Therefore, the function which is the entry point into the part of the
6150 program that uses the global register variable must explicitly save and
6151 restore the value which belongs to its caller.
6153 @cindex register variable after @code{longjmp}
6154 @cindex global register after @code{longjmp}
6155 @cindex value after @code{longjmp}
6158 On most machines, @code{longjmp} will restore to each global register
6159 variable the value it had at the time of the @code{setjmp}. On some
6160 machines, however, @code{longjmp} will not change the value of global
6161 register variables. To be portable, the function that called @code{setjmp}
6162 should make other arrangements to save the values of the global register
6163 variables, and to restore them in a @code{longjmp}. This way, the same
6164 thing will happen regardless of what @code{longjmp} does.
6166 All global register variable declarations must precede all function
6167 definitions. If such a declaration could appear after function
6168 definitions, the declaration would be too late to prevent the register from
6169 being used for other purposes in the preceding functions.
6171 Global register variables may not have initial values, because an
6172 executable file has no means to supply initial contents for a register.
6174 On the SPARC, there are reports that g3 @dots{} g7 are suitable
6175 registers, but certain library functions, such as @code{getwd}, as well
6176 as the subroutines for division and remainder, modify g3 and g4. g1 and
6177 g2 are local temporaries.
6179 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
6180 Of course, it will not do to use more than a few of those.
6182 @node Local Reg Vars
6183 @subsection Specifying Registers for Local Variables
6184 @cindex local variables, specifying registers
6185 @cindex specifying registers for local variables
6186 @cindex registers for local variables
6188 You can define a local register variable with a specified register
6192 register int *foo asm ("a5");
6196 Here @code{a5} is the name of the register which should be used. Note
6197 that this is the same syntax used for defining global register
6198 variables, but for a local variable it would appear within a function.
6200 Naturally the register name is cpu-dependent, but this is not a
6201 problem, since specific registers are most often useful with explicit
6202 assembler instructions (@pxref{Extended Asm}). Both of these things
6203 generally require that you conditionalize your program according to
6206 In addition, operating systems on one type of cpu may differ in how they
6207 name the registers; then you would need additional conditionals. For
6208 example, some 68000 operating systems call this register @code{%a5}.
6210 Defining such a register variable does not reserve the register; it
6211 remains available for other uses in places where flow control determines
6212 the variable's value is not live.
6214 This option does not guarantee that GCC will generate code that has
6215 this variable in the register you specify at all times. You may not
6216 code an explicit reference to this register in the @emph{assembler
6217 instruction template} part of an @code{asm} statement and assume it will
6218 always refer to this variable. However, using the variable as an
6219 @code{asm} @emph{operand} guarantees that the specified register is used
6222 Stores into local register variables may be deleted when they appear to be dead
6223 according to dataflow analysis. References to local register variables may
6224 be deleted or moved or simplified.
6226 As for global register variables, it's recommended that you choose a
6227 register which is normally saved and restored by function calls on
6228 your machine, so that library routines will not clobber it. A common
6229 pitfall is to initialize multiple call-clobbered registers with
6230 arbitrary expressions, where a function call or library call for an
6231 arithmetic operator will overwrite a register value from a previous
6232 assignment, for example @code{r0} below:
6234 register int *p1 asm ("r0") = @dots{};
6235 register int *p2 asm ("r1") = @dots{};
6237 In those cases, a solution is to use a temporary variable for
6238 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
6240 @node Alternate Keywords
6241 @section Alternate Keywords
6242 @cindex alternate keywords
6243 @cindex keywords, alternate
6245 @option{-ansi} and the various @option{-std} options disable certain
6246 keywords. This causes trouble when you want to use GNU C extensions, or
6247 a general-purpose header file that should be usable by all programs,
6248 including ISO C programs. The keywords @code{asm}, @code{typeof} and
6249 @code{inline} are not available in programs compiled with
6250 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
6251 program compiled with @option{-std=c99} or @option{-std=c1x}). The
6253 @code{restrict} is only available when @option{-std=gnu99} (which will
6254 eventually be the default) or @option{-std=c99} (or the equivalent
6255 @option{-std=iso9899:1999}), or an option for a later standard
6258 The way to solve these problems is to put @samp{__} at the beginning and
6259 end of each problematical keyword. For example, use @code{__asm__}
6260 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
6262 Other C compilers won't accept these alternative keywords; if you want to
6263 compile with another compiler, you can define the alternate keywords as
6264 macros to replace them with the customary keywords. It looks like this:
6272 @findex __extension__
6274 @option{-pedantic} and other options cause warnings for many GNU C extensions.
6276 prevent such warnings within one expression by writing
6277 @code{__extension__} before the expression. @code{__extension__} has no
6278 effect aside from this.
6280 @node Incomplete Enums
6281 @section Incomplete @code{enum} Types
6283 You can define an @code{enum} tag without specifying its possible values.
6284 This results in an incomplete type, much like what you get if you write
6285 @code{struct foo} without describing the elements. A later declaration
6286 which does specify the possible values completes the type.
6288 You can't allocate variables or storage using the type while it is
6289 incomplete. However, you can work with pointers to that type.
6291 This extension may not be very useful, but it makes the handling of
6292 @code{enum} more consistent with the way @code{struct} and @code{union}
6295 This extension is not supported by GNU C++.
6297 @node Function Names
6298 @section Function Names as Strings
6299 @cindex @code{__func__} identifier
6300 @cindex @code{__FUNCTION__} identifier
6301 @cindex @code{__PRETTY_FUNCTION__} identifier
6303 GCC provides three magic variables which hold the name of the current
6304 function, as a string. The first of these is @code{__func__}, which
6305 is part of the C99 standard:
6307 The identifier @code{__func__} is implicitly declared by the translator
6308 as if, immediately following the opening brace of each function
6309 definition, the declaration
6312 static const char __func__[] = "function-name";
6316 appeared, where function-name is the name of the lexically-enclosing
6317 function. This name is the unadorned name of the function.
6319 @code{__FUNCTION__} is another name for @code{__func__}. Older
6320 versions of GCC recognize only this name. However, it is not
6321 standardized. For maximum portability, we recommend you use
6322 @code{__func__}, but provide a fallback definition with the
6326 #if __STDC_VERSION__ < 199901L
6328 # define __func__ __FUNCTION__
6330 # define __func__ "<unknown>"
6335 In C, @code{__PRETTY_FUNCTION__} is yet another name for
6336 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
6337 the type signature of the function as well as its bare name. For
6338 example, this program:
6342 extern int printf (char *, ...);
6349 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
6350 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
6368 __PRETTY_FUNCTION__ = void a::sub(int)
6371 These identifiers are not preprocessor macros. In GCC 3.3 and
6372 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
6373 were treated as string literals; they could be used to initialize
6374 @code{char} arrays, and they could be concatenated with other string
6375 literals. GCC 3.4 and later treat them as variables, like
6376 @code{__func__}. In C++, @code{__FUNCTION__} and
6377 @code{__PRETTY_FUNCTION__} have always been variables.
6379 @node Return Address
6380 @section Getting the Return or Frame Address of a Function
6382 These functions may be used to get information about the callers of a
6385 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
6386 This function returns the return address of the current function, or of
6387 one of its callers. The @var{level} argument is number of frames to
6388 scan up the call stack. A value of @code{0} yields the return address
6389 of the current function, a value of @code{1} yields the return address
6390 of the caller of the current function, and so forth. When inlining
6391 the expected behavior is that the function will return the address of
6392 the function that will be returned to. To work around this behavior use
6393 the @code{noinline} function attribute.
6395 The @var{level} argument must be a constant integer.
6397 On some machines it may be impossible to determine the return address of
6398 any function other than the current one; in such cases, or when the top
6399 of the stack has been reached, this function will return @code{0} or a
6400 random value. In addition, @code{__builtin_frame_address} may be used
6401 to determine if the top of the stack has been reached.
6403 Additional post-processing of the returned value may be needed, see
6404 @code{__builtin_extract_return_address}.
6406 This function should only be used with a nonzero argument for debugging
6410 @deftypefn {Built-in Function} {void *} __builtin_extract_return_address (void *@var{addr})
6411 The address as returned by @code{__builtin_return_address} may have to be fed
6412 through this function to get the actual encoded address. For example, on the
6413 31-bit S/390 platform the highest bit has to be masked out, or on SPARC
6414 platforms an offset has to be added for the true next instruction to be
6417 If no fixup is needed, this function simply passes through @var{addr}.
6420 @deftypefn {Built-in Function} {void *} __builtin_frob_return_address (void *@var{addr})
6421 This function does the reverse of @code{__builtin_extract_return_address}.
6424 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
6425 This function is similar to @code{__builtin_return_address}, but it
6426 returns the address of the function frame rather than the return address
6427 of the function. Calling @code{__builtin_frame_address} with a value of
6428 @code{0} yields the frame address of the current function, a value of
6429 @code{1} yields the frame address of the caller of the current function,
6432 The frame is the area on the stack which holds local variables and saved
6433 registers. The frame address is normally the address of the first word
6434 pushed on to the stack by the function. However, the exact definition
6435 depends upon the processor and the calling convention. If the processor
6436 has a dedicated frame pointer register, and the function has a frame,
6437 then @code{__builtin_frame_address} will return the value of the frame
6440 On some machines it may be impossible to determine the frame address of
6441 any function other than the current one; in such cases, or when the top
6442 of the stack has been reached, this function will return @code{0} if
6443 the first frame pointer is properly initialized by the startup code.
6445 This function should only be used with a nonzero argument for debugging
6449 @node Vector Extensions
6450 @section Using vector instructions through built-in functions
6452 On some targets, the instruction set contains SIMD vector instructions that
6453 operate on multiple values contained in one large register at the same time.
6454 For example, on the i386 the MMX, 3DNow!@: and SSE extensions can be used
6457 The first step in using these extensions is to provide the necessary data
6458 types. This should be done using an appropriate @code{typedef}:
6461 typedef int v4si __attribute__ ((vector_size (16)));
6464 The @code{int} type specifies the base type, while the attribute specifies
6465 the vector size for the variable, measured in bytes. For example, the
6466 declaration above causes the compiler to set the mode for the @code{v4si}
6467 type to be 16 bytes wide and divided into @code{int} sized units. For
6468 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
6469 corresponding mode of @code{foo} will be @acronym{V4SI}.
6471 The @code{vector_size} attribute is only applicable to integral and
6472 float scalars, although arrays, pointers, and function return values
6473 are allowed in conjunction with this construct.
6475 All the basic integer types can be used as base types, both as signed
6476 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
6477 @code{long long}. In addition, @code{float} and @code{double} can be
6478 used to build floating-point vector types.
6480 Specifying a combination that is not valid for the current architecture
6481 will cause GCC to synthesize the instructions using a narrower mode.
6482 For example, if you specify a variable of type @code{V4SI} and your
6483 architecture does not allow for this specific SIMD type, GCC will
6484 produce code that uses 4 @code{SIs}.
6486 The types defined in this manner can be used with a subset of normal C
6487 operations. Currently, GCC will allow using the following operators
6488 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~, %}@.
6490 The operations behave like C++ @code{valarrays}. Addition is defined as
6491 the addition of the corresponding elements of the operands. For
6492 example, in the code below, each of the 4 elements in @var{a} will be
6493 added to the corresponding 4 elements in @var{b} and the resulting
6494 vector will be stored in @var{c}.
6497 typedef int v4si __attribute__ ((vector_size (16)));
6504 Subtraction, multiplication, division, and the logical operations
6505 operate in a similar manner. Likewise, the result of using the unary
6506 minus or complement operators on a vector type is a vector whose
6507 elements are the negative or complemented values of the corresponding
6508 elements in the operand.
6510 In C it is possible to use shifting operators @code{<<}, @code{>>} on
6511 integer-type vectors. The operation is defined as following: @code{@{a0,
6512 a1, @dots{}, an@} >> @{b0, b1, @dots{}, bn@} == @{a0 >> b0, a1 >> b1,
6513 @dots{}, an >> bn@}}@. Vector operands must have the same number of
6514 elements. Additionally second operands can be a scalar integer in which
6515 case the scalar is converted to the type used by the vector operand (with
6516 possible truncation) and each element of this new vector is the scalar's
6518 Consider the following code.
6521 typedef int v4si __attribute__ ((vector_size (16)));
6525 b = a >> 1; /* b = a >> @{1,1,1,1@}; */
6528 In C vectors can be subscripted as if the vector were an array with
6529 the same number of elements and base type. Out of bound accesses
6530 invoke undefined behavior at runtime. Warnings for out of bound
6531 accesses for vector subscription can be enabled with
6532 @option{-Warray-bounds}.
6534 You can declare variables and use them in function calls and returns, as
6535 well as in assignments and some casts. You can specify a vector type as
6536 a return type for a function. Vector types can also be used as function
6537 arguments. It is possible to cast from one vector type to another,
6538 provided they are of the same size (in fact, you can also cast vectors
6539 to and from other datatypes of the same size).
6541 You cannot operate between vectors of different lengths or different
6542 signedness without a cast.
6544 A port that supports hardware vector operations, usually provides a set
6545 of built-in functions that can be used to operate on vectors. For
6546 example, a function to add two vectors and multiply the result by a
6547 third could look like this:
6550 v4si f (v4si a, v4si b, v4si c)
6552 v4si tmp = __builtin_addv4si (a, b);
6553 return __builtin_mulv4si (tmp, c);
6560 @findex __builtin_offsetof
6562 GCC implements for both C and C++ a syntactic extension to implement
6563 the @code{offsetof} macro.
6567 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
6569 offsetof_member_designator:
6571 | offsetof_member_designator "." @code{identifier}
6572 | offsetof_member_designator "[" @code{expr} "]"
6575 This extension is sufficient such that
6578 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
6581 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
6582 may be dependent. In either case, @var{member} may consist of a single
6583 identifier, or a sequence of member accesses and array references.
6585 @node Atomic Builtins
6586 @section Built-in functions for atomic memory access
6588 The following builtins are intended to be compatible with those described
6589 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
6590 section 7.4. As such, they depart from the normal GCC practice of using
6591 the ``__builtin_'' prefix, and further that they are overloaded such that
6592 they work on multiple types.
6594 The definition given in the Intel documentation allows only for the use of
6595 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
6596 counterparts. GCC will allow any integral scalar or pointer type that is
6597 1, 2, 4 or 8 bytes in length.
6599 Not all operations are supported by all target processors. If a particular
6600 operation cannot be implemented on the target processor, a warning will be
6601 generated and a call an external function will be generated. The external
6602 function will carry the same name as the builtin, with an additional suffix
6603 @samp{_@var{n}} where @var{n} is the size of the data type.
6605 @c ??? Should we have a mechanism to suppress this warning? This is almost
6606 @c useful for implementing the operation under the control of an external
6609 In most cases, these builtins are considered a @dfn{full barrier}. That is,
6610 no memory operand will be moved across the operation, either forward or
6611 backward. Further, instructions will be issued as necessary to prevent the
6612 processor from speculating loads across the operation and from queuing stores
6613 after the operation.
6615 All of the routines are described in the Intel documentation to take
6616 ``an optional list of variables protected by the memory barrier''. It's
6617 not clear what is meant by that; it could mean that @emph{only} the
6618 following variables are protected, or it could mean that these variables
6619 should in addition be protected. At present GCC ignores this list and
6620 protects all variables which are globally accessible. If in the future
6621 we make some use of this list, an empty list will continue to mean all
6622 globally accessible variables.
6625 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
6626 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
6627 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
6628 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
6629 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
6630 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
6631 @findex __sync_fetch_and_add
6632 @findex __sync_fetch_and_sub
6633 @findex __sync_fetch_and_or
6634 @findex __sync_fetch_and_and
6635 @findex __sync_fetch_and_xor
6636 @findex __sync_fetch_and_nand
6637 These builtins perform the operation suggested by the name, and
6638 returns the value that had previously been in memory. That is,
6641 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
6642 @{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; @} // nand
6645 @emph{Note:} GCC 4.4 and later implement @code{__sync_fetch_and_nand}
6646 builtin as @code{*ptr = ~(tmp & value)} instead of @code{*ptr = ~tmp & value}.
6648 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
6649 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
6650 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
6651 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
6652 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
6653 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
6654 @findex __sync_add_and_fetch
6655 @findex __sync_sub_and_fetch
6656 @findex __sync_or_and_fetch
6657 @findex __sync_and_and_fetch
6658 @findex __sync_xor_and_fetch
6659 @findex __sync_nand_and_fetch
6660 These builtins perform the operation suggested by the name, and
6661 return the new value. That is,
6664 @{ *ptr @var{op}= value; return *ptr; @}
6665 @{ *ptr = ~(*ptr & value); return *ptr; @} // nand
6668 @emph{Note:} GCC 4.4 and later implement @code{__sync_nand_and_fetch}
6669 builtin as @code{*ptr = ~(*ptr & value)} instead of
6670 @code{*ptr = ~*ptr & value}.
6672 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
6673 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
6674 @findex __sync_bool_compare_and_swap
6675 @findex __sync_val_compare_and_swap
6676 These builtins perform an atomic compare and swap. That is, if the current
6677 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
6680 The ``bool'' version returns true if the comparison is successful and
6681 @var{newval} was written. The ``val'' version returns the contents
6682 of @code{*@var{ptr}} before the operation.
6684 @item __sync_synchronize (...)
6685 @findex __sync_synchronize
6686 This builtin issues a full memory barrier.
6688 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
6689 @findex __sync_lock_test_and_set
6690 This builtin, as described by Intel, is not a traditional test-and-set
6691 operation, but rather an atomic exchange operation. It writes @var{value}
6692 into @code{*@var{ptr}}, and returns the previous contents of
6695 Many targets have only minimal support for such locks, and do not support
6696 a full exchange operation. In this case, a target may support reduced
6697 functionality here by which the @emph{only} valid value to store is the
6698 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
6699 is implementation defined.
6701 This builtin is not a full barrier, but rather an @dfn{acquire barrier}.
6702 This means that references after the builtin cannot move to (or be
6703 speculated to) before the builtin, but previous memory stores may not
6704 be globally visible yet, and previous memory loads may not yet be
6707 @item void __sync_lock_release (@var{type} *ptr, ...)
6708 @findex __sync_lock_release
6709 This builtin releases the lock acquired by @code{__sync_lock_test_and_set}.
6710 Normally this means writing the constant 0 to @code{*@var{ptr}}.
6712 This builtin is not a full barrier, but rather a @dfn{release barrier}.
6713 This means that all previous memory stores are globally visible, and all
6714 previous memory loads have been satisfied, but following memory reads
6715 are not prevented from being speculated to before the barrier.
6718 @node Object Size Checking
6719 @section Object Size Checking Builtins
6720 @findex __builtin_object_size
6721 @findex __builtin___memcpy_chk
6722 @findex __builtin___mempcpy_chk
6723 @findex __builtin___memmove_chk
6724 @findex __builtin___memset_chk
6725 @findex __builtin___strcpy_chk
6726 @findex __builtin___stpcpy_chk
6727 @findex __builtin___strncpy_chk
6728 @findex __builtin___strcat_chk
6729 @findex __builtin___strncat_chk
6730 @findex __builtin___sprintf_chk
6731 @findex __builtin___snprintf_chk
6732 @findex __builtin___vsprintf_chk
6733 @findex __builtin___vsnprintf_chk
6734 @findex __builtin___printf_chk
6735 @findex __builtin___vprintf_chk
6736 @findex __builtin___fprintf_chk
6737 @findex __builtin___vfprintf_chk
6739 GCC implements a limited buffer overflow protection mechanism
6740 that can prevent some buffer overflow attacks.
6742 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
6743 is a built-in construct that returns a constant number of bytes from
6744 @var{ptr} to the end of the object @var{ptr} pointer points to
6745 (if known at compile time). @code{__builtin_object_size} never evaluates
6746 its arguments for side-effects. If there are any side-effects in them, it
6747 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
6748 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
6749 point to and all of them are known at compile time, the returned number
6750 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
6751 0 and minimum if nonzero. If it is not possible to determine which objects
6752 @var{ptr} points to at compile time, @code{__builtin_object_size} should
6753 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
6754 for @var{type} 2 or 3.
6756 @var{type} is an integer constant from 0 to 3. If the least significant
6757 bit is clear, objects are whole variables, if it is set, a closest
6758 surrounding subobject is considered the object a pointer points to.
6759 The second bit determines if maximum or minimum of remaining bytes
6763 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
6764 char *p = &var.buf1[1], *q = &var.b;
6766 /* Here the object p points to is var. */
6767 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
6768 /* The subobject p points to is var.buf1. */
6769 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
6770 /* The object q points to is var. */
6771 assert (__builtin_object_size (q, 0)
6772 == (char *) (&var + 1) - (char *) &var.b);
6773 /* The subobject q points to is var.b. */
6774 assert (__builtin_object_size (q, 1) == sizeof (var.b));
6778 There are built-in functions added for many common string operation
6779 functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk}
6780 built-in is provided. This built-in has an additional last argument,
6781 which is the number of bytes remaining in object the @var{dest}
6782 argument points to or @code{(size_t) -1} if the size is not known.
6784 The built-in functions are optimized into the normal string functions
6785 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
6786 it is known at compile time that the destination object will not
6787 be overflown. If the compiler can determine at compile time the
6788 object will be always overflown, it issues a warning.
6790 The intended use can be e.g.
6794 #define bos0(dest) __builtin_object_size (dest, 0)
6795 #define memcpy(dest, src, n) \
6796 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
6800 /* It is unknown what object p points to, so this is optimized
6801 into plain memcpy - no checking is possible. */
6802 memcpy (p, "abcde", n);
6803 /* Destination is known and length too. It is known at compile
6804 time there will be no overflow. */
6805 memcpy (&buf[5], "abcde", 5);
6806 /* Destination is known, but the length is not known at compile time.
6807 This will result in __memcpy_chk call that can check for overflow
6809 memcpy (&buf[5], "abcde", n);
6810 /* Destination is known and it is known at compile time there will
6811 be overflow. There will be a warning and __memcpy_chk call that
6812 will abort the program at runtime. */
6813 memcpy (&buf[6], "abcde", 5);
6816 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
6817 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
6818 @code{strcat} and @code{strncat}.
6820 There are also checking built-in functions for formatted output functions.
6822 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
6823 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
6824 const char *fmt, ...);
6825 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
6827 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
6828 const char *fmt, va_list ap);
6831 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
6832 etc.@: functions and can contain implementation specific flags on what
6833 additional security measures the checking function might take, such as
6834 handling @code{%n} differently.
6836 The @var{os} argument is the object size @var{s} points to, like in the
6837 other built-in functions. There is a small difference in the behavior
6838 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
6839 optimized into the non-checking functions only if @var{flag} is 0, otherwise
6840 the checking function is called with @var{os} argument set to
6843 In addition to this, there are checking built-in functions
6844 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
6845 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
6846 These have just one additional argument, @var{flag}, right before
6847 format string @var{fmt}. If the compiler is able to optimize them to
6848 @code{fputc} etc.@: functions, it will, otherwise the checking function
6849 should be called and the @var{flag} argument passed to it.
6851 @node Other Builtins
6852 @section Other built-in functions provided by GCC
6853 @cindex built-in functions
6854 @findex __builtin_fpclassify
6855 @findex __builtin_isfinite
6856 @findex __builtin_isnormal
6857 @findex __builtin_isgreater
6858 @findex __builtin_isgreaterequal
6859 @findex __builtin_isinf_sign
6860 @findex __builtin_isless
6861 @findex __builtin_islessequal
6862 @findex __builtin_islessgreater
6863 @findex __builtin_isunordered
6864 @findex __builtin_powi
6865 @findex __builtin_powif
6866 @findex __builtin_powil
7024 @findex fprintf_unlocked
7026 @findex fputs_unlocked
7143 @findex printf_unlocked
7175 @findex significandf
7176 @findex significandl
7247 GCC provides a large number of built-in functions other than the ones
7248 mentioned above. Some of these are for internal use in the processing
7249 of exceptions or variable-length argument lists and will not be
7250 documented here because they may change from time to time; we do not
7251 recommend general use of these functions.
7253 The remaining functions are provided for optimization purposes.
7255 @opindex fno-builtin
7256 GCC includes built-in versions of many of the functions in the standard
7257 C library. The versions prefixed with @code{__builtin_} will always be
7258 treated as having the same meaning as the C library function even if you
7259 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
7260 Many of these functions are only optimized in certain cases; if they are
7261 not optimized in a particular case, a call to the library function will
7266 Outside strict ISO C mode (@option{-ansi}, @option{-std=c90},
7267 @option{-std=c99} or @option{-std=c1x}), the functions
7268 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
7269 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
7270 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
7271 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
7272 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
7273 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
7274 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
7275 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
7276 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
7277 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
7278 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
7279 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
7280 @code{signbitd64}, @code{signbitd128}, @code{significandf},
7281 @code{significandl}, @code{significand}, @code{sincosf},
7282 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
7283 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
7284 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
7285 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
7287 may be handled as built-in functions.
7288 All these functions have corresponding versions
7289 prefixed with @code{__builtin_}, which may be used even in strict C90
7292 The ISO C99 functions
7293 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
7294 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
7295 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
7296 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
7297 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
7298 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
7299 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
7300 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
7301 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
7302 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
7303 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
7304 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
7305 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
7306 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
7307 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
7308 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
7309 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
7310 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
7311 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
7312 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
7313 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
7314 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
7315 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
7316 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
7317 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
7318 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
7319 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
7320 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
7321 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
7322 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
7323 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
7324 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
7325 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
7326 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
7327 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
7328 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
7329 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
7330 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
7331 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
7332 are handled as built-in functions
7333 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
7335 There are also built-in versions of the ISO C99 functions
7336 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
7337 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
7338 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
7339 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
7340 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
7341 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
7342 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
7343 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
7344 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
7345 that are recognized in any mode since ISO C90 reserves these names for
7346 the purpose to which ISO C99 puts them. All these functions have
7347 corresponding versions prefixed with @code{__builtin_}.
7349 The ISO C94 functions
7350 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
7351 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
7352 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
7354 are handled as built-in functions
7355 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
7357 The ISO C90 functions
7358 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
7359 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
7360 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
7361 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
7362 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
7363 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
7364 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
7365 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
7366 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
7367 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
7368 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
7369 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
7370 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
7371 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
7372 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
7373 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
7374 are all recognized as built-in functions unless
7375 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
7376 is specified for an individual function). All of these functions have
7377 corresponding versions prefixed with @code{__builtin_}.
7379 GCC provides built-in versions of the ISO C99 floating point comparison
7380 macros that avoid raising exceptions for unordered operands. They have
7381 the same names as the standard macros ( @code{isgreater},
7382 @code{isgreaterequal}, @code{isless}, @code{islessequal},
7383 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
7384 prefixed. We intend for a library implementor to be able to simply
7385 @code{#define} each standard macro to its built-in equivalent.
7386 In the same fashion, GCC provides @code{fpclassify}, @code{isfinite},
7387 @code{isinf_sign} and @code{isnormal} built-ins used with
7388 @code{__builtin_} prefixed. The @code{isinf} and @code{isnan}
7389 builtins appear both with and without the @code{__builtin_} prefix.
7391 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
7393 You can use the built-in function @code{__builtin_types_compatible_p} to
7394 determine whether two types are the same.
7396 This built-in function returns 1 if the unqualified versions of the
7397 types @var{type1} and @var{type2} (which are types, not expressions) are
7398 compatible, 0 otherwise. The result of this built-in function can be
7399 used in integer constant expressions.
7401 This built-in function ignores top level qualifiers (e.g., @code{const},
7402 @code{volatile}). For example, @code{int} is equivalent to @code{const
7405 The type @code{int[]} and @code{int[5]} are compatible. On the other
7406 hand, @code{int} and @code{char *} are not compatible, even if the size
7407 of their types, on the particular architecture are the same. Also, the
7408 amount of pointer indirection is taken into account when determining
7409 similarity. Consequently, @code{short *} is not similar to
7410 @code{short **}. Furthermore, two types that are typedefed are
7411 considered compatible if their underlying types are compatible.
7413 An @code{enum} type is not considered to be compatible with another
7414 @code{enum} type even if both are compatible with the same integer
7415 type; this is what the C standard specifies.
7416 For example, @code{enum @{foo, bar@}} is not similar to
7417 @code{enum @{hot, dog@}}.
7419 You would typically use this function in code whose execution varies
7420 depending on the arguments' types. For example:
7425 typeof (x) tmp = (x); \
7426 if (__builtin_types_compatible_p (typeof (x), long double)) \
7427 tmp = foo_long_double (tmp); \
7428 else if (__builtin_types_compatible_p (typeof (x), double)) \
7429 tmp = foo_double (tmp); \
7430 else if (__builtin_types_compatible_p (typeof (x), float)) \
7431 tmp = foo_float (tmp); \
7438 @emph{Note:} This construct is only available for C@.
7442 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
7444 You can use the built-in function @code{__builtin_choose_expr} to
7445 evaluate code depending on the value of a constant expression. This
7446 built-in function returns @var{exp1} if @var{const_exp}, which is an
7447 integer constant expression, is nonzero. Otherwise it returns @var{exp2}.
7449 This built-in function is analogous to the @samp{? :} operator in C,
7450 except that the expression returned has its type unaltered by promotion
7451 rules. Also, the built-in function does not evaluate the expression
7452 that was not chosen. For example, if @var{const_exp} evaluates to true,
7453 @var{exp2} is not evaluated even if it has side-effects.
7455 This built-in function can return an lvalue if the chosen argument is an
7458 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
7459 type. Similarly, if @var{exp2} is returned, its return type is the same
7466 __builtin_choose_expr ( \
7467 __builtin_types_compatible_p (typeof (x), double), \
7469 __builtin_choose_expr ( \
7470 __builtin_types_compatible_p (typeof (x), float), \
7472 /* @r{The void expression results in a compile-time error} \
7473 @r{when assigning the result to something.} */ \
7477 @emph{Note:} This construct is only available for C@. Furthermore, the
7478 unused expression (@var{exp1} or @var{exp2} depending on the value of
7479 @var{const_exp}) may still generate syntax errors. This may change in
7484 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
7485 You can use the built-in function @code{__builtin_constant_p} to
7486 determine if a value is known to be constant at compile-time and hence
7487 that GCC can perform constant-folding on expressions involving that
7488 value. The argument of the function is the value to test. The function
7489 returns the integer 1 if the argument is known to be a compile-time
7490 constant and 0 if it is not known to be a compile-time constant. A
7491 return of 0 does not indicate that the value is @emph{not} a constant,
7492 but merely that GCC cannot prove it is a constant with the specified
7493 value of the @option{-O} option.
7495 You would typically use this function in an embedded application where
7496 memory was a critical resource. If you have some complex calculation,
7497 you may want it to be folded if it involves constants, but need to call
7498 a function if it does not. For example:
7501 #define Scale_Value(X) \
7502 (__builtin_constant_p (X) \
7503 ? ((X) * SCALE + OFFSET) : Scale (X))
7506 You may use this built-in function in either a macro or an inline
7507 function. However, if you use it in an inlined function and pass an
7508 argument of the function as the argument to the built-in, GCC will
7509 never return 1 when you call the inline function with a string constant
7510 or compound literal (@pxref{Compound Literals}) and will not return 1
7511 when you pass a constant numeric value to the inline function unless you
7512 specify the @option{-O} option.
7514 You may also use @code{__builtin_constant_p} in initializers for static
7515 data. For instance, you can write
7518 static const int table[] = @{
7519 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
7525 This is an acceptable initializer even if @var{EXPRESSION} is not a
7526 constant expression, including the case where
7527 @code{__builtin_constant_p} returns 1 because @var{EXPRESSION} can be
7528 folded to a constant but @var{EXPRESSION} contains operands that would
7529 not otherwise be permitted in a static initializer (for example,
7530 @code{0 && foo ()}). GCC must be more conservative about evaluating the
7531 built-in in this case, because it has no opportunity to perform
7534 Previous versions of GCC did not accept this built-in in data
7535 initializers. The earliest version where it is completely safe is
7539 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
7540 @opindex fprofile-arcs
7541 You may use @code{__builtin_expect} to provide the compiler with
7542 branch prediction information. In general, you should prefer to
7543 use actual profile feedback for this (@option{-fprofile-arcs}), as
7544 programmers are notoriously bad at predicting how their programs
7545 actually perform. However, there are applications in which this
7546 data is hard to collect.
7548 The return value is the value of @var{exp}, which should be an integral
7549 expression. The semantics of the built-in are that it is expected that
7550 @var{exp} == @var{c}. For example:
7553 if (__builtin_expect (x, 0))
7558 would indicate that we do not expect to call @code{foo}, since
7559 we expect @code{x} to be zero. Since you are limited to integral
7560 expressions for @var{exp}, you should use constructions such as
7563 if (__builtin_expect (ptr != NULL, 1))
7568 when testing pointer or floating-point values.
7571 @deftypefn {Built-in Function} void __builtin_trap (void)
7572 This function causes the program to exit abnormally. GCC implements
7573 this function by using a target-dependent mechanism (such as
7574 intentionally executing an illegal instruction) or by calling
7575 @code{abort}. The mechanism used may vary from release to release so
7576 you should not rely on any particular implementation.
7579 @deftypefn {Built-in Function} void __builtin_unreachable (void)
7580 If control flow reaches the point of the @code{__builtin_unreachable},
7581 the program is undefined. It is useful in situations where the
7582 compiler cannot deduce the unreachability of the code.
7584 One such case is immediately following an @code{asm} statement that
7585 will either never terminate, or one that transfers control elsewhere
7586 and never returns. In this example, without the
7587 @code{__builtin_unreachable}, GCC would issue a warning that control
7588 reaches the end of a non-void function. It would also generate code
7589 to return after the @code{asm}.
7592 int f (int c, int v)
7600 asm("jmp error_handler");
7601 __builtin_unreachable ();
7606 Because the @code{asm} statement unconditionally transfers control out
7607 of the function, control will never reach the end of the function
7608 body. The @code{__builtin_unreachable} is in fact unreachable and
7609 communicates this fact to the compiler.
7611 Another use for @code{__builtin_unreachable} is following a call a
7612 function that never returns but that is not declared
7613 @code{__attribute__((noreturn))}, as in this example:
7616 void function_that_never_returns (void);
7626 function_that_never_returns ();
7627 __builtin_unreachable ();
7634 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
7635 This function is used to flush the processor's instruction cache for
7636 the region of memory between @var{begin} inclusive and @var{end}
7637 exclusive. Some targets require that the instruction cache be
7638 flushed, after modifying memory containing code, in order to obtain
7639 deterministic behavior.
7641 If the target does not require instruction cache flushes,
7642 @code{__builtin___clear_cache} has no effect. Otherwise either
7643 instructions are emitted in-line to clear the instruction cache or a
7644 call to the @code{__clear_cache} function in libgcc is made.
7647 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
7648 This function is used to minimize cache-miss latency by moving data into
7649 a cache before it is accessed.
7650 You can insert calls to @code{__builtin_prefetch} into code for which
7651 you know addresses of data in memory that is likely to be accessed soon.
7652 If the target supports them, data prefetch instructions will be generated.
7653 If the prefetch is done early enough before the access then the data will
7654 be in the cache by the time it is accessed.
7656 The value of @var{addr} is the address of the memory to prefetch.
7657 There are two optional arguments, @var{rw} and @var{locality}.
7658 The value of @var{rw} is a compile-time constant one or zero; one
7659 means that the prefetch is preparing for a write to the memory address
7660 and zero, the default, means that the prefetch is preparing for a read.
7661 The value @var{locality} must be a compile-time constant integer between
7662 zero and three. A value of zero means that the data has no temporal
7663 locality, so it need not be left in the cache after the access. A value
7664 of three means that the data has a high degree of temporal locality and
7665 should be left in all levels of cache possible. Values of one and two
7666 mean, respectively, a low or moderate degree of temporal locality. The
7670 for (i = 0; i < n; i++)
7673 __builtin_prefetch (&a[i+j], 1, 1);
7674 __builtin_prefetch (&b[i+j], 0, 1);
7679 Data prefetch does not generate faults if @var{addr} is invalid, but
7680 the address expression itself must be valid. For example, a prefetch
7681 of @code{p->next} will not fault if @code{p->next} is not a valid
7682 address, but evaluation will fault if @code{p} is not a valid address.
7684 If the target does not support data prefetch, the address expression
7685 is evaluated if it includes side effects but no other code is generated
7686 and GCC does not issue a warning.
7689 @deftypefn {Built-in Function} double __builtin_huge_val (void)
7690 Returns a positive infinity, if supported by the floating-point format,
7691 else @code{DBL_MAX}. This function is suitable for implementing the
7692 ISO C macro @code{HUGE_VAL}.
7695 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
7696 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
7699 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
7700 Similar to @code{__builtin_huge_val}, except the return
7701 type is @code{long double}.
7704 @deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...)
7705 This built-in implements the C99 fpclassify functionality. The first
7706 five int arguments should be the target library's notion of the
7707 possible FP classes and are used for return values. They must be
7708 constant values and they must appear in this order: @code{FP_NAN},
7709 @code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and
7710 @code{FP_ZERO}. The ellipsis is for exactly one floating point value
7711 to classify. GCC treats the last argument as type-generic, which
7712 means it does not do default promotion from float to double.
7715 @deftypefn {Built-in Function} double __builtin_inf (void)
7716 Similar to @code{__builtin_huge_val}, except a warning is generated
7717 if the target floating-point format does not support infinities.
7720 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
7721 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
7724 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
7725 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
7728 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
7729 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
7732 @deftypefn {Built-in Function} float __builtin_inff (void)
7733 Similar to @code{__builtin_inf}, except the return type is @code{float}.
7734 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
7737 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
7738 Similar to @code{__builtin_inf}, except the return
7739 type is @code{long double}.
7742 @deftypefn {Built-in Function} int __builtin_isinf_sign (...)
7743 Similar to @code{isinf}, except the return value will be negative for
7744 an argument of @code{-Inf}. Note while the parameter list is an
7745 ellipsis, this function only accepts exactly one floating point
7746 argument. GCC treats this parameter as type-generic, which means it
7747 does not do default promotion from float to double.
7750 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
7751 This is an implementation of the ISO C99 function @code{nan}.
7753 Since ISO C99 defines this function in terms of @code{strtod}, which we
7754 do not implement, a description of the parsing is in order. The string
7755 is parsed as by @code{strtol}; that is, the base is recognized by
7756 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
7757 in the significand such that the least significant bit of the number
7758 is at the least significant bit of the significand. The number is
7759 truncated to fit the significand field provided. The significand is
7760 forced to be a quiet NaN@.
7762 This function, if given a string literal all of which would have been
7763 consumed by strtol, is evaluated early enough that it is considered a
7764 compile-time constant.
7767 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
7768 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
7771 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
7772 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
7775 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
7776 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
7779 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
7780 Similar to @code{__builtin_nan}, except the return type is @code{float}.
7783 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
7784 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
7787 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
7788 Similar to @code{__builtin_nan}, except the significand is forced
7789 to be a signaling NaN@. The @code{nans} function is proposed by
7790 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
7793 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
7794 Similar to @code{__builtin_nans}, except the return type is @code{float}.
7797 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
7798 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
7801 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
7802 Returns one plus the index of the least significant 1-bit of @var{x}, or
7803 if @var{x} is zero, returns zero.
7806 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
7807 Returns the number of leading 0-bits in @var{x}, starting at the most
7808 significant bit position. If @var{x} is 0, the result is undefined.
7811 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
7812 Returns the number of trailing 0-bits in @var{x}, starting at the least
7813 significant bit position. If @var{x} is 0, the result is undefined.
7816 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
7817 Returns the number of 1-bits in @var{x}.
7820 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
7821 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
7825 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
7826 Similar to @code{__builtin_ffs}, except the argument type is
7827 @code{unsigned long}.
7830 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
7831 Similar to @code{__builtin_clz}, except the argument type is
7832 @code{unsigned long}.
7835 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
7836 Similar to @code{__builtin_ctz}, except the argument type is
7837 @code{unsigned long}.
7840 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
7841 Similar to @code{__builtin_popcount}, except the argument type is
7842 @code{unsigned long}.
7845 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
7846 Similar to @code{__builtin_parity}, except the argument type is
7847 @code{unsigned long}.
7850 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
7851 Similar to @code{__builtin_ffs}, except the argument type is
7852 @code{unsigned long long}.
7855 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
7856 Similar to @code{__builtin_clz}, except the argument type is
7857 @code{unsigned long long}.
7860 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
7861 Similar to @code{__builtin_ctz}, except the argument type is
7862 @code{unsigned long long}.
7865 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
7866 Similar to @code{__builtin_popcount}, except the argument type is
7867 @code{unsigned long long}.
7870 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
7871 Similar to @code{__builtin_parity}, except the argument type is
7872 @code{unsigned long long}.
7875 @deftypefn {Built-in Function} double __builtin_powi (double, int)
7876 Returns the first argument raised to the power of the second. Unlike the
7877 @code{pow} function no guarantees about precision and rounding are made.
7880 @deftypefn {Built-in Function} float __builtin_powif (float, int)
7881 Similar to @code{__builtin_powi}, except the argument and return types
7885 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
7886 Similar to @code{__builtin_powi}, except the argument and return types
7887 are @code{long double}.
7890 @deftypefn {Built-in Function} int32_t __builtin_bswap32 (int32_t x)
7891 Returns @var{x} with the order of the bytes reversed; for example,
7892 @code{0xaabbccdd} becomes @code{0xddccbbaa}. Byte here always means
7896 @deftypefn {Built-in Function} int64_t __builtin_bswap64 (int64_t x)
7897 Similar to @code{__builtin_bswap32}, except the argument and return types
7901 @node Target Builtins
7902 @section Built-in Functions Specific to Particular Target Machines
7904 On some target machines, GCC supports many built-in functions specific
7905 to those machines. Generally these generate calls to specific machine
7906 instructions, but allow the compiler to schedule those calls.
7909 * Alpha Built-in Functions::
7910 * ARM iWMMXt Built-in Functions::
7911 * ARM NEON Intrinsics::
7912 * Blackfin Built-in Functions::
7913 * FR-V Built-in Functions::
7914 * X86 Built-in Functions::
7915 * MIPS DSP Built-in Functions::
7916 * MIPS Paired-Single Support::
7917 * MIPS Loongson Built-in Functions::
7918 * Other MIPS Built-in Functions::
7919 * picoChip Built-in Functions::
7920 * PowerPC AltiVec/VSX Built-in Functions::
7921 * RX Built-in Functions::
7922 * SPARC VIS Built-in Functions::
7923 * SPU Built-in Functions::
7926 @node Alpha Built-in Functions
7927 @subsection Alpha Built-in Functions
7929 These built-in functions are available for the Alpha family of
7930 processors, depending on the command-line switches used.
7932 The following built-in functions are always available. They
7933 all generate the machine instruction that is part of the name.
7936 long __builtin_alpha_implver (void)
7937 long __builtin_alpha_rpcc (void)
7938 long __builtin_alpha_amask (long)
7939 long __builtin_alpha_cmpbge (long, long)
7940 long __builtin_alpha_extbl (long, long)
7941 long __builtin_alpha_extwl (long, long)
7942 long __builtin_alpha_extll (long, long)
7943 long __builtin_alpha_extql (long, long)
7944 long __builtin_alpha_extwh (long, long)
7945 long __builtin_alpha_extlh (long, long)
7946 long __builtin_alpha_extqh (long, long)
7947 long __builtin_alpha_insbl (long, long)
7948 long __builtin_alpha_inswl (long, long)
7949 long __builtin_alpha_insll (long, long)
7950 long __builtin_alpha_insql (long, long)
7951 long __builtin_alpha_inswh (long, long)
7952 long __builtin_alpha_inslh (long, long)
7953 long __builtin_alpha_insqh (long, long)
7954 long __builtin_alpha_mskbl (long, long)
7955 long __builtin_alpha_mskwl (long, long)
7956 long __builtin_alpha_mskll (long, long)
7957 long __builtin_alpha_mskql (long, long)
7958 long __builtin_alpha_mskwh (long, long)
7959 long __builtin_alpha_msklh (long, long)
7960 long __builtin_alpha_mskqh (long, long)
7961 long __builtin_alpha_umulh (long, long)
7962 long __builtin_alpha_zap (long, long)
7963 long __builtin_alpha_zapnot (long, long)
7966 The following built-in functions are always with @option{-mmax}
7967 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
7968 later. They all generate the machine instruction that is part
7972 long __builtin_alpha_pklb (long)
7973 long __builtin_alpha_pkwb (long)
7974 long __builtin_alpha_unpkbl (long)
7975 long __builtin_alpha_unpkbw (long)
7976 long __builtin_alpha_minub8 (long, long)
7977 long __builtin_alpha_minsb8 (long, long)
7978 long __builtin_alpha_minuw4 (long, long)
7979 long __builtin_alpha_minsw4 (long, long)
7980 long __builtin_alpha_maxub8 (long, long)
7981 long __builtin_alpha_maxsb8 (long, long)
7982 long __builtin_alpha_maxuw4 (long, long)
7983 long __builtin_alpha_maxsw4 (long, long)
7984 long __builtin_alpha_perr (long, long)
7987 The following built-in functions are always with @option{-mcix}
7988 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
7989 later. They all generate the machine instruction that is part
7993 long __builtin_alpha_cttz (long)
7994 long __builtin_alpha_ctlz (long)
7995 long __builtin_alpha_ctpop (long)
7998 The following builtins are available on systems that use the OSF/1
7999 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
8000 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
8001 @code{rdval} and @code{wrval}.
8004 void *__builtin_thread_pointer (void)
8005 void __builtin_set_thread_pointer (void *)
8008 @node ARM iWMMXt Built-in Functions
8009 @subsection ARM iWMMXt Built-in Functions
8011 These built-in functions are available for the ARM family of
8012 processors when the @option{-mcpu=iwmmxt} switch is used:
8015 typedef int v2si __attribute__ ((vector_size (8)));
8016 typedef short v4hi __attribute__ ((vector_size (8)));
8017 typedef char v8qi __attribute__ ((vector_size (8)));
8019 int __builtin_arm_getwcx (int)
8020 void __builtin_arm_setwcx (int, int)
8021 int __builtin_arm_textrmsb (v8qi, int)
8022 int __builtin_arm_textrmsh (v4hi, int)
8023 int __builtin_arm_textrmsw (v2si, int)
8024 int __builtin_arm_textrmub (v8qi, int)
8025 int __builtin_arm_textrmuh (v4hi, int)
8026 int __builtin_arm_textrmuw (v2si, int)
8027 v8qi __builtin_arm_tinsrb (v8qi, int)
8028 v4hi __builtin_arm_tinsrh (v4hi, int)
8029 v2si __builtin_arm_tinsrw (v2si, int)
8030 long long __builtin_arm_tmia (long long, int, int)
8031 long long __builtin_arm_tmiabb (long long, int, int)
8032 long long __builtin_arm_tmiabt (long long, int, int)
8033 long long __builtin_arm_tmiaph (long long, int, int)
8034 long long __builtin_arm_tmiatb (long long, int, int)
8035 long long __builtin_arm_tmiatt (long long, int, int)
8036 int __builtin_arm_tmovmskb (v8qi)
8037 int __builtin_arm_tmovmskh (v4hi)
8038 int __builtin_arm_tmovmskw (v2si)
8039 long long __builtin_arm_waccb (v8qi)
8040 long long __builtin_arm_wacch (v4hi)
8041 long long __builtin_arm_waccw (v2si)
8042 v8qi __builtin_arm_waddb (v8qi, v8qi)
8043 v8qi __builtin_arm_waddbss (v8qi, v8qi)
8044 v8qi __builtin_arm_waddbus (v8qi, v8qi)
8045 v4hi __builtin_arm_waddh (v4hi, v4hi)
8046 v4hi __builtin_arm_waddhss (v4hi, v4hi)
8047 v4hi __builtin_arm_waddhus (v4hi, v4hi)
8048 v2si __builtin_arm_waddw (v2si, v2si)
8049 v2si __builtin_arm_waddwss (v2si, v2si)
8050 v2si __builtin_arm_waddwus (v2si, v2si)
8051 v8qi __builtin_arm_walign (v8qi, v8qi, int)
8052 long long __builtin_arm_wand(long long, long long)
8053 long long __builtin_arm_wandn (long long, long long)
8054 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
8055 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
8056 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
8057 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
8058 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
8059 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
8060 v2si __builtin_arm_wcmpeqw (v2si, v2si)
8061 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
8062 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
8063 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
8064 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
8065 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
8066 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
8067 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
8068 long long __builtin_arm_wmacsz (v4hi, v4hi)
8069 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
8070 long long __builtin_arm_wmacuz (v4hi, v4hi)
8071 v4hi __builtin_arm_wmadds (v4hi, v4hi)
8072 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
8073 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
8074 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
8075 v2si __builtin_arm_wmaxsw (v2si, v2si)
8076 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
8077 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
8078 v2si __builtin_arm_wmaxuw (v2si, v2si)
8079 v8qi __builtin_arm_wminsb (v8qi, v8qi)
8080 v4hi __builtin_arm_wminsh (v4hi, v4hi)
8081 v2si __builtin_arm_wminsw (v2si, v2si)
8082 v8qi __builtin_arm_wminub (v8qi, v8qi)
8083 v4hi __builtin_arm_wminuh (v4hi, v4hi)
8084 v2si __builtin_arm_wminuw (v2si, v2si)
8085 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
8086 v4hi __builtin_arm_wmulul (v4hi, v4hi)
8087 v4hi __builtin_arm_wmulum (v4hi, v4hi)
8088 long long __builtin_arm_wor (long long, long long)
8089 v2si __builtin_arm_wpackdss (long long, long long)
8090 v2si __builtin_arm_wpackdus (long long, long long)
8091 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
8092 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
8093 v4hi __builtin_arm_wpackwss (v2si, v2si)
8094 v4hi __builtin_arm_wpackwus (v2si, v2si)
8095 long long __builtin_arm_wrord (long long, long long)
8096 long long __builtin_arm_wrordi (long long, int)
8097 v4hi __builtin_arm_wrorh (v4hi, long long)
8098 v4hi __builtin_arm_wrorhi (v4hi, int)
8099 v2si __builtin_arm_wrorw (v2si, long long)
8100 v2si __builtin_arm_wrorwi (v2si, int)
8101 v2si __builtin_arm_wsadb (v8qi, v8qi)
8102 v2si __builtin_arm_wsadbz (v8qi, v8qi)
8103 v2si __builtin_arm_wsadh (v4hi, v4hi)
8104 v2si __builtin_arm_wsadhz (v4hi, v4hi)
8105 v4hi __builtin_arm_wshufh (v4hi, int)
8106 long long __builtin_arm_wslld (long long, long long)
8107 long long __builtin_arm_wslldi (long long, int)
8108 v4hi __builtin_arm_wsllh (v4hi, long long)
8109 v4hi __builtin_arm_wsllhi (v4hi, int)
8110 v2si __builtin_arm_wsllw (v2si, long long)
8111 v2si __builtin_arm_wsllwi (v2si, int)
8112 long long __builtin_arm_wsrad (long long, long long)
8113 long long __builtin_arm_wsradi (long long, int)
8114 v4hi __builtin_arm_wsrah (v4hi, long long)
8115 v4hi __builtin_arm_wsrahi (v4hi, int)
8116 v2si __builtin_arm_wsraw (v2si, long long)
8117 v2si __builtin_arm_wsrawi (v2si, int)
8118 long long __builtin_arm_wsrld (long long, long long)
8119 long long __builtin_arm_wsrldi (long long, int)
8120 v4hi __builtin_arm_wsrlh (v4hi, long long)
8121 v4hi __builtin_arm_wsrlhi (v4hi, int)
8122 v2si __builtin_arm_wsrlw (v2si, long long)
8123 v2si __builtin_arm_wsrlwi (v2si, int)
8124 v8qi __builtin_arm_wsubb (v8qi, v8qi)
8125 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
8126 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
8127 v4hi __builtin_arm_wsubh (v4hi, v4hi)
8128 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
8129 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
8130 v2si __builtin_arm_wsubw (v2si, v2si)
8131 v2si __builtin_arm_wsubwss (v2si, v2si)
8132 v2si __builtin_arm_wsubwus (v2si, v2si)
8133 v4hi __builtin_arm_wunpckehsb (v8qi)
8134 v2si __builtin_arm_wunpckehsh (v4hi)
8135 long long __builtin_arm_wunpckehsw (v2si)
8136 v4hi __builtin_arm_wunpckehub (v8qi)
8137 v2si __builtin_arm_wunpckehuh (v4hi)
8138 long long __builtin_arm_wunpckehuw (v2si)
8139 v4hi __builtin_arm_wunpckelsb (v8qi)
8140 v2si __builtin_arm_wunpckelsh (v4hi)
8141 long long __builtin_arm_wunpckelsw (v2si)
8142 v4hi __builtin_arm_wunpckelub (v8qi)
8143 v2si __builtin_arm_wunpckeluh (v4hi)
8144 long long __builtin_arm_wunpckeluw (v2si)
8145 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
8146 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
8147 v2si __builtin_arm_wunpckihw (v2si, v2si)
8148 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
8149 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
8150 v2si __builtin_arm_wunpckilw (v2si, v2si)
8151 long long __builtin_arm_wxor (long long, long long)
8152 long long __builtin_arm_wzero ()
8155 @node ARM NEON Intrinsics
8156 @subsection ARM NEON Intrinsics
8158 These built-in intrinsics for the ARM Advanced SIMD extension are available
8159 when the @option{-mfpu=neon} switch is used:
8161 @include arm-neon-intrinsics.texi
8163 @node Blackfin Built-in Functions
8164 @subsection Blackfin Built-in Functions
8166 Currently, there are two Blackfin-specific built-in functions. These are
8167 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
8168 using inline assembly; by using these built-in functions the compiler can
8169 automatically add workarounds for hardware errata involving these
8170 instructions. These functions are named as follows:
8173 void __builtin_bfin_csync (void)
8174 void __builtin_bfin_ssync (void)
8177 @node FR-V Built-in Functions
8178 @subsection FR-V Built-in Functions
8180 GCC provides many FR-V-specific built-in functions. In general,
8181 these functions are intended to be compatible with those described
8182 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
8183 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
8184 @code{__MBTOHE}, the gcc forms of which pass 128-bit values by
8185 pointer rather than by value.
8187 Most of the functions are named after specific FR-V instructions.
8188 Such functions are said to be ``directly mapped'' and are summarized
8189 here in tabular form.
8193 * Directly-mapped Integer Functions::
8194 * Directly-mapped Media Functions::
8195 * Raw read/write Functions::
8196 * Other Built-in Functions::
8199 @node Argument Types
8200 @subsubsection Argument Types
8202 The arguments to the built-in functions can be divided into three groups:
8203 register numbers, compile-time constants and run-time values. In order
8204 to make this classification clear at a glance, the arguments and return
8205 values are given the following pseudo types:
8207 @multitable @columnfractions .20 .30 .15 .35
8208 @item Pseudo type @tab Real C type @tab Constant? @tab Description
8209 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
8210 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
8211 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
8212 @item @code{uw2} @tab @code{unsigned long long} @tab No
8213 @tab an unsigned doubleword
8214 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
8215 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
8216 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
8217 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
8220 These pseudo types are not defined by GCC, they are simply a notational
8221 convenience used in this manual.
8223 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
8224 and @code{sw2} are evaluated at run time. They correspond to
8225 register operands in the underlying FR-V instructions.
8227 @code{const} arguments represent immediate operands in the underlying
8228 FR-V instructions. They must be compile-time constants.
8230 @code{acc} arguments are evaluated at compile time and specify the number
8231 of an accumulator register. For example, an @code{acc} argument of 2
8232 will select the ACC2 register.
8234 @code{iacc} arguments are similar to @code{acc} arguments but specify the
8235 number of an IACC register. See @pxref{Other Built-in Functions}
8238 @node Directly-mapped Integer Functions
8239 @subsubsection Directly-mapped Integer Functions
8241 The functions listed below map directly to FR-V I-type instructions.
8243 @multitable @columnfractions .45 .32 .23
8244 @item Function prototype @tab Example usage @tab Assembly output
8245 @item @code{sw1 __ADDSS (sw1, sw1)}
8246 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
8247 @tab @code{ADDSS @var{a},@var{b},@var{c}}
8248 @item @code{sw1 __SCAN (sw1, sw1)}
8249 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
8250 @tab @code{SCAN @var{a},@var{b},@var{c}}
8251 @item @code{sw1 __SCUTSS (sw1)}
8252 @tab @code{@var{b} = __SCUTSS (@var{a})}
8253 @tab @code{SCUTSS @var{a},@var{b}}
8254 @item @code{sw1 __SLASS (sw1, sw1)}
8255 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
8256 @tab @code{SLASS @var{a},@var{b},@var{c}}
8257 @item @code{void __SMASS (sw1, sw1)}
8258 @tab @code{__SMASS (@var{a}, @var{b})}
8259 @tab @code{SMASS @var{a},@var{b}}
8260 @item @code{void __SMSSS (sw1, sw1)}
8261 @tab @code{__SMSSS (@var{a}, @var{b})}
8262 @tab @code{SMSSS @var{a},@var{b}}
8263 @item @code{void __SMU (sw1, sw1)}
8264 @tab @code{__SMU (@var{a}, @var{b})}
8265 @tab @code{SMU @var{a},@var{b}}
8266 @item @code{sw2 __SMUL (sw1, sw1)}
8267 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
8268 @tab @code{SMUL @var{a},@var{b},@var{c}}
8269 @item @code{sw1 __SUBSS (sw1, sw1)}
8270 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
8271 @tab @code{SUBSS @var{a},@var{b},@var{c}}
8272 @item @code{uw2 __UMUL (uw1, uw1)}
8273 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
8274 @tab @code{UMUL @var{a},@var{b},@var{c}}
8277 @node Directly-mapped Media Functions
8278 @subsubsection Directly-mapped Media Functions
8280 The functions listed below map directly to FR-V M-type instructions.
8282 @multitable @columnfractions .45 .32 .23
8283 @item Function prototype @tab Example usage @tab Assembly output
8284 @item @code{uw1 __MABSHS (sw1)}
8285 @tab @code{@var{b} = __MABSHS (@var{a})}
8286 @tab @code{MABSHS @var{a},@var{b}}
8287 @item @code{void __MADDACCS (acc, acc)}
8288 @tab @code{__MADDACCS (@var{b}, @var{a})}
8289 @tab @code{MADDACCS @var{a},@var{b}}
8290 @item @code{sw1 __MADDHSS (sw1, sw1)}
8291 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
8292 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
8293 @item @code{uw1 __MADDHUS (uw1, uw1)}
8294 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
8295 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
8296 @item @code{uw1 __MAND (uw1, uw1)}
8297 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
8298 @tab @code{MAND @var{a},@var{b},@var{c}}
8299 @item @code{void __MASACCS (acc, acc)}
8300 @tab @code{__MASACCS (@var{b}, @var{a})}
8301 @tab @code{MASACCS @var{a},@var{b}}
8302 @item @code{uw1 __MAVEH (uw1, uw1)}
8303 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
8304 @tab @code{MAVEH @var{a},@var{b},@var{c}}
8305 @item @code{uw2 __MBTOH (uw1)}
8306 @tab @code{@var{b} = __MBTOH (@var{a})}
8307 @tab @code{MBTOH @var{a},@var{b}}
8308 @item @code{void __MBTOHE (uw1 *, uw1)}
8309 @tab @code{__MBTOHE (&@var{b}, @var{a})}
8310 @tab @code{MBTOHE @var{a},@var{b}}
8311 @item @code{void __MCLRACC (acc)}
8312 @tab @code{__MCLRACC (@var{a})}
8313 @tab @code{MCLRACC @var{a}}
8314 @item @code{void __MCLRACCA (void)}
8315 @tab @code{__MCLRACCA ()}
8316 @tab @code{MCLRACCA}
8317 @item @code{uw1 __Mcop1 (uw1, uw1)}
8318 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
8319 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
8320 @item @code{uw1 __Mcop2 (uw1, uw1)}
8321 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
8322 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
8323 @item @code{uw1 __MCPLHI (uw2, const)}
8324 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
8325 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
8326 @item @code{uw1 __MCPLI (uw2, const)}
8327 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
8328 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
8329 @item @code{void __MCPXIS (acc, sw1, sw1)}
8330 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
8331 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
8332 @item @code{void __MCPXIU (acc, uw1, uw1)}
8333 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
8334 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
8335 @item @code{void __MCPXRS (acc, sw1, sw1)}
8336 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
8337 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
8338 @item @code{void __MCPXRU (acc, uw1, uw1)}
8339 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
8340 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
8341 @item @code{uw1 __MCUT (acc, uw1)}
8342 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
8343 @tab @code{MCUT @var{a},@var{b},@var{c}}
8344 @item @code{uw1 __MCUTSS (acc, sw1)}
8345 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
8346 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
8347 @item @code{void __MDADDACCS (acc, acc)}
8348 @tab @code{__MDADDACCS (@var{b}, @var{a})}
8349 @tab @code{MDADDACCS @var{a},@var{b}}
8350 @item @code{void __MDASACCS (acc, acc)}
8351 @tab @code{__MDASACCS (@var{b}, @var{a})}
8352 @tab @code{MDASACCS @var{a},@var{b}}
8353 @item @code{uw2 __MDCUTSSI (acc, const)}
8354 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
8355 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
8356 @item @code{uw2 __MDPACKH (uw2, uw2)}
8357 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
8358 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
8359 @item @code{uw2 __MDROTLI (uw2, const)}
8360 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
8361 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
8362 @item @code{void __MDSUBACCS (acc, acc)}
8363 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
8364 @tab @code{MDSUBACCS @var{a},@var{b}}
8365 @item @code{void __MDUNPACKH (uw1 *, uw2)}
8366 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
8367 @tab @code{MDUNPACKH @var{a},@var{b}}
8368 @item @code{uw2 __MEXPDHD (uw1, const)}
8369 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
8370 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
8371 @item @code{uw1 __MEXPDHW (uw1, const)}
8372 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
8373 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
8374 @item @code{uw1 __MHDSETH (uw1, const)}
8375 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
8376 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
8377 @item @code{sw1 __MHDSETS (const)}
8378 @tab @code{@var{b} = __MHDSETS (@var{a})}
8379 @tab @code{MHDSETS #@var{a},@var{b}}
8380 @item @code{uw1 __MHSETHIH (uw1, const)}
8381 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
8382 @tab @code{MHSETHIH #@var{a},@var{b}}
8383 @item @code{sw1 __MHSETHIS (sw1, const)}
8384 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
8385 @tab @code{MHSETHIS #@var{a},@var{b}}
8386 @item @code{uw1 __MHSETLOH (uw1, const)}
8387 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
8388 @tab @code{MHSETLOH #@var{a},@var{b}}
8389 @item @code{sw1 __MHSETLOS (sw1, const)}
8390 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
8391 @tab @code{MHSETLOS #@var{a},@var{b}}
8392 @item @code{uw1 __MHTOB (uw2)}
8393 @tab @code{@var{b} = __MHTOB (@var{a})}
8394 @tab @code{MHTOB @var{a},@var{b}}
8395 @item @code{void __MMACHS (acc, sw1, sw1)}
8396 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
8397 @tab @code{MMACHS @var{a},@var{b},@var{c}}
8398 @item @code{void __MMACHU (acc, uw1, uw1)}
8399 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
8400 @tab @code{MMACHU @var{a},@var{b},@var{c}}
8401 @item @code{void __MMRDHS (acc, sw1, sw1)}
8402 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
8403 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
8404 @item @code{void __MMRDHU (acc, uw1, uw1)}
8405 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
8406 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
8407 @item @code{void __MMULHS (acc, sw1, sw1)}
8408 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
8409 @tab @code{MMULHS @var{a},@var{b},@var{c}}
8410 @item @code{void __MMULHU (acc, uw1, uw1)}
8411 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
8412 @tab @code{MMULHU @var{a},@var{b},@var{c}}
8413 @item @code{void __MMULXHS (acc, sw1, sw1)}
8414 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
8415 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
8416 @item @code{void __MMULXHU (acc, uw1, uw1)}
8417 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
8418 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
8419 @item @code{uw1 __MNOT (uw1)}
8420 @tab @code{@var{b} = __MNOT (@var{a})}
8421 @tab @code{MNOT @var{a},@var{b}}
8422 @item @code{uw1 __MOR (uw1, uw1)}
8423 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
8424 @tab @code{MOR @var{a},@var{b},@var{c}}
8425 @item @code{uw1 __MPACKH (uh, uh)}
8426 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
8427 @tab @code{MPACKH @var{a},@var{b},@var{c}}
8428 @item @code{sw2 __MQADDHSS (sw2, sw2)}
8429 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
8430 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
8431 @item @code{uw2 __MQADDHUS (uw2, uw2)}
8432 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
8433 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
8434 @item @code{void __MQCPXIS (acc, sw2, sw2)}
8435 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
8436 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
8437 @item @code{void __MQCPXIU (acc, uw2, uw2)}
8438 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
8439 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
8440 @item @code{void __MQCPXRS (acc, sw2, sw2)}
8441 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
8442 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
8443 @item @code{void __MQCPXRU (acc, uw2, uw2)}
8444 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
8445 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
8446 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
8447 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
8448 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
8449 @item @code{sw2 __MQLMTHS (sw2, sw2)}
8450 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
8451 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
8452 @item @code{void __MQMACHS (acc, sw2, sw2)}
8453 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
8454 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
8455 @item @code{void __MQMACHU (acc, uw2, uw2)}
8456 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
8457 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
8458 @item @code{void __MQMACXHS (acc, sw2, sw2)}
8459 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
8460 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
8461 @item @code{void __MQMULHS (acc, sw2, sw2)}
8462 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
8463 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
8464 @item @code{void __MQMULHU (acc, uw2, uw2)}
8465 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
8466 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
8467 @item @code{void __MQMULXHS (acc, sw2, sw2)}
8468 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
8469 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
8470 @item @code{void __MQMULXHU (acc, uw2, uw2)}
8471 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
8472 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
8473 @item @code{sw2 __MQSATHS (sw2, sw2)}
8474 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
8475 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
8476 @item @code{uw2 __MQSLLHI (uw2, int)}
8477 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
8478 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
8479 @item @code{sw2 __MQSRAHI (sw2, int)}
8480 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
8481 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
8482 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
8483 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
8484 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
8485 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
8486 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
8487 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
8488 @item @code{void __MQXMACHS (acc, sw2, sw2)}
8489 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
8490 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
8491 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
8492 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
8493 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
8494 @item @code{uw1 __MRDACC (acc)}
8495 @tab @code{@var{b} = __MRDACC (@var{a})}
8496 @tab @code{MRDACC @var{a},@var{b}}
8497 @item @code{uw1 __MRDACCG (acc)}
8498 @tab @code{@var{b} = __MRDACCG (@var{a})}
8499 @tab @code{MRDACCG @var{a},@var{b}}
8500 @item @code{uw1 __MROTLI (uw1, const)}
8501 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
8502 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
8503 @item @code{uw1 __MROTRI (uw1, const)}
8504 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
8505 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
8506 @item @code{sw1 __MSATHS (sw1, sw1)}
8507 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
8508 @tab @code{MSATHS @var{a},@var{b},@var{c}}
8509 @item @code{uw1 __MSATHU (uw1, uw1)}
8510 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
8511 @tab @code{MSATHU @var{a},@var{b},@var{c}}
8512 @item @code{uw1 __MSLLHI (uw1, const)}
8513 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
8514 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
8515 @item @code{sw1 __MSRAHI (sw1, const)}
8516 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
8517 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
8518 @item @code{uw1 __MSRLHI (uw1, const)}
8519 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
8520 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
8521 @item @code{void __MSUBACCS (acc, acc)}
8522 @tab @code{__MSUBACCS (@var{b}, @var{a})}
8523 @tab @code{MSUBACCS @var{a},@var{b}}
8524 @item @code{sw1 __MSUBHSS (sw1, sw1)}
8525 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
8526 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
8527 @item @code{uw1 __MSUBHUS (uw1, uw1)}
8528 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
8529 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
8530 @item @code{void __MTRAP (void)}
8531 @tab @code{__MTRAP ()}
8533 @item @code{uw2 __MUNPACKH (uw1)}
8534 @tab @code{@var{b} = __MUNPACKH (@var{a})}
8535 @tab @code{MUNPACKH @var{a},@var{b}}
8536 @item @code{uw1 __MWCUT (uw2, uw1)}
8537 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
8538 @tab @code{MWCUT @var{a},@var{b},@var{c}}
8539 @item @code{void __MWTACC (acc, uw1)}
8540 @tab @code{__MWTACC (@var{b}, @var{a})}
8541 @tab @code{MWTACC @var{a},@var{b}}
8542 @item @code{void __MWTACCG (acc, uw1)}
8543 @tab @code{__MWTACCG (@var{b}, @var{a})}
8544 @tab @code{MWTACCG @var{a},@var{b}}
8545 @item @code{uw1 __MXOR (uw1, uw1)}
8546 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
8547 @tab @code{MXOR @var{a},@var{b},@var{c}}
8550 @node Raw read/write Functions
8551 @subsubsection Raw read/write Functions
8553 This sections describes built-in functions related to read and write
8554 instructions to access memory. These functions generate
8555 @code{membar} instructions to flush the I/O load and stores where
8556 appropriate, as described in Fujitsu's manual described above.
8560 @item unsigned char __builtin_read8 (void *@var{data})
8561 @item unsigned short __builtin_read16 (void *@var{data})
8562 @item unsigned long __builtin_read32 (void *@var{data})
8563 @item unsigned long long __builtin_read64 (void *@var{data})
8565 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
8566 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
8567 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
8568 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
8571 @node Other Built-in Functions
8572 @subsubsection Other Built-in Functions
8574 This section describes built-in functions that are not named after
8575 a specific FR-V instruction.
8578 @item sw2 __IACCreadll (iacc @var{reg})
8579 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
8580 for future expansion and must be 0.
8582 @item sw1 __IACCreadl (iacc @var{reg})
8583 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
8584 Other values of @var{reg} are rejected as invalid.
8586 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
8587 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
8588 is reserved for future expansion and must be 0.
8590 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
8591 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
8592 is 1. Other values of @var{reg} are rejected as invalid.
8594 @item void __data_prefetch0 (const void *@var{x})
8595 Use the @code{dcpl} instruction to load the contents of address @var{x}
8596 into the data cache.
8598 @item void __data_prefetch (const void *@var{x})
8599 Use the @code{nldub} instruction to load the contents of address @var{x}
8600 into the data cache. The instruction will be issued in slot I1@.
8603 @node X86 Built-in Functions
8604 @subsection X86 Built-in Functions
8606 These built-in functions are available for the i386 and x86-64 family
8607 of computers, depending on the command-line switches used.
8609 Note that, if you specify command-line switches such as @option{-msse},
8610 the compiler could use the extended instruction sets even if the built-ins
8611 are not used explicitly in the program. For this reason, applications
8612 which perform runtime CPU detection must compile separate files for each
8613 supported architecture, using the appropriate flags. In particular,
8614 the file containing the CPU detection code should be compiled without
8617 The following machine modes are available for use with MMX built-in functions
8618 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
8619 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
8620 vector of eight 8-bit integers. Some of the built-in functions operate on
8621 MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode.
8623 If 3DNow!@: extensions are enabled, @code{V2SF} is used as a mode for a vector
8624 of two 32-bit floating point values.
8626 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
8627 floating point values. Some instructions use a vector of four 32-bit
8628 integers, these use @code{V4SI}. Finally, some instructions operate on an
8629 entire vector register, interpreting it as a 128-bit integer, these use mode
8632 In 64-bit mode, the x86-64 family of processors uses additional built-in
8633 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
8634 floating point and @code{TC} 128-bit complex floating point values.
8636 The following floating point built-in functions are available in 64-bit
8637 mode. All of them implement the function that is part of the name.
8640 __float128 __builtin_fabsq (__float128)
8641 __float128 __builtin_copysignq (__float128, __float128)
8644 The following floating point built-in functions are made available in the
8648 @item __float128 __builtin_infq (void)
8649 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
8650 @findex __builtin_infq
8652 @item __float128 __builtin_huge_valq (void)
8653 Similar to @code{__builtin_huge_val}, except the return type is @code{__float128}.
8654 @findex __builtin_huge_valq
8657 The following built-in functions are made available by @option{-mmmx}.
8658 All of them generate the machine instruction that is part of the name.
8661 v8qi __builtin_ia32_paddb (v8qi, v8qi)
8662 v4hi __builtin_ia32_paddw (v4hi, v4hi)
8663 v2si __builtin_ia32_paddd (v2si, v2si)
8664 v8qi __builtin_ia32_psubb (v8qi, v8qi)
8665 v4hi __builtin_ia32_psubw (v4hi, v4hi)
8666 v2si __builtin_ia32_psubd (v2si, v2si)
8667 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
8668 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
8669 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
8670 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
8671 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
8672 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
8673 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
8674 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
8675 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
8676 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
8677 di __builtin_ia32_pand (di, di)
8678 di __builtin_ia32_pandn (di,di)
8679 di __builtin_ia32_por (di, di)
8680 di __builtin_ia32_pxor (di, di)
8681 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
8682 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
8683 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
8684 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
8685 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
8686 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
8687 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
8688 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
8689 v2si __builtin_ia32_punpckhdq (v2si, v2si)
8690 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
8691 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
8692 v2si __builtin_ia32_punpckldq (v2si, v2si)
8693 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
8694 v4hi __builtin_ia32_packssdw (v2si, v2si)
8695 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
8697 v4hi __builtin_ia32_psllw (v4hi, v4hi)
8698 v2si __builtin_ia32_pslld (v2si, v2si)
8699 v1di __builtin_ia32_psllq (v1di, v1di)
8700 v4hi __builtin_ia32_psrlw (v4hi, v4hi)
8701 v2si __builtin_ia32_psrld (v2si, v2si)
8702 v1di __builtin_ia32_psrlq (v1di, v1di)
8703 v4hi __builtin_ia32_psraw (v4hi, v4hi)
8704 v2si __builtin_ia32_psrad (v2si, v2si)
8705 v4hi __builtin_ia32_psllwi (v4hi, int)
8706 v2si __builtin_ia32_pslldi (v2si, int)
8707 v1di __builtin_ia32_psllqi (v1di, int)
8708 v4hi __builtin_ia32_psrlwi (v4hi, int)
8709 v2si __builtin_ia32_psrldi (v2si, int)
8710 v1di __builtin_ia32_psrlqi (v1di, int)
8711 v4hi __builtin_ia32_psrawi (v4hi, int)
8712 v2si __builtin_ia32_psradi (v2si, int)
8716 The following built-in functions are made available either with
8717 @option{-msse}, or with a combination of @option{-m3dnow} and
8718 @option{-march=athlon}. All of them generate the machine
8719 instruction that is part of the name.
8722 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
8723 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
8724 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
8725 v1di __builtin_ia32_psadbw (v8qi, v8qi)
8726 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
8727 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
8728 v8qi __builtin_ia32_pminub (v8qi, v8qi)
8729 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
8730 int __builtin_ia32_pextrw (v4hi, int)
8731 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
8732 int __builtin_ia32_pmovmskb (v8qi)
8733 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
8734 void __builtin_ia32_movntq (di *, di)
8735 void __builtin_ia32_sfence (void)
8738 The following built-in functions are available when @option{-msse} is used.
8739 All of them generate the machine instruction that is part of the name.
8742 int __builtin_ia32_comieq (v4sf, v4sf)
8743 int __builtin_ia32_comineq (v4sf, v4sf)
8744 int __builtin_ia32_comilt (v4sf, v4sf)
8745 int __builtin_ia32_comile (v4sf, v4sf)
8746 int __builtin_ia32_comigt (v4sf, v4sf)
8747 int __builtin_ia32_comige (v4sf, v4sf)
8748 int __builtin_ia32_ucomieq (v4sf, v4sf)
8749 int __builtin_ia32_ucomineq (v4sf, v4sf)
8750 int __builtin_ia32_ucomilt (v4sf, v4sf)
8751 int __builtin_ia32_ucomile (v4sf, v4sf)
8752 int __builtin_ia32_ucomigt (v4sf, v4sf)
8753 int __builtin_ia32_ucomige (v4sf, v4sf)
8754 v4sf __builtin_ia32_addps (v4sf, v4sf)
8755 v4sf __builtin_ia32_subps (v4sf, v4sf)
8756 v4sf __builtin_ia32_mulps (v4sf, v4sf)
8757 v4sf __builtin_ia32_divps (v4sf, v4sf)
8758 v4sf __builtin_ia32_addss (v4sf, v4sf)
8759 v4sf __builtin_ia32_subss (v4sf, v4sf)
8760 v4sf __builtin_ia32_mulss (v4sf, v4sf)
8761 v4sf __builtin_ia32_divss (v4sf, v4sf)
8762 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
8763 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
8764 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
8765 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
8766 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
8767 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
8768 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
8769 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
8770 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
8771 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
8772 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
8773 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
8774 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
8775 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
8776 v4si __builtin_ia32_cmpless (v4sf, v4sf)
8777 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
8778 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
8779 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
8780 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
8781 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
8782 v4sf __builtin_ia32_maxps (v4sf, v4sf)
8783 v4sf __builtin_ia32_maxss (v4sf, v4sf)
8784 v4sf __builtin_ia32_minps (v4sf, v4sf)
8785 v4sf __builtin_ia32_minss (v4sf, v4sf)
8786 v4sf __builtin_ia32_andps (v4sf, v4sf)
8787 v4sf __builtin_ia32_andnps (v4sf, v4sf)
8788 v4sf __builtin_ia32_orps (v4sf, v4sf)
8789 v4sf __builtin_ia32_xorps (v4sf, v4sf)
8790 v4sf __builtin_ia32_movss (v4sf, v4sf)
8791 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
8792 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
8793 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
8794 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
8795 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
8796 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
8797 v2si __builtin_ia32_cvtps2pi (v4sf)
8798 int __builtin_ia32_cvtss2si (v4sf)
8799 v2si __builtin_ia32_cvttps2pi (v4sf)
8800 int __builtin_ia32_cvttss2si (v4sf)
8801 v4sf __builtin_ia32_rcpps (v4sf)
8802 v4sf __builtin_ia32_rsqrtps (v4sf)
8803 v4sf __builtin_ia32_sqrtps (v4sf)
8804 v4sf __builtin_ia32_rcpss (v4sf)
8805 v4sf __builtin_ia32_rsqrtss (v4sf)
8806 v4sf __builtin_ia32_sqrtss (v4sf)
8807 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
8808 void __builtin_ia32_movntps (float *, v4sf)
8809 int __builtin_ia32_movmskps (v4sf)
8812 The following built-in functions are available when @option{-msse} is used.
8815 @item v4sf __builtin_ia32_loadaps (float *)
8816 Generates the @code{movaps} machine instruction as a load from memory.
8817 @item void __builtin_ia32_storeaps (float *, v4sf)
8818 Generates the @code{movaps} machine instruction as a store to memory.
8819 @item v4sf __builtin_ia32_loadups (float *)
8820 Generates the @code{movups} machine instruction as a load from memory.
8821 @item void __builtin_ia32_storeups (float *, v4sf)
8822 Generates the @code{movups} machine instruction as a store to memory.
8823 @item v4sf __builtin_ia32_loadsss (float *)
8824 Generates the @code{movss} machine instruction as a load from memory.
8825 @item void __builtin_ia32_storess (float *, v4sf)
8826 Generates the @code{movss} machine instruction as a store to memory.
8827 @item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)
8828 Generates the @code{movhps} machine instruction as a load from memory.
8829 @item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)
8830 Generates the @code{movlps} machine instruction as a load from memory
8831 @item void __builtin_ia32_storehps (v2sf *, v4sf)
8832 Generates the @code{movhps} machine instruction as a store to memory.
8833 @item void __builtin_ia32_storelps (v2sf *, v4sf)
8834 Generates the @code{movlps} machine instruction as a store to memory.
8837 The following built-in functions are available when @option{-msse2} is used.
8838 All of them generate the machine instruction that is part of the name.
8841 int __builtin_ia32_comisdeq (v2df, v2df)
8842 int __builtin_ia32_comisdlt (v2df, v2df)
8843 int __builtin_ia32_comisdle (v2df, v2df)
8844 int __builtin_ia32_comisdgt (v2df, v2df)
8845 int __builtin_ia32_comisdge (v2df, v2df)
8846 int __builtin_ia32_comisdneq (v2df, v2df)
8847 int __builtin_ia32_ucomisdeq (v2df, v2df)
8848 int __builtin_ia32_ucomisdlt (v2df, v2df)
8849 int __builtin_ia32_ucomisdle (v2df, v2df)
8850 int __builtin_ia32_ucomisdgt (v2df, v2df)
8851 int __builtin_ia32_ucomisdge (v2df, v2df)
8852 int __builtin_ia32_ucomisdneq (v2df, v2df)
8853 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
8854 v2df __builtin_ia32_cmpltpd (v2df, v2df)
8855 v2df __builtin_ia32_cmplepd (v2df, v2df)
8856 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
8857 v2df __builtin_ia32_cmpgepd (v2df, v2df)
8858 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
8859 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
8860 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
8861 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
8862 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
8863 v2df __builtin_ia32_cmpngepd (v2df, v2df)
8864 v2df __builtin_ia32_cmpordpd (v2df, v2df)
8865 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
8866 v2df __builtin_ia32_cmpltsd (v2df, v2df)
8867 v2df __builtin_ia32_cmplesd (v2df, v2df)
8868 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
8869 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
8870 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
8871 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
8872 v2df __builtin_ia32_cmpordsd (v2df, v2df)
8873 v2di __builtin_ia32_paddq (v2di, v2di)
8874 v2di __builtin_ia32_psubq (v2di, v2di)
8875 v2df __builtin_ia32_addpd (v2df, v2df)
8876 v2df __builtin_ia32_subpd (v2df, v2df)
8877 v2df __builtin_ia32_mulpd (v2df, v2df)
8878 v2df __builtin_ia32_divpd (v2df, v2df)
8879 v2df __builtin_ia32_addsd (v2df, v2df)
8880 v2df __builtin_ia32_subsd (v2df, v2df)
8881 v2df __builtin_ia32_mulsd (v2df, v2df)
8882 v2df __builtin_ia32_divsd (v2df, v2df)
8883 v2df __builtin_ia32_minpd (v2df, v2df)
8884 v2df __builtin_ia32_maxpd (v2df, v2df)
8885 v2df __builtin_ia32_minsd (v2df, v2df)
8886 v2df __builtin_ia32_maxsd (v2df, v2df)
8887 v2df __builtin_ia32_andpd (v2df, v2df)
8888 v2df __builtin_ia32_andnpd (v2df, v2df)
8889 v2df __builtin_ia32_orpd (v2df, v2df)
8890 v2df __builtin_ia32_xorpd (v2df, v2df)
8891 v2df __builtin_ia32_movsd (v2df, v2df)
8892 v2df __builtin_ia32_unpckhpd (v2df, v2df)
8893 v2df __builtin_ia32_unpcklpd (v2df, v2df)
8894 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
8895 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
8896 v4si __builtin_ia32_paddd128 (v4si, v4si)
8897 v2di __builtin_ia32_paddq128 (v2di, v2di)
8898 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
8899 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
8900 v4si __builtin_ia32_psubd128 (v4si, v4si)
8901 v2di __builtin_ia32_psubq128 (v2di, v2di)
8902 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
8903 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
8904 v2di __builtin_ia32_pand128 (v2di, v2di)
8905 v2di __builtin_ia32_pandn128 (v2di, v2di)
8906 v2di __builtin_ia32_por128 (v2di, v2di)
8907 v2di __builtin_ia32_pxor128 (v2di, v2di)
8908 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
8909 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
8910 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
8911 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
8912 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
8913 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
8914 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
8915 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
8916 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
8917 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
8918 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
8919 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
8920 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
8921 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
8922 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
8923 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
8924 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
8925 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
8926 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
8927 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
8928 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi)
8929 v8hi __builtin_ia32_packssdw128 (v4si, v4si)
8930 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi)
8931 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
8932 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
8933 v2df __builtin_ia32_loadupd (double *)
8934 void __builtin_ia32_storeupd (double *, v2df)
8935 v2df __builtin_ia32_loadhpd (v2df, double const *)
8936 v2df __builtin_ia32_loadlpd (v2df, double const *)
8937 int __builtin_ia32_movmskpd (v2df)
8938 int __builtin_ia32_pmovmskb128 (v16qi)
8939 void __builtin_ia32_movnti (int *, int)
8940 void __builtin_ia32_movntpd (double *, v2df)
8941 void __builtin_ia32_movntdq (v2df *, v2df)
8942 v4si __builtin_ia32_pshufd (v4si, int)
8943 v8hi __builtin_ia32_pshuflw (v8hi, int)
8944 v8hi __builtin_ia32_pshufhw (v8hi, int)
8945 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
8946 v2df __builtin_ia32_sqrtpd (v2df)
8947 v2df __builtin_ia32_sqrtsd (v2df)
8948 v2df __builtin_ia32_shufpd (v2df, v2df, int)
8949 v2df __builtin_ia32_cvtdq2pd (v4si)
8950 v4sf __builtin_ia32_cvtdq2ps (v4si)
8951 v4si __builtin_ia32_cvtpd2dq (v2df)
8952 v2si __builtin_ia32_cvtpd2pi (v2df)
8953 v4sf __builtin_ia32_cvtpd2ps (v2df)
8954 v4si __builtin_ia32_cvttpd2dq (v2df)
8955 v2si __builtin_ia32_cvttpd2pi (v2df)
8956 v2df __builtin_ia32_cvtpi2pd (v2si)
8957 int __builtin_ia32_cvtsd2si (v2df)
8958 int __builtin_ia32_cvttsd2si (v2df)
8959 long long __builtin_ia32_cvtsd2si64 (v2df)
8960 long long __builtin_ia32_cvttsd2si64 (v2df)
8961 v4si __builtin_ia32_cvtps2dq (v4sf)
8962 v2df __builtin_ia32_cvtps2pd (v4sf)
8963 v4si __builtin_ia32_cvttps2dq (v4sf)
8964 v2df __builtin_ia32_cvtsi2sd (v2df, int)
8965 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
8966 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
8967 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
8968 void __builtin_ia32_clflush (const void *)
8969 void __builtin_ia32_lfence (void)
8970 void __builtin_ia32_mfence (void)
8971 v16qi __builtin_ia32_loaddqu (const char *)
8972 void __builtin_ia32_storedqu (char *, v16qi)
8973 v1di __builtin_ia32_pmuludq (v2si, v2si)
8974 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
8975 v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
8976 v4si __builtin_ia32_pslld128 (v4si, v4si)
8977 v2di __builtin_ia32_psllq128 (v2di, v2di)
8978 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
8979 v4si __builtin_ia32_psrld128 (v4si, v4si)
8980 v2di __builtin_ia32_psrlq128 (v2di, v2di)
8981 v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
8982 v4si __builtin_ia32_psrad128 (v4si, v4si)
8983 v2di __builtin_ia32_pslldqi128 (v2di, int)
8984 v8hi __builtin_ia32_psllwi128 (v8hi, int)
8985 v4si __builtin_ia32_pslldi128 (v4si, int)
8986 v2di __builtin_ia32_psllqi128 (v2di, int)
8987 v2di __builtin_ia32_psrldqi128 (v2di, int)
8988 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
8989 v4si __builtin_ia32_psrldi128 (v4si, int)
8990 v2di __builtin_ia32_psrlqi128 (v2di, int)
8991 v8hi __builtin_ia32_psrawi128 (v8hi, int)
8992 v4si __builtin_ia32_psradi128 (v4si, int)
8993 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
8994 v2di __builtin_ia32_movq128 (v2di)
8997 The following built-in functions are available when @option{-msse3} is used.
8998 All of them generate the machine instruction that is part of the name.
9001 v2df __builtin_ia32_addsubpd (v2df, v2df)
9002 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
9003 v2df __builtin_ia32_haddpd (v2df, v2df)
9004 v4sf __builtin_ia32_haddps (v4sf, v4sf)
9005 v2df __builtin_ia32_hsubpd (v2df, v2df)
9006 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
9007 v16qi __builtin_ia32_lddqu (char const *)
9008 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
9009 v2df __builtin_ia32_movddup (v2df)
9010 v4sf __builtin_ia32_movshdup (v4sf)
9011 v4sf __builtin_ia32_movsldup (v4sf)
9012 void __builtin_ia32_mwait (unsigned int, unsigned int)
9015 The following built-in functions are available when @option{-msse3} is used.
9018 @item v2df __builtin_ia32_loadddup (double const *)
9019 Generates the @code{movddup} machine instruction as a load from memory.
9022 The following built-in functions are available when @option{-mssse3} is used.
9023 All of them generate the machine instruction that is part of the name
9027 v2si __builtin_ia32_phaddd (v2si, v2si)
9028 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
9029 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
9030 v2si __builtin_ia32_phsubd (v2si, v2si)
9031 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
9032 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
9033 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi)
9034 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
9035 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
9036 v8qi __builtin_ia32_psignb (v8qi, v8qi)
9037 v2si __builtin_ia32_psignd (v2si, v2si)
9038 v4hi __builtin_ia32_psignw (v4hi, v4hi)
9039 v1di __builtin_ia32_palignr (v1di, v1di, int)
9040 v8qi __builtin_ia32_pabsb (v8qi)
9041 v2si __builtin_ia32_pabsd (v2si)
9042 v4hi __builtin_ia32_pabsw (v4hi)
9045 The following built-in functions are available when @option{-mssse3} is used.
9046 All of them generate the machine instruction that is part of the name
9050 v4si __builtin_ia32_phaddd128 (v4si, v4si)
9051 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
9052 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
9053 v4si __builtin_ia32_phsubd128 (v4si, v4si)
9054 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
9055 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
9056 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
9057 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
9058 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
9059 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
9060 v4si __builtin_ia32_psignd128 (v4si, v4si)
9061 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
9062 v2di __builtin_ia32_palignr128 (v2di, v2di, int)
9063 v16qi __builtin_ia32_pabsb128 (v16qi)
9064 v4si __builtin_ia32_pabsd128 (v4si)
9065 v8hi __builtin_ia32_pabsw128 (v8hi)
9068 The following built-in functions are available when @option{-msse4.1} is
9069 used. All of them generate the machine instruction that is part of the
9073 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
9074 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
9075 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
9076 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
9077 v2df __builtin_ia32_dppd (v2df, v2df, const int)
9078 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
9079 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
9080 v2di __builtin_ia32_movntdqa (v2di *);
9081 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
9082 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
9083 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
9084 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
9085 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
9086 v8hi __builtin_ia32_phminposuw128 (v8hi)
9087 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
9088 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
9089 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
9090 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
9091 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
9092 v4si __builtin_ia32_pminsd128 (v4si, v4si)
9093 v4si __builtin_ia32_pminud128 (v4si, v4si)
9094 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
9095 v4si __builtin_ia32_pmovsxbd128 (v16qi)
9096 v2di __builtin_ia32_pmovsxbq128 (v16qi)
9097 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
9098 v2di __builtin_ia32_pmovsxdq128 (v4si)
9099 v4si __builtin_ia32_pmovsxwd128 (v8hi)
9100 v2di __builtin_ia32_pmovsxwq128 (v8hi)
9101 v4si __builtin_ia32_pmovzxbd128 (v16qi)
9102 v2di __builtin_ia32_pmovzxbq128 (v16qi)
9103 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
9104 v2di __builtin_ia32_pmovzxdq128 (v4si)
9105 v4si __builtin_ia32_pmovzxwd128 (v8hi)
9106 v2di __builtin_ia32_pmovzxwq128 (v8hi)
9107 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
9108 v4si __builtin_ia32_pmulld128 (v4si, v4si)
9109 int __builtin_ia32_ptestc128 (v2di, v2di)
9110 int __builtin_ia32_ptestnzc128 (v2di, v2di)
9111 int __builtin_ia32_ptestz128 (v2di, v2di)
9112 v2df __builtin_ia32_roundpd (v2df, const int)
9113 v4sf __builtin_ia32_roundps (v4sf, const int)
9114 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
9115 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
9118 The following built-in functions are available when @option{-msse4.1} is
9122 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
9123 Generates the @code{insertps} machine instruction.
9124 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
9125 Generates the @code{pextrb} machine instruction.
9126 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
9127 Generates the @code{pinsrb} machine instruction.
9128 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
9129 Generates the @code{pinsrd} machine instruction.
9130 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
9131 Generates the @code{pinsrq} machine instruction in 64bit mode.
9134 The following built-in functions are changed to generate new SSE4.1
9135 instructions when @option{-msse4.1} is used.
9138 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
9139 Generates the @code{extractps} machine instruction.
9140 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
9141 Generates the @code{pextrd} machine instruction.
9142 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
9143 Generates the @code{pextrq} machine instruction in 64bit mode.
9146 The following built-in functions are available when @option{-msse4.2} is
9147 used. All of them generate the machine instruction that is part of the
9151 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
9152 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
9153 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
9154 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
9155 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
9156 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
9157 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
9158 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
9159 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
9160 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
9161 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
9162 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
9163 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
9164 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
9165 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
9168 The following built-in functions are available when @option{-msse4.2} is
9172 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
9173 Generates the @code{crc32b} machine instruction.
9174 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
9175 Generates the @code{crc32w} machine instruction.
9176 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
9177 Generates the @code{crc32l} machine instruction.
9178 @item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)
9179 Generates the @code{crc32q} machine instruction.
9182 The following built-in functions are changed to generate new SSE4.2
9183 instructions when @option{-msse4.2} is used.
9186 @item int __builtin_popcount (unsigned int)
9187 Generates the @code{popcntl} machine instruction.
9188 @item int __builtin_popcountl (unsigned long)
9189 Generates the @code{popcntl} or @code{popcntq} machine instruction,
9190 depending on the size of @code{unsigned long}.
9191 @item int __builtin_popcountll (unsigned long long)
9192 Generates the @code{popcntq} machine instruction.
9195 The following built-in functions are available when @option{-mavx} is
9196 used. All of them generate the machine instruction that is part of the
9200 v4df __builtin_ia32_addpd256 (v4df,v4df)
9201 v8sf __builtin_ia32_addps256 (v8sf,v8sf)
9202 v4df __builtin_ia32_addsubpd256 (v4df,v4df)
9203 v8sf __builtin_ia32_addsubps256 (v8sf,v8sf)
9204 v4df __builtin_ia32_andnpd256 (v4df,v4df)
9205 v8sf __builtin_ia32_andnps256 (v8sf,v8sf)
9206 v4df __builtin_ia32_andpd256 (v4df,v4df)
9207 v8sf __builtin_ia32_andps256 (v8sf,v8sf)
9208 v4df __builtin_ia32_blendpd256 (v4df,v4df,int)
9209 v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int)
9210 v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df)
9211 v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf)
9212 v2df __builtin_ia32_cmppd (v2df,v2df,int)
9213 v4df __builtin_ia32_cmppd256 (v4df,v4df,int)
9214 v4sf __builtin_ia32_cmpps (v4sf,v4sf,int)
9215 v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int)
9216 v2df __builtin_ia32_cmpsd (v2df,v2df,int)
9217 v4sf __builtin_ia32_cmpss (v4sf,v4sf,int)
9218 v4df __builtin_ia32_cvtdq2pd256 (v4si)
9219 v8sf __builtin_ia32_cvtdq2ps256 (v8si)
9220 v4si __builtin_ia32_cvtpd2dq256 (v4df)
9221 v4sf __builtin_ia32_cvtpd2ps256 (v4df)
9222 v8si __builtin_ia32_cvtps2dq256 (v8sf)
9223 v4df __builtin_ia32_cvtps2pd256 (v4sf)
9224 v4si __builtin_ia32_cvttpd2dq256 (v4df)
9225 v8si __builtin_ia32_cvttps2dq256 (v8sf)
9226 v4df __builtin_ia32_divpd256 (v4df,v4df)
9227 v8sf __builtin_ia32_divps256 (v8sf,v8sf)
9228 v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int)
9229 v4df __builtin_ia32_haddpd256 (v4df,v4df)
9230 v8sf __builtin_ia32_haddps256 (v8sf,v8sf)
9231 v4df __builtin_ia32_hsubpd256 (v4df,v4df)
9232 v8sf __builtin_ia32_hsubps256 (v8sf,v8sf)
9233 v32qi __builtin_ia32_lddqu256 (pcchar)
9234 v32qi __builtin_ia32_loaddqu256 (pcchar)
9235 v4df __builtin_ia32_loadupd256 (pcdouble)
9236 v8sf __builtin_ia32_loadups256 (pcfloat)
9237 v2df __builtin_ia32_maskloadpd (pcv2df,v2df)
9238 v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df)
9239 v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf)
9240 v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf)
9241 void __builtin_ia32_maskstorepd (pv2df,v2df,v2df)
9242 void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df)
9243 void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf)
9244 void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf)
9245 v4df __builtin_ia32_maxpd256 (v4df,v4df)
9246 v8sf __builtin_ia32_maxps256 (v8sf,v8sf)
9247 v4df __builtin_ia32_minpd256 (v4df,v4df)
9248 v8sf __builtin_ia32_minps256 (v8sf,v8sf)
9249 v4df __builtin_ia32_movddup256 (v4df)
9250 int __builtin_ia32_movmskpd256 (v4df)
9251 int __builtin_ia32_movmskps256 (v8sf)
9252 v8sf __builtin_ia32_movshdup256 (v8sf)
9253 v8sf __builtin_ia32_movsldup256 (v8sf)
9254 v4df __builtin_ia32_mulpd256 (v4df,v4df)
9255 v8sf __builtin_ia32_mulps256 (v8sf,v8sf)
9256 v4df __builtin_ia32_orpd256 (v4df,v4df)
9257 v8sf __builtin_ia32_orps256 (v8sf,v8sf)
9258 v2df __builtin_ia32_pd_pd256 (v4df)
9259 v4df __builtin_ia32_pd256_pd (v2df)
9260 v4sf __builtin_ia32_ps_ps256 (v8sf)
9261 v8sf __builtin_ia32_ps256_ps (v4sf)
9262 int __builtin_ia32_ptestc256 (v4di,v4di,ptest)
9263 int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest)
9264 int __builtin_ia32_ptestz256 (v4di,v4di,ptest)
9265 v8sf __builtin_ia32_rcpps256 (v8sf)
9266 v4df __builtin_ia32_roundpd256 (v4df,int)
9267 v8sf __builtin_ia32_roundps256 (v8sf,int)
9268 v8sf __builtin_ia32_rsqrtps_nr256 (v8sf)
9269 v8sf __builtin_ia32_rsqrtps256 (v8sf)
9270 v4df __builtin_ia32_shufpd256 (v4df,v4df,int)
9271 v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int)
9272 v4si __builtin_ia32_si_si256 (v8si)
9273 v8si __builtin_ia32_si256_si (v4si)
9274 v4df __builtin_ia32_sqrtpd256 (v4df)
9275 v8sf __builtin_ia32_sqrtps_nr256 (v8sf)
9276 v8sf __builtin_ia32_sqrtps256 (v8sf)
9277 void __builtin_ia32_storedqu256 (pchar,v32qi)
9278 void __builtin_ia32_storeupd256 (pdouble,v4df)
9279 void __builtin_ia32_storeups256 (pfloat,v8sf)
9280 v4df __builtin_ia32_subpd256 (v4df,v4df)
9281 v8sf __builtin_ia32_subps256 (v8sf,v8sf)
9282 v4df __builtin_ia32_unpckhpd256 (v4df,v4df)
9283 v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf)
9284 v4df __builtin_ia32_unpcklpd256 (v4df,v4df)
9285 v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf)
9286 v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df)
9287 v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf)
9288 v4df __builtin_ia32_vbroadcastsd256 (pcdouble)
9289 v4sf __builtin_ia32_vbroadcastss (pcfloat)
9290 v8sf __builtin_ia32_vbroadcastss256 (pcfloat)
9291 v2df __builtin_ia32_vextractf128_pd256 (v4df,int)
9292 v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int)
9293 v4si __builtin_ia32_vextractf128_si256 (v8si,int)
9294 v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int)
9295 v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int)
9296 v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int)
9297 v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int)
9298 v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int)
9299 v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int)
9300 v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int)
9301 v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int)
9302 v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int)
9303 v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int)
9304 v2df __builtin_ia32_vpermilpd (v2df,int)
9305 v4df __builtin_ia32_vpermilpd256 (v4df,int)
9306 v4sf __builtin_ia32_vpermilps (v4sf,int)
9307 v8sf __builtin_ia32_vpermilps256 (v8sf,int)
9308 v2df __builtin_ia32_vpermilvarpd (v2df,v2di)
9309 v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di)
9310 v4sf __builtin_ia32_vpermilvarps (v4sf,v4si)
9311 v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si)
9312 int __builtin_ia32_vtestcpd (v2df,v2df,ptest)
9313 int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest)
9314 int __builtin_ia32_vtestcps (v4sf,v4sf,ptest)
9315 int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest)
9316 int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest)
9317 int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest)
9318 int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest)
9319 int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest)
9320 int __builtin_ia32_vtestzpd (v2df,v2df,ptest)
9321 int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest)
9322 int __builtin_ia32_vtestzps (v4sf,v4sf,ptest)
9323 int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest)
9324 void __builtin_ia32_vzeroall (void)
9325 void __builtin_ia32_vzeroupper (void)
9326 v4df __builtin_ia32_xorpd256 (v4df,v4df)
9327 v8sf __builtin_ia32_xorps256 (v8sf,v8sf)
9330 The following built-in functions are available when @option{-maes} is
9331 used. All of them generate the machine instruction that is part of the
9335 v2di __builtin_ia32_aesenc128 (v2di, v2di)
9336 v2di __builtin_ia32_aesenclast128 (v2di, v2di)
9337 v2di __builtin_ia32_aesdec128 (v2di, v2di)
9338 v2di __builtin_ia32_aesdeclast128 (v2di, v2di)
9339 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int)
9340 v2di __builtin_ia32_aesimc128 (v2di)
9343 The following built-in function is available when @option{-mpclmul} is
9347 @item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)
9348 Generates the @code{pclmulqdq} machine instruction.
9351 The following built-in function is available when @option{-mfsgsbase} is
9352 used. All of them generate the machine instruction that is part of the
9356 unsigned int __builtin_ia32_rdfsbase32 (void)
9357 unsigned long long __builtin_ia32_rdfsbase64 (void)
9358 unsigned int __builtin_ia32_rdgsbase32 (void)
9359 unsigned long long __builtin_ia32_rdgsbase64 (void)
9360 void _writefsbase_u32 (unsigned int)
9361 void _writefsbase_u64 (unsigned long long)
9362 void _writegsbase_u32 (unsigned int)
9363 void _writegsbase_u64 (unsigned long long)
9366 The following built-in function is available when @option{-mrdrnd} is
9367 used. All of them generate the machine instruction that is part of the
9371 unsigned short __builtin_ia32_rdrand16 (void)
9372 unsigned int __builtin_ia32_rdrand32 (void)
9373 unsigned long long __builtin_ia32_rdrand64 (void)
9376 The following built-in functions are available when @option{-msse4a} is used.
9377 All of them generate the machine instruction that is part of the name.
9380 void __builtin_ia32_movntsd (double *, v2df)
9381 void __builtin_ia32_movntss (float *, v4sf)
9382 v2di __builtin_ia32_extrq (v2di, v16qi)
9383 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
9384 v2di __builtin_ia32_insertq (v2di, v2di)
9385 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
9388 The following built-in functions are available when @option{-mxop} is used.
9390 v2df __builtin_ia32_vfrczpd (v2df)
9391 v4sf __builtin_ia32_vfrczps (v4sf)
9392 v2df __builtin_ia32_vfrczsd (v2df, v2df)
9393 v4sf __builtin_ia32_vfrczss (v4sf, v4sf)
9394 v4df __builtin_ia32_vfrczpd256 (v4df)
9395 v8sf __builtin_ia32_vfrczps256 (v8sf)
9396 v2di __builtin_ia32_vpcmov (v2di, v2di, v2di)
9397 v2di __builtin_ia32_vpcmov_v2di (v2di, v2di, v2di)
9398 v4si __builtin_ia32_vpcmov_v4si (v4si, v4si, v4si)
9399 v8hi __builtin_ia32_vpcmov_v8hi (v8hi, v8hi, v8hi)
9400 v16qi __builtin_ia32_vpcmov_v16qi (v16qi, v16qi, v16qi)
9401 v2df __builtin_ia32_vpcmov_v2df (v2df, v2df, v2df)
9402 v4sf __builtin_ia32_vpcmov_v4sf (v4sf, v4sf, v4sf)
9403 v4di __builtin_ia32_vpcmov_v4di256 (v4di, v4di, v4di)
9404 v8si __builtin_ia32_vpcmov_v8si256 (v8si, v8si, v8si)
9405 v16hi __builtin_ia32_vpcmov_v16hi256 (v16hi, v16hi, v16hi)
9406 v32qi __builtin_ia32_vpcmov_v32qi256 (v32qi, v32qi, v32qi)
9407 v4df __builtin_ia32_vpcmov_v4df256 (v4df, v4df, v4df)
9408 v8sf __builtin_ia32_vpcmov_v8sf256 (v8sf, v8sf, v8sf)
9409 v16qi __builtin_ia32_vpcomeqb (v16qi, v16qi)
9410 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
9411 v4si __builtin_ia32_vpcomeqd (v4si, v4si)
9412 v2di __builtin_ia32_vpcomeqq (v2di, v2di)
9413 v16qi __builtin_ia32_vpcomequb (v16qi, v16qi)
9414 v4si __builtin_ia32_vpcomequd (v4si, v4si)
9415 v2di __builtin_ia32_vpcomequq (v2di, v2di)
9416 v8hi __builtin_ia32_vpcomequw (v8hi, v8hi)
9417 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
9418 v16qi __builtin_ia32_vpcomfalseb (v16qi, v16qi)
9419 v4si __builtin_ia32_vpcomfalsed (v4si, v4si)
9420 v2di __builtin_ia32_vpcomfalseq (v2di, v2di)
9421 v16qi __builtin_ia32_vpcomfalseub (v16qi, v16qi)
9422 v4si __builtin_ia32_vpcomfalseud (v4si, v4si)
9423 v2di __builtin_ia32_vpcomfalseuq (v2di, v2di)
9424 v8hi __builtin_ia32_vpcomfalseuw (v8hi, v8hi)
9425 v8hi __builtin_ia32_vpcomfalsew (v8hi, v8hi)
9426 v16qi __builtin_ia32_vpcomgeb (v16qi, v16qi)
9427 v4si __builtin_ia32_vpcomged (v4si, v4si)
9428 v2di __builtin_ia32_vpcomgeq (v2di, v2di)
9429 v16qi __builtin_ia32_vpcomgeub (v16qi, v16qi)
9430 v4si __builtin_ia32_vpcomgeud (v4si, v4si)
9431 v2di __builtin_ia32_vpcomgeuq (v2di, v2di)
9432 v8hi __builtin_ia32_vpcomgeuw (v8hi, v8hi)
9433 v8hi __builtin_ia32_vpcomgew (v8hi, v8hi)
9434 v16qi __builtin_ia32_vpcomgtb (v16qi, v16qi)
9435 v4si __builtin_ia32_vpcomgtd (v4si, v4si)
9436 v2di __builtin_ia32_vpcomgtq (v2di, v2di)
9437 v16qi __builtin_ia32_vpcomgtub (v16qi, v16qi)
9438 v4si __builtin_ia32_vpcomgtud (v4si, v4si)
9439 v2di __builtin_ia32_vpcomgtuq (v2di, v2di)
9440 v8hi __builtin_ia32_vpcomgtuw (v8hi, v8hi)
9441 v8hi __builtin_ia32_vpcomgtw (v8hi, v8hi)
9442 v16qi __builtin_ia32_vpcomleb (v16qi, v16qi)
9443 v4si __builtin_ia32_vpcomled (v4si, v4si)
9444 v2di __builtin_ia32_vpcomleq (v2di, v2di)
9445 v16qi __builtin_ia32_vpcomleub (v16qi, v16qi)
9446 v4si __builtin_ia32_vpcomleud (v4si, v4si)
9447 v2di __builtin_ia32_vpcomleuq (v2di, v2di)
9448 v8hi __builtin_ia32_vpcomleuw (v8hi, v8hi)
9449 v8hi __builtin_ia32_vpcomlew (v8hi, v8hi)
9450 v16qi __builtin_ia32_vpcomltb (v16qi, v16qi)
9451 v4si __builtin_ia32_vpcomltd (v4si, v4si)
9452 v2di __builtin_ia32_vpcomltq (v2di, v2di)
9453 v16qi __builtin_ia32_vpcomltub (v16qi, v16qi)
9454 v4si __builtin_ia32_vpcomltud (v4si, v4si)
9455 v2di __builtin_ia32_vpcomltuq (v2di, v2di)
9456 v8hi __builtin_ia32_vpcomltuw (v8hi, v8hi)
9457 v8hi __builtin_ia32_vpcomltw (v8hi, v8hi)
9458 v16qi __builtin_ia32_vpcomneb (v16qi, v16qi)
9459 v4si __builtin_ia32_vpcomned (v4si, v4si)
9460 v2di __builtin_ia32_vpcomneq (v2di, v2di)
9461 v16qi __builtin_ia32_vpcomneub (v16qi, v16qi)
9462 v4si __builtin_ia32_vpcomneud (v4si, v4si)
9463 v2di __builtin_ia32_vpcomneuq (v2di, v2di)
9464 v8hi __builtin_ia32_vpcomneuw (v8hi, v8hi)
9465 v8hi __builtin_ia32_vpcomnew (v8hi, v8hi)
9466 v16qi __builtin_ia32_vpcomtrueb (v16qi, v16qi)
9467 v4si __builtin_ia32_vpcomtrued (v4si, v4si)
9468 v2di __builtin_ia32_vpcomtrueq (v2di, v2di)
9469 v16qi __builtin_ia32_vpcomtrueub (v16qi, v16qi)
9470 v4si __builtin_ia32_vpcomtrueud (v4si, v4si)
9471 v2di __builtin_ia32_vpcomtrueuq (v2di, v2di)
9472 v8hi __builtin_ia32_vpcomtrueuw (v8hi, v8hi)
9473 v8hi __builtin_ia32_vpcomtruew (v8hi, v8hi)
9474 v4si __builtin_ia32_vphaddbd (v16qi)
9475 v2di __builtin_ia32_vphaddbq (v16qi)
9476 v8hi __builtin_ia32_vphaddbw (v16qi)
9477 v2di __builtin_ia32_vphadddq (v4si)
9478 v4si __builtin_ia32_vphaddubd (v16qi)
9479 v2di __builtin_ia32_vphaddubq (v16qi)
9480 v8hi __builtin_ia32_vphaddubw (v16qi)
9481 v2di __builtin_ia32_vphaddudq (v4si)
9482 v4si __builtin_ia32_vphadduwd (v8hi)
9483 v2di __builtin_ia32_vphadduwq (v8hi)
9484 v4si __builtin_ia32_vphaddwd (v8hi)
9485 v2di __builtin_ia32_vphaddwq (v8hi)
9486 v8hi __builtin_ia32_vphsubbw (v16qi)
9487 v2di __builtin_ia32_vphsubdq (v4si)
9488 v4si __builtin_ia32_vphsubwd (v8hi)
9489 v4si __builtin_ia32_vpmacsdd (v4si, v4si, v4si)
9490 v2di __builtin_ia32_vpmacsdqh (v4si, v4si, v2di)
9491 v2di __builtin_ia32_vpmacsdql (v4si, v4si, v2di)
9492 v4si __builtin_ia32_vpmacssdd (v4si, v4si, v4si)
9493 v2di __builtin_ia32_vpmacssdqh (v4si, v4si, v2di)
9494 v2di __builtin_ia32_vpmacssdql (v4si, v4si, v2di)
9495 v4si __builtin_ia32_vpmacsswd (v8hi, v8hi, v4si)
9496 v8hi __builtin_ia32_vpmacssww (v8hi, v8hi, v8hi)
9497 v4si __builtin_ia32_vpmacswd (v8hi, v8hi, v4si)
9498 v8hi __builtin_ia32_vpmacsww (v8hi, v8hi, v8hi)
9499 v4si __builtin_ia32_vpmadcsswd (v8hi, v8hi, v4si)
9500 v4si __builtin_ia32_vpmadcswd (v8hi, v8hi, v4si)
9501 v16qi __builtin_ia32_vpperm (v16qi, v16qi, v16qi)
9502 v16qi __builtin_ia32_vprotb (v16qi, v16qi)
9503 v4si __builtin_ia32_vprotd (v4si, v4si)
9504 v2di __builtin_ia32_vprotq (v2di, v2di)
9505 v8hi __builtin_ia32_vprotw (v8hi, v8hi)
9506 v16qi __builtin_ia32_vpshab (v16qi, v16qi)
9507 v4si __builtin_ia32_vpshad (v4si, v4si)
9508 v2di __builtin_ia32_vpshaq (v2di, v2di)
9509 v8hi __builtin_ia32_vpshaw (v8hi, v8hi)
9510 v16qi __builtin_ia32_vpshlb (v16qi, v16qi)
9511 v4si __builtin_ia32_vpshld (v4si, v4si)
9512 v2di __builtin_ia32_vpshlq (v2di, v2di)
9513 v8hi __builtin_ia32_vpshlw (v8hi, v8hi)
9516 The following built-in functions are available when @option{-mfma4} is used.
9517 All of them generate the machine instruction that is part of the name
9521 v2df __builtin_ia32_fmaddpd (v2df, v2df, v2df)
9522 v4sf __builtin_ia32_fmaddps (v4sf, v4sf, v4sf)
9523 v2df __builtin_ia32_fmaddsd (v2df, v2df, v2df)
9524 v4sf __builtin_ia32_fmaddss (v4sf, v4sf, v4sf)
9525 v2df __builtin_ia32_fmsubpd (v2df, v2df, v2df)
9526 v4sf __builtin_ia32_fmsubps (v4sf, v4sf, v4sf)
9527 v2df __builtin_ia32_fmsubsd (v2df, v2df, v2df)
9528 v4sf __builtin_ia32_fmsubss (v4sf, v4sf, v4sf)
9529 v2df __builtin_ia32_fnmaddpd (v2df, v2df, v2df)
9530 v4sf __builtin_ia32_fnmaddps (v4sf, v4sf, v4sf)
9531 v2df __builtin_ia32_fnmaddsd (v2df, v2df, v2df)
9532 v4sf __builtin_ia32_fnmaddss (v4sf, v4sf, v4sf)
9533 v2df __builtin_ia32_fnmsubpd (v2df, v2df, v2df)
9534 v4sf __builtin_ia32_fnmsubps (v4sf, v4sf, v4sf)
9535 v2df __builtin_ia32_fnmsubsd (v2df, v2df, v2df)
9536 v4sf __builtin_ia32_fnmsubss (v4sf, v4sf, v4sf)
9537 v2df __builtin_ia32_fmaddsubpd (v2df, v2df, v2df)
9538 v4sf __builtin_ia32_fmaddsubps (v4sf, v4sf, v4sf)
9539 v2df __builtin_ia32_fmsubaddpd (v2df, v2df, v2df)
9540 v4sf __builtin_ia32_fmsubaddps (v4sf, v4sf, v4sf)
9541 v4df __builtin_ia32_fmaddpd256 (v4df, v4df, v4df)
9542 v8sf __builtin_ia32_fmaddps256 (v8sf, v8sf, v8sf)
9543 v4df __builtin_ia32_fmsubpd256 (v4df, v4df, v4df)
9544 v8sf __builtin_ia32_fmsubps256 (v8sf, v8sf, v8sf)
9545 v4df __builtin_ia32_fnmaddpd256 (v4df, v4df, v4df)
9546 v8sf __builtin_ia32_fnmaddps256 (v8sf, v8sf, v8sf)
9547 v4df __builtin_ia32_fnmsubpd256 (v4df, v4df, v4df)
9548 v8sf __builtin_ia32_fnmsubps256 (v8sf, v8sf, v8sf)
9549 v4df __builtin_ia32_fmaddsubpd256 (v4df, v4df, v4df)
9550 v8sf __builtin_ia32_fmaddsubps256 (v8sf, v8sf, v8sf)
9551 v4df __builtin_ia32_fmsubaddpd256 (v4df, v4df, v4df)
9552 v8sf __builtin_ia32_fmsubaddps256 (v8sf, v8sf, v8sf)
9556 The following built-in functions are available when @option{-mlwp} is used.
9559 void __builtin_ia32_llwpcb16 (void *);
9560 void __builtin_ia32_llwpcb32 (void *);
9561 void __builtin_ia32_llwpcb64 (void *);
9562 void * __builtin_ia32_llwpcb16 (void);
9563 void * __builtin_ia32_llwpcb32 (void);
9564 void * __builtin_ia32_llwpcb64 (void);
9565 void __builtin_ia32_lwpval16 (unsigned short, unsigned int, unsigned short)
9566 void __builtin_ia32_lwpval32 (unsigned int, unsigned int, unsigned int)
9567 void __builtin_ia32_lwpval64 (unsigned __int64, unsigned int, unsigned int)
9568 unsigned char __builtin_ia32_lwpins16 (unsigned short, unsigned int, unsigned short)
9569 unsigned char __builtin_ia32_lwpins32 (unsigned int, unsigned int, unsigned int)
9570 unsigned char __builtin_ia32_lwpins64 (unsigned __int64, unsigned int, unsigned int)
9573 The following built-in functions are available when @option{-mbmi} is used.
9574 All of them generate the machine instruction that is part of the name.
9576 unsigned int __builtin_ia32_bextr_u32(unsigned int, unsigned int);
9577 unsigned long long __builtin_ia32_bextr_u64 (unsigned long long, unsigned long long);
9578 unsigned short __builtin_ia32_lzcnt_16(unsigned short);
9579 unsigned int __builtin_ia32_lzcnt_u32(unsigned int);
9580 unsigned long long __builtin_ia32_lzcnt_u64 (unsigned long long);
9583 The following built-in functions are available when @option{-mtbm} is used.
9584 Both of them generate the immediate form of the bextr machine instruction.
9586 unsigned int __builtin_ia32_bextri_u32 (unsigned int, const unsigned int);
9587 unsigned long long __builtin_ia32_bextri_u64 (unsigned long long, const unsigned long long);
9591 The following built-in functions are available when @option{-m3dnow} is used.
9592 All of them generate the machine instruction that is part of the name.
9595 void __builtin_ia32_femms (void)
9596 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
9597 v2si __builtin_ia32_pf2id (v2sf)
9598 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
9599 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
9600 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
9601 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
9602 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
9603 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
9604 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
9605 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
9606 v2sf __builtin_ia32_pfrcp (v2sf)
9607 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
9608 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
9609 v2sf __builtin_ia32_pfrsqrt (v2sf)
9610 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
9611 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
9612 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
9613 v2sf __builtin_ia32_pi2fd (v2si)
9614 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
9617 The following built-in functions are available when both @option{-m3dnow}
9618 and @option{-march=athlon} are used. All of them generate the machine
9619 instruction that is part of the name.
9622 v2si __builtin_ia32_pf2iw (v2sf)
9623 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
9624 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
9625 v2sf __builtin_ia32_pi2fw (v2si)
9626 v2sf __builtin_ia32_pswapdsf (v2sf)
9627 v2si __builtin_ia32_pswapdsi (v2si)
9630 @node MIPS DSP Built-in Functions
9631 @subsection MIPS DSP Built-in Functions
9633 The MIPS DSP Application-Specific Extension (ASE) includes new
9634 instructions that are designed to improve the performance of DSP and
9635 media applications. It provides instructions that operate on packed
9636 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
9638 GCC supports MIPS DSP operations using both the generic
9639 vector extensions (@pxref{Vector Extensions}) and a collection of
9640 MIPS-specific built-in functions. Both kinds of support are
9641 enabled by the @option{-mdsp} command-line option.
9643 Revision 2 of the ASE was introduced in the second half of 2006.
9644 This revision adds extra instructions to the original ASE, but is
9645 otherwise backwards-compatible with it. You can select revision 2
9646 using the command-line option @option{-mdspr2}; this option implies
9649 The SCOUNT and POS bits of the DSP control register are global. The
9650 WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and
9651 POS bits. During optimization, the compiler will not delete these
9652 instructions and it will not delete calls to functions containing
9655 At present, GCC only provides support for operations on 32-bit
9656 vectors. The vector type associated with 8-bit integer data is
9657 usually called @code{v4i8}, the vector type associated with Q7
9658 is usually called @code{v4q7}, the vector type associated with 16-bit
9659 integer data is usually called @code{v2i16}, and the vector type
9660 associated with Q15 is usually called @code{v2q15}. They can be
9661 defined in C as follows:
9664 typedef signed char v4i8 __attribute__ ((vector_size(4)));
9665 typedef signed char v4q7 __attribute__ ((vector_size(4)));
9666 typedef short v2i16 __attribute__ ((vector_size(4)));
9667 typedef short v2q15 __attribute__ ((vector_size(4)));
9670 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
9671 initialized in the same way as aggregates. For example:
9674 v4i8 a = @{1, 2, 3, 4@};
9676 b = (v4i8) @{5, 6, 7, 8@};
9678 v2q15 c = @{0x0fcb, 0x3a75@};
9680 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
9683 @emph{Note:} The CPU's endianness determines the order in which values
9684 are packed. On little-endian targets, the first value is the least
9685 significant and the last value is the most significant. The opposite
9686 order applies to big-endian targets. For example, the code above will
9687 set the lowest byte of @code{a} to @code{1} on little-endian targets
9688 and @code{4} on big-endian targets.
9690 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
9691 representation. As shown in this example, the integer representation
9692 of a Q7 value can be obtained by multiplying the fractional value by
9693 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
9694 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
9697 The table below lists the @code{v4i8} and @code{v2q15} operations for which
9698 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
9699 and @code{c} and @code{d} are @code{v2q15} values.
9701 @multitable @columnfractions .50 .50
9702 @item C code @tab MIPS instruction
9703 @item @code{a + b} @tab @code{addu.qb}
9704 @item @code{c + d} @tab @code{addq.ph}
9705 @item @code{a - b} @tab @code{subu.qb}
9706 @item @code{c - d} @tab @code{subq.ph}
9709 The table below lists the @code{v2i16} operation for which
9710 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
9711 @code{v2i16} values.
9713 @multitable @columnfractions .50 .50
9714 @item C code @tab MIPS instruction
9715 @item @code{e * f} @tab @code{mul.ph}
9718 It is easier to describe the DSP built-in functions if we first define
9719 the following types:
9724 typedef unsigned int ui32;
9725 typedef long long a64;
9728 @code{q31} and @code{i32} are actually the same as @code{int}, but we
9729 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
9730 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
9731 @code{long long}, but we use @code{a64} to indicate values that will
9732 be placed in one of the four DSP accumulators (@code{$ac0},
9733 @code{$ac1}, @code{$ac2} or @code{$ac3}).
9735 Also, some built-in functions prefer or require immediate numbers as
9736 parameters, because the corresponding DSP instructions accept both immediate
9737 numbers and register operands, or accept immediate numbers only. The
9738 immediate parameters are listed as follows.
9747 imm_n32_31: -32 to 31.
9748 imm_n512_511: -512 to 511.
9751 The following built-in functions map directly to a particular MIPS DSP
9752 instruction. Please refer to the architecture specification
9753 for details on what each instruction does.
9756 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
9757 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
9758 q31 __builtin_mips_addq_s_w (q31, q31)
9759 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
9760 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
9761 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
9762 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
9763 q31 __builtin_mips_subq_s_w (q31, q31)
9764 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
9765 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
9766 i32 __builtin_mips_addsc (i32, i32)
9767 i32 __builtin_mips_addwc (i32, i32)
9768 i32 __builtin_mips_modsub (i32, i32)
9769 i32 __builtin_mips_raddu_w_qb (v4i8)
9770 v2q15 __builtin_mips_absq_s_ph (v2q15)
9771 q31 __builtin_mips_absq_s_w (q31)
9772 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
9773 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
9774 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
9775 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
9776 q31 __builtin_mips_preceq_w_phl (v2q15)
9777 q31 __builtin_mips_preceq_w_phr (v2q15)
9778 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
9779 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
9780 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
9781 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
9782 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
9783 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
9784 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
9785 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
9786 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
9787 v4i8 __builtin_mips_shll_qb (v4i8, i32)
9788 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
9789 v2q15 __builtin_mips_shll_ph (v2q15, i32)
9790 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
9791 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
9792 q31 __builtin_mips_shll_s_w (q31, imm0_31)
9793 q31 __builtin_mips_shll_s_w (q31, i32)
9794 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
9795 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
9796 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
9797 v2q15 __builtin_mips_shra_ph (v2q15, i32)
9798 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
9799 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
9800 q31 __builtin_mips_shra_r_w (q31, imm0_31)
9801 q31 __builtin_mips_shra_r_w (q31, i32)
9802 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
9803 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
9804 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
9805 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
9806 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
9807 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
9808 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
9809 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
9810 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
9811 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
9812 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
9813 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
9814 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
9815 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
9816 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
9817 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
9818 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
9819 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
9820 i32 __builtin_mips_bitrev (i32)
9821 i32 __builtin_mips_insv (i32, i32)
9822 v4i8 __builtin_mips_repl_qb (imm0_255)
9823 v4i8 __builtin_mips_repl_qb (i32)
9824 v2q15 __builtin_mips_repl_ph (imm_n512_511)
9825 v2q15 __builtin_mips_repl_ph (i32)
9826 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
9827 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
9828 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
9829 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
9830 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
9831 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
9832 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
9833 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
9834 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
9835 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
9836 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
9837 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
9838 i32 __builtin_mips_extr_w (a64, imm0_31)
9839 i32 __builtin_mips_extr_w (a64, i32)
9840 i32 __builtin_mips_extr_r_w (a64, imm0_31)
9841 i32 __builtin_mips_extr_s_h (a64, i32)
9842 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
9843 i32 __builtin_mips_extr_rs_w (a64, i32)
9844 i32 __builtin_mips_extr_s_h (a64, imm0_31)
9845 i32 __builtin_mips_extr_r_w (a64, i32)
9846 i32 __builtin_mips_extp (a64, imm0_31)
9847 i32 __builtin_mips_extp (a64, i32)
9848 i32 __builtin_mips_extpdp (a64, imm0_31)
9849 i32 __builtin_mips_extpdp (a64, i32)
9850 a64 __builtin_mips_shilo (a64, imm_n32_31)
9851 a64 __builtin_mips_shilo (a64, i32)
9852 a64 __builtin_mips_mthlip (a64, i32)
9853 void __builtin_mips_wrdsp (i32, imm0_63)
9854 i32 __builtin_mips_rddsp (imm0_63)
9855 i32 __builtin_mips_lbux (void *, i32)
9856 i32 __builtin_mips_lhx (void *, i32)
9857 i32 __builtin_mips_lwx (void *, i32)
9858 i32 __builtin_mips_bposge32 (void)
9859 a64 __builtin_mips_madd (a64, i32, i32);
9860 a64 __builtin_mips_maddu (a64, ui32, ui32);
9861 a64 __builtin_mips_msub (a64, i32, i32);
9862 a64 __builtin_mips_msubu (a64, ui32, ui32);
9863 a64 __builtin_mips_mult (i32, i32);
9864 a64 __builtin_mips_multu (ui32, ui32);
9867 The following built-in functions map directly to a particular MIPS DSP REV 2
9868 instruction. Please refer to the architecture specification
9869 for details on what each instruction does.
9872 v4q7 __builtin_mips_absq_s_qb (v4q7);
9873 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
9874 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
9875 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
9876 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
9877 i32 __builtin_mips_append (i32, i32, imm0_31);
9878 i32 __builtin_mips_balign (i32, i32, imm0_3);
9879 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
9880 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
9881 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
9882 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
9883 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
9884 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
9885 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
9886 q31 __builtin_mips_mulq_rs_w (q31, q31);
9887 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
9888 q31 __builtin_mips_mulq_s_w (q31, q31);
9889 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
9890 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
9891 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
9892 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
9893 i32 __builtin_mips_prepend (i32, i32, imm0_31);
9894 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
9895 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
9896 v4i8 __builtin_mips_shra_qb (v4i8, i32);
9897 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
9898 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
9899 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
9900 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
9901 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
9902 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
9903 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
9904 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
9905 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
9906 q31 __builtin_mips_addqh_w (q31, q31);
9907 q31 __builtin_mips_addqh_r_w (q31, q31);
9908 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
9909 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
9910 q31 __builtin_mips_subqh_w (q31, q31);
9911 q31 __builtin_mips_subqh_r_w (q31, q31);
9912 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
9913 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
9914 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
9915 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
9916 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
9917 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
9921 @node MIPS Paired-Single Support
9922 @subsection MIPS Paired-Single Support
9924 The MIPS64 architecture includes a number of instructions that
9925 operate on pairs of single-precision floating-point values.
9926 Each pair is packed into a 64-bit floating-point register,
9927 with one element being designated the ``upper half'' and
9928 the other being designated the ``lower half''.
9930 GCC supports paired-single operations using both the generic
9931 vector extensions (@pxref{Vector Extensions}) and a collection of
9932 MIPS-specific built-in functions. Both kinds of support are
9933 enabled by the @option{-mpaired-single} command-line option.
9935 The vector type associated with paired-single values is usually
9936 called @code{v2sf}. It can be defined in C as follows:
9939 typedef float v2sf __attribute__ ((vector_size (8)));
9942 @code{v2sf} values are initialized in the same way as aggregates.
9946 v2sf a = @{1.5, 9.1@};
9949 b = (v2sf) @{e, f@};
9952 @emph{Note:} The CPU's endianness determines which value is stored in
9953 the upper half of a register and which value is stored in the lower half.
9954 On little-endian targets, the first value is the lower one and the second
9955 value is the upper one. The opposite order applies to big-endian targets.
9956 For example, the code above will set the lower half of @code{a} to
9957 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
9959 @node MIPS Loongson Built-in Functions
9960 @subsection MIPS Loongson Built-in Functions
9962 GCC provides intrinsics to access the SIMD instructions provided by the
9963 ST Microelectronics Loongson-2E and -2F processors. These intrinsics,
9964 available after inclusion of the @code{loongson.h} header file,
9965 operate on the following 64-bit vector types:
9968 @item @code{uint8x8_t}, a vector of eight unsigned 8-bit integers;
9969 @item @code{uint16x4_t}, a vector of four unsigned 16-bit integers;
9970 @item @code{uint32x2_t}, a vector of two unsigned 32-bit integers;
9971 @item @code{int8x8_t}, a vector of eight signed 8-bit integers;
9972 @item @code{int16x4_t}, a vector of four signed 16-bit integers;
9973 @item @code{int32x2_t}, a vector of two signed 32-bit integers.
9976 The intrinsics provided are listed below; each is named after the
9977 machine instruction to which it corresponds, with suffixes added as
9978 appropriate to distinguish intrinsics that expand to the same machine
9979 instruction yet have different argument types. Refer to the architecture
9980 documentation for a description of the functionality of each
9984 int16x4_t packsswh (int32x2_t s, int32x2_t t);
9985 int8x8_t packsshb (int16x4_t s, int16x4_t t);
9986 uint8x8_t packushb (uint16x4_t s, uint16x4_t t);
9987 uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t);
9988 uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t);
9989 uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t);
9990 int32x2_t paddw_s (int32x2_t s, int32x2_t t);
9991 int16x4_t paddh_s (int16x4_t s, int16x4_t t);
9992 int8x8_t paddb_s (int8x8_t s, int8x8_t t);
9993 uint64_t paddd_u (uint64_t s, uint64_t t);
9994 int64_t paddd_s (int64_t s, int64_t t);
9995 int16x4_t paddsh (int16x4_t s, int16x4_t t);
9996 int8x8_t paddsb (int8x8_t s, int8x8_t t);
9997 uint16x4_t paddush (uint16x4_t s, uint16x4_t t);
9998 uint8x8_t paddusb (uint8x8_t s, uint8x8_t t);
9999 uint64_t pandn_ud (uint64_t s, uint64_t t);
10000 uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t);
10001 uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t);
10002 uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t);
10003 int64_t pandn_sd (int64_t s, int64_t t);
10004 int32x2_t pandn_sw (int32x2_t s, int32x2_t t);
10005 int16x4_t pandn_sh (int16x4_t s, int16x4_t t);
10006 int8x8_t pandn_sb (int8x8_t s, int8x8_t t);
10007 uint16x4_t pavgh (uint16x4_t s, uint16x4_t t);
10008 uint8x8_t pavgb (uint8x8_t s, uint8x8_t t);
10009 uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t);
10010 uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t);
10011 uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t);
10012 int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t);
10013 int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t);
10014 int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t);
10015 uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t);
10016 uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t);
10017 uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t);
10018 int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t);
10019 int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t);
10020 int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t);
10021 uint16x4_t pextrh_u (uint16x4_t s, int field);
10022 int16x4_t pextrh_s (int16x4_t s, int field);
10023 uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t);
10024 uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t);
10025 uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t);
10026 uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t);
10027 int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t);
10028 int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t);
10029 int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t);
10030 int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t);
10031 int32x2_t pmaddhw (int16x4_t s, int16x4_t t);
10032 int16x4_t pmaxsh (int16x4_t s, int16x4_t t);
10033 uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t);
10034 int16x4_t pminsh (int16x4_t s, int16x4_t t);
10035 uint8x8_t pminub (uint8x8_t s, uint8x8_t t);
10036 uint8x8_t pmovmskb_u (uint8x8_t s);
10037 int8x8_t pmovmskb_s (int8x8_t s);
10038 uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t);
10039 int16x4_t pmulhh (int16x4_t s, int16x4_t t);
10040 int16x4_t pmullh (int16x4_t s, int16x4_t t);
10041 int64_t pmuluw (uint32x2_t s, uint32x2_t t);
10042 uint8x8_t pasubub (uint8x8_t s, uint8x8_t t);
10043 uint16x4_t biadd (uint8x8_t s);
10044 uint16x4_t psadbh (uint8x8_t s, uint8x8_t t);
10045 uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order);
10046 int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order);
10047 uint16x4_t psllh_u (uint16x4_t s, uint8_t amount);
10048 int16x4_t psllh_s (int16x4_t s, uint8_t amount);
10049 uint32x2_t psllw_u (uint32x2_t s, uint8_t amount);
10050 int32x2_t psllw_s (int32x2_t s, uint8_t amount);
10051 uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount);
10052 int16x4_t psrlh_s (int16x4_t s, uint8_t amount);
10053 uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount);
10054 int32x2_t psrlw_s (int32x2_t s, uint8_t amount);
10055 uint16x4_t psrah_u (uint16x4_t s, uint8_t amount);
10056 int16x4_t psrah_s (int16x4_t s, uint8_t amount);
10057 uint32x2_t psraw_u (uint32x2_t s, uint8_t amount);
10058 int32x2_t psraw_s (int32x2_t s, uint8_t amount);
10059 uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t);
10060 uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t);
10061 uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t);
10062 int32x2_t psubw_s (int32x2_t s, int32x2_t t);
10063 int16x4_t psubh_s (int16x4_t s, int16x4_t t);
10064 int8x8_t psubb_s (int8x8_t s, int8x8_t t);
10065 uint64_t psubd_u (uint64_t s, uint64_t t);
10066 int64_t psubd_s (int64_t s, int64_t t);
10067 int16x4_t psubsh (int16x4_t s, int16x4_t t);
10068 int8x8_t psubsb (int8x8_t s, int8x8_t t);
10069 uint16x4_t psubush (uint16x4_t s, uint16x4_t t);
10070 uint8x8_t psubusb (uint8x8_t s, uint8x8_t t);
10071 uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t);
10072 uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t);
10073 uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t);
10074 int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t);
10075 int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t);
10076 int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t);
10077 uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t);
10078 uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t);
10079 uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t);
10080 int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t);
10081 int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t);
10082 int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t);
10086 * Paired-Single Arithmetic::
10087 * Paired-Single Built-in Functions::
10088 * MIPS-3D Built-in Functions::
10091 @node Paired-Single Arithmetic
10092 @subsubsection Paired-Single Arithmetic
10094 The table below lists the @code{v2sf} operations for which hardware
10095 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
10096 values and @code{x} is an integral value.
10098 @multitable @columnfractions .50 .50
10099 @item C code @tab MIPS instruction
10100 @item @code{a + b} @tab @code{add.ps}
10101 @item @code{a - b} @tab @code{sub.ps}
10102 @item @code{-a} @tab @code{neg.ps}
10103 @item @code{a * b} @tab @code{mul.ps}
10104 @item @code{a * b + c} @tab @code{madd.ps}
10105 @item @code{a * b - c} @tab @code{msub.ps}
10106 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
10107 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
10108 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
10111 Note that the multiply-accumulate instructions can be disabled
10112 using the command-line option @code{-mno-fused-madd}.
10114 @node Paired-Single Built-in Functions
10115 @subsubsection Paired-Single Built-in Functions
10117 The following paired-single functions map directly to a particular
10118 MIPS instruction. Please refer to the architecture specification
10119 for details on what each instruction does.
10122 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
10123 Pair lower lower (@code{pll.ps}).
10125 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
10126 Pair upper lower (@code{pul.ps}).
10128 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
10129 Pair lower upper (@code{plu.ps}).
10131 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
10132 Pair upper upper (@code{puu.ps}).
10134 @item v2sf __builtin_mips_cvt_ps_s (float, float)
10135 Convert pair to paired single (@code{cvt.ps.s}).
10137 @item float __builtin_mips_cvt_s_pl (v2sf)
10138 Convert pair lower to single (@code{cvt.s.pl}).
10140 @item float __builtin_mips_cvt_s_pu (v2sf)
10141 Convert pair upper to single (@code{cvt.s.pu}).
10143 @item v2sf __builtin_mips_abs_ps (v2sf)
10144 Absolute value (@code{abs.ps}).
10146 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
10147 Align variable (@code{alnv.ps}).
10149 @emph{Note:} The value of the third parameter must be 0 or 4
10150 modulo 8, otherwise the result will be unpredictable. Please read the
10151 instruction description for details.
10154 The following multi-instruction functions are also available.
10155 In each case, @var{cond} can be any of the 16 floating-point conditions:
10156 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
10157 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
10158 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
10161 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10162 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10163 Conditional move based on floating point comparison (@code{c.@var{cond}.ps},
10164 @code{movt.ps}/@code{movf.ps}).
10166 The @code{movt} functions return the value @var{x} computed by:
10169 c.@var{cond}.ps @var{cc},@var{a},@var{b}
10170 mov.ps @var{x},@var{c}
10171 movt.ps @var{x},@var{d},@var{cc}
10174 The @code{movf} functions are similar but use @code{movf.ps} instead
10177 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10178 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10179 Comparison of two paired-single values (@code{c.@var{cond}.ps},
10180 @code{bc1t}/@code{bc1f}).
10182 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
10183 and return either the upper or lower half of the result. For example:
10187 if (__builtin_mips_upper_c_eq_ps (a, b))
10188 upper_halves_are_equal ();
10190 upper_halves_are_unequal ();
10192 if (__builtin_mips_lower_c_eq_ps (a, b))
10193 lower_halves_are_equal ();
10195 lower_halves_are_unequal ();
10199 @node MIPS-3D Built-in Functions
10200 @subsubsection MIPS-3D Built-in Functions
10202 The MIPS-3D Application-Specific Extension (ASE) includes additional
10203 paired-single instructions that are designed to improve the performance
10204 of 3D graphics operations. Support for these instructions is controlled
10205 by the @option{-mips3d} command-line option.
10207 The functions listed below map directly to a particular MIPS-3D
10208 instruction. Please refer to the architecture specification for
10209 more details on what each instruction does.
10212 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
10213 Reduction add (@code{addr.ps}).
10215 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
10216 Reduction multiply (@code{mulr.ps}).
10218 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
10219 Convert paired single to paired word (@code{cvt.pw.ps}).
10221 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
10222 Convert paired word to paired single (@code{cvt.ps.pw}).
10224 @item float __builtin_mips_recip1_s (float)
10225 @itemx double __builtin_mips_recip1_d (double)
10226 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
10227 Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
10229 @item float __builtin_mips_recip2_s (float, float)
10230 @itemx double __builtin_mips_recip2_d (double, double)
10231 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
10232 Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
10234 @item float __builtin_mips_rsqrt1_s (float)
10235 @itemx double __builtin_mips_rsqrt1_d (double)
10236 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
10237 Reduced precision reciprocal square root (sequence step 1)
10238 (@code{rsqrt1.@var{fmt}}).
10240 @item float __builtin_mips_rsqrt2_s (float, float)
10241 @itemx double __builtin_mips_rsqrt2_d (double, double)
10242 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
10243 Reduced precision reciprocal square root (sequence step 2)
10244 (@code{rsqrt2.@var{fmt}}).
10247 The following multi-instruction functions are also available.
10248 In each case, @var{cond} can be any of the 16 floating-point conditions:
10249 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
10250 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
10251 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
10254 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
10255 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
10256 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
10257 @code{bc1t}/@code{bc1f}).
10259 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
10260 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
10265 if (__builtin_mips_cabs_eq_s (a, b))
10271 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10272 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10273 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
10274 @code{bc1t}/@code{bc1f}).
10276 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
10277 and return either the upper or lower half of the result. For example:
10281 if (__builtin_mips_upper_cabs_eq_ps (a, b))
10282 upper_halves_are_equal ();
10284 upper_halves_are_unequal ();
10286 if (__builtin_mips_lower_cabs_eq_ps (a, b))
10287 lower_halves_are_equal ();
10289 lower_halves_are_unequal ();
10292 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10293 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10294 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
10295 @code{movt.ps}/@code{movf.ps}).
10297 The @code{movt} functions return the value @var{x} computed by:
10300 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
10301 mov.ps @var{x},@var{c}
10302 movt.ps @var{x},@var{d},@var{cc}
10305 The @code{movf} functions are similar but use @code{movf.ps} instead
10308 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10309 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10310 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10311 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10312 Comparison of two paired-single values
10313 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
10314 @code{bc1any2t}/@code{bc1any2f}).
10316 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
10317 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
10318 result is true and the @code{all} forms return true if both results are true.
10323 if (__builtin_mips_any_c_eq_ps (a, b))
10328 if (__builtin_mips_all_c_eq_ps (a, b))
10334 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10335 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10336 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10337 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10338 Comparison of four paired-single values
10339 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
10340 @code{bc1any4t}/@code{bc1any4f}).
10342 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
10343 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
10344 The @code{any} forms return true if any of the four results are true
10345 and the @code{all} forms return true if all four results are true.
10350 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
10355 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
10362 @node picoChip Built-in Functions
10363 @subsection picoChip Built-in Functions
10365 GCC provides an interface to selected machine instructions from the
10366 picoChip instruction set.
10369 @item int __builtin_sbc (int @var{value})
10370 Sign bit count. Return the number of consecutive bits in @var{value}
10371 which have the same value as the sign-bit. The result is the number of
10372 leading sign bits minus one, giving the number of redundant sign bits in
10375 @item int __builtin_byteswap (int @var{value})
10376 Byte swap. Return the result of swapping the upper and lower bytes of
10379 @item int __builtin_brev (int @var{value})
10380 Bit reversal. Return the result of reversing the bits in
10381 @var{value}. Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1,
10384 @item int __builtin_adds (int @var{x}, int @var{y})
10385 Saturating addition. Return the result of adding @var{x} and @var{y},
10386 storing the value 32767 if the result overflows.
10388 @item int __builtin_subs (int @var{x}, int @var{y})
10389 Saturating subtraction. Return the result of subtracting @var{y} from
10390 @var{x}, storing the value @minus{}32768 if the result overflows.
10392 @item void __builtin_halt (void)
10393 Halt. The processor will stop execution. This built-in is useful for
10394 implementing assertions.
10398 @node Other MIPS Built-in Functions
10399 @subsection Other MIPS Built-in Functions
10401 GCC provides other MIPS-specific built-in functions:
10404 @item void __builtin_mips_cache (int @var{op}, const volatile void *@var{addr})
10405 Insert a @samp{cache} instruction with operands @var{op} and @var{addr}.
10406 GCC defines the preprocessor macro @code{___GCC_HAVE_BUILTIN_MIPS_CACHE}
10407 when this function is available.
10410 @node PowerPC AltiVec/VSX Built-in Functions
10411 @subsection PowerPC AltiVec Built-in Functions
10413 GCC provides an interface for the PowerPC family of processors to access
10414 the AltiVec operations described in Motorola's AltiVec Programming
10415 Interface Manual. The interface is made available by including
10416 @code{<altivec.h>} and using @option{-maltivec} and
10417 @option{-mabi=altivec}. The interface supports the following vector
10421 vector unsigned char
10425 vector unsigned short
10426 vector signed short
10430 vector unsigned int
10436 If @option{-mvsx} is used the following additional vector types are
10440 vector unsigned long
10445 The long types are only implemented for 64-bit code generation, and
10446 the long type is only used in the floating point/integer conversion
10449 GCC's implementation of the high-level language interface available from
10450 C and C++ code differs from Motorola's documentation in several ways.
10455 A vector constant is a list of constant expressions within curly braces.
10458 A vector initializer requires no cast if the vector constant is of the
10459 same type as the variable it is initializing.
10462 If @code{signed} or @code{unsigned} is omitted, the signedness of the
10463 vector type is the default signedness of the base type. The default
10464 varies depending on the operating system, so a portable program should
10465 always specify the signedness.
10468 Compiling with @option{-maltivec} adds keywords @code{__vector},
10469 @code{vector}, @code{__pixel}, @code{pixel}, @code{__bool} and
10470 @code{bool}. When compiling ISO C, the context-sensitive substitution
10471 of the keywords @code{vector}, @code{pixel} and @code{bool} is
10472 disabled. To use them, you must include @code{<altivec.h>} instead.
10475 GCC allows using a @code{typedef} name as the type specifier for a
10479 For C, overloaded functions are implemented with macros so the following
10483 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
10486 Since @code{vec_add} is a macro, the vector constant in the example
10487 is treated as four separate arguments. Wrap the entire argument in
10488 parentheses for this to work.
10491 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
10492 Internally, GCC uses built-in functions to achieve the functionality in
10493 the aforementioned header file, but they are not supported and are
10494 subject to change without notice.
10496 The following interfaces are supported for the generic and specific
10497 AltiVec operations and the AltiVec predicates. In cases where there
10498 is a direct mapping between generic and specific operations, only the
10499 generic names are shown here, although the specific operations can also
10502 Arguments that are documented as @code{const int} require literal
10503 integral values within the range required for that operation.
10506 vector signed char vec_abs (vector signed char);
10507 vector signed short vec_abs (vector signed short);
10508 vector signed int vec_abs (vector signed int);
10509 vector float vec_abs (vector float);
10511 vector signed char vec_abss (vector signed char);
10512 vector signed short vec_abss (vector signed short);
10513 vector signed int vec_abss (vector signed int);
10515 vector signed char vec_add (vector bool char, vector signed char);
10516 vector signed char vec_add (vector signed char, vector bool char);
10517 vector signed char vec_add (vector signed char, vector signed char);
10518 vector unsigned char vec_add (vector bool char, vector unsigned char);
10519 vector unsigned char vec_add (vector unsigned char, vector bool char);
10520 vector unsigned char vec_add (vector unsigned char,
10521 vector unsigned char);
10522 vector signed short vec_add (vector bool short, vector signed short);
10523 vector signed short vec_add (vector signed short, vector bool short);
10524 vector signed short vec_add (vector signed short, vector signed short);
10525 vector unsigned short vec_add (vector bool short,
10526 vector unsigned short);
10527 vector unsigned short vec_add (vector unsigned short,
10528 vector bool short);
10529 vector unsigned short vec_add (vector unsigned short,
10530 vector unsigned short);
10531 vector signed int vec_add (vector bool int, vector signed int);
10532 vector signed int vec_add (vector signed int, vector bool int);
10533 vector signed int vec_add (vector signed int, vector signed int);
10534 vector unsigned int vec_add (vector bool int, vector unsigned int);
10535 vector unsigned int vec_add (vector unsigned int, vector bool int);
10536 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
10537 vector float vec_add (vector float, vector float);
10539 vector float vec_vaddfp (vector float, vector float);
10541 vector signed int vec_vadduwm (vector bool int, vector signed int);
10542 vector signed int vec_vadduwm (vector signed int, vector bool int);
10543 vector signed int vec_vadduwm (vector signed int, vector signed int);
10544 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
10545 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
10546 vector unsigned int vec_vadduwm (vector unsigned int,
10547 vector unsigned int);
10549 vector signed short vec_vadduhm (vector bool short,
10550 vector signed short);
10551 vector signed short vec_vadduhm (vector signed short,
10552 vector bool short);
10553 vector signed short vec_vadduhm (vector signed short,
10554 vector signed short);
10555 vector unsigned short vec_vadduhm (vector bool short,
10556 vector unsigned short);
10557 vector unsigned short vec_vadduhm (vector unsigned short,
10558 vector bool short);
10559 vector unsigned short vec_vadduhm (vector unsigned short,
10560 vector unsigned short);
10562 vector signed char vec_vaddubm (vector bool char, vector signed char);
10563 vector signed char vec_vaddubm (vector signed char, vector bool char);
10564 vector signed char vec_vaddubm (vector signed char, vector signed char);
10565 vector unsigned char vec_vaddubm (vector bool char,
10566 vector unsigned char);
10567 vector unsigned char vec_vaddubm (vector unsigned char,
10569 vector unsigned char vec_vaddubm (vector unsigned char,
10570 vector unsigned char);
10572 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
10574 vector unsigned char vec_adds (vector bool char, vector unsigned char);
10575 vector unsigned char vec_adds (vector unsigned char, vector bool char);
10576 vector unsigned char vec_adds (vector unsigned char,
10577 vector unsigned char);
10578 vector signed char vec_adds (vector bool char, vector signed char);
10579 vector signed char vec_adds (vector signed char, vector bool char);
10580 vector signed char vec_adds (vector signed char, vector signed char);
10581 vector unsigned short vec_adds (vector bool short,
10582 vector unsigned short);
10583 vector unsigned short vec_adds (vector unsigned short,
10584 vector bool short);
10585 vector unsigned short vec_adds (vector unsigned short,
10586 vector unsigned short);
10587 vector signed short vec_adds (vector bool short, vector signed short);
10588 vector signed short vec_adds (vector signed short, vector bool short);
10589 vector signed short vec_adds (vector signed short, vector signed short);
10590 vector unsigned int vec_adds (vector bool int, vector unsigned int);
10591 vector unsigned int vec_adds (vector unsigned int, vector bool int);
10592 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
10593 vector signed int vec_adds (vector bool int, vector signed int);
10594 vector signed int vec_adds (vector signed int, vector bool int);
10595 vector signed int vec_adds (vector signed int, vector signed int);
10597 vector signed int vec_vaddsws (vector bool int, vector signed int);
10598 vector signed int vec_vaddsws (vector signed int, vector bool int);
10599 vector signed int vec_vaddsws (vector signed int, vector signed int);
10601 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
10602 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
10603 vector unsigned int vec_vadduws (vector unsigned int,
10604 vector unsigned int);
10606 vector signed short vec_vaddshs (vector bool short,
10607 vector signed short);
10608 vector signed short vec_vaddshs (vector signed short,
10609 vector bool short);
10610 vector signed short vec_vaddshs (vector signed short,
10611 vector signed short);
10613 vector unsigned short vec_vadduhs (vector bool short,
10614 vector unsigned short);
10615 vector unsigned short vec_vadduhs (vector unsigned short,
10616 vector bool short);
10617 vector unsigned short vec_vadduhs (vector unsigned short,
10618 vector unsigned short);
10620 vector signed char vec_vaddsbs (vector bool char, vector signed char);
10621 vector signed char vec_vaddsbs (vector signed char, vector bool char);
10622 vector signed char vec_vaddsbs (vector signed char, vector signed char);
10624 vector unsigned char vec_vaddubs (vector bool char,
10625 vector unsigned char);
10626 vector unsigned char vec_vaddubs (vector unsigned char,
10628 vector unsigned char vec_vaddubs (vector unsigned char,
10629 vector unsigned char);
10631 vector float vec_and (vector float, vector float);
10632 vector float vec_and (vector float, vector bool int);
10633 vector float vec_and (vector bool int, vector float);
10634 vector bool int vec_and (vector bool int, vector bool int);
10635 vector signed int vec_and (vector bool int, vector signed int);
10636 vector signed int vec_and (vector signed int, vector bool int);
10637 vector signed int vec_and (vector signed int, vector signed int);
10638 vector unsigned int vec_and (vector bool int, vector unsigned int);
10639 vector unsigned int vec_and (vector unsigned int, vector bool int);
10640 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
10641 vector bool short vec_and (vector bool short, vector bool short);
10642 vector signed short vec_and (vector bool short, vector signed short);
10643 vector signed short vec_and (vector signed short, vector bool short);
10644 vector signed short vec_and (vector signed short, vector signed short);
10645 vector unsigned short vec_and (vector bool short,
10646 vector unsigned short);
10647 vector unsigned short vec_and (vector unsigned short,
10648 vector bool short);
10649 vector unsigned short vec_and (vector unsigned short,
10650 vector unsigned short);
10651 vector signed char vec_and (vector bool char, vector signed char);
10652 vector bool char vec_and (vector bool char, vector bool char);
10653 vector signed char vec_and (vector signed char, vector bool char);
10654 vector signed char vec_and (vector signed char, vector signed char);
10655 vector unsigned char vec_and (vector bool char, vector unsigned char);
10656 vector unsigned char vec_and (vector unsigned char, vector bool char);
10657 vector unsigned char vec_and (vector unsigned char,
10658 vector unsigned char);
10660 vector float vec_andc (vector float, vector float);
10661 vector float vec_andc (vector float, vector bool int);
10662 vector float vec_andc (vector bool int, vector float);
10663 vector bool int vec_andc (vector bool int, vector bool int);
10664 vector signed int vec_andc (vector bool int, vector signed int);
10665 vector signed int vec_andc (vector signed int, vector bool int);
10666 vector signed int vec_andc (vector signed int, vector signed int);
10667 vector unsigned int vec_andc (vector bool int, vector unsigned int);
10668 vector unsigned int vec_andc (vector unsigned int, vector bool int);
10669 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
10670 vector bool short vec_andc (vector bool short, vector bool short);
10671 vector signed short vec_andc (vector bool short, vector signed short);
10672 vector signed short vec_andc (vector signed short, vector bool short);
10673 vector signed short vec_andc (vector signed short, vector signed short);
10674 vector unsigned short vec_andc (vector bool short,
10675 vector unsigned short);
10676 vector unsigned short vec_andc (vector unsigned short,
10677 vector bool short);
10678 vector unsigned short vec_andc (vector unsigned short,
10679 vector unsigned short);
10680 vector signed char vec_andc (vector bool char, vector signed char);
10681 vector bool char vec_andc (vector bool char, vector bool char);
10682 vector signed char vec_andc (vector signed char, vector bool char);
10683 vector signed char vec_andc (vector signed char, vector signed char);
10684 vector unsigned char vec_andc (vector bool char, vector unsigned char);
10685 vector unsigned char vec_andc (vector unsigned char, vector bool char);
10686 vector unsigned char vec_andc (vector unsigned char,
10687 vector unsigned char);
10689 vector unsigned char vec_avg (vector unsigned char,
10690 vector unsigned char);
10691 vector signed char vec_avg (vector signed char, vector signed char);
10692 vector unsigned short vec_avg (vector unsigned short,
10693 vector unsigned short);
10694 vector signed short vec_avg (vector signed short, vector signed short);
10695 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
10696 vector signed int vec_avg (vector signed int, vector signed int);
10698 vector signed int vec_vavgsw (vector signed int, vector signed int);
10700 vector unsigned int vec_vavguw (vector unsigned int,
10701 vector unsigned int);
10703 vector signed short vec_vavgsh (vector signed short,
10704 vector signed short);
10706 vector unsigned short vec_vavguh (vector unsigned short,
10707 vector unsigned short);
10709 vector signed char vec_vavgsb (vector signed char, vector signed char);
10711 vector unsigned char vec_vavgub (vector unsigned char,
10712 vector unsigned char);
10714 vector float vec_copysign (vector float);
10716 vector float vec_ceil (vector float);
10718 vector signed int vec_cmpb (vector float, vector float);
10720 vector bool char vec_cmpeq (vector signed char, vector signed char);
10721 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
10722 vector bool short vec_cmpeq (vector signed short, vector signed short);
10723 vector bool short vec_cmpeq (vector unsigned short,
10724 vector unsigned short);
10725 vector bool int vec_cmpeq (vector signed int, vector signed int);
10726 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
10727 vector bool int vec_cmpeq (vector float, vector float);
10729 vector bool int vec_vcmpeqfp (vector float, vector float);
10731 vector bool int vec_vcmpequw (vector signed int, vector signed int);
10732 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
10734 vector bool short vec_vcmpequh (vector signed short,
10735 vector signed short);
10736 vector bool short vec_vcmpequh (vector unsigned short,
10737 vector unsigned short);
10739 vector bool char vec_vcmpequb (vector signed char, vector signed char);
10740 vector bool char vec_vcmpequb (vector unsigned char,
10741 vector unsigned char);
10743 vector bool int vec_cmpge (vector float, vector float);
10745 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
10746 vector bool char vec_cmpgt (vector signed char, vector signed char);
10747 vector bool short vec_cmpgt (vector unsigned short,
10748 vector unsigned short);
10749 vector bool short vec_cmpgt (vector signed short, vector signed short);
10750 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
10751 vector bool int vec_cmpgt (vector signed int, vector signed int);
10752 vector bool int vec_cmpgt (vector float, vector float);
10754 vector bool int vec_vcmpgtfp (vector float, vector float);
10756 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
10758 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
10760 vector bool short vec_vcmpgtsh (vector signed short,
10761 vector signed short);
10763 vector bool short vec_vcmpgtuh (vector unsigned short,
10764 vector unsigned short);
10766 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
10768 vector bool char vec_vcmpgtub (vector unsigned char,
10769 vector unsigned char);
10771 vector bool int vec_cmple (vector float, vector float);
10773 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
10774 vector bool char vec_cmplt (vector signed char, vector signed char);
10775 vector bool short vec_cmplt (vector unsigned short,
10776 vector unsigned short);
10777 vector bool short vec_cmplt (vector signed short, vector signed short);
10778 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
10779 vector bool int vec_cmplt (vector signed int, vector signed int);
10780 vector bool int vec_cmplt (vector float, vector float);
10782 vector float vec_ctf (vector unsigned int, const int);
10783 vector float vec_ctf (vector signed int, const int);
10785 vector float vec_vcfsx (vector signed int, const int);
10787 vector float vec_vcfux (vector unsigned int, const int);
10789 vector signed int vec_cts (vector float, const int);
10791 vector unsigned int vec_ctu (vector float, const int);
10793 void vec_dss (const int);
10795 void vec_dssall (void);
10797 void vec_dst (const vector unsigned char *, int, const int);
10798 void vec_dst (const vector signed char *, int, const int);
10799 void vec_dst (const vector bool char *, int, const int);
10800 void vec_dst (const vector unsigned short *, int, const int);
10801 void vec_dst (const vector signed short *, int, const int);
10802 void vec_dst (const vector bool short *, int, const int);
10803 void vec_dst (const vector pixel *, int, const int);
10804 void vec_dst (const vector unsigned int *, int, const int);
10805 void vec_dst (const vector signed int *, int, const int);
10806 void vec_dst (const vector bool int *, int, const int);
10807 void vec_dst (const vector float *, int, const int);
10808 void vec_dst (const unsigned char *, int, const int);
10809 void vec_dst (const signed char *, int, const int);
10810 void vec_dst (const unsigned short *, int, const int);
10811 void vec_dst (const short *, int, const int);
10812 void vec_dst (const unsigned int *, int, const int);
10813 void vec_dst (const int *, int, const int);
10814 void vec_dst (const unsigned long *, int, const int);
10815 void vec_dst (const long *, int, const int);
10816 void vec_dst (const float *, int, const int);
10818 void vec_dstst (const vector unsigned char *, int, const int);
10819 void vec_dstst (const vector signed char *, int, const int);
10820 void vec_dstst (const vector bool char *, int, const int);
10821 void vec_dstst (const vector unsigned short *, int, const int);
10822 void vec_dstst (const vector signed short *, int, const int);
10823 void vec_dstst (const vector bool short *, int, const int);
10824 void vec_dstst (const vector pixel *, int, const int);
10825 void vec_dstst (const vector unsigned int *, int, const int);
10826 void vec_dstst (const vector signed int *, int, const int);
10827 void vec_dstst (const vector bool int *, int, const int);
10828 void vec_dstst (const vector float *, int, const int);
10829 void vec_dstst (const unsigned char *, int, const int);
10830 void vec_dstst (const signed char *, int, const int);
10831 void vec_dstst (const unsigned short *, int, const int);
10832 void vec_dstst (const short *, int, const int);
10833 void vec_dstst (const unsigned int *, int, const int);
10834 void vec_dstst (const int *, int, const int);
10835 void vec_dstst (const unsigned long *, int, const int);
10836 void vec_dstst (const long *, int, const int);
10837 void vec_dstst (const float *, int, const int);
10839 void vec_dststt (const vector unsigned char *, int, const int);
10840 void vec_dststt (const vector signed char *, int, const int);
10841 void vec_dststt (const vector bool char *, int, const int);
10842 void vec_dststt (const vector unsigned short *, int, const int);
10843 void vec_dststt (const vector signed short *, int, const int);
10844 void vec_dststt (const vector bool short *, int, const int);
10845 void vec_dststt (const vector pixel *, int, const int);
10846 void vec_dststt (const vector unsigned int *, int, const int);
10847 void vec_dststt (const vector signed int *, int, const int);
10848 void vec_dststt (const vector bool int *, int, const int);
10849 void vec_dststt (const vector float *, int, const int);
10850 void vec_dststt (const unsigned char *, int, const int);
10851 void vec_dststt (const signed char *, int, const int);
10852 void vec_dststt (const unsigned short *, int, const int);
10853 void vec_dststt (const short *, int, const int);
10854 void vec_dststt (const unsigned int *, int, const int);
10855 void vec_dststt (const int *, int, const int);
10856 void vec_dststt (const unsigned long *, int, const int);
10857 void vec_dststt (const long *, int, const int);
10858 void vec_dststt (const float *, int, const int);
10860 void vec_dstt (const vector unsigned char *, int, const int);
10861 void vec_dstt (const vector signed char *, int, const int);
10862 void vec_dstt (const vector bool char *, int, const int);
10863 void vec_dstt (const vector unsigned short *, int, const int);
10864 void vec_dstt (const vector signed short *, int, const int);
10865 void vec_dstt (const vector bool short *, int, const int);
10866 void vec_dstt (const vector pixel *, int, const int);
10867 void vec_dstt (const vector unsigned int *, int, const int);
10868 void vec_dstt (const vector signed int *, int, const int);
10869 void vec_dstt (const vector bool int *, int, const int);
10870 void vec_dstt (const vector float *, int, const int);
10871 void vec_dstt (const unsigned char *, int, const int);
10872 void vec_dstt (const signed char *, int, const int);
10873 void vec_dstt (const unsigned short *, int, const int);
10874 void vec_dstt (const short *, int, const int);
10875 void vec_dstt (const unsigned int *, int, const int);
10876 void vec_dstt (const int *, int, const int);
10877 void vec_dstt (const unsigned long *, int, const int);
10878 void vec_dstt (const long *, int, const int);
10879 void vec_dstt (const float *, int, const int);
10881 vector float vec_expte (vector float);
10883 vector float vec_floor (vector float);
10885 vector float vec_ld (int, const vector float *);
10886 vector float vec_ld (int, const float *);
10887 vector bool int vec_ld (int, const vector bool int *);
10888 vector signed int vec_ld (int, const vector signed int *);
10889 vector signed int vec_ld (int, const int *);
10890 vector signed int vec_ld (int, const long *);
10891 vector unsigned int vec_ld (int, const vector unsigned int *);
10892 vector unsigned int vec_ld (int, const unsigned int *);
10893 vector unsigned int vec_ld (int, const unsigned long *);
10894 vector bool short vec_ld (int, const vector bool short *);
10895 vector pixel vec_ld (int, const vector pixel *);
10896 vector signed short vec_ld (int, const vector signed short *);
10897 vector signed short vec_ld (int, const short *);
10898 vector unsigned short vec_ld (int, const vector unsigned short *);
10899 vector unsigned short vec_ld (int, const unsigned short *);
10900 vector bool char vec_ld (int, const vector bool char *);
10901 vector signed char vec_ld (int, const vector signed char *);
10902 vector signed char vec_ld (int, const signed char *);
10903 vector unsigned char vec_ld (int, const vector unsigned char *);
10904 vector unsigned char vec_ld (int, const unsigned char *);
10906 vector signed char vec_lde (int, const signed char *);
10907 vector unsigned char vec_lde (int, const unsigned char *);
10908 vector signed short vec_lde (int, const short *);
10909 vector unsigned short vec_lde (int, const unsigned short *);
10910 vector float vec_lde (int, const float *);
10911 vector signed int vec_lde (int, const int *);
10912 vector unsigned int vec_lde (int, const unsigned int *);
10913 vector signed int vec_lde (int, const long *);
10914 vector unsigned int vec_lde (int, const unsigned long *);
10916 vector float vec_lvewx (int, float *);
10917 vector signed int vec_lvewx (int, int *);
10918 vector unsigned int vec_lvewx (int, unsigned int *);
10919 vector signed int vec_lvewx (int, long *);
10920 vector unsigned int vec_lvewx (int, unsigned long *);
10922 vector signed short vec_lvehx (int, short *);
10923 vector unsigned short vec_lvehx (int, unsigned short *);
10925 vector signed char vec_lvebx (int, char *);
10926 vector unsigned char vec_lvebx (int, unsigned char *);
10928 vector float vec_ldl (int, const vector float *);
10929 vector float vec_ldl (int, const float *);
10930 vector bool int vec_ldl (int, const vector bool int *);
10931 vector signed int vec_ldl (int, const vector signed int *);
10932 vector signed int vec_ldl (int, const int *);
10933 vector signed int vec_ldl (int, const long *);
10934 vector unsigned int vec_ldl (int, const vector unsigned int *);
10935 vector unsigned int vec_ldl (int, const unsigned int *);
10936 vector unsigned int vec_ldl (int, const unsigned long *);
10937 vector bool short vec_ldl (int, const vector bool short *);
10938 vector pixel vec_ldl (int, const vector pixel *);
10939 vector signed short vec_ldl (int, const vector signed short *);
10940 vector signed short vec_ldl (int, const short *);
10941 vector unsigned short vec_ldl (int, const vector unsigned short *);
10942 vector unsigned short vec_ldl (int, const unsigned short *);
10943 vector bool char vec_ldl (int, const vector bool char *);
10944 vector signed char vec_ldl (int, const vector signed char *);
10945 vector signed char vec_ldl (int, const signed char *);
10946 vector unsigned char vec_ldl (int, const vector unsigned char *);
10947 vector unsigned char vec_ldl (int, const unsigned char *);
10949 vector float vec_loge (vector float);
10951 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
10952 vector unsigned char vec_lvsl (int, const volatile signed char *);
10953 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
10954 vector unsigned char vec_lvsl (int, const volatile short *);
10955 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
10956 vector unsigned char vec_lvsl (int, const volatile int *);
10957 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
10958 vector unsigned char vec_lvsl (int, const volatile long *);
10959 vector unsigned char vec_lvsl (int, const volatile float *);
10961 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
10962 vector unsigned char vec_lvsr (int, const volatile signed char *);
10963 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
10964 vector unsigned char vec_lvsr (int, const volatile short *);
10965 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
10966 vector unsigned char vec_lvsr (int, const volatile int *);
10967 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
10968 vector unsigned char vec_lvsr (int, const volatile long *);
10969 vector unsigned char vec_lvsr (int, const volatile float *);
10971 vector float vec_madd (vector float, vector float, vector float);
10973 vector signed short vec_madds (vector signed short,
10974 vector signed short,
10975 vector signed short);
10977 vector unsigned char vec_max (vector bool char, vector unsigned char);
10978 vector unsigned char vec_max (vector unsigned char, vector bool char);
10979 vector unsigned char vec_max (vector unsigned char,
10980 vector unsigned char);
10981 vector signed char vec_max (vector bool char, vector signed char);
10982 vector signed char vec_max (vector signed char, vector bool char);
10983 vector signed char vec_max (vector signed char, vector signed char);
10984 vector unsigned short vec_max (vector bool short,
10985 vector unsigned short);
10986 vector unsigned short vec_max (vector unsigned short,
10987 vector bool short);
10988 vector unsigned short vec_max (vector unsigned short,
10989 vector unsigned short);
10990 vector signed short vec_max (vector bool short, vector signed short);
10991 vector signed short vec_max (vector signed short, vector bool short);
10992 vector signed short vec_max (vector signed short, vector signed short);
10993 vector unsigned int vec_max (vector bool int, vector unsigned int);
10994 vector unsigned int vec_max (vector unsigned int, vector bool int);
10995 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
10996 vector signed int vec_max (vector bool int, vector signed int);
10997 vector signed int vec_max (vector signed int, vector bool int);
10998 vector signed int vec_max (vector signed int, vector signed int);
10999 vector float vec_max (vector float, vector float);
11001 vector float vec_vmaxfp (vector float, vector float);
11003 vector signed int vec_vmaxsw (vector bool int, vector signed int);
11004 vector signed int vec_vmaxsw (vector signed int, vector bool int);
11005 vector signed int vec_vmaxsw (vector signed int, vector signed int);
11007 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
11008 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
11009 vector unsigned int vec_vmaxuw (vector unsigned int,
11010 vector unsigned int);
11012 vector signed short vec_vmaxsh (vector bool short, vector signed short);
11013 vector signed short vec_vmaxsh (vector signed short, vector bool short);
11014 vector signed short vec_vmaxsh (vector signed short,
11015 vector signed short);
11017 vector unsigned short vec_vmaxuh (vector bool short,
11018 vector unsigned short);
11019 vector unsigned short vec_vmaxuh (vector unsigned short,
11020 vector bool short);
11021 vector unsigned short vec_vmaxuh (vector unsigned short,
11022 vector unsigned short);
11024 vector signed char vec_vmaxsb (vector bool char, vector signed char);
11025 vector signed char vec_vmaxsb (vector signed char, vector bool char);
11026 vector signed char vec_vmaxsb (vector signed char, vector signed char);
11028 vector unsigned char vec_vmaxub (vector bool char,
11029 vector unsigned char);
11030 vector unsigned char vec_vmaxub (vector unsigned char,
11032 vector unsigned char vec_vmaxub (vector unsigned char,
11033 vector unsigned char);
11035 vector bool char vec_mergeh (vector bool char, vector bool char);
11036 vector signed char vec_mergeh (vector signed char, vector signed char);
11037 vector unsigned char vec_mergeh (vector unsigned char,
11038 vector unsigned char);
11039 vector bool short vec_mergeh (vector bool short, vector bool short);
11040 vector pixel vec_mergeh (vector pixel, vector pixel);
11041 vector signed short vec_mergeh (vector signed short,
11042 vector signed short);
11043 vector unsigned short vec_mergeh (vector unsigned short,
11044 vector unsigned short);
11045 vector float vec_mergeh (vector float, vector float);
11046 vector bool int vec_mergeh (vector bool int, vector bool int);
11047 vector signed int vec_mergeh (vector signed int, vector signed int);
11048 vector unsigned int vec_mergeh (vector unsigned int,
11049 vector unsigned int);
11051 vector float vec_vmrghw (vector float, vector float);
11052 vector bool int vec_vmrghw (vector bool int, vector bool int);
11053 vector signed int vec_vmrghw (vector signed int, vector signed int);
11054 vector unsigned int vec_vmrghw (vector unsigned int,
11055 vector unsigned int);
11057 vector bool short vec_vmrghh (vector bool short, vector bool short);
11058 vector signed short vec_vmrghh (vector signed short,
11059 vector signed short);
11060 vector unsigned short vec_vmrghh (vector unsigned short,
11061 vector unsigned short);
11062 vector pixel vec_vmrghh (vector pixel, vector pixel);
11064 vector bool char vec_vmrghb (vector bool char, vector bool char);
11065 vector signed char vec_vmrghb (vector signed char, vector signed char);
11066 vector unsigned char vec_vmrghb (vector unsigned char,
11067 vector unsigned char);
11069 vector bool char vec_mergel (vector bool char, vector bool char);
11070 vector signed char vec_mergel (vector signed char, vector signed char);
11071 vector unsigned char vec_mergel (vector unsigned char,
11072 vector unsigned char);
11073 vector bool short vec_mergel (vector bool short, vector bool short);
11074 vector pixel vec_mergel (vector pixel, vector pixel);
11075 vector signed short vec_mergel (vector signed short,
11076 vector signed short);
11077 vector unsigned short vec_mergel (vector unsigned short,
11078 vector unsigned short);
11079 vector float vec_mergel (vector float, vector float);
11080 vector bool int vec_mergel (vector bool int, vector bool int);
11081 vector signed int vec_mergel (vector signed int, vector signed int);
11082 vector unsigned int vec_mergel (vector unsigned int,
11083 vector unsigned int);
11085 vector float vec_vmrglw (vector float, vector float);
11086 vector signed int vec_vmrglw (vector signed int, vector signed int);
11087 vector unsigned int vec_vmrglw (vector unsigned int,
11088 vector unsigned int);
11089 vector bool int vec_vmrglw (vector bool int, vector bool int);
11091 vector bool short vec_vmrglh (vector bool short, vector bool short);
11092 vector signed short vec_vmrglh (vector signed short,
11093 vector signed short);
11094 vector unsigned short vec_vmrglh (vector unsigned short,
11095 vector unsigned short);
11096 vector pixel vec_vmrglh (vector pixel, vector pixel);
11098 vector bool char vec_vmrglb (vector bool char, vector bool char);
11099 vector signed char vec_vmrglb (vector signed char, vector signed char);
11100 vector unsigned char vec_vmrglb (vector unsigned char,
11101 vector unsigned char);
11103 vector unsigned short vec_mfvscr (void);
11105 vector unsigned char vec_min (vector bool char, vector unsigned char);
11106 vector unsigned char vec_min (vector unsigned char, vector bool char);
11107 vector unsigned char vec_min (vector unsigned char,
11108 vector unsigned char);
11109 vector signed char vec_min (vector bool char, vector signed char);
11110 vector signed char vec_min (vector signed char, vector bool char);
11111 vector signed char vec_min (vector signed char, vector signed char);
11112 vector unsigned short vec_min (vector bool short,
11113 vector unsigned short);
11114 vector unsigned short vec_min (vector unsigned short,
11115 vector bool short);
11116 vector unsigned short vec_min (vector unsigned short,
11117 vector unsigned short);
11118 vector signed short vec_min (vector bool short, vector signed short);
11119 vector signed short vec_min (vector signed short, vector bool short);
11120 vector signed short vec_min (vector signed short, vector signed short);
11121 vector unsigned int vec_min (vector bool int, vector unsigned int);
11122 vector unsigned int vec_min (vector unsigned int, vector bool int);
11123 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
11124 vector signed int vec_min (vector bool int, vector signed int);
11125 vector signed int vec_min (vector signed int, vector bool int);
11126 vector signed int vec_min (vector signed int, vector signed int);
11127 vector float vec_min (vector float, vector float);
11129 vector float vec_vminfp (vector float, vector float);
11131 vector signed int vec_vminsw (vector bool int, vector signed int);
11132 vector signed int vec_vminsw (vector signed int, vector bool int);
11133 vector signed int vec_vminsw (vector signed int, vector signed int);
11135 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
11136 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
11137 vector unsigned int vec_vminuw (vector unsigned int,
11138 vector unsigned int);
11140 vector signed short vec_vminsh (vector bool short, vector signed short);
11141 vector signed short vec_vminsh (vector signed short, vector bool short);
11142 vector signed short vec_vminsh (vector signed short,
11143 vector signed short);
11145 vector unsigned short vec_vminuh (vector bool short,
11146 vector unsigned short);
11147 vector unsigned short vec_vminuh (vector unsigned short,
11148 vector bool short);
11149 vector unsigned short vec_vminuh (vector unsigned short,
11150 vector unsigned short);
11152 vector signed char vec_vminsb (vector bool char, vector signed char);
11153 vector signed char vec_vminsb (vector signed char, vector bool char);
11154 vector signed char vec_vminsb (vector signed char, vector signed char);
11156 vector unsigned char vec_vminub (vector bool char,
11157 vector unsigned char);
11158 vector unsigned char vec_vminub (vector unsigned char,
11160 vector unsigned char vec_vminub (vector unsigned char,
11161 vector unsigned char);
11163 vector signed short vec_mladd (vector signed short,
11164 vector signed short,
11165 vector signed short);
11166 vector signed short vec_mladd (vector signed short,
11167 vector unsigned short,
11168 vector unsigned short);
11169 vector signed short vec_mladd (vector unsigned short,
11170 vector signed short,
11171 vector signed short);
11172 vector unsigned short vec_mladd (vector unsigned short,
11173 vector unsigned short,
11174 vector unsigned short);
11176 vector signed short vec_mradds (vector signed short,
11177 vector signed short,
11178 vector signed short);
11180 vector unsigned int vec_msum (vector unsigned char,
11181 vector unsigned char,
11182 vector unsigned int);
11183 vector signed int vec_msum (vector signed char,
11184 vector unsigned char,
11185 vector signed int);
11186 vector unsigned int vec_msum (vector unsigned short,
11187 vector unsigned short,
11188 vector unsigned int);
11189 vector signed int vec_msum (vector signed short,
11190 vector signed short,
11191 vector signed int);
11193 vector signed int vec_vmsumshm (vector signed short,
11194 vector signed short,
11195 vector signed int);
11197 vector unsigned int vec_vmsumuhm (vector unsigned short,
11198 vector unsigned short,
11199 vector unsigned int);
11201 vector signed int vec_vmsummbm (vector signed char,
11202 vector unsigned char,
11203 vector signed int);
11205 vector unsigned int vec_vmsumubm (vector unsigned char,
11206 vector unsigned char,
11207 vector unsigned int);
11209 vector unsigned int vec_msums (vector unsigned short,
11210 vector unsigned short,
11211 vector unsigned int);
11212 vector signed int vec_msums (vector signed short,
11213 vector signed short,
11214 vector signed int);
11216 vector signed int vec_vmsumshs (vector signed short,
11217 vector signed short,
11218 vector signed int);
11220 vector unsigned int vec_vmsumuhs (vector unsigned short,
11221 vector unsigned short,
11222 vector unsigned int);
11224 void vec_mtvscr (vector signed int);
11225 void vec_mtvscr (vector unsigned int);
11226 void vec_mtvscr (vector bool int);
11227 void vec_mtvscr (vector signed short);
11228 void vec_mtvscr (vector unsigned short);
11229 void vec_mtvscr (vector bool short);
11230 void vec_mtvscr (vector pixel);
11231 void vec_mtvscr (vector signed char);
11232 void vec_mtvscr (vector unsigned char);
11233 void vec_mtvscr (vector bool char);
11235 vector unsigned short vec_mule (vector unsigned char,
11236 vector unsigned char);
11237 vector signed short vec_mule (vector signed char,
11238 vector signed char);
11239 vector unsigned int vec_mule (vector unsigned short,
11240 vector unsigned short);
11241 vector signed int vec_mule (vector signed short, vector signed short);
11243 vector signed int vec_vmulesh (vector signed short,
11244 vector signed short);
11246 vector unsigned int vec_vmuleuh (vector unsigned short,
11247 vector unsigned short);
11249 vector signed short vec_vmulesb (vector signed char,
11250 vector signed char);
11252 vector unsigned short vec_vmuleub (vector unsigned char,
11253 vector unsigned char);
11255 vector unsigned short vec_mulo (vector unsigned char,
11256 vector unsigned char);
11257 vector signed short vec_mulo (vector signed char, vector signed char);
11258 vector unsigned int vec_mulo (vector unsigned short,
11259 vector unsigned short);
11260 vector signed int vec_mulo (vector signed short, vector signed short);
11262 vector signed int vec_vmulosh (vector signed short,
11263 vector signed short);
11265 vector unsigned int vec_vmulouh (vector unsigned short,
11266 vector unsigned short);
11268 vector signed short vec_vmulosb (vector signed char,
11269 vector signed char);
11271 vector unsigned short vec_vmuloub (vector unsigned char,
11272 vector unsigned char);
11274 vector float vec_nmsub (vector float, vector float, vector float);
11276 vector float vec_nor (vector float, vector float);
11277 vector signed int vec_nor (vector signed int, vector signed int);
11278 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
11279 vector bool int vec_nor (vector bool int, vector bool int);
11280 vector signed short vec_nor (vector signed short, vector signed short);
11281 vector unsigned short vec_nor (vector unsigned short,
11282 vector unsigned short);
11283 vector bool short vec_nor (vector bool short, vector bool short);
11284 vector signed char vec_nor (vector signed char, vector signed char);
11285 vector unsigned char vec_nor (vector unsigned char,
11286 vector unsigned char);
11287 vector bool char vec_nor (vector bool char, vector bool char);
11289 vector float vec_or (vector float, vector float);
11290 vector float vec_or (vector float, vector bool int);
11291 vector float vec_or (vector bool int, vector float);
11292 vector bool int vec_or (vector bool int, vector bool int);
11293 vector signed int vec_or (vector bool int, vector signed int);
11294 vector signed int vec_or (vector signed int, vector bool int);
11295 vector signed int vec_or (vector signed int, vector signed int);
11296 vector unsigned int vec_or (vector bool int, vector unsigned int);
11297 vector unsigned int vec_or (vector unsigned int, vector bool int);
11298 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
11299 vector bool short vec_or (vector bool short, vector bool short);
11300 vector signed short vec_or (vector bool short, vector signed short);
11301 vector signed short vec_or (vector signed short, vector bool short);
11302 vector signed short vec_or (vector signed short, vector signed short);
11303 vector unsigned short vec_or (vector bool short, vector unsigned short);
11304 vector unsigned short vec_or (vector unsigned short, vector bool short);
11305 vector unsigned short vec_or (vector unsigned short,
11306 vector unsigned short);
11307 vector signed char vec_or (vector bool char, vector signed char);
11308 vector bool char vec_or (vector bool char, vector bool char);
11309 vector signed char vec_or (vector signed char, vector bool char);
11310 vector signed char vec_or (vector signed char, vector signed char);
11311 vector unsigned char vec_or (vector bool char, vector unsigned char);
11312 vector unsigned char vec_or (vector unsigned char, vector bool char);
11313 vector unsigned char vec_or (vector unsigned char,
11314 vector unsigned char);
11316 vector signed char vec_pack (vector signed short, vector signed short);
11317 vector unsigned char vec_pack (vector unsigned short,
11318 vector unsigned short);
11319 vector bool char vec_pack (vector bool short, vector bool short);
11320 vector signed short vec_pack (vector signed int, vector signed int);
11321 vector unsigned short vec_pack (vector unsigned int,
11322 vector unsigned int);
11323 vector bool short vec_pack (vector bool int, vector bool int);
11325 vector bool short vec_vpkuwum (vector bool int, vector bool int);
11326 vector signed short vec_vpkuwum (vector signed int, vector signed int);
11327 vector unsigned short vec_vpkuwum (vector unsigned int,
11328 vector unsigned int);
11330 vector bool char vec_vpkuhum (vector bool short, vector bool short);
11331 vector signed char vec_vpkuhum (vector signed short,
11332 vector signed short);
11333 vector unsigned char vec_vpkuhum (vector unsigned short,
11334 vector unsigned short);
11336 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
11338 vector unsigned char vec_packs (vector unsigned short,
11339 vector unsigned short);
11340 vector signed char vec_packs (vector signed short, vector signed short);
11341 vector unsigned short vec_packs (vector unsigned int,
11342 vector unsigned int);
11343 vector signed short vec_packs (vector signed int, vector signed int);
11345 vector signed short vec_vpkswss (vector signed int, vector signed int);
11347 vector unsigned short vec_vpkuwus (vector unsigned int,
11348 vector unsigned int);
11350 vector signed char vec_vpkshss (vector signed short,
11351 vector signed short);
11353 vector unsigned char vec_vpkuhus (vector unsigned short,
11354 vector unsigned short);
11356 vector unsigned char vec_packsu (vector unsigned short,
11357 vector unsigned short);
11358 vector unsigned char vec_packsu (vector signed short,
11359 vector signed short);
11360 vector unsigned short vec_packsu (vector unsigned int,
11361 vector unsigned int);
11362 vector unsigned short vec_packsu (vector signed int, vector signed int);
11364 vector unsigned short vec_vpkswus (vector signed int,
11365 vector signed int);
11367 vector unsigned char vec_vpkshus (vector signed short,
11368 vector signed short);
11370 vector float vec_perm (vector float,
11372 vector unsigned char);
11373 vector signed int vec_perm (vector signed int,
11375 vector unsigned char);
11376 vector unsigned int vec_perm (vector unsigned int,
11377 vector unsigned int,
11378 vector unsigned char);
11379 vector bool int vec_perm (vector bool int,
11381 vector unsigned char);
11382 vector signed short vec_perm (vector signed short,
11383 vector signed short,
11384 vector unsigned char);
11385 vector unsigned short vec_perm (vector unsigned short,
11386 vector unsigned short,
11387 vector unsigned char);
11388 vector bool short vec_perm (vector bool short,
11390 vector unsigned char);
11391 vector pixel vec_perm (vector pixel,
11393 vector unsigned char);
11394 vector signed char vec_perm (vector signed char,
11395 vector signed char,
11396 vector unsigned char);
11397 vector unsigned char vec_perm (vector unsigned char,
11398 vector unsigned char,
11399 vector unsigned char);
11400 vector bool char vec_perm (vector bool char,
11402 vector unsigned char);
11404 vector float vec_re (vector float);
11406 vector signed char vec_rl (vector signed char,
11407 vector unsigned char);
11408 vector unsigned char vec_rl (vector unsigned char,
11409 vector unsigned char);
11410 vector signed short vec_rl (vector signed short, vector unsigned short);
11411 vector unsigned short vec_rl (vector unsigned short,
11412 vector unsigned short);
11413 vector signed int vec_rl (vector signed int, vector unsigned int);
11414 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
11416 vector signed int vec_vrlw (vector signed int, vector unsigned int);
11417 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
11419 vector signed short vec_vrlh (vector signed short,
11420 vector unsigned short);
11421 vector unsigned short vec_vrlh (vector unsigned short,
11422 vector unsigned short);
11424 vector signed char vec_vrlb (vector signed char, vector unsigned char);
11425 vector unsigned char vec_vrlb (vector unsigned char,
11426 vector unsigned char);
11428 vector float vec_round (vector float);
11430 vector float vec_recip (vector float, vector float);
11432 vector float vec_rsqrt (vector float);
11434 vector float vec_rsqrte (vector float);
11436 vector float vec_sel (vector float, vector float, vector bool int);
11437 vector float vec_sel (vector float, vector float, vector unsigned int);
11438 vector signed int vec_sel (vector signed int,
11441 vector signed int vec_sel (vector signed int,
11443 vector unsigned int);
11444 vector unsigned int vec_sel (vector unsigned int,
11445 vector unsigned int,
11447 vector unsigned int vec_sel (vector unsigned int,
11448 vector unsigned int,
11449 vector unsigned int);
11450 vector bool int vec_sel (vector bool int,
11453 vector bool int vec_sel (vector bool int,
11455 vector unsigned int);
11456 vector signed short vec_sel (vector signed short,
11457 vector signed short,
11458 vector bool short);
11459 vector signed short vec_sel (vector signed short,
11460 vector signed short,
11461 vector unsigned short);
11462 vector unsigned short vec_sel (vector unsigned short,
11463 vector unsigned short,
11464 vector bool short);
11465 vector unsigned short vec_sel (vector unsigned short,
11466 vector unsigned short,
11467 vector unsigned short);
11468 vector bool short vec_sel (vector bool short,
11470 vector bool short);
11471 vector bool short vec_sel (vector bool short,
11473 vector unsigned short);
11474 vector signed char vec_sel (vector signed char,
11475 vector signed char,
11477 vector signed char vec_sel (vector signed char,
11478 vector signed char,
11479 vector unsigned char);
11480 vector unsigned char vec_sel (vector unsigned char,
11481 vector unsigned char,
11483 vector unsigned char vec_sel (vector unsigned char,
11484 vector unsigned char,
11485 vector unsigned char);
11486 vector bool char vec_sel (vector bool char,
11489 vector bool char vec_sel (vector bool char,
11491 vector unsigned char);
11493 vector signed char vec_sl (vector signed char,
11494 vector unsigned char);
11495 vector unsigned char vec_sl (vector unsigned char,
11496 vector unsigned char);
11497 vector signed short vec_sl (vector signed short, vector unsigned short);
11498 vector unsigned short vec_sl (vector unsigned short,
11499 vector unsigned short);
11500 vector signed int vec_sl (vector signed int, vector unsigned int);
11501 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
11503 vector signed int vec_vslw (vector signed int, vector unsigned int);
11504 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
11506 vector signed short vec_vslh (vector signed short,
11507 vector unsigned short);
11508 vector unsigned short vec_vslh (vector unsigned short,
11509 vector unsigned short);
11511 vector signed char vec_vslb (vector signed char, vector unsigned char);
11512 vector unsigned char vec_vslb (vector unsigned char,
11513 vector unsigned char);
11515 vector float vec_sld (vector float, vector float, const int);
11516 vector signed int vec_sld (vector signed int,
11519 vector unsigned int vec_sld (vector unsigned int,
11520 vector unsigned int,
11522 vector bool int vec_sld (vector bool int,
11525 vector signed short vec_sld (vector signed short,
11526 vector signed short,
11528 vector unsigned short vec_sld (vector unsigned short,
11529 vector unsigned short,
11531 vector bool short vec_sld (vector bool short,
11534 vector pixel vec_sld (vector pixel,
11537 vector signed char vec_sld (vector signed char,
11538 vector signed char,
11540 vector unsigned char vec_sld (vector unsigned char,
11541 vector unsigned char,
11543 vector bool char vec_sld (vector bool char,
11547 vector signed int vec_sll (vector signed int,
11548 vector unsigned int);
11549 vector signed int vec_sll (vector signed int,
11550 vector unsigned short);
11551 vector signed int vec_sll (vector signed int,
11552 vector unsigned char);
11553 vector unsigned int vec_sll (vector unsigned int,
11554 vector unsigned int);
11555 vector unsigned int vec_sll (vector unsigned int,
11556 vector unsigned short);
11557 vector unsigned int vec_sll (vector unsigned int,
11558 vector unsigned char);
11559 vector bool int vec_sll (vector bool int,
11560 vector unsigned int);
11561 vector bool int vec_sll (vector bool int,
11562 vector unsigned short);
11563 vector bool int vec_sll (vector bool int,
11564 vector unsigned char);
11565 vector signed short vec_sll (vector signed short,
11566 vector unsigned int);
11567 vector signed short vec_sll (vector signed short,
11568 vector unsigned short);
11569 vector signed short vec_sll (vector signed short,
11570 vector unsigned char);
11571 vector unsigned short vec_sll (vector unsigned short,
11572 vector unsigned int);
11573 vector unsigned short vec_sll (vector unsigned short,
11574 vector unsigned short);
11575 vector unsigned short vec_sll (vector unsigned short,
11576 vector unsigned char);
11577 vector bool short vec_sll (vector bool short, vector unsigned int);
11578 vector bool short vec_sll (vector bool short, vector unsigned short);
11579 vector bool short vec_sll (vector bool short, vector unsigned char);
11580 vector pixel vec_sll (vector pixel, vector unsigned int);
11581 vector pixel vec_sll (vector pixel, vector unsigned short);
11582 vector pixel vec_sll (vector pixel, vector unsigned char);
11583 vector signed char vec_sll (vector signed char, vector unsigned int);
11584 vector signed char vec_sll (vector signed char, vector unsigned short);
11585 vector signed char vec_sll (vector signed char, vector unsigned char);
11586 vector unsigned char vec_sll (vector unsigned char,
11587 vector unsigned int);
11588 vector unsigned char vec_sll (vector unsigned char,
11589 vector unsigned short);
11590 vector unsigned char vec_sll (vector unsigned char,
11591 vector unsigned char);
11592 vector bool char vec_sll (vector bool char, vector unsigned int);
11593 vector bool char vec_sll (vector bool char, vector unsigned short);
11594 vector bool char vec_sll (vector bool char, vector unsigned char);
11596 vector float vec_slo (vector float, vector signed char);
11597 vector float vec_slo (vector float, vector unsigned char);
11598 vector signed int vec_slo (vector signed int, vector signed char);
11599 vector signed int vec_slo (vector signed int, vector unsigned char);
11600 vector unsigned int vec_slo (vector unsigned int, vector signed char);
11601 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
11602 vector signed short vec_slo (vector signed short, vector signed char);
11603 vector signed short vec_slo (vector signed short, vector unsigned char);
11604 vector unsigned short vec_slo (vector unsigned short,
11605 vector signed char);
11606 vector unsigned short vec_slo (vector unsigned short,
11607 vector unsigned char);
11608 vector pixel vec_slo (vector pixel, vector signed char);
11609 vector pixel vec_slo (vector pixel, vector unsigned char);
11610 vector signed char vec_slo (vector signed char, vector signed char);
11611 vector signed char vec_slo (vector signed char, vector unsigned char);
11612 vector unsigned char vec_slo (vector unsigned char, vector signed char);
11613 vector unsigned char vec_slo (vector unsigned char,
11614 vector unsigned char);
11616 vector signed char vec_splat (vector signed char, const int);
11617 vector unsigned char vec_splat (vector unsigned char, const int);
11618 vector bool char vec_splat (vector bool char, const int);
11619 vector signed short vec_splat (vector signed short, const int);
11620 vector unsigned short vec_splat (vector unsigned short, const int);
11621 vector bool short vec_splat (vector bool short, const int);
11622 vector pixel vec_splat (vector pixel, const int);
11623 vector float vec_splat (vector float, const int);
11624 vector signed int vec_splat (vector signed int, const int);
11625 vector unsigned int vec_splat (vector unsigned int, const int);
11626 vector bool int vec_splat (vector bool int, const int);
11628 vector float vec_vspltw (vector float, const int);
11629 vector signed int vec_vspltw (vector signed int, const int);
11630 vector unsigned int vec_vspltw (vector unsigned int, const int);
11631 vector bool int vec_vspltw (vector bool int, const int);
11633 vector bool short vec_vsplth (vector bool short, const int);
11634 vector signed short vec_vsplth (vector signed short, const int);
11635 vector unsigned short vec_vsplth (vector unsigned short, const int);
11636 vector pixel vec_vsplth (vector pixel, const int);
11638 vector signed char vec_vspltb (vector signed char, const int);
11639 vector unsigned char vec_vspltb (vector unsigned char, const int);
11640 vector bool char vec_vspltb (vector bool char, const int);
11642 vector signed char vec_splat_s8 (const int);
11644 vector signed short vec_splat_s16 (const int);
11646 vector signed int vec_splat_s32 (const int);
11648 vector unsigned char vec_splat_u8 (const int);
11650 vector unsigned short vec_splat_u16 (const int);
11652 vector unsigned int vec_splat_u32 (const int);
11654 vector signed char vec_sr (vector signed char, vector unsigned char);
11655 vector unsigned char vec_sr (vector unsigned char,
11656 vector unsigned char);
11657 vector signed short vec_sr (vector signed short,
11658 vector unsigned short);
11659 vector unsigned short vec_sr (vector unsigned short,
11660 vector unsigned short);
11661 vector signed int vec_sr (vector signed int, vector unsigned int);
11662 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
11664 vector signed int vec_vsrw (vector signed int, vector unsigned int);
11665 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
11667 vector signed short vec_vsrh (vector signed short,
11668 vector unsigned short);
11669 vector unsigned short vec_vsrh (vector unsigned short,
11670 vector unsigned short);
11672 vector signed char vec_vsrb (vector signed char, vector unsigned char);
11673 vector unsigned char vec_vsrb (vector unsigned char,
11674 vector unsigned char);
11676 vector signed char vec_sra (vector signed char, vector unsigned char);
11677 vector unsigned char vec_sra (vector unsigned char,
11678 vector unsigned char);
11679 vector signed short vec_sra (vector signed short,
11680 vector unsigned short);
11681 vector unsigned short vec_sra (vector unsigned short,
11682 vector unsigned short);
11683 vector signed int vec_sra (vector signed int, vector unsigned int);
11684 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
11686 vector signed int vec_vsraw (vector signed int, vector unsigned int);
11687 vector unsigned int vec_vsraw (vector unsigned int,
11688 vector unsigned int);
11690 vector signed short vec_vsrah (vector signed short,
11691 vector unsigned short);
11692 vector unsigned short vec_vsrah (vector unsigned short,
11693 vector unsigned short);
11695 vector signed char vec_vsrab (vector signed char, vector unsigned char);
11696 vector unsigned char vec_vsrab (vector unsigned char,
11697 vector unsigned char);
11699 vector signed int vec_srl (vector signed int, vector unsigned int);
11700 vector signed int vec_srl (vector signed int, vector unsigned short);
11701 vector signed int vec_srl (vector signed int, vector unsigned char);
11702 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
11703 vector unsigned int vec_srl (vector unsigned int,
11704 vector unsigned short);
11705 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
11706 vector bool int vec_srl (vector bool int, vector unsigned int);
11707 vector bool int vec_srl (vector bool int, vector unsigned short);
11708 vector bool int vec_srl (vector bool int, vector unsigned char);
11709 vector signed short vec_srl (vector signed short, vector unsigned int);
11710 vector signed short vec_srl (vector signed short,
11711 vector unsigned short);
11712 vector signed short vec_srl (vector signed short, vector unsigned char);
11713 vector unsigned short vec_srl (vector unsigned short,
11714 vector unsigned int);
11715 vector unsigned short vec_srl (vector unsigned short,
11716 vector unsigned short);
11717 vector unsigned short vec_srl (vector unsigned short,
11718 vector unsigned char);
11719 vector bool short vec_srl (vector bool short, vector unsigned int);
11720 vector bool short vec_srl (vector bool short, vector unsigned short);
11721 vector bool short vec_srl (vector bool short, vector unsigned char);
11722 vector pixel vec_srl (vector pixel, vector unsigned int);
11723 vector pixel vec_srl (vector pixel, vector unsigned short);
11724 vector pixel vec_srl (vector pixel, vector unsigned char);
11725 vector signed char vec_srl (vector signed char, vector unsigned int);
11726 vector signed char vec_srl (vector signed char, vector unsigned short);
11727 vector signed char vec_srl (vector signed char, vector unsigned char);
11728 vector unsigned char vec_srl (vector unsigned char,
11729 vector unsigned int);
11730 vector unsigned char vec_srl (vector unsigned char,
11731 vector unsigned short);
11732 vector unsigned char vec_srl (vector unsigned char,
11733 vector unsigned char);
11734 vector bool char vec_srl (vector bool char, vector unsigned int);
11735 vector bool char vec_srl (vector bool char, vector unsigned short);
11736 vector bool char vec_srl (vector bool char, vector unsigned char);
11738 vector float vec_sro (vector float, vector signed char);
11739 vector float vec_sro (vector float, vector unsigned char);
11740 vector signed int vec_sro (vector signed int, vector signed char);
11741 vector signed int vec_sro (vector signed int, vector unsigned char);
11742 vector unsigned int vec_sro (vector unsigned int, vector signed char);
11743 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
11744 vector signed short vec_sro (vector signed short, vector signed char);
11745 vector signed short vec_sro (vector signed short, vector unsigned char);
11746 vector unsigned short vec_sro (vector unsigned short,
11747 vector signed char);
11748 vector unsigned short vec_sro (vector unsigned short,
11749 vector unsigned char);
11750 vector pixel vec_sro (vector pixel, vector signed char);
11751 vector pixel vec_sro (vector pixel, vector unsigned char);
11752 vector signed char vec_sro (vector signed char, vector signed char);
11753 vector signed char vec_sro (vector signed char, vector unsigned char);
11754 vector unsigned char vec_sro (vector unsigned char, vector signed char);
11755 vector unsigned char vec_sro (vector unsigned char,
11756 vector unsigned char);
11758 void vec_st (vector float, int, vector float *);
11759 void vec_st (vector float, int, float *);
11760 void vec_st (vector signed int, int, vector signed int *);
11761 void vec_st (vector signed int, int, int *);
11762 void vec_st (vector unsigned int, int, vector unsigned int *);
11763 void vec_st (vector unsigned int, int, unsigned int *);
11764 void vec_st (vector bool int, int, vector bool int *);
11765 void vec_st (vector bool int, int, unsigned int *);
11766 void vec_st (vector bool int, int, int *);
11767 void vec_st (vector signed short, int, vector signed short *);
11768 void vec_st (vector signed short, int, short *);
11769 void vec_st (vector unsigned short, int, vector unsigned short *);
11770 void vec_st (vector unsigned short, int, unsigned short *);
11771 void vec_st (vector bool short, int, vector bool short *);
11772 void vec_st (vector bool short, int, unsigned short *);
11773 void vec_st (vector pixel, int, vector pixel *);
11774 void vec_st (vector pixel, int, unsigned short *);
11775 void vec_st (vector pixel, int, short *);
11776 void vec_st (vector bool short, int, short *);
11777 void vec_st (vector signed char, int, vector signed char *);
11778 void vec_st (vector signed char, int, signed char *);
11779 void vec_st (vector unsigned char, int, vector unsigned char *);
11780 void vec_st (vector unsigned char, int, unsigned char *);
11781 void vec_st (vector bool char, int, vector bool char *);
11782 void vec_st (vector bool char, int, unsigned char *);
11783 void vec_st (vector bool char, int, signed char *);
11785 void vec_ste (vector signed char, int, signed char *);
11786 void vec_ste (vector unsigned char, int, unsigned char *);
11787 void vec_ste (vector bool char, int, signed char *);
11788 void vec_ste (vector bool char, int, unsigned char *);
11789 void vec_ste (vector signed short, int, short *);
11790 void vec_ste (vector unsigned short, int, unsigned short *);
11791 void vec_ste (vector bool short, int, short *);
11792 void vec_ste (vector bool short, int, unsigned short *);
11793 void vec_ste (vector pixel, int, short *);
11794 void vec_ste (vector pixel, int, unsigned short *);
11795 void vec_ste (vector float, int, float *);
11796 void vec_ste (vector signed int, int, int *);
11797 void vec_ste (vector unsigned int, int, unsigned int *);
11798 void vec_ste (vector bool int, int, int *);
11799 void vec_ste (vector bool int, int, unsigned int *);
11801 void vec_stvewx (vector float, int, float *);
11802 void vec_stvewx (vector signed int, int, int *);
11803 void vec_stvewx (vector unsigned int, int, unsigned int *);
11804 void vec_stvewx (vector bool int, int, int *);
11805 void vec_stvewx (vector bool int, int, unsigned int *);
11807 void vec_stvehx (vector signed short, int, short *);
11808 void vec_stvehx (vector unsigned short, int, unsigned short *);
11809 void vec_stvehx (vector bool short, int, short *);
11810 void vec_stvehx (vector bool short, int, unsigned short *);
11811 void vec_stvehx (vector pixel, int, short *);
11812 void vec_stvehx (vector pixel, int, unsigned short *);
11814 void vec_stvebx (vector signed char, int, signed char *);
11815 void vec_stvebx (vector unsigned char, int, unsigned char *);
11816 void vec_stvebx (vector bool char, int, signed char *);
11817 void vec_stvebx (vector bool char, int, unsigned char *);
11819 void vec_stl (vector float, int, vector float *);
11820 void vec_stl (vector float, int, float *);
11821 void vec_stl (vector signed int, int, vector signed int *);
11822 void vec_stl (vector signed int, int, int *);
11823 void vec_stl (vector unsigned int, int, vector unsigned int *);
11824 void vec_stl (vector unsigned int, int, unsigned int *);
11825 void vec_stl (vector bool int, int, vector bool int *);
11826 void vec_stl (vector bool int, int, unsigned int *);
11827 void vec_stl (vector bool int, int, int *);
11828 void vec_stl (vector signed short, int, vector signed short *);
11829 void vec_stl (vector signed short, int, short *);
11830 void vec_stl (vector unsigned short, int, vector unsigned short *);
11831 void vec_stl (vector unsigned short, int, unsigned short *);
11832 void vec_stl (vector bool short, int, vector bool short *);
11833 void vec_stl (vector bool short, int, unsigned short *);
11834 void vec_stl (vector bool short, int, short *);
11835 void vec_stl (vector pixel, int, vector pixel *);
11836 void vec_stl (vector pixel, int, unsigned short *);
11837 void vec_stl (vector pixel, int, short *);
11838 void vec_stl (vector signed char, int, vector signed char *);
11839 void vec_stl (vector signed char, int, signed char *);
11840 void vec_stl (vector unsigned char, int, vector unsigned char *);
11841 void vec_stl (vector unsigned char, int, unsigned char *);
11842 void vec_stl (vector bool char, int, vector bool char *);
11843 void vec_stl (vector bool char, int, unsigned char *);
11844 void vec_stl (vector bool char, int, signed char *);
11846 vector signed char vec_sub (vector bool char, vector signed char);
11847 vector signed char vec_sub (vector signed char, vector bool char);
11848 vector signed char vec_sub (vector signed char, vector signed char);
11849 vector unsigned char vec_sub (vector bool char, vector unsigned char);
11850 vector unsigned char vec_sub (vector unsigned char, vector bool char);
11851 vector unsigned char vec_sub (vector unsigned char,
11852 vector unsigned char);
11853 vector signed short vec_sub (vector bool short, vector signed short);
11854 vector signed short vec_sub (vector signed short, vector bool short);
11855 vector signed short vec_sub (vector signed short, vector signed short);
11856 vector unsigned short vec_sub (vector bool short,
11857 vector unsigned short);
11858 vector unsigned short vec_sub (vector unsigned short,
11859 vector bool short);
11860 vector unsigned short vec_sub (vector unsigned short,
11861 vector unsigned short);
11862 vector signed int vec_sub (vector bool int, vector signed int);
11863 vector signed int vec_sub (vector signed int, vector bool int);
11864 vector signed int vec_sub (vector signed int, vector signed int);
11865 vector unsigned int vec_sub (vector bool int, vector unsigned int);
11866 vector unsigned int vec_sub (vector unsigned int, vector bool int);
11867 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
11868 vector float vec_sub (vector float, vector float);
11870 vector float vec_vsubfp (vector float, vector float);
11872 vector signed int vec_vsubuwm (vector bool int, vector signed int);
11873 vector signed int vec_vsubuwm (vector signed int, vector bool int);
11874 vector signed int vec_vsubuwm (vector signed int, vector signed int);
11875 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
11876 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
11877 vector unsigned int vec_vsubuwm (vector unsigned int,
11878 vector unsigned int);
11880 vector signed short vec_vsubuhm (vector bool short,
11881 vector signed short);
11882 vector signed short vec_vsubuhm (vector signed short,
11883 vector bool short);
11884 vector signed short vec_vsubuhm (vector signed short,
11885 vector signed short);
11886 vector unsigned short vec_vsubuhm (vector bool short,
11887 vector unsigned short);
11888 vector unsigned short vec_vsubuhm (vector unsigned short,
11889 vector bool short);
11890 vector unsigned short vec_vsubuhm (vector unsigned short,
11891 vector unsigned short);
11893 vector signed char vec_vsububm (vector bool char, vector signed char);
11894 vector signed char vec_vsububm (vector signed char, vector bool char);
11895 vector signed char vec_vsububm (vector signed char, vector signed char);
11896 vector unsigned char vec_vsububm (vector bool char,
11897 vector unsigned char);
11898 vector unsigned char vec_vsububm (vector unsigned char,
11900 vector unsigned char vec_vsububm (vector unsigned char,
11901 vector unsigned char);
11903 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
11905 vector unsigned char vec_subs (vector bool char, vector unsigned char);
11906 vector unsigned char vec_subs (vector unsigned char, vector bool char);
11907 vector unsigned char vec_subs (vector unsigned char,
11908 vector unsigned char);
11909 vector signed char vec_subs (vector bool char, vector signed char);
11910 vector signed char vec_subs (vector signed char, vector bool char);
11911 vector signed char vec_subs (vector signed char, vector signed char);
11912 vector unsigned short vec_subs (vector bool short,
11913 vector unsigned short);
11914 vector unsigned short vec_subs (vector unsigned short,
11915 vector bool short);
11916 vector unsigned short vec_subs (vector unsigned short,
11917 vector unsigned short);
11918 vector signed short vec_subs (vector bool short, vector signed short);
11919 vector signed short vec_subs (vector signed short, vector bool short);
11920 vector signed short vec_subs (vector signed short, vector signed short);
11921 vector unsigned int vec_subs (vector bool int, vector unsigned int);
11922 vector unsigned int vec_subs (vector unsigned int, vector bool int);
11923 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
11924 vector signed int vec_subs (vector bool int, vector signed int);
11925 vector signed int vec_subs (vector signed int, vector bool int);
11926 vector signed int vec_subs (vector signed int, vector signed int);
11928 vector signed int vec_vsubsws (vector bool int, vector signed int);
11929 vector signed int vec_vsubsws (vector signed int, vector bool int);
11930 vector signed int vec_vsubsws (vector signed int, vector signed int);
11932 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
11933 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
11934 vector unsigned int vec_vsubuws (vector unsigned int,
11935 vector unsigned int);
11937 vector signed short vec_vsubshs (vector bool short,
11938 vector signed short);
11939 vector signed short vec_vsubshs (vector signed short,
11940 vector bool short);
11941 vector signed short vec_vsubshs (vector signed short,
11942 vector signed short);
11944 vector unsigned short vec_vsubuhs (vector bool short,
11945 vector unsigned short);
11946 vector unsigned short vec_vsubuhs (vector unsigned short,
11947 vector bool short);
11948 vector unsigned short vec_vsubuhs (vector unsigned short,
11949 vector unsigned short);
11951 vector signed char vec_vsubsbs (vector bool char, vector signed char);
11952 vector signed char vec_vsubsbs (vector signed char, vector bool char);
11953 vector signed char vec_vsubsbs (vector signed char, vector signed char);
11955 vector unsigned char vec_vsububs (vector bool char,
11956 vector unsigned char);
11957 vector unsigned char vec_vsububs (vector unsigned char,
11959 vector unsigned char vec_vsububs (vector unsigned char,
11960 vector unsigned char);
11962 vector unsigned int vec_sum4s (vector unsigned char,
11963 vector unsigned int);
11964 vector signed int vec_sum4s (vector signed char, vector signed int);
11965 vector signed int vec_sum4s (vector signed short, vector signed int);
11967 vector signed int vec_vsum4shs (vector signed short, vector signed int);
11969 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
11971 vector unsigned int vec_vsum4ubs (vector unsigned char,
11972 vector unsigned int);
11974 vector signed int vec_sum2s (vector signed int, vector signed int);
11976 vector signed int vec_sums (vector signed int, vector signed int);
11978 vector float vec_trunc (vector float);
11980 vector signed short vec_unpackh (vector signed char);
11981 vector bool short vec_unpackh (vector bool char);
11982 vector signed int vec_unpackh (vector signed short);
11983 vector bool int vec_unpackh (vector bool short);
11984 vector unsigned int vec_unpackh (vector pixel);
11986 vector bool int vec_vupkhsh (vector bool short);
11987 vector signed int vec_vupkhsh (vector signed short);
11989 vector unsigned int vec_vupkhpx (vector pixel);
11991 vector bool short vec_vupkhsb (vector bool char);
11992 vector signed short vec_vupkhsb (vector signed char);
11994 vector signed short vec_unpackl (vector signed char);
11995 vector bool short vec_unpackl (vector bool char);
11996 vector unsigned int vec_unpackl (vector pixel);
11997 vector signed int vec_unpackl (vector signed short);
11998 vector bool int vec_unpackl (vector bool short);
12000 vector unsigned int vec_vupklpx (vector pixel);
12002 vector bool int vec_vupklsh (vector bool short);
12003 vector signed int vec_vupklsh (vector signed short);
12005 vector bool short vec_vupklsb (vector bool char);
12006 vector signed short vec_vupklsb (vector signed char);
12008 vector float vec_xor (vector float, vector float);
12009 vector float vec_xor (vector float, vector bool int);
12010 vector float vec_xor (vector bool int, vector float);
12011 vector bool int vec_xor (vector bool int, vector bool int);
12012 vector signed int vec_xor (vector bool int, vector signed int);
12013 vector signed int vec_xor (vector signed int, vector bool int);
12014 vector signed int vec_xor (vector signed int, vector signed int);
12015 vector unsigned int vec_xor (vector bool int, vector unsigned int);
12016 vector unsigned int vec_xor (vector unsigned int, vector bool int);
12017 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
12018 vector bool short vec_xor (vector bool short, vector bool short);
12019 vector signed short vec_xor (vector bool short, vector signed short);
12020 vector signed short vec_xor (vector signed short, vector bool short);
12021 vector signed short vec_xor (vector signed short, vector signed short);
12022 vector unsigned short vec_xor (vector bool short,
12023 vector unsigned short);
12024 vector unsigned short vec_xor (vector unsigned short,
12025 vector bool short);
12026 vector unsigned short vec_xor (vector unsigned short,
12027 vector unsigned short);
12028 vector signed char vec_xor (vector bool char, vector signed char);
12029 vector bool char vec_xor (vector bool char, vector bool char);
12030 vector signed char vec_xor (vector signed char, vector bool char);
12031 vector signed char vec_xor (vector signed char, vector signed char);
12032 vector unsigned char vec_xor (vector bool char, vector unsigned char);
12033 vector unsigned char vec_xor (vector unsigned char, vector bool char);
12034 vector unsigned char vec_xor (vector unsigned char,
12035 vector unsigned char);
12037 int vec_all_eq (vector signed char, vector bool char);
12038 int vec_all_eq (vector signed char, vector signed char);
12039 int vec_all_eq (vector unsigned char, vector bool char);
12040 int vec_all_eq (vector unsigned char, vector unsigned char);
12041 int vec_all_eq (vector bool char, vector bool char);
12042 int vec_all_eq (vector bool char, vector unsigned char);
12043 int vec_all_eq (vector bool char, vector signed char);
12044 int vec_all_eq (vector signed short, vector bool short);
12045 int vec_all_eq (vector signed short, vector signed short);
12046 int vec_all_eq (vector unsigned short, vector bool short);
12047 int vec_all_eq (vector unsigned short, vector unsigned short);
12048 int vec_all_eq (vector bool short, vector bool short);
12049 int vec_all_eq (vector bool short, vector unsigned short);
12050 int vec_all_eq (vector bool short, vector signed short);
12051 int vec_all_eq (vector pixel, vector pixel);
12052 int vec_all_eq (vector signed int, vector bool int);
12053 int vec_all_eq (vector signed int, vector signed int);
12054 int vec_all_eq (vector unsigned int, vector bool int);
12055 int vec_all_eq (vector unsigned int, vector unsigned int);
12056 int vec_all_eq (vector bool int, vector bool int);
12057 int vec_all_eq (vector bool int, vector unsigned int);
12058 int vec_all_eq (vector bool int, vector signed int);
12059 int vec_all_eq (vector float, vector float);
12061 int vec_all_ge (vector bool char, vector unsigned char);
12062 int vec_all_ge (vector unsigned char, vector bool char);
12063 int vec_all_ge (vector unsigned char, vector unsigned char);
12064 int vec_all_ge (vector bool char, vector signed char);
12065 int vec_all_ge (vector signed char, vector bool char);
12066 int vec_all_ge (vector signed char, vector signed char);
12067 int vec_all_ge (vector bool short, vector unsigned short);
12068 int vec_all_ge (vector unsigned short, vector bool short);
12069 int vec_all_ge (vector unsigned short, vector unsigned short);
12070 int vec_all_ge (vector signed short, vector signed short);
12071 int vec_all_ge (vector bool short, vector signed short);
12072 int vec_all_ge (vector signed short, vector bool short);
12073 int vec_all_ge (vector bool int, vector unsigned int);
12074 int vec_all_ge (vector unsigned int, vector bool int);
12075 int vec_all_ge (vector unsigned int, vector unsigned int);
12076 int vec_all_ge (vector bool int, vector signed int);
12077 int vec_all_ge (vector signed int, vector bool int);
12078 int vec_all_ge (vector signed int, vector signed int);
12079 int vec_all_ge (vector float, vector float);
12081 int vec_all_gt (vector bool char, vector unsigned char);
12082 int vec_all_gt (vector unsigned char, vector bool char);
12083 int vec_all_gt (vector unsigned char, vector unsigned char);
12084 int vec_all_gt (vector bool char, vector signed char);
12085 int vec_all_gt (vector signed char, vector bool char);
12086 int vec_all_gt (vector signed char, vector signed char);
12087 int vec_all_gt (vector bool short, vector unsigned short);
12088 int vec_all_gt (vector unsigned short, vector bool short);
12089 int vec_all_gt (vector unsigned short, vector unsigned short);
12090 int vec_all_gt (vector bool short, vector signed short);
12091 int vec_all_gt (vector signed short, vector bool short);
12092 int vec_all_gt (vector signed short, vector signed short);
12093 int vec_all_gt (vector bool int, vector unsigned int);
12094 int vec_all_gt (vector unsigned int, vector bool int);
12095 int vec_all_gt (vector unsigned int, vector unsigned int);
12096 int vec_all_gt (vector bool int, vector signed int);
12097 int vec_all_gt (vector signed int, vector bool int);
12098 int vec_all_gt (vector signed int, vector signed int);
12099 int vec_all_gt (vector float, vector float);
12101 int vec_all_in (vector float, vector float);
12103 int vec_all_le (vector bool char, vector unsigned char);
12104 int vec_all_le (vector unsigned char, vector bool char);
12105 int vec_all_le (vector unsigned char, vector unsigned char);
12106 int vec_all_le (vector bool char, vector signed char);
12107 int vec_all_le (vector signed char, vector bool char);
12108 int vec_all_le (vector signed char, vector signed char);
12109 int vec_all_le (vector bool short, vector unsigned short);
12110 int vec_all_le (vector unsigned short, vector bool short);
12111 int vec_all_le (vector unsigned short, vector unsigned short);
12112 int vec_all_le (vector bool short, vector signed short);
12113 int vec_all_le (vector signed short, vector bool short);
12114 int vec_all_le (vector signed short, vector signed short);
12115 int vec_all_le (vector bool int, vector unsigned int);
12116 int vec_all_le (vector unsigned int, vector bool int);
12117 int vec_all_le (vector unsigned int, vector unsigned int);
12118 int vec_all_le (vector bool int, vector signed int);
12119 int vec_all_le (vector signed int, vector bool int);
12120 int vec_all_le (vector signed int, vector signed int);
12121 int vec_all_le (vector float, vector float);
12123 int vec_all_lt (vector bool char, vector unsigned char);
12124 int vec_all_lt (vector unsigned char, vector bool char);
12125 int vec_all_lt (vector unsigned char, vector unsigned char);
12126 int vec_all_lt (vector bool char, vector signed char);
12127 int vec_all_lt (vector signed char, vector bool char);
12128 int vec_all_lt (vector signed char, vector signed char);
12129 int vec_all_lt (vector bool short, vector unsigned short);
12130 int vec_all_lt (vector unsigned short, vector bool short);
12131 int vec_all_lt (vector unsigned short, vector unsigned short);
12132 int vec_all_lt (vector bool short, vector signed short);
12133 int vec_all_lt (vector signed short, vector bool short);
12134 int vec_all_lt (vector signed short, vector signed short);
12135 int vec_all_lt (vector bool int, vector unsigned int);
12136 int vec_all_lt (vector unsigned int, vector bool int);
12137 int vec_all_lt (vector unsigned int, vector unsigned int);
12138 int vec_all_lt (vector bool int, vector signed int);
12139 int vec_all_lt (vector signed int, vector bool int);
12140 int vec_all_lt (vector signed int, vector signed int);
12141 int vec_all_lt (vector float, vector float);
12143 int vec_all_nan (vector float);
12145 int vec_all_ne (vector signed char, vector bool char);
12146 int vec_all_ne (vector signed char, vector signed char);
12147 int vec_all_ne (vector unsigned char, vector bool char);
12148 int vec_all_ne (vector unsigned char, vector unsigned char);
12149 int vec_all_ne (vector bool char, vector bool char);
12150 int vec_all_ne (vector bool char, vector unsigned char);
12151 int vec_all_ne (vector bool char, vector signed char);
12152 int vec_all_ne (vector signed short, vector bool short);
12153 int vec_all_ne (vector signed short, vector signed short);
12154 int vec_all_ne (vector unsigned short, vector bool short);
12155 int vec_all_ne (vector unsigned short, vector unsigned short);
12156 int vec_all_ne (vector bool short, vector bool short);
12157 int vec_all_ne (vector bool short, vector unsigned short);
12158 int vec_all_ne (vector bool short, vector signed short);
12159 int vec_all_ne (vector pixel, vector pixel);
12160 int vec_all_ne (vector signed int, vector bool int);
12161 int vec_all_ne (vector signed int, vector signed int);
12162 int vec_all_ne (vector unsigned int, vector bool int);
12163 int vec_all_ne (vector unsigned int, vector unsigned int);
12164 int vec_all_ne (vector bool int, vector bool int);
12165 int vec_all_ne (vector bool int, vector unsigned int);
12166 int vec_all_ne (vector bool int, vector signed int);
12167 int vec_all_ne (vector float, vector float);
12169 int vec_all_nge (vector float, vector float);
12171 int vec_all_ngt (vector float, vector float);
12173 int vec_all_nle (vector float, vector float);
12175 int vec_all_nlt (vector float, vector float);
12177 int vec_all_numeric (vector float);
12179 int vec_any_eq (vector signed char, vector bool char);
12180 int vec_any_eq (vector signed char, vector signed char);
12181 int vec_any_eq (vector unsigned char, vector bool char);
12182 int vec_any_eq (vector unsigned char, vector unsigned char);
12183 int vec_any_eq (vector bool char, vector bool char);
12184 int vec_any_eq (vector bool char, vector unsigned char);
12185 int vec_any_eq (vector bool char, vector signed char);
12186 int vec_any_eq (vector signed short, vector bool short);
12187 int vec_any_eq (vector signed short, vector signed short);
12188 int vec_any_eq (vector unsigned short, vector bool short);
12189 int vec_any_eq (vector unsigned short, vector unsigned short);
12190 int vec_any_eq (vector bool short, vector bool short);
12191 int vec_any_eq (vector bool short, vector unsigned short);
12192 int vec_any_eq (vector bool short, vector signed short);
12193 int vec_any_eq (vector pixel, vector pixel);
12194 int vec_any_eq (vector signed int, vector bool int);
12195 int vec_any_eq (vector signed int, vector signed int);
12196 int vec_any_eq (vector unsigned int, vector bool int);
12197 int vec_any_eq (vector unsigned int, vector unsigned int);
12198 int vec_any_eq (vector bool int, vector bool int);
12199 int vec_any_eq (vector bool int, vector unsigned int);
12200 int vec_any_eq (vector bool int, vector signed int);
12201 int vec_any_eq (vector float, vector float);
12203 int vec_any_ge (vector signed char, vector bool char);
12204 int vec_any_ge (vector unsigned char, vector bool char);
12205 int vec_any_ge (vector unsigned char, vector unsigned char);
12206 int vec_any_ge (vector signed char, vector signed char);
12207 int vec_any_ge (vector bool char, vector unsigned char);
12208 int vec_any_ge (vector bool char, vector signed char);
12209 int vec_any_ge (vector unsigned short, vector bool short);
12210 int vec_any_ge (vector unsigned short, vector unsigned short);
12211 int vec_any_ge (vector signed short, vector signed short);
12212 int vec_any_ge (vector signed short, vector bool short);
12213 int vec_any_ge (vector bool short, vector unsigned short);
12214 int vec_any_ge (vector bool short, vector signed short);
12215 int vec_any_ge (vector signed int, vector bool int);
12216 int vec_any_ge (vector unsigned int, vector bool int);
12217 int vec_any_ge (vector unsigned int, vector unsigned int);
12218 int vec_any_ge (vector signed int, vector signed int);
12219 int vec_any_ge (vector bool int, vector unsigned int);
12220 int vec_any_ge (vector bool int, vector signed int);
12221 int vec_any_ge (vector float, vector float);
12223 int vec_any_gt (vector bool char, vector unsigned char);
12224 int vec_any_gt (vector unsigned char, vector bool char);
12225 int vec_any_gt (vector unsigned char, vector unsigned char);
12226 int vec_any_gt (vector bool char, vector signed char);
12227 int vec_any_gt (vector signed char, vector bool char);
12228 int vec_any_gt (vector signed char, vector signed char);
12229 int vec_any_gt (vector bool short, vector unsigned short);
12230 int vec_any_gt (vector unsigned short, vector bool short);
12231 int vec_any_gt (vector unsigned short, vector unsigned short);
12232 int vec_any_gt (vector bool short, vector signed short);
12233 int vec_any_gt (vector signed short, vector bool short);
12234 int vec_any_gt (vector signed short, vector signed short);
12235 int vec_any_gt (vector bool int, vector unsigned int);
12236 int vec_any_gt (vector unsigned int, vector bool int);
12237 int vec_any_gt (vector unsigned int, vector unsigned int);
12238 int vec_any_gt (vector bool int, vector signed int);
12239 int vec_any_gt (vector signed int, vector bool int);
12240 int vec_any_gt (vector signed int, vector signed int);
12241 int vec_any_gt (vector float, vector float);
12243 int vec_any_le (vector bool char, vector unsigned char);
12244 int vec_any_le (vector unsigned char, vector bool char);
12245 int vec_any_le (vector unsigned char, vector unsigned char);
12246 int vec_any_le (vector bool char, vector signed char);
12247 int vec_any_le (vector signed char, vector bool char);
12248 int vec_any_le (vector signed char, vector signed char);
12249 int vec_any_le (vector bool short, vector unsigned short);
12250 int vec_any_le (vector unsigned short, vector bool short);
12251 int vec_any_le (vector unsigned short, vector unsigned short);
12252 int vec_any_le (vector bool short, vector signed short);
12253 int vec_any_le (vector signed short, vector bool short);
12254 int vec_any_le (vector signed short, vector signed short);
12255 int vec_any_le (vector bool int, vector unsigned int);
12256 int vec_any_le (vector unsigned int, vector bool int);
12257 int vec_any_le (vector unsigned int, vector unsigned int);
12258 int vec_any_le (vector bool int, vector signed int);
12259 int vec_any_le (vector signed int, vector bool int);
12260 int vec_any_le (vector signed int, vector signed int);
12261 int vec_any_le (vector float, vector float);
12263 int vec_any_lt (vector bool char, vector unsigned char);
12264 int vec_any_lt (vector unsigned char, vector bool char);
12265 int vec_any_lt (vector unsigned char, vector unsigned char);
12266 int vec_any_lt (vector bool char, vector signed char);
12267 int vec_any_lt (vector signed char, vector bool char);
12268 int vec_any_lt (vector signed char, vector signed char);
12269 int vec_any_lt (vector bool short, vector unsigned short);
12270 int vec_any_lt (vector unsigned short, vector bool short);
12271 int vec_any_lt (vector unsigned short, vector unsigned short);
12272 int vec_any_lt (vector bool short, vector signed short);
12273 int vec_any_lt (vector signed short, vector bool short);
12274 int vec_any_lt (vector signed short, vector signed short);
12275 int vec_any_lt (vector bool int, vector unsigned int);
12276 int vec_any_lt (vector unsigned int, vector bool int);
12277 int vec_any_lt (vector unsigned int, vector unsigned int);
12278 int vec_any_lt (vector bool int, vector signed int);
12279 int vec_any_lt (vector signed int, vector bool int);
12280 int vec_any_lt (vector signed int, vector signed int);
12281 int vec_any_lt (vector float, vector float);
12283 int vec_any_nan (vector float);
12285 int vec_any_ne (vector signed char, vector bool char);
12286 int vec_any_ne (vector signed char, vector signed char);
12287 int vec_any_ne (vector unsigned char, vector bool char);
12288 int vec_any_ne (vector unsigned char, vector unsigned char);
12289 int vec_any_ne (vector bool char, vector bool char);
12290 int vec_any_ne (vector bool char, vector unsigned char);
12291 int vec_any_ne (vector bool char, vector signed char);
12292 int vec_any_ne (vector signed short, vector bool short);
12293 int vec_any_ne (vector signed short, vector signed short);
12294 int vec_any_ne (vector unsigned short, vector bool short);
12295 int vec_any_ne (vector unsigned short, vector unsigned short);
12296 int vec_any_ne (vector bool short, vector bool short);
12297 int vec_any_ne (vector bool short, vector unsigned short);
12298 int vec_any_ne (vector bool short, vector signed short);
12299 int vec_any_ne (vector pixel, vector pixel);
12300 int vec_any_ne (vector signed int, vector bool int);
12301 int vec_any_ne (vector signed int, vector signed int);
12302 int vec_any_ne (vector unsigned int, vector bool int);
12303 int vec_any_ne (vector unsigned int, vector unsigned int);
12304 int vec_any_ne (vector bool int, vector bool int);
12305 int vec_any_ne (vector bool int, vector unsigned int);
12306 int vec_any_ne (vector bool int, vector signed int);
12307 int vec_any_ne (vector float, vector float);
12309 int vec_any_nge (vector float, vector float);
12311 int vec_any_ngt (vector float, vector float);
12313 int vec_any_nle (vector float, vector float);
12315 int vec_any_nlt (vector float, vector float);
12317 int vec_any_numeric (vector float);
12319 int vec_any_out (vector float, vector float);
12322 If the vector/scalar (VSX) instruction set is available, the following
12323 additional functions are available:
12326 vector double vec_abs (vector double);
12327 vector double vec_add (vector double, vector double);
12328 vector double vec_and (vector double, vector double);
12329 vector double vec_and (vector double, vector bool long);
12330 vector double vec_and (vector bool long, vector double);
12331 vector double vec_andc (vector double, vector double);
12332 vector double vec_andc (vector double, vector bool long);
12333 vector double vec_andc (vector bool long, vector double);
12334 vector double vec_ceil (vector double);
12335 vector bool long vec_cmpeq (vector double, vector double);
12336 vector bool long vec_cmpge (vector double, vector double);
12337 vector bool long vec_cmpgt (vector double, vector double);
12338 vector bool long vec_cmple (vector double, vector double);
12339 vector bool long vec_cmplt (vector double, vector double);
12340 vector float vec_div (vector float, vector float);
12341 vector double vec_div (vector double, vector double);
12342 vector double vec_floor (vector double);
12343 vector double vec_madd (vector double, vector double, vector double);
12344 vector double vec_max (vector double, vector double);
12345 vector double vec_min (vector double, vector double);
12346 vector float vec_msub (vector float, vector float, vector float);
12347 vector double vec_msub (vector double, vector double, vector double);
12348 vector float vec_mul (vector float, vector float);
12349 vector double vec_mul (vector double, vector double);
12350 vector float vec_nearbyint (vector float);
12351 vector double vec_nearbyint (vector double);
12352 vector float vec_nmadd (vector float, vector float, vector float);
12353 vector double vec_nmadd (vector double, vector double, vector double);
12354 vector double vec_nmsub (vector double, vector double, vector double);
12355 vector double vec_nor (vector double, vector double);
12356 vector double vec_or (vector double, vector double);
12357 vector double vec_or (vector double, vector bool long);
12358 vector double vec_or (vector bool long, vector double);
12359 vector double vec_perm (vector double,
12361 vector unsigned char);
12362 vector double vec_rint (vector double);
12363 vector double vec_recip (vector double, vector double);
12364 vector double vec_rsqrt (vector double);
12365 vector double vec_rsqrte (vector double);
12366 vector double vec_sel (vector double, vector double, vector bool long);
12367 vector double vec_sel (vector double, vector double, vector unsigned long);
12368 vector double vec_sub (vector double, vector double);
12369 vector float vec_sqrt (vector float);
12370 vector double vec_sqrt (vector double);
12371 vector double vec_trunc (vector double);
12372 vector double vec_xor (vector double, vector double);
12373 vector double vec_xor (vector double, vector bool long);
12374 vector double vec_xor (vector bool long, vector double);
12375 int vec_all_eq (vector double, vector double);
12376 int vec_all_ge (vector double, vector double);
12377 int vec_all_gt (vector double, vector double);
12378 int vec_all_le (vector double, vector double);
12379 int vec_all_lt (vector double, vector double);
12380 int vec_all_nan (vector double);
12381 int vec_all_ne (vector double, vector double);
12382 int vec_all_nge (vector double, vector double);
12383 int vec_all_ngt (vector double, vector double);
12384 int vec_all_nle (vector double, vector double);
12385 int vec_all_nlt (vector double, vector double);
12386 int vec_all_numeric (vector double);
12387 int vec_any_eq (vector double, vector double);
12388 int vec_any_ge (vector double, vector double);
12389 int vec_any_gt (vector double, vector double);
12390 int vec_any_le (vector double, vector double);
12391 int vec_any_lt (vector double, vector double);
12392 int vec_any_nan (vector double);
12393 int vec_any_ne (vector double, vector double);
12394 int vec_any_nge (vector double, vector double);
12395 int vec_any_ngt (vector double, vector double);
12396 int vec_any_nle (vector double, vector double);
12397 int vec_any_nlt (vector double, vector double);
12398 int vec_any_numeric (vector double);
12401 GCC provides a few other builtins on Powerpc to access certain instructions:
12403 float __builtin_recipdivf (float, float);
12404 float __builtin_rsqrtf (float);
12405 double __builtin_recipdiv (double, double);
12406 double __builtin_rsqrt (double);
12407 long __builtin_bpermd (long, long);
12408 int __builtin_bswap16 (int);
12411 The @code{vec_rsqrt}, @code{__builtin_rsqrt}, and
12412 @code{__builtin_rsqrtf} functions generate multiple instructions to
12413 implement the reciprocal sqrt functionality using reciprocal sqrt
12414 estimate instructions.
12416 The @code{__builtin_recipdiv}, and @code{__builtin_recipdivf}
12417 functions generate multiple instructions to implement division using
12418 the reciprocal estimate instructions.
12420 @node RX Built-in Functions
12421 @subsection RX Built-in Functions
12422 GCC supports some of the RX instructions which cannot be expressed in
12423 the C programming language via the use of built-in functions. The
12424 following functions are supported:
12426 @deftypefn {Built-in Function} void __builtin_rx_brk (void)
12427 Generates the @code{brk} machine instruction.
12430 @deftypefn {Built-in Function} void __builtin_rx_clrpsw (int)
12431 Generates the @code{clrpsw} machine instruction to clear the specified
12432 bit in the processor status word.
12435 @deftypefn {Built-in Function} void __builtin_rx_int (int)
12436 Generates the @code{int} machine instruction to generate an interrupt
12437 with the specified value.
12440 @deftypefn {Built-in Function} void __builtin_rx_machi (int, int)
12441 Generates the @code{machi} machine instruction to add the result of
12442 multiplying the top 16-bits of the two arguments into the
12446 @deftypefn {Built-in Function} void __builtin_rx_maclo (int, int)
12447 Generates the @code{maclo} machine instruction to add the result of
12448 multiplying the bottom 16-bits of the two arguments into the
12452 @deftypefn {Built-in Function} void __builtin_rx_mulhi (int, int)
12453 Generates the @code{mulhi} machine instruction to place the result of
12454 multiplying the top 16-bits of the two arguments into the
12458 @deftypefn {Built-in Function} void __builtin_rx_mullo (int, int)
12459 Generates the @code{mullo} machine instruction to place the result of
12460 multiplying the bottom 16-bits of the two arguments into the
12464 @deftypefn {Built-in Function} int __builtin_rx_mvfachi (void)
12465 Generates the @code{mvfachi} machine instruction to read the top
12466 32-bits of the accumulator.
12469 @deftypefn {Built-in Function} int __builtin_rx_mvfacmi (void)
12470 Generates the @code{mvfacmi} machine instruction to read the middle
12471 32-bits of the accumulator.
12474 @deftypefn {Built-in Function} int __builtin_rx_mvfc (int)
12475 Generates the @code{mvfc} machine instruction which reads the control
12476 register specified in its argument and returns its value.
12479 @deftypefn {Built-in Function} void __builtin_rx_mvtachi (int)
12480 Generates the @code{mvtachi} machine instruction to set the top
12481 32-bits of the accumulator.
12484 @deftypefn {Built-in Function} void __builtin_rx_mvtaclo (int)
12485 Generates the @code{mvtaclo} machine instruction to set the bottom
12486 32-bits of the accumulator.
12489 @deftypefn {Built-in Function} void __builtin_rx_mvtc (int reg, int val)
12490 Generates the @code{mvtc} machine instruction which sets control
12491 register number @code{reg} to @code{val}.
12494 @deftypefn {Built-in Function} void __builtin_rx_mvtipl (int)
12495 Generates the @code{mvtipl} machine instruction set the interrupt
12499 @deftypefn {Built-in Function} void __builtin_rx_racw (int)
12500 Generates the @code{racw} machine instruction to round the accumulator
12501 according to the specified mode.
12504 @deftypefn {Built-in Function} int __builtin_rx_revw (int)
12505 Generates the @code{revw} machine instruction which swaps the bytes in
12506 the argument so that bits 0--7 now occupy bits 8--15 and vice versa,
12507 and also bits 16--23 occupy bits 24--31 and vice versa.
12510 @deftypefn {Built-in Function} void __builtin_rx_rmpa (void)
12511 Generates the @code{rmpa} machine instruction which initiates a
12512 repeated multiply and accumulate sequence.
12515 @deftypefn {Built-in Function} void __builtin_rx_round (float)
12516 Generates the @code{round} machine instruction which returns the
12517 floating point argument rounded according to the current rounding mode
12518 set in the floating point status word register.
12521 @deftypefn {Built-in Function} int __builtin_rx_sat (int)
12522 Generates the @code{sat} machine instruction which returns the
12523 saturated value of the argument.
12526 @deftypefn {Built-in Function} void __builtin_rx_setpsw (int)
12527 Generates the @code{setpsw} machine instruction to set the specified
12528 bit in the processor status word.
12531 @deftypefn {Built-in Function} void __builtin_rx_wait (void)
12532 Generates the @code{wait} machine instruction.
12535 @node SPARC VIS Built-in Functions
12536 @subsection SPARC VIS Built-in Functions
12538 GCC supports SIMD operations on the SPARC using both the generic vector
12539 extensions (@pxref{Vector Extensions}) as well as built-in functions for
12540 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
12541 switch, the VIS extension is exposed as the following built-in functions:
12544 typedef int v2si __attribute__ ((vector_size (8)));
12545 typedef short v4hi __attribute__ ((vector_size (8)));
12546 typedef short v2hi __attribute__ ((vector_size (4)));
12547 typedef char v8qi __attribute__ ((vector_size (8)));
12548 typedef char v4qi __attribute__ ((vector_size (4)));
12550 void * __builtin_vis_alignaddr (void *, long);
12551 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
12552 v2si __builtin_vis_faligndatav2si (v2si, v2si);
12553 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
12554 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
12556 v4hi __builtin_vis_fexpand (v4qi);
12558 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
12559 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
12560 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
12561 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
12562 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
12563 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
12564 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
12566 v4qi __builtin_vis_fpack16 (v4hi);
12567 v8qi __builtin_vis_fpack32 (v2si, v2si);
12568 v2hi __builtin_vis_fpackfix (v2si);
12569 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
12571 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
12574 @node SPU Built-in Functions
12575 @subsection SPU Built-in Functions
12577 GCC provides extensions for the SPU processor as described in the
12578 Sony/Toshiba/IBM SPU Language Extensions Specification, which can be
12579 found at @uref{http://cell.scei.co.jp/} or
12580 @uref{http://www.ibm.com/developerworks/power/cell/}. GCC's
12581 implementation differs in several ways.
12586 The optional extension of specifying vector constants in parentheses is
12590 A vector initializer requires no cast if the vector constant is of the
12591 same type as the variable it is initializing.
12594 If @code{signed} or @code{unsigned} is omitted, the signedness of the
12595 vector type is the default signedness of the base type. The default
12596 varies depending on the operating system, so a portable program should
12597 always specify the signedness.
12600 By default, the keyword @code{__vector} is added. The macro
12601 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
12605 GCC allows using a @code{typedef} name as the type specifier for a
12609 For C, overloaded functions are implemented with macros so the following
12613 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
12616 Since @code{spu_add} is a macro, the vector constant in the example
12617 is treated as four separate arguments. Wrap the entire argument in
12618 parentheses for this to work.
12621 The extended version of @code{__builtin_expect} is not supported.
12625 @emph{Note:} Only the interface described in the aforementioned
12626 specification is supported. Internally, GCC uses built-in functions to
12627 implement the required functionality, but these are not supported and
12628 are subject to change without notice.
12630 @node Target Format Checks
12631 @section Format Checks Specific to Particular Target Machines
12633 For some target machines, GCC supports additional options to the
12635 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
12638 * Solaris Format Checks::
12639 * Darwin Format Checks::
12642 @node Solaris Format Checks
12643 @subsection Solaris Format Checks
12645 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
12646 check. @code{cmn_err} accepts a subset of the standard @code{printf}
12647 conversions, and the two-argument @code{%b} conversion for displaying
12648 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
12650 @node Darwin Format Checks
12651 @subsection Darwin Format Checks
12653 Darwin targets support the @code{CFString} (or @code{__CFString__}) in the format
12654 attribute context. Declarations made with such attribution will be parsed for correct syntax
12655 and format argument types. However, parsing of the format string itself is currently undefined
12656 and will not be carried out by this version of the compiler.
12658 Additionally, @code{CFStringRefs} (defined by the @code{CoreFoundation} headers) may
12659 also be used as format arguments. Note that the relevant headers are only likely to be
12660 available on Darwin (OSX) installations. On such installations, the XCode and system
12661 documentation provide descriptions of @code{CFString}, @code{CFStringRefs} and
12662 associated functions.
12665 @section Pragmas Accepted by GCC
12667 @cindex @code{#pragma}
12669 GCC supports several types of pragmas, primarily in order to compile
12670 code originally written for other compilers. Note that in general
12671 we do not recommend the use of pragmas; @xref{Function Attributes},
12672 for further explanation.
12678 * RS/6000 and PowerPC Pragmas::
12680 * Solaris Pragmas::
12681 * Symbol-Renaming Pragmas::
12682 * Structure-Packing Pragmas::
12684 * Diagnostic Pragmas::
12685 * Visibility Pragmas::
12686 * Push/Pop Macro Pragmas::
12687 * Function Specific Option Pragmas::
12691 @subsection ARM Pragmas
12693 The ARM target defines pragmas for controlling the default addition of
12694 @code{long_call} and @code{short_call} attributes to functions.
12695 @xref{Function Attributes}, for information about the effects of these
12700 @cindex pragma, long_calls
12701 Set all subsequent functions to have the @code{long_call} attribute.
12703 @item no_long_calls
12704 @cindex pragma, no_long_calls
12705 Set all subsequent functions to have the @code{short_call} attribute.
12707 @item long_calls_off
12708 @cindex pragma, long_calls_off
12709 Do not affect the @code{long_call} or @code{short_call} attributes of
12710 subsequent functions.
12714 @subsection M32C Pragmas
12717 @item GCC memregs @var{number}
12718 @cindex pragma, memregs
12719 Overrides the command-line option @code{-memregs=} for the current
12720 file. Use with care! This pragma must be before any function in the
12721 file, and mixing different memregs values in different objects may
12722 make them incompatible. This pragma is useful when a
12723 performance-critical function uses a memreg for temporary values,
12724 as it may allow you to reduce the number of memregs used.
12726 @item ADDRESS @var{name} @var{address}
12727 @cindex pragma, address
12728 For any declared symbols matching @var{name}, this does three things
12729 to that symbol: it forces the symbol to be located at the given
12730 address (a number), it forces the symbol to be volatile, and it
12731 changes the symbol's scope to be static. This pragma exists for
12732 compatibility with other compilers, but note that the common
12733 @code{1234H} numeric syntax is not supported (use @code{0x1234}
12737 #pragma ADDRESS port3 0x103
12744 @subsection MeP Pragmas
12748 @item custom io_volatile (on|off)
12749 @cindex pragma, custom io_volatile
12750 Overrides the command line option @code{-mio-volatile} for the current
12751 file. Note that for compatibility with future GCC releases, this
12752 option should only be used once before any @code{io} variables in each
12755 @item GCC coprocessor available @var{registers}
12756 @cindex pragma, coprocessor available
12757 Specifies which coprocessor registers are available to the register
12758 allocator. @var{registers} may be a single register, register range
12759 separated by ellipses, or comma-separated list of those. Example:
12762 #pragma GCC coprocessor available $c0...$c10, $c28
12765 @item GCC coprocessor call_saved @var{registers}
12766 @cindex pragma, coprocessor call_saved
12767 Specifies which coprocessor registers are to be saved and restored by
12768 any function using them. @var{registers} may be a single register,
12769 register range separated by ellipses, or comma-separated list of
12773 #pragma GCC coprocessor call_saved $c4...$c6, $c31
12776 @item GCC coprocessor subclass '(A|B|C|D)' = @var{registers}
12777 @cindex pragma, coprocessor subclass
12778 Creates and defines a register class. These register classes can be
12779 used by inline @code{asm} constructs. @var{registers} may be a single
12780 register, register range separated by ellipses, or comma-separated
12781 list of those. Example:
12784 #pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6
12786 asm ("cpfoo %0" : "=B" (x));
12789 @item GCC disinterrupt @var{name} , @var{name} @dots{}
12790 @cindex pragma, disinterrupt
12791 For the named functions, the compiler adds code to disable interrupts
12792 for the duration of those functions. Any functions so named, which
12793 are not encountered in the source, cause a warning that the pragma was
12794 not used. Examples:
12797 #pragma disinterrupt foo
12798 #pragma disinterrupt bar, grill
12799 int foo () @{ @dots{} @}
12802 @item GCC call @var{name} , @var{name} @dots{}
12803 @cindex pragma, call
12804 For the named functions, the compiler always uses a register-indirect
12805 call model when calling the named functions. Examples:
12814 @node RS/6000 and PowerPC Pragmas
12815 @subsection RS/6000 and PowerPC Pragmas
12817 The RS/6000 and PowerPC targets define one pragma for controlling
12818 whether or not the @code{longcall} attribute is added to function
12819 declarations by default. This pragma overrides the @option{-mlongcall}
12820 option, but not the @code{longcall} and @code{shortcall} attributes.
12821 @xref{RS/6000 and PowerPC Options}, for more information about when long
12822 calls are and are not necessary.
12826 @cindex pragma, longcall
12827 Apply the @code{longcall} attribute to all subsequent function
12831 Do not apply the @code{longcall} attribute to subsequent function
12835 @c Describe h8300 pragmas here.
12836 @c Describe sh pragmas here.
12837 @c Describe v850 pragmas here.
12839 @node Darwin Pragmas
12840 @subsection Darwin Pragmas
12842 The following pragmas are available for all architectures running the
12843 Darwin operating system. These are useful for compatibility with other
12847 @item mark @var{tokens}@dots{}
12848 @cindex pragma, mark
12849 This pragma is accepted, but has no effect.
12851 @item options align=@var{alignment}
12852 @cindex pragma, options align
12853 This pragma sets the alignment of fields in structures. The values of
12854 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
12855 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
12856 properly; to restore the previous setting, use @code{reset} for the
12859 @item segment @var{tokens}@dots{}
12860 @cindex pragma, segment
12861 This pragma is accepted, but has no effect.
12863 @item unused (@var{var} [, @var{var}]@dots{})
12864 @cindex pragma, unused
12865 This pragma declares variables to be possibly unused. GCC will not
12866 produce warnings for the listed variables. The effect is similar to
12867 that of the @code{unused} attribute, except that this pragma may appear
12868 anywhere within the variables' scopes.
12871 @node Solaris Pragmas
12872 @subsection Solaris Pragmas
12874 The Solaris target supports @code{#pragma redefine_extname}
12875 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
12876 @code{#pragma} directives for compatibility with the system compiler.
12879 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
12880 @cindex pragma, align
12882 Increase the minimum alignment of each @var{variable} to @var{alignment}.
12883 This is the same as GCC's @code{aligned} attribute @pxref{Variable
12884 Attributes}). Macro expansion occurs on the arguments to this pragma
12885 when compiling C and Objective-C@. It does not currently occur when
12886 compiling C++, but this is a bug which may be fixed in a future
12889 @item fini (@var{function} [, @var{function}]...)
12890 @cindex pragma, fini
12892 This pragma causes each listed @var{function} to be called after
12893 main, or during shared module unloading, by adding a call to the
12894 @code{.fini} section.
12896 @item init (@var{function} [, @var{function}]...)
12897 @cindex pragma, init
12899 This pragma causes each listed @var{function} to be called during
12900 initialization (before @code{main}) or during shared module loading, by
12901 adding a call to the @code{.init} section.
12905 @node Symbol-Renaming Pragmas
12906 @subsection Symbol-Renaming Pragmas
12908 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
12909 supports two @code{#pragma} directives which change the name used in
12910 assembly for a given declaration. @code{#pragma extern_prefix} is only
12911 available on platforms whose system headers need it. To get this effect
12912 on all platforms supported by GCC, use the asm labels extension (@pxref{Asm
12916 @item redefine_extname @var{oldname} @var{newname}
12917 @cindex pragma, redefine_extname
12919 This pragma gives the C function @var{oldname} the assembly symbol
12920 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
12921 will be defined if this pragma is available (currently on all platforms).
12923 @item extern_prefix @var{string}
12924 @cindex pragma, extern_prefix
12926 This pragma causes all subsequent external function and variable
12927 declarations to have @var{string} prepended to their assembly symbols.
12928 This effect may be terminated with another @code{extern_prefix} pragma
12929 whose argument is an empty string. The preprocessor macro
12930 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
12931 available (currently only on Tru64 UNIX)@.
12934 These pragmas and the asm labels extension interact in a complicated
12935 manner. Here are some corner cases you may want to be aware of.
12938 @item Both pragmas silently apply only to declarations with external
12939 linkage. Asm labels do not have this restriction.
12941 @item In C++, both pragmas silently apply only to declarations with
12942 ``C'' linkage. Again, asm labels do not have this restriction.
12944 @item If any of the three ways of changing the assembly name of a
12945 declaration is applied to a declaration whose assembly name has
12946 already been determined (either by a previous use of one of these
12947 features, or because the compiler needed the assembly name in order to
12948 generate code), and the new name is different, a warning issues and
12949 the name does not change.
12951 @item The @var{oldname} used by @code{#pragma redefine_extname} is
12952 always the C-language name.
12954 @item If @code{#pragma extern_prefix} is in effect, and a declaration
12955 occurs with an asm label attached, the prefix is silently ignored for
12958 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
12959 apply to the same declaration, whichever triggered first wins, and a
12960 warning issues if they contradict each other. (We would like to have
12961 @code{#pragma redefine_extname} always win, for consistency with asm
12962 labels, but if @code{#pragma extern_prefix} triggers first we have no
12963 way of knowing that that happened.)
12966 @node Structure-Packing Pragmas
12967 @subsection Structure-Packing Pragmas
12969 For compatibility with Microsoft Windows compilers, GCC supports a
12970 set of @code{#pragma} directives which change the maximum alignment of
12971 members of structures (other than zero-width bitfields), unions, and
12972 classes subsequently defined. The @var{n} value below always is required
12973 to be a small power of two and specifies the new alignment in bytes.
12976 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
12977 @item @code{#pragma pack()} sets the alignment to the one that was in
12978 effect when compilation started (see also command-line option
12979 @option{-fpack-struct[=<n>]} @pxref{Code Gen Options}).
12980 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
12981 setting on an internal stack and then optionally sets the new alignment.
12982 @item @code{#pragma pack(pop)} restores the alignment setting to the one
12983 saved at the top of the internal stack (and removes that stack entry).
12984 Note that @code{#pragma pack([@var{n}])} does not influence this internal
12985 stack; thus it is possible to have @code{#pragma pack(push)} followed by
12986 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
12987 @code{#pragma pack(pop)}.
12990 Some targets, e.g.@: i386 and powerpc, support the @code{ms_struct}
12991 @code{#pragma} which lays out a structure as the documented
12992 @code{__attribute__ ((ms_struct))}.
12994 @item @code{#pragma ms_struct on} turns on the layout for structures
12996 @item @code{#pragma ms_struct off} turns off the layout for structures
12998 @item @code{#pragma ms_struct reset} goes back to the default layout.
13002 @subsection Weak Pragmas
13004 For compatibility with SVR4, GCC supports a set of @code{#pragma}
13005 directives for declaring symbols to be weak, and defining weak
13009 @item #pragma weak @var{symbol}
13010 @cindex pragma, weak
13011 This pragma declares @var{symbol} to be weak, as if the declaration
13012 had the attribute of the same name. The pragma may appear before
13013 or after the declaration of @var{symbol}, but must appear before
13014 either its first use or its definition. It is not an error for
13015 @var{symbol} to never be defined at all.
13017 @item #pragma weak @var{symbol1} = @var{symbol2}
13018 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
13019 It is an error if @var{symbol2} is not defined in the current
13023 @node Diagnostic Pragmas
13024 @subsection Diagnostic Pragmas
13026 GCC allows the user to selectively enable or disable certain types of
13027 diagnostics, and change the kind of the diagnostic. For example, a
13028 project's policy might require that all sources compile with
13029 @option{-Werror} but certain files might have exceptions allowing
13030 specific types of warnings. Or, a project might selectively enable
13031 diagnostics and treat them as errors depending on which preprocessor
13032 macros are defined.
13035 @item #pragma GCC diagnostic @var{kind} @var{option}
13036 @cindex pragma, diagnostic
13038 Modifies the disposition of a diagnostic. Note that not all
13039 diagnostics are modifiable; at the moment only warnings (normally
13040 controlled by @samp{-W@dots{}}) can be controlled, and not all of them.
13041 Use @option{-fdiagnostics-show-option} to determine which diagnostics
13042 are controllable and which option controls them.
13044 @var{kind} is @samp{error} to treat this diagnostic as an error,
13045 @samp{warning} to treat it like a warning (even if @option{-Werror} is
13046 in effect), or @samp{ignored} if the diagnostic is to be ignored.
13047 @var{option} is a double quoted string which matches the command-line
13051 #pragma GCC diagnostic warning "-Wformat"
13052 #pragma GCC diagnostic error "-Wformat"
13053 #pragma GCC diagnostic ignored "-Wformat"
13056 Note that these pragmas override any command-line options. GCC keeps
13057 track of the location of each pragma, and issues diagnostics according
13058 to the state as of that point in the source file. Thus, pragmas occurring
13059 after a line do not affect diagnostics caused by that line.
13061 @item #pragma GCC diagnostic push
13062 @itemx #pragma GCC diagnostic pop
13064 Causes GCC to remember the state of the diagnostics as of each
13065 @code{push}, and restore to that point at each @code{pop}. If a
13066 @code{pop} has no matching @code{push}, the command line options are
13070 #pragma GCC diagnostic error "-Wuninitialized"
13071 foo(a); /* error is given for this one */
13072 #pragma GCC diagnostic push
13073 #pragma GCC diagnostic ignored "-Wuninitialized"
13074 foo(b); /* no diagnostic for this one */
13075 #pragma GCC diagnostic pop
13076 foo(c); /* error is given for this one */
13077 #pragma GCC diagnostic pop
13078 foo(d); /* depends on command line options */
13083 GCC also offers a simple mechanism for printing messages during
13087 @item #pragma message @var{string}
13088 @cindex pragma, diagnostic
13090 Prints @var{string} as a compiler message on compilation. The message
13091 is informational only, and is neither a compilation warning nor an error.
13094 #pragma message "Compiling " __FILE__ "..."
13097 @var{string} may be parenthesized, and is printed with location
13098 information. For example,
13101 #define DO_PRAGMA(x) _Pragma (#x)
13102 #define TODO(x) DO_PRAGMA(message ("TODO - " #x))
13104 TODO(Remember to fix this)
13107 prints @samp{/tmp/file.c:4: note: #pragma message:
13108 TODO - Remember to fix this}.
13112 @node Visibility Pragmas
13113 @subsection Visibility Pragmas
13116 @item #pragma GCC visibility push(@var{visibility})
13117 @itemx #pragma GCC visibility pop
13118 @cindex pragma, visibility
13120 This pragma allows the user to set the visibility for multiple
13121 declarations without having to give each a visibility attribute
13122 @xref{Function Attributes}, for more information about visibility and
13123 the attribute syntax.
13125 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
13126 declarations. Class members and template specializations are not
13127 affected; if you want to override the visibility for a particular
13128 member or instantiation, you must use an attribute.
13133 @node Push/Pop Macro Pragmas
13134 @subsection Push/Pop Macro Pragmas
13136 For compatibility with Microsoft Windows compilers, GCC supports
13137 @samp{#pragma push_macro(@var{"macro_name"})}
13138 and @samp{#pragma pop_macro(@var{"macro_name"})}.
13141 @item #pragma push_macro(@var{"macro_name"})
13142 @cindex pragma, push_macro
13143 This pragma saves the value of the macro named as @var{macro_name} to
13144 the top of the stack for this macro.
13146 @item #pragma pop_macro(@var{"macro_name"})
13147 @cindex pragma, pop_macro
13148 This pragma sets the value of the macro named as @var{macro_name} to
13149 the value on top of the stack for this macro. If the stack for
13150 @var{macro_name} is empty, the value of the macro remains unchanged.
13157 #pragma push_macro("X")
13160 #pragma pop_macro("X")
13164 In this example, the definition of X as 1 is saved by @code{#pragma
13165 push_macro} and restored by @code{#pragma pop_macro}.
13167 @node Function Specific Option Pragmas
13168 @subsection Function Specific Option Pragmas
13171 @item #pragma GCC target (@var{"string"}...)
13172 @cindex pragma GCC target
13174 This pragma allows you to set target specific options for functions
13175 defined later in the source file. One or more strings can be
13176 specified. Each function that is defined after this point will be as
13177 if @code{attribute((target("STRING")))} was specified for that
13178 function. The parenthesis around the options is optional.
13179 @xref{Function Attributes}, for more information about the
13180 @code{target} attribute and the attribute syntax.
13182 The @code{#pragma GCC target} attribute is not implemented in GCC versions earlier
13183 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC backends. At
13184 present, it is not implemented for other backends.
13188 @item #pragma GCC optimize (@var{"string"}...)
13189 @cindex pragma GCC optimize
13191 This pragma allows you to set global optimization options for functions
13192 defined later in the source file. One or more strings can be
13193 specified. Each function that is defined after this point will be as
13194 if @code{attribute((optimize("STRING")))} was specified for that
13195 function. The parenthesis around the options is optional.
13196 @xref{Function Attributes}, for more information about the
13197 @code{optimize} attribute and the attribute syntax.
13199 The @samp{#pragma GCC optimize} pragma is not implemented in GCC
13200 versions earlier than 4.4.
13204 @item #pragma GCC push_options
13205 @itemx #pragma GCC pop_options
13206 @cindex pragma GCC push_options
13207 @cindex pragma GCC pop_options
13209 These pragmas maintain a stack of the current target and optimization
13210 options. It is intended for include files where you temporarily want
13211 to switch to using a different @samp{#pragma GCC target} or
13212 @samp{#pragma GCC optimize} and then to pop back to the previous
13215 The @samp{#pragma GCC push_options} and @samp{#pragma GCC pop_options}
13216 pragmas are not implemented in GCC versions earlier than 4.4.
13220 @item #pragma GCC reset_options
13221 @cindex pragma GCC reset_options
13223 This pragma clears the current @code{#pragma GCC target} and
13224 @code{#pragma GCC optimize} to use the default switches as specified
13225 on the command line.
13227 The @samp{#pragma GCC reset_options} pragma is not implemented in GCC
13228 versions earlier than 4.4.
13231 @node Unnamed Fields
13232 @section Unnamed struct/union fields within structs/unions
13233 @cindex @code{struct}
13234 @cindex @code{union}
13236 As permitted by ISO C1X and for compatibility with other compilers,
13237 GCC allows you to define
13238 a structure or union that contains, as fields, structures and unions
13239 without names. For example:
13252 In this example, the user would be able to access members of the unnamed
13253 union with code like @samp{foo.b}. Note that only unnamed structs and
13254 unions are allowed, you may not have, for example, an unnamed
13257 You must never create such structures that cause ambiguous field definitions.
13258 For example, this structure:
13269 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
13270 The compiler gives errors for such constructs.
13272 @opindex fms-extensions
13273 Unless @option{-fms-extensions} is used, the unnamed field must be a
13274 structure or union definition without a tag (for example, @samp{struct
13275 @{ int a; @};}), or a @code{typedef} name for such a structure or
13276 union. If @option{-fms-extensions} is used, the field may
13277 also be a definition with a tag such as @samp{struct foo @{ int a;
13278 @};}, a reference to a previously defined structure or union such as
13279 @samp{struct foo;}, or a reference to a @code{typedef} name for a
13280 previously defined structure or union type with a tag.
13282 @opindex fplan9-extensions
13283 The option @option{-fplan9-extensions} enables
13284 @option{-fms-extensions} as well as two other extensions. First, a
13285 pointer to a structure is automatically converted to a pointer to an
13286 anonymous field for assignments and function calls. For example:
13289 struct s1 @{ int a; @};
13290 struct s2 @{ struct s1; @};
13291 extern void f1 (struct s1 *);
13292 void f2 (struct s2 *p) @{ f1 (p); @}
13295 In the call to @code{f1} inside @code{f2}, the pointer @code{p} is
13296 converted into a pointer to the anonymous field.
13298 Second, when the type of an anonymous field is a @code{typedef} for a
13299 @code{struct} or @code{union}, code may refer to the field using the
13300 name of the @code{typedef}.
13303 typedef struct @{ int a; @} s1;
13304 struct s2 @{ s1; @};
13305 s1 f1 (struct s2 *p) @{ return p->s1; @}
13308 These usages are only permitted when they are not ambiguous.
13311 @section Thread-Local Storage
13312 @cindex Thread-Local Storage
13313 @cindex @acronym{TLS}
13314 @cindex @code{__thread}
13316 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
13317 are allocated such that there is one instance of the variable per extant
13318 thread. The run-time model GCC uses to implement this originates
13319 in the IA-64 processor-specific ABI, but has since been migrated
13320 to other processors as well. It requires significant support from
13321 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
13322 system libraries (@file{libc.so} and @file{libpthread.so}), so it
13323 is not available everywhere.
13325 At the user level, the extension is visible with a new storage
13326 class keyword: @code{__thread}. For example:
13330 extern __thread struct state s;
13331 static __thread char *p;
13334 The @code{__thread} specifier may be used alone, with the @code{extern}
13335 or @code{static} specifiers, but with no other storage class specifier.
13336 When used with @code{extern} or @code{static}, @code{__thread} must appear
13337 immediately after the other storage class specifier.
13339 The @code{__thread} specifier may be applied to any global, file-scoped
13340 static, function-scoped static, or static data member of a class. It may
13341 not be applied to block-scoped automatic or non-static data member.
13343 When the address-of operator is applied to a thread-local variable, it is
13344 evaluated at run-time and returns the address of the current thread's
13345 instance of that variable. An address so obtained may be used by any
13346 thread. When a thread terminates, any pointers to thread-local variables
13347 in that thread become invalid.
13349 No static initialization may refer to the address of a thread-local variable.
13351 In C++, if an initializer is present for a thread-local variable, it must
13352 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
13355 See @uref{http://people.redhat.com/drepper/tls.pdf,
13356 ELF Handling For Thread-Local Storage} for a detailed explanation of
13357 the four thread-local storage addressing models, and how the run-time
13358 is expected to function.
13361 * C99 Thread-Local Edits::
13362 * C++98 Thread-Local Edits::
13365 @node C99 Thread-Local Edits
13366 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
13368 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
13369 that document the exact semantics of the language extension.
13373 @cite{5.1.2 Execution environments}
13375 Add new text after paragraph 1
13378 Within either execution environment, a @dfn{thread} is a flow of
13379 control within a program. It is implementation defined whether
13380 or not there may be more than one thread associated with a program.
13381 It is implementation defined how threads beyond the first are
13382 created, the name and type of the function called at thread
13383 startup, and how threads may be terminated. However, objects
13384 with thread storage duration shall be initialized before thread
13389 @cite{6.2.4 Storage durations of objects}
13391 Add new text before paragraph 3
13394 An object whose identifier is declared with the storage-class
13395 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
13396 Its lifetime is the entire execution of the thread, and its
13397 stored value is initialized only once, prior to thread startup.
13401 @cite{6.4.1 Keywords}
13403 Add @code{__thread}.
13406 @cite{6.7.1 Storage-class specifiers}
13408 Add @code{__thread} to the list of storage class specifiers in
13411 Change paragraph 2 to
13414 With the exception of @code{__thread}, at most one storage-class
13415 specifier may be given [@dots{}]. The @code{__thread} specifier may
13416 be used alone, or immediately following @code{extern} or
13420 Add new text after paragraph 6
13423 The declaration of an identifier for a variable that has
13424 block scope that specifies @code{__thread} shall also
13425 specify either @code{extern} or @code{static}.
13427 The @code{__thread} specifier shall be used only with
13432 @node C++98 Thread-Local Edits
13433 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
13435 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
13436 that document the exact semantics of the language extension.
13440 @b{[intro.execution]}
13442 New text after paragraph 4
13445 A @dfn{thread} is a flow of control within the abstract machine.
13446 It is implementation defined whether or not there may be more than
13450 New text after paragraph 7
13453 It is unspecified whether additional action must be taken to
13454 ensure when and whether side effects are visible to other threads.
13460 Add @code{__thread}.
13463 @b{[basic.start.main]}
13465 Add after paragraph 5
13468 The thread that begins execution at the @code{main} function is called
13469 the @dfn{main thread}. It is implementation defined how functions
13470 beginning threads other than the main thread are designated or typed.
13471 A function so designated, as well as the @code{main} function, is called
13472 a @dfn{thread startup function}. It is implementation defined what
13473 happens if a thread startup function returns. It is implementation
13474 defined what happens to other threads when any thread calls @code{exit}.
13478 @b{[basic.start.init]}
13480 Add after paragraph 4
13483 The storage for an object of thread storage duration shall be
13484 statically initialized before the first statement of the thread startup
13485 function. An object of thread storage duration shall not require
13486 dynamic initialization.
13490 @b{[basic.start.term]}
13492 Add after paragraph 3
13495 The type of an object with thread storage duration shall not have a
13496 non-trivial destructor, nor shall it be an array type whose elements
13497 (directly or indirectly) have non-trivial destructors.
13503 Add ``thread storage duration'' to the list in paragraph 1.
13508 Thread, static, and automatic storage durations are associated with
13509 objects introduced by declarations [@dots{}].
13512 Add @code{__thread} to the list of specifiers in paragraph 3.
13515 @b{[basic.stc.thread]}
13517 New section before @b{[basic.stc.static]}
13520 The keyword @code{__thread} applied to a non-local object gives the
13521 object thread storage duration.
13523 A local variable or class data member declared both @code{static}
13524 and @code{__thread} gives the variable or member thread storage
13529 @b{[basic.stc.static]}
13534 All objects which have neither thread storage duration, dynamic
13535 storage duration nor are local [@dots{}].
13541 Add @code{__thread} to the list in paragraph 1.
13546 With the exception of @code{__thread}, at most one
13547 @var{storage-class-specifier} shall appear in a given
13548 @var{decl-specifier-seq}. The @code{__thread} specifier may
13549 be used alone, or immediately following the @code{extern} or
13550 @code{static} specifiers. [@dots{}]
13553 Add after paragraph 5
13556 The @code{__thread} specifier can be applied only to the names of objects
13557 and to anonymous unions.
13563 Add after paragraph 6
13566 Non-@code{static} members shall not be @code{__thread}.
13570 @node Binary constants
13571 @section Binary constants using the @samp{0b} prefix
13572 @cindex Binary constants using the @samp{0b} prefix
13574 Integer constants can be written as binary constants, consisting of a
13575 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
13576 @samp{0B}. This is particularly useful in environments that operate a
13577 lot on the bit-level (like microcontrollers).
13579 The following statements are identical:
13588 The type of these constants follows the same rules as for octal or
13589 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
13592 @node C++ Extensions
13593 @chapter Extensions to the C++ Language
13594 @cindex extensions, C++ language
13595 @cindex C++ language extensions
13597 The GNU compiler provides these extensions to the C++ language (and you
13598 can also use most of the C language extensions in your C++ programs). If you
13599 want to write code that checks whether these features are available, you can
13600 test for the GNU compiler the same way as for C programs: check for a
13601 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
13602 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
13603 Predefined Macros,cpp,The GNU C Preprocessor}).
13606 * C++ Volatiles:: What constitutes an access to a volatile object.
13607 * Restricted Pointers:: C99 restricted pointers and references.
13608 * Vague Linkage:: Where G++ puts inlines, vtables and such.
13609 * C++ Interface:: You can use a single C++ header file for both
13610 declarations and definitions.
13611 * Template Instantiation:: Methods for ensuring that exactly one copy of
13612 each needed template instantiation is emitted.
13613 * Bound member functions:: You can extract a function pointer to the
13614 method denoted by a @samp{->*} or @samp{.*} expression.
13615 * C++ Attributes:: Variable, function, and type attributes for C++ only.
13616 * Namespace Association:: Strong using-directives for namespace association.
13617 * Type Traits:: Compiler support for type traits
13618 * Java Exceptions:: Tweaking exception handling to work with Java.
13619 * Deprecated Features:: Things will disappear from g++.
13620 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
13623 @node C++ Volatiles
13624 @section When is a Volatile C++ Object Accessed?
13625 @cindex accessing volatiles
13626 @cindex volatile read
13627 @cindex volatile write
13628 @cindex volatile access
13630 The C++ standard differs from the C standard in its treatment of
13631 volatile objects. It fails to specify what constitutes a volatile
13632 access, except to say that C++ should behave in a similar manner to C
13633 with respect to volatiles, where possible. However, the different
13634 lvalueness of expressions between C and C++ complicate the behaviour.
13635 G++ behaves the same as GCC for volatile access, @xref{C
13636 Extensions,,Volatiles}, for a description of GCC's behaviour.
13638 The C and C++ language specifications differ when an object is
13639 accessed in a void context:
13642 volatile int *src = @var{somevalue};
13646 The C++ standard specifies that such expressions do not undergo lvalue
13647 to rvalue conversion, and that the type of the dereferenced object may
13648 be incomplete. The C++ standard does not specify explicitly that it
13649 is lvalue to rvalue conversion which is responsible for causing an
13650 access. There is reason to believe that it is, because otherwise
13651 certain simple expressions become undefined. However, because it
13652 would surprise most programmers, G++ treats dereferencing a pointer to
13653 volatile object of complete type as GCC would do for an equivalent
13654 type in C@. When the object has incomplete type, G++ issues a
13655 warning; if you wish to force an error, you must force a conversion to
13656 rvalue with, for instance, a static cast.
13658 When using a reference to volatile, G++ does not treat equivalent
13659 expressions as accesses to volatiles, but instead issues a warning that
13660 no volatile is accessed. The rationale for this is that otherwise it
13661 becomes difficult to determine where volatile access occur, and not
13662 possible to ignore the return value from functions returning volatile
13663 references. Again, if you wish to force a read, cast the reference to
13666 G++ implements the same behaviour as GCC does when assigning to a
13667 volatile object -- there is no reread of the assigned-to object, the
13668 assigned rvalue is reused. Note that in C++ assignment expressions
13669 are lvalues, and if used as an lvalue, the volatile object will be
13670 referred to. For instance, @var{vref} will refer to @var{vobj}, as
13671 expected, in the following example:
13675 volatile int &vref = vobj = @var{something};
13678 @node Restricted Pointers
13679 @section Restricting Pointer Aliasing
13680 @cindex restricted pointers
13681 @cindex restricted references
13682 @cindex restricted this pointer
13684 As with the C front end, G++ understands the C99 feature of restricted pointers,
13685 specified with the @code{__restrict__}, or @code{__restrict} type
13686 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
13687 language flag, @code{restrict} is not a keyword in C++.
13689 In addition to allowing restricted pointers, you can specify restricted
13690 references, which indicate that the reference is not aliased in the local
13694 void fn (int *__restrict__ rptr, int &__restrict__ rref)
13701 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
13702 @var{rref} refers to a (different) unaliased integer.
13704 You may also specify whether a member function's @var{this} pointer is
13705 unaliased by using @code{__restrict__} as a member function qualifier.
13708 void T::fn () __restrict__
13715 Within the body of @code{T::fn}, @var{this} will have the effective
13716 definition @code{T *__restrict__ const this}. Notice that the
13717 interpretation of a @code{__restrict__} member function qualifier is
13718 different to that of @code{const} or @code{volatile} qualifier, in that it
13719 is applied to the pointer rather than the object. This is consistent with
13720 other compilers which implement restricted pointers.
13722 As with all outermost parameter qualifiers, @code{__restrict__} is
13723 ignored in function definition matching. This means you only need to
13724 specify @code{__restrict__} in a function definition, rather than
13725 in a function prototype as well.
13727 @node Vague Linkage
13728 @section Vague Linkage
13729 @cindex vague linkage
13731 There are several constructs in C++ which require space in the object
13732 file but are not clearly tied to a single translation unit. We say that
13733 these constructs have ``vague linkage''. Typically such constructs are
13734 emitted wherever they are needed, though sometimes we can be more
13738 @item Inline Functions
13739 Inline functions are typically defined in a header file which can be
13740 included in many different compilations. Hopefully they can usually be
13741 inlined, but sometimes an out-of-line copy is necessary, if the address
13742 of the function is taken or if inlining fails. In general, we emit an
13743 out-of-line copy in all translation units where one is needed. As an
13744 exception, we only emit inline virtual functions with the vtable, since
13745 it will always require a copy.
13747 Local static variables and string constants used in an inline function
13748 are also considered to have vague linkage, since they must be shared
13749 between all inlined and out-of-line instances of the function.
13753 C++ virtual functions are implemented in most compilers using a lookup
13754 table, known as a vtable. The vtable contains pointers to the virtual
13755 functions provided by a class, and each object of the class contains a
13756 pointer to its vtable (or vtables, in some multiple-inheritance
13757 situations). If the class declares any non-inline, non-pure virtual
13758 functions, the first one is chosen as the ``key method'' for the class,
13759 and the vtable is only emitted in the translation unit where the key
13762 @emph{Note:} If the chosen key method is later defined as inline, the
13763 vtable will still be emitted in every translation unit which defines it.
13764 Make sure that any inline virtuals are declared inline in the class
13765 body, even if they are not defined there.
13767 @item @code{type_info} objects
13768 @cindex @code{type_info}
13770 C++ requires information about types to be written out in order to
13771 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
13772 For polymorphic classes (classes with virtual functions), the @samp{type_info}
13773 object is written out along with the vtable so that @samp{dynamic_cast}
13774 can determine the dynamic type of a class object at runtime. For all
13775 other types, we write out the @samp{type_info} object when it is used: when
13776 applying @samp{typeid} to an expression, throwing an object, or
13777 referring to a type in a catch clause or exception specification.
13779 @item Template Instantiations
13780 Most everything in this section also applies to template instantiations,
13781 but there are other options as well.
13782 @xref{Template Instantiation,,Where's the Template?}.
13786 When used with GNU ld version 2.8 or later on an ELF system such as
13787 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
13788 these constructs will be discarded at link time. This is known as
13791 On targets that don't support COMDAT, but do support weak symbols, GCC
13792 will use them. This way one copy will override all the others, but
13793 the unused copies will still take up space in the executable.
13795 For targets which do not support either COMDAT or weak symbols,
13796 most entities with vague linkage will be emitted as local symbols to
13797 avoid duplicate definition errors from the linker. This will not happen
13798 for local statics in inlines, however, as having multiple copies will
13799 almost certainly break things.
13801 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
13802 another way to control placement of these constructs.
13804 @node C++ Interface
13805 @section #pragma interface and implementation
13807 @cindex interface and implementation headers, C++
13808 @cindex C++ interface and implementation headers
13809 @cindex pragmas, interface and implementation
13811 @code{#pragma interface} and @code{#pragma implementation} provide the
13812 user with a way of explicitly directing the compiler to emit entities
13813 with vague linkage (and debugging information) in a particular
13816 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
13817 most cases, because of COMDAT support and the ``key method'' heuristic
13818 mentioned in @ref{Vague Linkage}. Using them can actually cause your
13819 program to grow due to unnecessary out-of-line copies of inline
13820 functions. Currently (3.4) the only benefit of these
13821 @code{#pragma}s is reduced duplication of debugging information, and
13822 that should be addressed soon on DWARF 2 targets with the use of
13826 @item #pragma interface
13827 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
13828 @kindex #pragma interface
13829 Use this directive in @emph{header files} that define object classes, to save
13830 space in most of the object files that use those classes. Normally,
13831 local copies of certain information (backup copies of inline member
13832 functions, debugging information, and the internal tables that implement
13833 virtual functions) must be kept in each object file that includes class
13834 definitions. You can use this pragma to avoid such duplication. When a
13835 header file containing @samp{#pragma interface} is included in a
13836 compilation, this auxiliary information will not be generated (unless
13837 the main input source file itself uses @samp{#pragma implementation}).
13838 Instead, the object files will contain references to be resolved at link
13841 The second form of this directive is useful for the case where you have
13842 multiple headers with the same name in different directories. If you
13843 use this form, you must specify the same string to @samp{#pragma
13846 @item #pragma implementation
13847 @itemx #pragma implementation "@var{objects}.h"
13848 @kindex #pragma implementation
13849 Use this pragma in a @emph{main input file}, when you want full output from
13850 included header files to be generated (and made globally visible). The
13851 included header file, in turn, should use @samp{#pragma interface}.
13852 Backup copies of inline member functions, debugging information, and the
13853 internal tables used to implement virtual functions are all generated in
13854 implementation files.
13856 @cindex implied @code{#pragma implementation}
13857 @cindex @code{#pragma implementation}, implied
13858 @cindex naming convention, implementation headers
13859 If you use @samp{#pragma implementation} with no argument, it applies to
13860 an include file with the same basename@footnote{A file's @dfn{basename}
13861 was the name stripped of all leading path information and of trailing
13862 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
13863 file. For example, in @file{allclass.cc}, giving just
13864 @samp{#pragma implementation}
13865 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
13867 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
13868 an implementation file whenever you would include it from
13869 @file{allclass.cc} even if you never specified @samp{#pragma
13870 implementation}. This was deemed to be more trouble than it was worth,
13871 however, and disabled.
13873 Use the string argument if you want a single implementation file to
13874 include code from multiple header files. (You must also use
13875 @samp{#include} to include the header file; @samp{#pragma
13876 implementation} only specifies how to use the file---it doesn't actually
13879 There is no way to split up the contents of a single header file into
13880 multiple implementation files.
13883 @cindex inlining and C++ pragmas
13884 @cindex C++ pragmas, effect on inlining
13885 @cindex pragmas in C++, effect on inlining
13886 @samp{#pragma implementation} and @samp{#pragma interface} also have an
13887 effect on function inlining.
13889 If you define a class in a header file marked with @samp{#pragma
13890 interface}, the effect on an inline function defined in that class is
13891 similar to an explicit @code{extern} declaration---the compiler emits
13892 no code at all to define an independent version of the function. Its
13893 definition is used only for inlining with its callers.
13895 @opindex fno-implement-inlines
13896 Conversely, when you include the same header file in a main source file
13897 that declares it as @samp{#pragma implementation}, the compiler emits
13898 code for the function itself; this defines a version of the function
13899 that can be found via pointers (or by callers compiled without
13900 inlining). If all calls to the function can be inlined, you can avoid
13901 emitting the function by compiling with @option{-fno-implement-inlines}.
13902 If any calls were not inlined, you will get linker errors.
13904 @node Template Instantiation
13905 @section Where's the Template?
13906 @cindex template instantiation
13908 C++ templates are the first language feature to require more
13909 intelligence from the environment than one usually finds on a UNIX
13910 system. Somehow the compiler and linker have to make sure that each
13911 template instance occurs exactly once in the executable if it is needed,
13912 and not at all otherwise. There are two basic approaches to this
13913 problem, which are referred to as the Borland model and the Cfront model.
13916 @item Borland model
13917 Borland C++ solved the template instantiation problem by adding the code
13918 equivalent of common blocks to their linker; the compiler emits template
13919 instances in each translation unit that uses them, and the linker
13920 collapses them together. The advantage of this model is that the linker
13921 only has to consider the object files themselves; there is no external
13922 complexity to worry about. This disadvantage is that compilation time
13923 is increased because the template code is being compiled repeatedly.
13924 Code written for this model tends to include definitions of all
13925 templates in the header file, since they must be seen to be
13929 The AT&T C++ translator, Cfront, solved the template instantiation
13930 problem by creating the notion of a template repository, an
13931 automatically maintained place where template instances are stored. A
13932 more modern version of the repository works as follows: As individual
13933 object files are built, the compiler places any template definitions and
13934 instantiations encountered in the repository. At link time, the link
13935 wrapper adds in the objects in the repository and compiles any needed
13936 instances that were not previously emitted. The advantages of this
13937 model are more optimal compilation speed and the ability to use the
13938 system linker; to implement the Borland model a compiler vendor also
13939 needs to replace the linker. The disadvantages are vastly increased
13940 complexity, and thus potential for error; for some code this can be
13941 just as transparent, but in practice it can been very difficult to build
13942 multiple programs in one directory and one program in multiple
13943 directories. Code written for this model tends to separate definitions
13944 of non-inline member templates into a separate file, which should be
13945 compiled separately.
13948 When used with GNU ld version 2.8 or later on an ELF system such as
13949 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
13950 Borland model. On other systems, G++ implements neither automatic
13953 A future version of G++ will support a hybrid model whereby the compiler
13954 will emit any instantiations for which the template definition is
13955 included in the compile, and store template definitions and
13956 instantiation context information into the object file for the rest.
13957 The link wrapper will extract that information as necessary and invoke
13958 the compiler to produce the remaining instantiations. The linker will
13959 then combine duplicate instantiations.
13961 In the mean time, you have the following options for dealing with
13962 template instantiations:
13967 Compile your template-using code with @option{-frepo}. The compiler will
13968 generate files with the extension @samp{.rpo} listing all of the
13969 template instantiations used in the corresponding object files which
13970 could be instantiated there; the link wrapper, @samp{collect2}, will
13971 then update the @samp{.rpo} files to tell the compiler where to place
13972 those instantiations and rebuild any affected object files. The
13973 link-time overhead is negligible after the first pass, as the compiler
13974 will continue to place the instantiations in the same files.
13976 This is your best option for application code written for the Borland
13977 model, as it will just work. Code written for the Cfront model will
13978 need to be modified so that the template definitions are available at
13979 one or more points of instantiation; usually this is as simple as adding
13980 @code{#include <tmethods.cc>} to the end of each template header.
13982 For library code, if you want the library to provide all of the template
13983 instantiations it needs, just try to link all of its object files
13984 together; the link will fail, but cause the instantiations to be
13985 generated as a side effect. Be warned, however, that this may cause
13986 conflicts if multiple libraries try to provide the same instantiations.
13987 For greater control, use explicit instantiation as described in the next
13991 @opindex fno-implicit-templates
13992 Compile your code with @option{-fno-implicit-templates} to disable the
13993 implicit generation of template instances, and explicitly instantiate
13994 all the ones you use. This approach requires more knowledge of exactly
13995 which instances you need than do the others, but it's less
13996 mysterious and allows greater control. You can scatter the explicit
13997 instantiations throughout your program, perhaps putting them in the
13998 translation units where the instances are used or the translation units
13999 that define the templates themselves; you can put all of the explicit
14000 instantiations you need into one big file; or you can create small files
14007 template class Foo<int>;
14008 template ostream& operator <<
14009 (ostream&, const Foo<int>&);
14012 for each of the instances you need, and create a template instantiation
14013 library from those.
14015 If you are using Cfront-model code, you can probably get away with not
14016 using @option{-fno-implicit-templates} when compiling files that don't
14017 @samp{#include} the member template definitions.
14019 If you use one big file to do the instantiations, you may want to
14020 compile it without @option{-fno-implicit-templates} so you get all of the
14021 instances required by your explicit instantiations (but not by any
14022 other files) without having to specify them as well.
14024 G++ has extended the template instantiation syntax given in the ISO
14025 standard to allow forward declaration of explicit instantiations
14026 (with @code{extern}), instantiation of the compiler support data for a
14027 template class (i.e.@: the vtable) without instantiating any of its
14028 members (with @code{inline}), and instantiation of only the static data
14029 members of a template class, without the support data or member
14030 functions (with (@code{static}):
14033 extern template int max (int, int);
14034 inline template class Foo<int>;
14035 static template class Foo<int>;
14039 Do nothing. Pretend G++ does implement automatic instantiation
14040 management. Code written for the Borland model will work fine, but
14041 each translation unit will contain instances of each of the templates it
14042 uses. In a large program, this can lead to an unacceptable amount of code
14046 @node Bound member functions
14047 @section Extracting the function pointer from a bound pointer to member function
14049 @cindex pointer to member function
14050 @cindex bound pointer to member function
14052 In C++, pointer to member functions (PMFs) are implemented using a wide
14053 pointer of sorts to handle all the possible call mechanisms; the PMF
14054 needs to store information about how to adjust the @samp{this} pointer,
14055 and if the function pointed to is virtual, where to find the vtable, and
14056 where in the vtable to look for the member function. If you are using
14057 PMFs in an inner loop, you should really reconsider that decision. If
14058 that is not an option, you can extract the pointer to the function that
14059 would be called for a given object/PMF pair and call it directly inside
14060 the inner loop, to save a bit of time.
14062 Note that you will still be paying the penalty for the call through a
14063 function pointer; on most modern architectures, such a call defeats the
14064 branch prediction features of the CPU@. This is also true of normal
14065 virtual function calls.
14067 The syntax for this extension is
14071 extern int (A::*fp)();
14072 typedef int (*fptr)(A *);
14074 fptr p = (fptr)(a.*fp);
14077 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
14078 no object is needed to obtain the address of the function. They can be
14079 converted to function pointers directly:
14082 fptr p1 = (fptr)(&A::foo);
14085 @opindex Wno-pmf-conversions
14086 You must specify @option{-Wno-pmf-conversions} to use this extension.
14088 @node C++ Attributes
14089 @section C++-Specific Variable, Function, and Type Attributes
14091 Some attributes only make sense for C++ programs.
14094 @item init_priority (@var{priority})
14095 @cindex @code{init_priority} attribute
14098 In Standard C++, objects defined at namespace scope are guaranteed to be
14099 initialized in an order in strict accordance with that of their definitions
14100 @emph{in a given translation unit}. No guarantee is made for initializations
14101 across translation units. However, GNU C++ allows users to control the
14102 order of initialization of objects defined at namespace scope with the
14103 @code{init_priority} attribute by specifying a relative @var{priority},
14104 a constant integral expression currently bounded between 101 and 65535
14105 inclusive. Lower numbers indicate a higher priority.
14107 In the following example, @code{A} would normally be created before
14108 @code{B}, but the @code{init_priority} attribute has reversed that order:
14111 Some_Class A __attribute__ ((init_priority (2000)));
14112 Some_Class B __attribute__ ((init_priority (543)));
14116 Note that the particular values of @var{priority} do not matter; only their
14119 @item java_interface
14120 @cindex @code{java_interface} attribute
14122 This type attribute informs C++ that the class is a Java interface. It may
14123 only be applied to classes declared within an @code{extern "Java"} block.
14124 Calls to methods declared in this interface will be dispatched using GCJ's
14125 interface table mechanism, instead of regular virtual table dispatch.
14129 See also @ref{Namespace Association}.
14131 @node Namespace Association
14132 @section Namespace Association
14134 @strong{Caution:} The semantics of this extension are not fully
14135 defined. Users should refrain from using this extension as its
14136 semantics may change subtly over time. It is possible that this
14137 extension will be removed in future versions of G++.
14139 A using-directive with @code{__attribute ((strong))} is stronger
14140 than a normal using-directive in two ways:
14144 Templates from the used namespace can be specialized and explicitly
14145 instantiated as though they were members of the using namespace.
14148 The using namespace is considered an associated namespace of all
14149 templates in the used namespace for purposes of argument-dependent
14153 The used namespace must be nested within the using namespace so that
14154 normal unqualified lookup works properly.
14156 This is useful for composing a namespace transparently from
14157 implementation namespaces. For example:
14162 template <class T> struct A @{ @};
14164 using namespace debug __attribute ((__strong__));
14165 template <> struct A<int> @{ @}; // @r{ok to specialize}
14167 template <class T> void f (A<T>);
14172 f (std::A<float>()); // @r{lookup finds} std::f
14178 @section Type Traits
14180 The C++ front-end implements syntactic extensions that allow to
14181 determine at compile time various characteristics of a type (or of a
14185 @item __has_nothrow_assign (type)
14186 If @code{type} is const qualified or is a reference type then the trait is
14187 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
14188 is true, else if @code{type} is a cv class or union type with copy assignment
14189 operators that are known not to throw an exception then the trait is true,
14190 else it is false. Requires: @code{type} shall be a complete type, an array
14191 type of unknown bound, or is a @code{void} type.
14193 @item __has_nothrow_copy (type)
14194 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
14195 @code{type} is a cv class or union type with copy constructors that
14196 are known not to throw an exception then the trait is true, else it is false.
14197 Requires: @code{type} shall be a complete type, an array type of
14198 unknown bound, or is a @code{void} type.
14200 @item __has_nothrow_constructor (type)
14201 If @code{__has_trivial_constructor (type)} is true then the trait is
14202 true, else if @code{type} is a cv class or union type (or array
14203 thereof) with a default constructor that is known not to throw an
14204 exception then the trait is true, else it is false. Requires:
14205 @code{type} shall be a complete type, an array type of unknown bound,
14206 or is a @code{void} type.
14208 @item __has_trivial_assign (type)
14209 If @code{type} is const qualified or is a reference type then the trait is
14210 false. Otherwise if @code{__is_pod (type)} is true then the trait is
14211 true, else if @code{type} is a cv class or union type with a trivial
14212 copy assignment ([class.copy]) then the trait is true, else it is
14213 false. Requires: @code{type} shall be a complete type, an array type
14214 of unknown bound, or is a @code{void} type.
14216 @item __has_trivial_copy (type)
14217 If @code{__is_pod (type)} is true or @code{type} is a reference type
14218 then the trait is true, else if @code{type} is a cv class or union type
14219 with a trivial copy constructor ([class.copy]) then the trait
14220 is true, else it is false. Requires: @code{type} shall be a complete
14221 type, an array type of unknown bound, or is a @code{void} type.
14223 @item __has_trivial_constructor (type)
14224 If @code{__is_pod (type)} is true then the trait is true, else if
14225 @code{type} is a cv class or union type (or array thereof) with a
14226 trivial default constructor ([class.ctor]) then the trait is true,
14227 else it is false. Requires: @code{type} shall be a complete type, an
14228 array type of unknown bound, or is a @code{void} type.
14230 @item __has_trivial_destructor (type)
14231 If @code{__is_pod (type)} is true or @code{type} is a reference type then
14232 the trait is true, else if @code{type} is a cv class or union type (or
14233 array thereof) with a trivial destructor ([class.dtor]) then the trait
14234 is true, else it is false. Requires: @code{type} shall be a complete
14235 type, an array type of unknown bound, or is a @code{void} type.
14237 @item __has_virtual_destructor (type)
14238 If @code{type} is a class type with a virtual destructor
14239 ([class.dtor]) then the trait is true, else it is false. Requires:
14240 @code{type} shall be a complete type, an array type of unknown bound,
14241 or is a @code{void} type.
14243 @item __is_abstract (type)
14244 If @code{type} is an abstract class ([class.abstract]) then the trait
14245 is true, else it is false. Requires: @code{type} shall be a complete
14246 type, an array type of unknown bound, or is a @code{void} type.
14248 @item __is_base_of (base_type, derived_type)
14249 If @code{base_type} is a base class of @code{derived_type}
14250 ([class.derived]) then the trait is true, otherwise it is false.
14251 Top-level cv qualifications of @code{base_type} and
14252 @code{derived_type} are ignored. For the purposes of this trait, a
14253 class type is considered is own base. Requires: if @code{__is_class
14254 (base_type)} and @code{__is_class (derived_type)} are true and
14255 @code{base_type} and @code{derived_type} are not the same type
14256 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
14257 type. Diagnostic is produced if this requirement is not met.
14259 @item __is_class (type)
14260 If @code{type} is a cv class type, and not a union type
14261 ([basic.compound]) the trait is true, else it is false.
14263 @item __is_empty (type)
14264 If @code{__is_class (type)} is false then the trait is false.
14265 Otherwise @code{type} is considered empty if and only if: @code{type}
14266 has no non-static data members, or all non-static data members, if
14267 any, are bit-fields of length 0, and @code{type} has no virtual
14268 members, and @code{type} has no virtual base classes, and @code{type}
14269 has no base classes @code{base_type} for which
14270 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
14271 be a complete type, an array type of unknown bound, or is a
14274 @item __is_enum (type)
14275 If @code{type} is a cv enumeration type ([basic.compound]) the trait is
14276 true, else it is false.
14278 @item __is_pod (type)
14279 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
14280 else it is false. Requires: @code{type} shall be a complete type,
14281 an array type of unknown bound, or is a @code{void} type.
14283 @item __is_polymorphic (type)
14284 If @code{type} is a polymorphic class ([class.virtual]) then the trait
14285 is true, else it is false. Requires: @code{type} shall be a complete
14286 type, an array type of unknown bound, or is a @code{void} type.
14288 @item __is_union (type)
14289 If @code{type} is a cv union type ([basic.compound]) the trait is
14290 true, else it is false.
14294 @node Java Exceptions
14295 @section Java Exceptions
14297 The Java language uses a slightly different exception handling model
14298 from C++. Normally, GNU C++ will automatically detect when you are
14299 writing C++ code that uses Java exceptions, and handle them
14300 appropriately. However, if C++ code only needs to execute destructors
14301 when Java exceptions are thrown through it, GCC will guess incorrectly.
14302 Sample problematic code is:
14305 struct S @{ ~S(); @};
14306 extern void bar(); // @r{is written in Java, and may throw exceptions}
14315 The usual effect of an incorrect guess is a link failure, complaining of
14316 a missing routine called @samp{__gxx_personality_v0}.
14318 You can inform the compiler that Java exceptions are to be used in a
14319 translation unit, irrespective of what it might think, by writing
14320 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
14321 @samp{#pragma} must appear before any functions that throw or catch
14322 exceptions, or run destructors when exceptions are thrown through them.
14324 You cannot mix Java and C++ exceptions in the same translation unit. It
14325 is believed to be safe to throw a C++ exception from one file through
14326 another file compiled for the Java exception model, or vice versa, but
14327 there may be bugs in this area.
14329 @node Deprecated Features
14330 @section Deprecated Features
14332 In the past, the GNU C++ compiler was extended to experiment with new
14333 features, at a time when the C++ language was still evolving. Now that
14334 the C++ standard is complete, some of those features are superseded by
14335 superior alternatives. Using the old features might cause a warning in
14336 some cases that the feature will be dropped in the future. In other
14337 cases, the feature might be gone already.
14339 While the list below is not exhaustive, it documents some of the options
14340 that are now deprecated:
14343 @item -fexternal-templates
14344 @itemx -falt-external-templates
14345 These are two of the many ways for G++ to implement template
14346 instantiation. @xref{Template Instantiation}. The C++ standard clearly
14347 defines how template definitions have to be organized across
14348 implementation units. G++ has an implicit instantiation mechanism that
14349 should work just fine for standard-conforming code.
14351 @item -fstrict-prototype
14352 @itemx -fno-strict-prototype
14353 Previously it was possible to use an empty prototype parameter list to
14354 indicate an unspecified number of parameters (like C), rather than no
14355 parameters, as C++ demands. This feature has been removed, except where
14356 it is required for backwards compatibility. @xref{Backwards Compatibility}.
14359 G++ allows a virtual function returning @samp{void *} to be overridden
14360 by one returning a different pointer type. This extension to the
14361 covariant return type rules is now deprecated and will be removed from a
14364 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
14365 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
14366 and are now removed from G++. Code using these operators should be
14367 modified to use @code{std::min} and @code{std::max} instead.
14369 The named return value extension has been deprecated, and is now
14372 The use of initializer lists with new expressions has been deprecated,
14373 and is now removed from G++.
14375 Floating and complex non-type template parameters have been deprecated,
14376 and are now removed from G++.
14378 The implicit typename extension has been deprecated and is now
14381 The use of default arguments in function pointers, function typedefs
14382 and other places where they are not permitted by the standard is
14383 deprecated and will be removed from a future version of G++.
14385 G++ allows floating-point literals to appear in integral constant expressions,
14386 e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} }
14387 This extension is deprecated and will be removed from a future version.
14389 G++ allows static data members of const floating-point type to be declared
14390 with an initializer in a class definition. The standard only allows
14391 initializers for static members of const integral types and const
14392 enumeration types so this extension has been deprecated and will be removed
14393 from a future version.
14395 @node Backwards Compatibility
14396 @section Backwards Compatibility
14397 @cindex Backwards Compatibility
14398 @cindex ARM [Annotated C++ Reference Manual]
14400 Now that there is a definitive ISO standard C++, G++ has a specification
14401 to adhere to. The C++ language evolved over time, and features that
14402 used to be acceptable in previous drafts of the standard, such as the ARM
14403 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
14404 compilation of C++ written to such drafts, G++ contains some backwards
14405 compatibilities. @emph{All such backwards compatibility features are
14406 liable to disappear in future versions of G++.} They should be considered
14407 deprecated. @xref{Deprecated Features}.
14411 If a variable is declared at for scope, it used to remain in scope until
14412 the end of the scope which contained the for statement (rather than just
14413 within the for scope). G++ retains this, but issues a warning, if such a
14414 variable is accessed outside the for scope.
14416 @item Implicit C language
14417 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
14418 scope to set the language. On such systems, all header files are
14419 implicitly scoped inside a C language scope. Also, an empty prototype
14420 @code{()} will be treated as an unspecified number of arguments, rather
14421 than no arguments, as C++ demands.