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, 2011
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 * __sync Builtins:: Legacy built-in functions for atomic memory access.
83 * __atomic Builtins:: Atomic built-in functions with memory model.
84 * Object Size Checking:: Built-in functions for limited buffer overflow
86 * Other Builtins:: Other built-in functions.
87 * Target Builtins:: Built-in functions specific to particular targets.
88 * Target Format Checks:: Format checks specific to particular targets.
89 * Pragmas:: Pragmas accepted by GCC.
90 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
91 * Thread-Local:: Per-thread variables.
92 * Binary constants:: Binary constants using the @samp{0b} prefix.
96 @section Statements and Declarations in Expressions
97 @cindex statements inside expressions
98 @cindex declarations inside expressions
99 @cindex expressions containing statements
100 @cindex macros, statements in expressions
102 @c the above section title wrapped and causes an underfull hbox.. i
103 @c changed it from "within" to "in". --mew 4feb93
104 A compound statement enclosed in parentheses may appear as an expression
105 in GNU C@. This allows you to use loops, switches, and local variables
106 within an expression.
108 Recall that a compound statement is a sequence of statements surrounded
109 by braces; in this construct, parentheses go around the braces. For
113 (@{ int y = foo (); int z;
120 is a valid (though slightly more complex than necessary) expression
121 for the absolute value of @code{foo ()}.
123 The last thing in the compound statement should be an expression
124 followed by a semicolon; the value of this subexpression serves as the
125 value of the entire construct. (If you use some other kind of statement
126 last within the braces, the construct has type @code{void}, and thus
127 effectively no value.)
129 This feature is especially useful in making macro definitions ``safe'' (so
130 that they evaluate each operand exactly once). For example, the
131 ``maximum'' function is commonly defined as a macro in standard C as
135 #define max(a,b) ((a) > (b) ? (a) : (b))
139 @cindex side effects, macro argument
140 But this definition computes either @var{a} or @var{b} twice, with bad
141 results if the operand has side effects. In GNU C, if you know the
142 type of the operands (here taken as @code{int}), you can define
143 the macro safely as follows:
146 #define maxint(a,b) \
147 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
150 Embedded statements are not allowed in constant expressions, such as
151 the value of an enumeration constant, the width of a bit-field, or
152 the initial value of a static variable.
154 If you don't know the type of the operand, you can still do this, but you
155 must use @code{typeof} (@pxref{Typeof}).
157 In G++, the result value of a statement expression undergoes array and
158 function pointer decay, and is returned by value to the enclosing
159 expression. For instance, if @code{A} is a class, then
168 will construct a temporary @code{A} object to hold the result of the
169 statement expression, and that will be used to invoke @code{Foo}.
170 Therefore the @code{this} pointer observed by @code{Foo} will not be the
173 Any temporaries created within a statement within a statement expression
174 will be destroyed at the statement's end. This makes statement
175 expressions inside macros slightly different from function calls. In
176 the latter case temporaries introduced during argument evaluation will
177 be destroyed at the end of the statement that includes the function
178 call. In the statement expression case they will be destroyed during
179 the statement expression. For instance,
182 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
183 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
193 will have different places where temporaries are destroyed. For the
194 @code{macro} case, the temporary @code{X} will be destroyed just after
195 the initialization of @code{b}. In the @code{function} case that
196 temporary will be destroyed when the function returns.
198 These considerations mean that it is probably a bad idea to use
199 statement-expressions of this form in header files that are designed to
200 work with C++. (Note that some versions of the GNU C Library contained
201 header files using statement-expression that lead to precisely this
204 Jumping into a statement expression with @code{goto} or using a
205 @code{switch} statement outside the statement expression with a
206 @code{case} or @code{default} label inside the statement expression is
207 not permitted. Jumping into a statement expression with a computed
208 @code{goto} (@pxref{Labels as Values}) yields undefined behavior.
209 Jumping out of a statement expression is permitted, but if the
210 statement expression is part of a larger expression then it is
211 unspecified which other subexpressions of that expression have been
212 evaluated except where the language definition requires certain
213 subexpressions to be evaluated before or after the statement
214 expression. In any case, as with a function call the evaluation of a
215 statement expression is not interleaved with the evaluation of other
216 parts of the containing expression. For example,
219 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
223 will call @code{foo} and @code{bar1} and will not call @code{baz} but
224 may or may not call @code{bar2}. If @code{bar2} is called, it will be
225 called after @code{foo} and before @code{bar1}
228 @section Locally Declared Labels
230 @cindex macros, local labels
232 GCC allows you to declare @dfn{local labels} in any nested block
233 scope. A local label is just like an ordinary label, but you can
234 only reference it (with a @code{goto} statement, or by taking its
235 address) within the block in which it was declared.
237 A local label declaration looks like this:
240 __label__ @var{label};
247 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
250 Local label declarations must come at the beginning of the block,
251 before any ordinary declarations or statements.
253 The label declaration defines the label @emph{name}, but does not define
254 the label itself. You must do this in the usual way, with
255 @code{@var{label}:}, within the statements of the statement expression.
257 The local label feature is useful for complex macros. If a macro
258 contains nested loops, a @code{goto} can be useful for breaking out of
259 them. However, an ordinary label whose scope is the whole function
260 cannot be used: if the macro can be expanded several times in one
261 function, the label will be multiply defined in that function. A
262 local label avoids this problem. For example:
265 #define SEARCH(value, array, target) \
268 typeof (target) _SEARCH_target = (target); \
269 typeof (*(array)) *_SEARCH_array = (array); \
272 for (i = 0; i < max; i++) \
273 for (j = 0; j < max; j++) \
274 if (_SEARCH_array[i][j] == _SEARCH_target) \
275 @{ (value) = i; goto found; @} \
281 This could also be written using a statement-expression:
284 #define SEARCH(array, target) \
287 typeof (target) _SEARCH_target = (target); \
288 typeof (*(array)) *_SEARCH_array = (array); \
291 for (i = 0; i < max; i++) \
292 for (j = 0; j < max; j++) \
293 if (_SEARCH_array[i][j] == _SEARCH_target) \
294 @{ value = i; goto found; @} \
301 Local label declarations also make the labels they declare visible to
302 nested functions, if there are any. @xref{Nested Functions}, for details.
304 @node Labels as Values
305 @section Labels as Values
306 @cindex labels as values
307 @cindex computed gotos
308 @cindex goto with computed label
309 @cindex address of a label
311 You can get the address of a label defined in the current function
312 (or a containing function) with the unary operator @samp{&&}. The
313 value has type @code{void *}. This value is a constant and can be used
314 wherever a constant of that type is valid. For example:
322 To use these values, you need to be able to jump to one. This is done
323 with the computed goto statement@footnote{The analogous feature in
324 Fortran is called an assigned goto, but that name seems inappropriate in
325 C, where one can do more than simply store label addresses in label
326 variables.}, @code{goto *@var{exp};}. For example,
333 Any expression of type @code{void *} is allowed.
335 One way of using these constants is in initializing a static array that
336 will serve as a jump table:
339 static void *array[] = @{ &&foo, &&bar, &&hack @};
342 Then you can select a label with indexing, like this:
349 Note that this does not check whether the subscript is in bounds---array
350 indexing in C never does that.
352 Such an array of label values serves a purpose much like that of the
353 @code{switch} statement. The @code{switch} statement is cleaner, so
354 use that rather than an array unless the problem does not fit a
355 @code{switch} statement very well.
357 Another use of label values is in an interpreter for threaded code.
358 The labels within the interpreter function can be stored in the
359 threaded code for super-fast dispatching.
361 You may not use this mechanism to jump to code in a different function.
362 If you do that, totally unpredictable things will happen. The best way to
363 avoid this is to store the label address only in automatic variables and
364 never pass it as an argument.
366 An alternate way to write the above example is
369 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
371 goto *(&&foo + array[i]);
375 This is more friendly to code living in shared libraries, as it reduces
376 the number of dynamic relocations that are needed, and by consequence,
377 allows the data to be read-only.
379 The @code{&&foo} expressions for the same label might have different
380 values if the containing function is inlined or cloned. If a program
381 relies on them being always the same,
382 @code{__attribute__((__noinline__,__noclone__))} should be used to
383 prevent inlining and cloning. If @code{&&foo} is used in a static
384 variable initializer, inlining and cloning is forbidden.
386 @node Nested Functions
387 @section Nested Functions
388 @cindex nested functions
389 @cindex downward funargs
392 A @dfn{nested function} is a function defined inside another function.
393 (Nested functions are not supported for GNU C++.) The nested function's
394 name is local to the block where it is defined. For example, here we
395 define a nested function named @code{square}, and call it twice:
399 foo (double a, double b)
401 double square (double z) @{ return z * z; @}
403 return square (a) + square (b);
408 The nested function can access all the variables of the containing
409 function that are visible at the point of its definition. This is
410 called @dfn{lexical scoping}. For example, here we show a nested
411 function which uses an inherited variable named @code{offset}:
415 bar (int *array, int offset, int size)
417 int access (int *array, int index)
418 @{ return array[index + offset]; @}
421 for (i = 0; i < size; i++)
422 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
427 Nested function definitions are permitted within functions in the places
428 where variable definitions are allowed; that is, in any block, mixed
429 with the other declarations and statements in the block.
431 It is possible to call the nested function from outside the scope of its
432 name by storing its address or passing the address to another function:
435 hack (int *array, int size)
437 void store (int index, int value)
438 @{ array[index] = value; @}
440 intermediate (store, size);
444 Here, the function @code{intermediate} receives the address of
445 @code{store} as an argument. If @code{intermediate} calls @code{store},
446 the arguments given to @code{store} are used to store into @code{array}.
447 But this technique works only so long as the containing function
448 (@code{hack}, in this example) does not exit.
450 If you try to call the nested function through its address after the
451 containing function has exited, all hell will break loose. If you try
452 to call it after a containing scope level has exited, and if it refers
453 to some of the variables that are no longer in scope, you may be lucky,
454 but it's not wise to take the risk. If, however, the nested function
455 does not refer to anything that has gone out of scope, you should be
458 GCC implements taking the address of a nested function using a technique
459 called @dfn{trampolines}. This technique was described in
460 @cite{Lexical Closures for C++} (Thomas M. Breuel, USENIX
461 C++ Conference Proceedings, October 17-21, 1988).
463 A nested function can jump to a label inherited from a containing
464 function, provided the label was explicitly declared in the containing
465 function (@pxref{Local Labels}). Such a jump returns instantly to the
466 containing function, exiting the nested function which did the
467 @code{goto} and any intermediate functions as well. Here is an example:
471 bar (int *array, int offset, int size)
474 int access (int *array, int index)
478 return array[index + offset];
482 for (i = 0; i < size; i++)
483 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
487 /* @r{Control comes here from @code{access}
488 if it detects an error.} */
495 A nested function always has no linkage. Declaring one with
496 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
497 before its definition, use @code{auto} (which is otherwise meaningless
498 for function declarations).
501 bar (int *array, int offset, int size)
504 auto int access (int *, int);
506 int access (int *array, int index)
510 return array[index + offset];
516 @node Constructing Calls
517 @section Constructing Function Calls
518 @cindex constructing calls
519 @cindex forwarding calls
521 Using the built-in functions described below, you can record
522 the arguments a function received, and call another function
523 with the same arguments, without knowing the number or types
526 You can also record the return value of that function call,
527 and later return that value, without knowing what data type
528 the function tried to return (as long as your caller expects
531 However, these built-in functions may interact badly with some
532 sophisticated features or other extensions of the language. It
533 is, therefore, not recommended to use them outside very simple
534 functions acting as mere forwarders for their arguments.
536 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
537 This built-in function returns a pointer to data
538 describing how to perform a call with the same arguments as were passed
539 to the current function.
541 The function saves the arg pointer register, structure value address,
542 and all registers that might be used to pass arguments to a function
543 into a block of memory allocated on the stack. Then it returns the
544 address of that block.
547 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
548 This built-in function invokes @var{function}
549 with a copy of the parameters described by @var{arguments}
552 The value of @var{arguments} should be the value returned by
553 @code{__builtin_apply_args}. The argument @var{size} specifies the size
554 of the stack argument data, in bytes.
556 This function returns a pointer to data describing
557 how to return whatever value was returned by @var{function}. The data
558 is saved in a block of memory allocated on the stack.
560 It is not always simple to compute the proper value for @var{size}. The
561 value is used by @code{__builtin_apply} to compute the amount of data
562 that should be pushed on the stack and copied from the incoming argument
566 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
567 This built-in function returns the value described by @var{result} from
568 the containing function. You should specify, for @var{result}, a value
569 returned by @code{__builtin_apply}.
572 @deftypefn {Built-in Function} {} __builtin_va_arg_pack ()
573 This built-in function represents all anonymous arguments of an inline
574 function. It can be used only in inline functions which will be always
575 inlined, never compiled as a separate function, such as those using
576 @code{__attribute__ ((__always_inline__))} or
577 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
578 It must be only passed as last argument to some other function
579 with variable arguments. This is useful for writing small wrapper
580 inlines for variable argument functions, when using preprocessor
581 macros is undesirable. For example:
583 extern int myprintf (FILE *f, const char *format, ...);
584 extern inline __attribute__ ((__gnu_inline__)) int
585 myprintf (FILE *f, const char *format, ...)
587 int r = fprintf (f, "myprintf: ");
590 int s = fprintf (f, format, __builtin_va_arg_pack ());
598 @deftypefn {Built-in Function} {size_t} __builtin_va_arg_pack_len ()
599 This built-in function returns the number of anonymous arguments of
600 an inline function. It can be used only in inline functions which
601 will be always inlined, never compiled as a separate function, such
602 as those using @code{__attribute__ ((__always_inline__))} or
603 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
604 For example following will do link or runtime checking of open
605 arguments for optimized code:
608 extern inline __attribute__((__gnu_inline__)) int
609 myopen (const char *path, int oflag, ...)
611 if (__builtin_va_arg_pack_len () > 1)
612 warn_open_too_many_arguments ();
614 if (__builtin_constant_p (oflag))
616 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1)
618 warn_open_missing_mode ();
619 return __open_2 (path, oflag);
621 return open (path, oflag, __builtin_va_arg_pack ());
624 if (__builtin_va_arg_pack_len () < 1)
625 return __open_2 (path, oflag);
627 return open (path, oflag, __builtin_va_arg_pack ());
634 @section Referring to a Type with @code{typeof}
637 @cindex macros, types of arguments
639 Another way to refer to the type of an expression is with @code{typeof}.
640 The syntax of using of this keyword looks like @code{sizeof}, but the
641 construct acts semantically like a type name defined with @code{typedef}.
643 There are two ways of writing the argument to @code{typeof}: with an
644 expression or with a type. Here is an example with an expression:
651 This assumes that @code{x} is an array of pointers to functions;
652 the type described is that of the values of the functions.
654 Here is an example with a typename as the argument:
661 Here the type described is that of pointers to @code{int}.
663 If you are writing a header file that must work when included in ISO C
664 programs, write @code{__typeof__} instead of @code{typeof}.
665 @xref{Alternate Keywords}.
667 A @code{typeof}-construct can be used anywhere a typedef name could be
668 used. For example, you can use it in a declaration, in a cast, or inside
669 of @code{sizeof} or @code{typeof}.
671 The operand of @code{typeof} is evaluated for its side effects if and
672 only if it is an expression of variably modified type or the name of
675 @code{typeof} is often useful in conjunction with the
676 statements-within-expressions feature. Here is how the two together can
677 be used to define a safe ``maximum'' macro that operates on any
678 arithmetic type and evaluates each of its arguments exactly once:
682 (@{ typeof (a) _a = (a); \
683 typeof (b) _b = (b); \
684 _a > _b ? _a : _b; @})
687 @cindex underscores in variables in macros
688 @cindex @samp{_} in variables in macros
689 @cindex local variables in macros
690 @cindex variables, local, in macros
691 @cindex macros, local variables in
693 The reason for using names that start with underscores for the local
694 variables is to avoid conflicts with variable names that occur within the
695 expressions that are substituted for @code{a} and @code{b}. Eventually we
696 hope to design a new form of declaration syntax that allows you to declare
697 variables whose scopes start only after their initializers; this will be a
698 more reliable way to prevent such conflicts.
701 Some more examples of the use of @code{typeof}:
705 This declares @code{y} with the type of what @code{x} points to.
712 This declares @code{y} as an array of such values.
719 This declares @code{y} as an array of pointers to characters:
722 typeof (typeof (char *)[4]) y;
726 It is equivalent to the following traditional C declaration:
732 To see the meaning of the declaration using @code{typeof}, and why it
733 might be a useful way to write, rewrite it with these macros:
736 #define pointer(T) typeof(T *)
737 #define array(T, N) typeof(T [N])
741 Now the declaration can be rewritten this way:
744 array (pointer (char), 4) y;
748 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
749 pointers to @code{char}.
752 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
753 a more limited extension which permitted one to write
756 typedef @var{T} = @var{expr};
760 with the effect of declaring @var{T} to have the type of the expression
761 @var{expr}. This extension does not work with GCC 3 (versions between
762 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
763 relies on it should be rewritten to use @code{typeof}:
766 typedef typeof(@var{expr}) @var{T};
770 This will work with all versions of GCC@.
773 @section Conditionals with Omitted Operands
774 @cindex conditional expressions, extensions
775 @cindex omitted middle-operands
776 @cindex middle-operands, omitted
777 @cindex extensions, @code{?:}
778 @cindex @code{?:} extensions
780 The middle operand in a conditional expression may be omitted. Then
781 if the first operand is nonzero, its value is the value of the conditional
784 Therefore, the expression
791 has the value of @code{x} if that is nonzero; otherwise, the value of
794 This example is perfectly equivalent to
800 @cindex side effect in @code{?:}
801 @cindex @code{?:} side effect
803 In this simple case, the ability to omit the middle operand is not
804 especially useful. When it becomes useful is when the first operand does,
805 or may (if it is a macro argument), contain a side effect. Then repeating
806 the operand in the middle would perform the side effect twice. Omitting
807 the middle operand uses the value already computed without the undesirable
808 effects of recomputing it.
811 @section 128-bits integers
812 @cindex @code{__int128} data types
814 As an extension the integer scalar type @code{__int128} is supported for
815 targets having an integer mode wide enough to hold 128-bit.
816 Simply write @code{__int128} for a signed 128-bit integer, or
817 @code{unsigned __int128} for an unsigned 128-bit integer. There is no
818 support in GCC to express an integer constant of type @code{__int128}
819 for targets having @code{long long} integer with less then 128 bit width.
822 @section Double-Word Integers
823 @cindex @code{long long} data types
824 @cindex double-word arithmetic
825 @cindex multiprecision arithmetic
826 @cindex @code{LL} integer suffix
827 @cindex @code{ULL} integer suffix
829 ISO C99 supports data types for integers that are at least 64 bits wide,
830 and as an extension GCC supports them in C90 mode and in C++.
831 Simply write @code{long long int} for a signed integer, or
832 @code{unsigned long long int} for an unsigned integer. To make an
833 integer constant of type @code{long long int}, add the suffix @samp{LL}
834 to the integer. To make an integer constant of type @code{unsigned long
835 long int}, add the suffix @samp{ULL} to the integer.
837 You can use these types in arithmetic like any other integer types.
838 Addition, subtraction, and bitwise boolean operations on these types
839 are open-coded on all types of machines. Multiplication is open-coded
840 if the machine supports fullword-to-doubleword a widening multiply
841 instruction. Division and shifts are open-coded only on machines that
842 provide special support. The operations that are not open-coded use
843 special library routines that come with GCC@.
845 There may be pitfalls when you use @code{long long} types for function
846 arguments, unless you declare function prototypes. If a function
847 expects type @code{int} for its argument, and you pass a value of type
848 @code{long long int}, confusion will result because the caller and the
849 subroutine will disagree about the number of bytes for the argument.
850 Likewise, if the function expects @code{long long int} and you pass
851 @code{int}. The best way to avoid such problems is to use prototypes.
854 @section Complex Numbers
855 @cindex complex numbers
856 @cindex @code{_Complex} keyword
857 @cindex @code{__complex__} keyword
859 ISO C99 supports complex floating data types, and as an extension GCC
860 supports them in C90 mode and in C++, and supports complex integer data
861 types which are not part of ISO C99. You can declare complex types
862 using the keyword @code{_Complex}. As an extension, the older GNU
863 keyword @code{__complex__} is also supported.
865 For example, @samp{_Complex double x;} declares @code{x} as a
866 variable whose real part and imaginary part are both of type
867 @code{double}. @samp{_Complex short int y;} declares @code{y} to
868 have real and imaginary parts of type @code{short int}; this is not
869 likely to be useful, but it shows that the set of complex types is
872 To write a constant with a complex data type, use the suffix @samp{i} or
873 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
874 has type @code{_Complex float} and @code{3i} has type
875 @code{_Complex int}. Such a constant always has a pure imaginary
876 value, but you can form any complex value you like by adding one to a
877 real constant. This is a GNU extension; if you have an ISO C99
878 conforming C library (such as GNU libc), and want to construct complex
879 constants of floating type, you should include @code{<complex.h>} and
880 use the macros @code{I} or @code{_Complex_I} instead.
882 @cindex @code{__real__} keyword
883 @cindex @code{__imag__} keyword
884 To extract the real part of a complex-valued expression @var{exp}, write
885 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
886 extract the imaginary part. This is a GNU extension; for values of
887 floating type, you should use the ISO C99 functions @code{crealf},
888 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
889 @code{cimagl}, declared in @code{<complex.h>} and also provided as
890 built-in functions by GCC@.
892 @cindex complex conjugation
893 The operator @samp{~} performs complex conjugation when used on a value
894 with a complex type. This is a GNU extension; for values of
895 floating type, you should use the ISO C99 functions @code{conjf},
896 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
897 provided as built-in functions by GCC@.
899 GCC can allocate complex automatic variables in a noncontiguous
900 fashion; it's even possible for the real part to be in a register while
901 the imaginary part is on the stack (or vice-versa). Only the DWARF2
902 debug info format can represent this, so use of DWARF2 is recommended.
903 If you are using the stabs debug info format, GCC describes a noncontiguous
904 complex variable as if it were two separate variables of noncomplex type.
905 If the variable's actual name is @code{foo}, the two fictitious
906 variables are named @code{foo$real} and @code{foo$imag}. You can
907 examine and set these two fictitious variables with your debugger.
910 @section Additional Floating Types
911 @cindex additional floating types
912 @cindex @code{__float80} data type
913 @cindex @code{__float128} data type
914 @cindex @code{w} floating point suffix
915 @cindex @code{q} floating point suffix
916 @cindex @code{W} floating point suffix
917 @cindex @code{Q} floating point suffix
919 As an extension, the GNU C compiler supports additional floating
920 types, @code{__float80} and @code{__float128} to support 80bit
921 (@code{XFmode}) and 128 bit (@code{TFmode}) floating types.
922 Support for additional types includes the arithmetic operators:
923 add, subtract, multiply, divide; unary arithmetic operators;
924 relational operators; equality operators; and conversions to and from
925 integer and other floating types. Use a suffix @samp{w} or @samp{W}
926 in a literal constant of type @code{__float80} and @samp{q} or @samp{Q}
927 for @code{_float128}. You can declare complex types using the
928 corresponding internal complex type, @code{XCmode} for @code{__float80}
929 type and @code{TCmode} for @code{__float128} type:
932 typedef _Complex float __attribute__((mode(TC))) _Complex128;
933 typedef _Complex float __attribute__((mode(XC))) _Complex80;
936 Not all targets support additional floating point types. @code{__float80}
937 and @code{__float128} types are supported on i386, x86_64 and ia64 targets.
938 The @code{__float128} type is supported on hppa HP-UX targets.
941 @section Half-Precision Floating Point
942 @cindex half-precision floating point
943 @cindex @code{__fp16} data type
945 On ARM targets, GCC supports half-precision (16-bit) floating point via
946 the @code{__fp16} type. You must enable this type explicitly
947 with the @option{-mfp16-format} command-line option in order to use it.
949 ARM supports two incompatible representations for half-precision
950 floating-point values. You must choose one of the representations and
951 use it consistently in your program.
953 Specifying @option{-mfp16-format=ieee} selects the IEEE 754-2008 format.
954 This format can represent normalized values in the range of @math{2^{-14}} to 65504.
955 There are 11 bits of significand precision, approximately 3
958 Specifying @option{-mfp16-format=alternative} selects the ARM
959 alternative format. This representation is similar to the IEEE
960 format, but does not support infinities or NaNs. Instead, the range
961 of exponents is extended, so that this format can represent normalized
962 values in the range of @math{2^{-14}} to 131008.
964 The @code{__fp16} type is a storage format only. For purposes
965 of arithmetic and other operations, @code{__fp16} values in C or C++
966 expressions are automatically promoted to @code{float}. In addition,
967 you cannot declare a function with a return value or parameters
968 of type @code{__fp16}.
970 Note that conversions from @code{double} to @code{__fp16}
971 involve an intermediate conversion to @code{float}. Because
972 of rounding, this can sometimes produce a different result than a
975 ARM provides hardware support for conversions between
976 @code{__fp16} and @code{float} values
977 as an extension to VFP and NEON (Advanced SIMD). GCC generates
978 code using these hardware instructions if you compile with
979 options to select an FPU that provides them;
980 for example, @option{-mfpu=neon-fp16 -mfloat-abi=softfp},
981 in addition to the @option{-mfp16-format} option to select
982 a half-precision format.
984 Language-level support for the @code{__fp16} data type is
985 independent of whether GCC generates code using hardware floating-point
986 instructions. In cases where hardware support is not specified, GCC
987 implements conversions between @code{__fp16} and @code{float} values
991 @section Decimal Floating Types
992 @cindex decimal floating types
993 @cindex @code{_Decimal32} data type
994 @cindex @code{_Decimal64} data type
995 @cindex @code{_Decimal128} data type
996 @cindex @code{df} integer suffix
997 @cindex @code{dd} integer suffix
998 @cindex @code{dl} integer suffix
999 @cindex @code{DF} integer suffix
1000 @cindex @code{DD} integer suffix
1001 @cindex @code{DL} integer suffix
1003 As an extension, the GNU C compiler supports decimal floating types as
1004 defined in the N1312 draft of ISO/IEC WDTR24732. Support for decimal
1005 floating types in GCC will evolve as the draft technical report changes.
1006 Calling conventions for any target might also change. Not all targets
1007 support decimal floating types.
1009 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
1010 @code{_Decimal128}. They use a radix of ten, unlike the floating types
1011 @code{float}, @code{double}, and @code{long double} whose radix is not
1012 specified by the C standard but is usually two.
1014 Support for decimal floating types includes the arithmetic operators
1015 add, subtract, multiply, divide; unary arithmetic operators;
1016 relational operators; equality operators; and conversions to and from
1017 integer and other floating types. Use a suffix @samp{df} or
1018 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
1019 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
1022 GCC support of decimal float as specified by the draft technical report
1027 When the value of a decimal floating type cannot be represented in the
1028 integer type to which it is being converted, the result is undefined
1029 rather than the result value specified by the draft technical report.
1032 GCC does not provide the C library functionality associated with
1033 @file{math.h}, @file{fenv.h}, @file{stdio.h}, @file{stdlib.h}, and
1034 @file{wchar.h}, which must come from a separate C library implementation.
1035 Because of this the GNU C compiler does not define macro
1036 @code{__STDC_DEC_FP__} to indicate that the implementation conforms to
1037 the technical report.
1040 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
1041 are supported by the DWARF2 debug information format.
1047 ISO C99 supports floating-point numbers written not only in the usual
1048 decimal notation, such as @code{1.55e1}, but also numbers such as
1049 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1050 supports this in C90 mode (except in some cases when strictly
1051 conforming) and in C++. In that format the
1052 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1053 mandatory. The exponent is a decimal number that indicates the power of
1054 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1061 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1062 is the same as @code{1.55e1}.
1064 Unlike for floating-point numbers in the decimal notation the exponent
1065 is always required in the hexadecimal notation. Otherwise the compiler
1066 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1067 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1068 extension for floating-point constants of type @code{float}.
1071 @section Fixed-Point Types
1072 @cindex fixed-point types
1073 @cindex @code{_Fract} data type
1074 @cindex @code{_Accum} data type
1075 @cindex @code{_Sat} data type
1076 @cindex @code{hr} fixed-suffix
1077 @cindex @code{r} fixed-suffix
1078 @cindex @code{lr} fixed-suffix
1079 @cindex @code{llr} fixed-suffix
1080 @cindex @code{uhr} fixed-suffix
1081 @cindex @code{ur} fixed-suffix
1082 @cindex @code{ulr} fixed-suffix
1083 @cindex @code{ullr} fixed-suffix
1084 @cindex @code{hk} fixed-suffix
1085 @cindex @code{k} fixed-suffix
1086 @cindex @code{lk} fixed-suffix
1087 @cindex @code{llk} fixed-suffix
1088 @cindex @code{uhk} fixed-suffix
1089 @cindex @code{uk} fixed-suffix
1090 @cindex @code{ulk} fixed-suffix
1091 @cindex @code{ullk} fixed-suffix
1092 @cindex @code{HR} fixed-suffix
1093 @cindex @code{R} fixed-suffix
1094 @cindex @code{LR} fixed-suffix
1095 @cindex @code{LLR} fixed-suffix
1096 @cindex @code{UHR} fixed-suffix
1097 @cindex @code{UR} fixed-suffix
1098 @cindex @code{ULR} fixed-suffix
1099 @cindex @code{ULLR} fixed-suffix
1100 @cindex @code{HK} fixed-suffix
1101 @cindex @code{K} fixed-suffix
1102 @cindex @code{LK} fixed-suffix
1103 @cindex @code{LLK} fixed-suffix
1104 @cindex @code{UHK} fixed-suffix
1105 @cindex @code{UK} fixed-suffix
1106 @cindex @code{ULK} fixed-suffix
1107 @cindex @code{ULLK} fixed-suffix
1109 As an extension, the GNU C compiler supports fixed-point types as
1110 defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point
1111 types in GCC will evolve as the draft technical report changes.
1112 Calling conventions for any target might also change. Not all targets
1113 support fixed-point types.
1115 The fixed-point types are
1116 @code{short _Fract},
1119 @code{long long _Fract},
1120 @code{unsigned short _Fract},
1121 @code{unsigned _Fract},
1122 @code{unsigned long _Fract},
1123 @code{unsigned long long _Fract},
1124 @code{_Sat short _Fract},
1126 @code{_Sat long _Fract},
1127 @code{_Sat long long _Fract},
1128 @code{_Sat unsigned short _Fract},
1129 @code{_Sat unsigned _Fract},
1130 @code{_Sat unsigned long _Fract},
1131 @code{_Sat unsigned long long _Fract},
1132 @code{short _Accum},
1135 @code{long long _Accum},
1136 @code{unsigned short _Accum},
1137 @code{unsigned _Accum},
1138 @code{unsigned long _Accum},
1139 @code{unsigned long long _Accum},
1140 @code{_Sat short _Accum},
1142 @code{_Sat long _Accum},
1143 @code{_Sat long long _Accum},
1144 @code{_Sat unsigned short _Accum},
1145 @code{_Sat unsigned _Accum},
1146 @code{_Sat unsigned long _Accum},
1147 @code{_Sat unsigned long long _Accum}.
1149 Fixed-point data values contain fractional and optional integral parts.
1150 The format of fixed-point data varies and depends on the target machine.
1152 Support for fixed-point types includes:
1155 prefix and postfix increment and decrement operators (@code{++}, @code{--})
1157 unary arithmetic operators (@code{+}, @code{-}, @code{!})
1159 binary arithmetic operators (@code{+}, @code{-}, @code{*}, @code{/})
1161 binary shift operators (@code{<<}, @code{>>})
1163 relational operators (@code{<}, @code{<=}, @code{>=}, @code{>})
1165 equality operators (@code{==}, @code{!=})
1167 assignment operators (@code{+=}, @code{-=}, @code{*=}, @code{/=},
1168 @code{<<=}, @code{>>=})
1170 conversions to and from integer, floating-point, or fixed-point types
1173 Use a suffix in a fixed-point literal constant:
1175 @item @samp{hr} or @samp{HR} for @code{short _Fract} and
1176 @code{_Sat short _Fract}
1177 @item @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract}
1178 @item @samp{lr} or @samp{LR} for @code{long _Fract} and
1179 @code{_Sat long _Fract}
1180 @item @samp{llr} or @samp{LLR} for @code{long long _Fract} and
1181 @code{_Sat long long _Fract}
1182 @item @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and
1183 @code{_Sat unsigned short _Fract}
1184 @item @samp{ur} or @samp{UR} for @code{unsigned _Fract} and
1185 @code{_Sat unsigned _Fract}
1186 @item @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and
1187 @code{_Sat unsigned long _Fract}
1188 @item @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract}
1189 and @code{_Sat unsigned long long _Fract}
1190 @item @samp{hk} or @samp{HK} for @code{short _Accum} and
1191 @code{_Sat short _Accum}
1192 @item @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum}
1193 @item @samp{lk} or @samp{LK} for @code{long _Accum} and
1194 @code{_Sat long _Accum}
1195 @item @samp{llk} or @samp{LLK} for @code{long long _Accum} and
1196 @code{_Sat long long _Accum}
1197 @item @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and
1198 @code{_Sat unsigned short _Accum}
1199 @item @samp{uk} or @samp{UK} for @code{unsigned _Accum} and
1200 @code{_Sat unsigned _Accum}
1201 @item @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and
1202 @code{_Sat unsigned long _Accum}
1203 @item @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum}
1204 and @code{_Sat unsigned long long _Accum}
1207 GCC support of fixed-point types as specified by the draft technical report
1212 Pragmas to control overflow and rounding behaviors are not implemented.
1215 Fixed-point types are supported by the DWARF2 debug information format.
1217 @node Named Address Spaces
1218 @section Named address spaces
1219 @cindex named address spaces
1221 As an extension, the GNU C compiler supports named address spaces as
1222 defined in the N1275 draft of ISO/IEC DTR 18037. Support for named
1223 address spaces in GCC will evolve as the draft technical report changes.
1224 Calling conventions for any target might also change. At present, only
1225 the SPU and M32C targets support other address spaces. On the SPU target, for
1226 example, variables may be declared as belonging to another address space
1227 by qualifying the type with the @code{__ea} address space identifier:
1233 When the variable @code{i} is accessed, the compiler will generate
1234 special code to access this variable. It may use runtime library
1235 support, or generate special machine instructions to access that address
1238 The @code{__ea} identifier may be used exactly like any other C type
1239 qualifier (e.g., @code{const} or @code{volatile}). See the N1275
1240 document for more details.
1242 On the M32C target, with the R8C and M16C cpu variants, variables
1243 qualified with @code{__far} are accessed using 32-bit addresses in
1244 order to access memory beyond the first 64k bytes. If @code{__far} is
1245 used with the M32CM or M32C cpu variants, it has no effect.
1248 @section Arrays of Length Zero
1249 @cindex arrays of length zero
1250 @cindex zero-length arrays
1251 @cindex length-zero arrays
1252 @cindex flexible array members
1254 Zero-length arrays are allowed in GNU C@. They are very useful as the
1255 last element of a structure which is really a header for a variable-length
1264 struct line *thisline = (struct line *)
1265 malloc (sizeof (struct line) + this_length);
1266 thisline->length = this_length;
1269 In ISO C90, you would have to give @code{contents} a length of 1, which
1270 means either you waste space or complicate the argument to @code{malloc}.
1272 In ISO C99, you would use a @dfn{flexible array member}, which is
1273 slightly different in syntax and semantics:
1277 Flexible array members are written as @code{contents[]} without
1281 Flexible array members have incomplete type, and so the @code{sizeof}
1282 operator may not be applied. As a quirk of the original implementation
1283 of zero-length arrays, @code{sizeof} evaluates to zero.
1286 Flexible array members may only appear as the last member of a
1287 @code{struct} that is otherwise non-empty.
1290 A structure containing a flexible array member, or a union containing
1291 such a structure (possibly recursively), may not be a member of a
1292 structure or an element of an array. (However, these uses are
1293 permitted by GCC as extensions.)
1296 GCC versions before 3.0 allowed zero-length arrays to be statically
1297 initialized, as if they were flexible arrays. In addition to those
1298 cases that were useful, it also allowed initializations in situations
1299 that would corrupt later data. Non-empty initialization of zero-length
1300 arrays is now treated like any case where there are more initializer
1301 elements than the array holds, in that a suitable warning about "excess
1302 elements in array" is given, and the excess elements (all of them, in
1303 this case) are ignored.
1305 Instead GCC allows static initialization of flexible array members.
1306 This is equivalent to defining a new structure containing the original
1307 structure followed by an array of sufficient size to contain the data.
1308 I.e.@: in the following, @code{f1} is constructed as if it were declared
1314 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1317 struct f1 f1; int data[3];
1318 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1322 The convenience of this extension is that @code{f1} has the desired
1323 type, eliminating the need to consistently refer to @code{f2.f1}.
1325 This has symmetry with normal static arrays, in that an array of
1326 unknown size is also written with @code{[]}.
1328 Of course, this extension only makes sense if the extra data comes at
1329 the end of a top-level object, as otherwise we would be overwriting
1330 data at subsequent offsets. To avoid undue complication and confusion
1331 with initialization of deeply nested arrays, we simply disallow any
1332 non-empty initialization except when the structure is the top-level
1333 object. For example:
1336 struct foo @{ int x; int y[]; @};
1337 struct bar @{ struct foo z; @};
1339 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1340 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1341 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1342 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1345 @node Empty Structures
1346 @section Structures With No Members
1347 @cindex empty structures
1348 @cindex zero-size structures
1350 GCC permits a C structure to have no members:
1357 The structure will have size zero. In C++, empty structures are part
1358 of the language. G++ treats empty structures as if they had a single
1359 member of type @code{char}.
1361 @node Variable Length
1362 @section Arrays of Variable Length
1363 @cindex variable-length arrays
1364 @cindex arrays of variable length
1367 Variable-length automatic arrays are allowed in ISO C99, and as an
1368 extension GCC accepts them in C90 mode and in C++. These arrays are
1369 declared like any other automatic arrays, but with a length that is not
1370 a constant expression. The storage is allocated at the point of
1371 declaration and deallocated when the brace-level is exited. For
1376 concat_fopen (char *s1, char *s2, char *mode)
1378 char str[strlen (s1) + strlen (s2) + 1];
1381 return fopen (str, mode);
1385 @cindex scope of a variable length array
1386 @cindex variable-length array scope
1387 @cindex deallocating variable length arrays
1388 Jumping or breaking out of the scope of the array name deallocates the
1389 storage. Jumping into the scope is not allowed; you get an error
1392 @cindex @code{alloca} vs variable-length arrays
1393 You can use the function @code{alloca} to get an effect much like
1394 variable-length arrays. The function @code{alloca} is available in
1395 many other C implementations (but not in all). On the other hand,
1396 variable-length arrays are more elegant.
1398 There are other differences between these two methods. Space allocated
1399 with @code{alloca} exists until the containing @emph{function} returns.
1400 The space for a variable-length array is deallocated as soon as the array
1401 name's scope ends. (If you use both variable-length arrays and
1402 @code{alloca} in the same function, deallocation of a variable-length array
1403 will also deallocate anything more recently allocated with @code{alloca}.)
1405 You can also use variable-length arrays as arguments to functions:
1409 tester (int len, char data[len][len])
1415 The length of an array is computed once when the storage is allocated
1416 and is remembered for the scope of the array in case you access it with
1419 If you want to pass the array first and the length afterward, you can
1420 use a forward declaration in the parameter list---another GNU extension.
1424 tester (int len; char data[len][len], int len)
1430 @cindex parameter forward declaration
1431 The @samp{int len} before the semicolon is a @dfn{parameter forward
1432 declaration}, and it serves the purpose of making the name @code{len}
1433 known when the declaration of @code{data} is parsed.
1435 You can write any number of such parameter forward declarations in the
1436 parameter list. They can be separated by commas or semicolons, but the
1437 last one must end with a semicolon, which is followed by the ``real''
1438 parameter declarations. Each forward declaration must match a ``real''
1439 declaration in parameter name and data type. ISO C99 does not support
1440 parameter forward declarations.
1442 @node Variadic Macros
1443 @section Macros with a Variable Number of Arguments.
1444 @cindex variable number of arguments
1445 @cindex macro with variable arguments
1446 @cindex rest argument (in macro)
1447 @cindex variadic macros
1449 In the ISO C standard of 1999, a macro can be declared to accept a
1450 variable number of arguments much as a function can. The syntax for
1451 defining the macro is similar to that of a function. Here is an
1455 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1458 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1459 such a macro, it represents the zero or more tokens until the closing
1460 parenthesis that ends the invocation, including any commas. This set of
1461 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1462 wherever it appears. See the CPP manual for more information.
1464 GCC has long supported variadic macros, and used a different syntax that
1465 allowed you to give a name to the variable arguments just like any other
1466 argument. Here is an example:
1469 #define debug(format, args...) fprintf (stderr, format, args)
1472 This is in all ways equivalent to the ISO C example above, but arguably
1473 more readable and descriptive.
1475 GNU CPP has two further variadic macro extensions, and permits them to
1476 be used with either of the above forms of macro definition.
1478 In standard C, you are not allowed to leave the variable argument out
1479 entirely; but you are allowed to pass an empty argument. For example,
1480 this invocation is invalid in ISO C, because there is no comma after
1487 GNU CPP permits you to completely omit the variable arguments in this
1488 way. In the above examples, the compiler would complain, though since
1489 the expansion of the macro still has the extra comma after the format
1492 To help solve this problem, CPP behaves specially for variable arguments
1493 used with the token paste operator, @samp{##}. If instead you write
1496 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1499 and if the variable arguments are omitted or empty, the @samp{##}
1500 operator causes the preprocessor to remove the comma before it. If you
1501 do provide some variable arguments in your macro invocation, GNU CPP
1502 does not complain about the paste operation and instead places the
1503 variable arguments after the comma. Just like any other pasted macro
1504 argument, these arguments are not macro expanded.
1506 @node Escaped Newlines
1507 @section Slightly Looser Rules for Escaped Newlines
1508 @cindex escaped newlines
1509 @cindex newlines (escaped)
1511 Recently, the preprocessor has relaxed its treatment of escaped
1512 newlines. Previously, the newline had to immediately follow a
1513 backslash. The current implementation allows whitespace in the form
1514 of spaces, horizontal and vertical tabs, and form feeds between the
1515 backslash and the subsequent newline. The preprocessor issues a
1516 warning, but treats it as a valid escaped newline and combines the two
1517 lines to form a single logical line. This works within comments and
1518 tokens, as well as between tokens. Comments are @emph{not} treated as
1519 whitespace for the purposes of this relaxation, since they have not
1520 yet been replaced with spaces.
1523 @section Non-Lvalue Arrays May Have Subscripts
1524 @cindex subscripting
1525 @cindex arrays, non-lvalue
1527 @cindex subscripting and function values
1528 In ISO C99, arrays that are not lvalues still decay to pointers, and
1529 may be subscripted, although they may not be modified or used after
1530 the next sequence point and the unary @samp{&} operator may not be
1531 applied to them. As an extension, GCC allows such arrays to be
1532 subscripted in C90 mode, though otherwise they do not decay to
1533 pointers outside C99 mode. For example,
1534 this is valid in GNU C though not valid in C90:
1538 struct foo @{int a[4];@};
1544 return f().a[index];
1550 @section Arithmetic on @code{void}- and Function-Pointers
1551 @cindex void pointers, arithmetic
1552 @cindex void, size of pointer to
1553 @cindex function pointers, arithmetic
1554 @cindex function, size of pointer to
1556 In GNU C, addition and subtraction operations are supported on pointers to
1557 @code{void} and on pointers to functions. This is done by treating the
1558 size of a @code{void} or of a function as 1.
1560 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1561 and on function types, and returns 1.
1563 @opindex Wpointer-arith
1564 The option @option{-Wpointer-arith} requests a warning if these extensions
1568 @section Non-Constant Initializers
1569 @cindex initializers, non-constant
1570 @cindex non-constant initializers
1572 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1573 automatic variable are not required to be constant expressions in GNU C@.
1574 Here is an example of an initializer with run-time varying elements:
1577 foo (float f, float g)
1579 float beat_freqs[2] = @{ f-g, f+g @};
1584 @node Compound Literals
1585 @section Compound Literals
1586 @cindex constructor expressions
1587 @cindex initializations in expressions
1588 @cindex structures, constructor expression
1589 @cindex expressions, constructor
1590 @cindex compound literals
1591 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1593 ISO C99 supports compound literals. A compound literal looks like
1594 a cast containing an initializer. Its value is an object of the
1595 type specified in the cast, containing the elements specified in
1596 the initializer; it is an lvalue. As an extension, GCC supports
1597 compound literals in C90 mode and in C++.
1599 Usually, the specified type is a structure. Assume that
1600 @code{struct foo} and @code{structure} are declared as shown:
1603 struct foo @{int a; char b[2];@} structure;
1607 Here is an example of constructing a @code{struct foo} with a compound literal:
1610 structure = ((struct foo) @{x + y, 'a', 0@});
1614 This is equivalent to writing the following:
1618 struct foo temp = @{x + y, 'a', 0@};
1623 You can also construct an array. If all the elements of the compound literal
1624 are (made up of) simple constant expressions, suitable for use in
1625 initializers of objects of static storage duration, then the compound
1626 literal can be coerced to a pointer to its first element and used in
1627 such an initializer, as shown here:
1630 char **foo = (char *[]) @{ "x", "y", "z" @};
1633 Compound literals for scalar types and union types are
1634 also allowed, but then the compound literal is equivalent
1637 As a GNU extension, GCC allows initialization of objects with static storage
1638 duration by compound literals (which is not possible in ISO C99, because
1639 the initializer is not a constant).
1640 It is handled as if the object was initialized only with the bracket
1641 enclosed list if the types of the compound literal and the object match.
1642 The initializer list of the compound literal must be constant.
1643 If the object being initialized has array type of unknown size, the size is
1644 determined by compound literal size.
1647 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1648 static int y[] = (int []) @{1, 2, 3@};
1649 static int z[] = (int [3]) @{1@};
1653 The above lines are equivalent to the following:
1655 static struct foo x = @{1, 'a', 'b'@};
1656 static int y[] = @{1, 2, 3@};
1657 static int z[] = @{1, 0, 0@};
1660 @node Designated Inits
1661 @section Designated Initializers
1662 @cindex initializers with labeled elements
1663 @cindex labeled elements in initializers
1664 @cindex case labels in initializers
1665 @cindex designated initializers
1667 Standard C90 requires the elements of an initializer to appear in a fixed
1668 order, the same as the order of the elements in the array or structure
1671 In ISO C99 you can give the elements in any order, specifying the array
1672 indices or structure field names they apply to, and GNU C allows this as
1673 an extension in C90 mode as well. This extension is not
1674 implemented in GNU C++.
1676 To specify an array index, write
1677 @samp{[@var{index}] =} before the element value. For example,
1680 int a[6] = @{ [4] = 29, [2] = 15 @};
1687 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1691 The index values must be constant expressions, even if the array being
1692 initialized is automatic.
1694 An alternative syntax for this which has been obsolete since GCC 2.5 but
1695 GCC still accepts is to write @samp{[@var{index}]} before the element
1696 value, with no @samp{=}.
1698 To initialize a range of elements to the same value, write
1699 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1700 extension. For example,
1703 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1707 If the value in it has side-effects, the side-effects will happen only once,
1708 not for each initialized field by the range initializer.
1711 Note that the length of the array is the highest value specified
1714 In a structure initializer, specify the name of a field to initialize
1715 with @samp{.@var{fieldname} =} before the element value. For example,
1716 given the following structure,
1719 struct point @{ int x, y; @};
1723 the following initialization
1726 struct point p = @{ .y = yvalue, .x = xvalue @};
1733 struct point p = @{ xvalue, yvalue @};
1736 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1737 @samp{@var{fieldname}:}, as shown here:
1740 struct point p = @{ y: yvalue, x: xvalue @};
1744 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1745 @dfn{designator}. You can also use a designator (or the obsolete colon
1746 syntax) when initializing a union, to specify which element of the union
1747 should be used. For example,
1750 union foo @{ int i; double d; @};
1752 union foo f = @{ .d = 4 @};
1756 will convert 4 to a @code{double} to store it in the union using
1757 the second element. By contrast, casting 4 to type @code{union foo}
1758 would store it into the union as the integer @code{i}, since it is
1759 an integer. (@xref{Cast to Union}.)
1761 You can combine this technique of naming elements with ordinary C
1762 initialization of successive elements. Each initializer element that
1763 does not have a designator applies to the next consecutive element of the
1764 array or structure. For example,
1767 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1774 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1777 Labeling the elements of an array initializer is especially useful
1778 when the indices are characters or belong to an @code{enum} type.
1783 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1784 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1787 @cindex designator lists
1788 You can also write a series of @samp{.@var{fieldname}} and
1789 @samp{[@var{index}]} designators before an @samp{=} to specify a
1790 nested subobject to initialize; the list is taken relative to the
1791 subobject corresponding to the closest surrounding brace pair. For
1792 example, with the @samp{struct point} declaration above:
1795 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1799 If the same field is initialized multiple times, it will have value from
1800 the last initialization. If any such overridden initialization has
1801 side-effect, it is unspecified whether the side-effect happens or not.
1802 Currently, GCC will discard them and issue a warning.
1805 @section Case Ranges
1807 @cindex ranges in case statements
1809 You can specify a range of consecutive values in a single @code{case} label,
1813 case @var{low} ... @var{high}:
1817 This has the same effect as the proper number of individual @code{case}
1818 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1820 This feature is especially useful for ranges of ASCII character codes:
1826 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1827 it may be parsed wrong when you use it with integer values. For example,
1842 @section Cast to a Union Type
1843 @cindex cast to a union
1844 @cindex union, casting to a
1846 A cast to union type is similar to other casts, except that the type
1847 specified is a union type. You can specify the type either with
1848 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1849 a constructor though, not a cast, and hence does not yield an lvalue like
1850 normal casts. (@xref{Compound Literals}.)
1852 The types that may be cast to the union type are those of the members
1853 of the union. Thus, given the following union and variables:
1856 union foo @{ int i; double d; @};
1862 both @code{x} and @code{y} can be cast to type @code{union foo}.
1864 Using the cast as the right-hand side of an assignment to a variable of
1865 union type is equivalent to storing in a member of the union:
1870 u = (union foo) x @equiv{} u.i = x
1871 u = (union foo) y @equiv{} u.d = y
1874 You can also use the union cast as a function argument:
1877 void hack (union foo);
1879 hack ((union foo) x);
1882 @node Mixed Declarations
1883 @section Mixed Declarations and Code
1884 @cindex mixed declarations and code
1885 @cindex declarations, mixed with code
1886 @cindex code, mixed with declarations
1888 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1889 within compound statements. As an extension, GCC also allows this in
1890 C90 mode. For example, you could do:
1899 Each identifier is visible from where it is declared until the end of
1900 the enclosing block.
1902 @node Function Attributes
1903 @section Declaring Attributes of Functions
1904 @cindex function attributes
1905 @cindex declaring attributes of functions
1906 @cindex functions that never return
1907 @cindex functions that return more than once
1908 @cindex functions that have no side effects
1909 @cindex functions in arbitrary sections
1910 @cindex functions that behave like malloc
1911 @cindex @code{volatile} applied to function
1912 @cindex @code{const} applied to function
1913 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1914 @cindex functions with non-null pointer arguments
1915 @cindex functions that are passed arguments in registers on the 386
1916 @cindex functions that pop the argument stack on the 386
1917 @cindex functions that do not pop the argument stack on the 386
1918 @cindex functions that have different compilation options on the 386
1919 @cindex functions that have different optimization options
1920 @cindex functions that are dynamically resolved
1922 In GNU C, you declare certain things about functions called in your program
1923 which help the compiler optimize function calls and check your code more
1926 The keyword @code{__attribute__} allows you to specify special
1927 attributes when making a declaration. This keyword is followed by an
1928 attribute specification inside double parentheses. The following
1929 attributes are currently defined for functions on all targets:
1930 @code{aligned}, @code{alloc_size}, @code{noreturn},
1931 @code{returns_twice}, @code{noinline}, @code{noclone},
1932 @code{always_inline}, @code{flatten}, @code{pure}, @code{const},
1933 @code{nothrow}, @code{sentinel}, @code{format}, @code{format_arg},
1934 @code{no_instrument_function}, @code{no_split_stack},
1935 @code{section}, @code{constructor},
1936 @code{destructor}, @code{used}, @code{unused}, @code{deprecated},
1937 @code{weak}, @code{malloc}, @code{alias}, @code{ifunc},
1938 @code{warn_unused_result}, @code{nonnull}, @code{gnu_inline},
1939 @code{externally_visible}, @code{hot}, @code{cold}, @code{artificial},
1940 @code{error} and @code{warning}. Several other attributes are defined
1941 for functions on particular target systems. Other attributes,
1942 including @code{section} are supported for variables declarations
1943 (@pxref{Variable Attributes}) and for types (@pxref{Type Attributes}).
1945 GCC plugins may provide their own attributes.
1947 You may also specify attributes with @samp{__} preceding and following
1948 each keyword. This allows you to use them in header files without
1949 being concerned about a possible macro of the same name. For example,
1950 you may use @code{__noreturn__} instead of @code{noreturn}.
1952 @xref{Attribute Syntax}, for details of the exact syntax for using
1956 @c Keep this table alphabetized by attribute name. Treat _ as space.
1958 @item alias ("@var{target}")
1959 @cindex @code{alias} attribute
1960 The @code{alias} attribute causes the declaration to be emitted as an
1961 alias for another symbol, which must be specified. For instance,
1964 void __f () @{ /* @r{Do something.} */; @}
1965 void f () __attribute__ ((weak, alias ("__f")));
1968 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
1969 mangled name for the target must be used. It is an error if @samp{__f}
1970 is not defined in the same translation unit.
1972 Not all target machines support this attribute.
1974 @item aligned (@var{alignment})
1975 @cindex @code{aligned} attribute
1976 This attribute specifies a minimum alignment for the function,
1979 You cannot use this attribute to decrease the alignment of a function,
1980 only to increase it. However, when you explicitly specify a function
1981 alignment this will override the effect of the
1982 @option{-falign-functions} (@pxref{Optimize Options}) option for this
1985 Note that the effectiveness of @code{aligned} attributes may be
1986 limited by inherent limitations in your linker. On many systems, the
1987 linker is only able to arrange for functions to be aligned up to a
1988 certain maximum alignment. (For some linkers, the maximum supported
1989 alignment may be very very small.) See your linker documentation for
1990 further information.
1992 The @code{aligned} attribute can also be used for variables and fields
1993 (@pxref{Variable Attributes}.)
1996 @cindex @code{alloc_size} attribute
1997 The @code{alloc_size} attribute is used to tell the compiler that the
1998 function return value points to memory, where the size is given by
1999 one or two of the functions parameters. GCC uses this
2000 information to improve the correctness of @code{__builtin_object_size}.
2002 The function parameter(s) denoting the allocated size are specified by
2003 one or two integer arguments supplied to the attribute. The allocated size
2004 is either the value of the single function argument specified or the product
2005 of the two function arguments specified. Argument numbering starts at
2011 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
2012 void my_realloc(void*, size_t) __attribute__((alloc_size(2)))
2015 declares that my_calloc will return memory of the size given by
2016 the product of parameter 1 and 2 and that my_realloc will return memory
2017 of the size given by parameter 2.
2020 @cindex @code{always_inline} function attribute
2021 Generally, functions are not inlined unless optimization is specified.
2022 For functions declared inline, this attribute inlines the function even
2023 if no optimization level was specified.
2026 @cindex @code{gnu_inline} function attribute
2027 This attribute should be used with a function which is also declared
2028 with the @code{inline} keyword. It directs GCC to treat the function
2029 as if it were defined in gnu90 mode even when compiling in C99 or
2032 If the function is declared @code{extern}, then this definition of the
2033 function is used only for inlining. In no case is the function
2034 compiled as a standalone function, not even if you take its address
2035 explicitly. Such an address becomes an external reference, as if you
2036 had only declared the function, and had not defined it. This has
2037 almost the effect of a macro. The way to use this is to put a
2038 function definition in a header file with this attribute, and put
2039 another copy of the function, without @code{extern}, in a library
2040 file. The definition in the header file will cause most calls to the
2041 function to be inlined. If any uses of the function remain, they will
2042 refer to the single copy in the library. Note that the two
2043 definitions of the functions need not be precisely the same, although
2044 if they do not have the same effect your program may behave oddly.
2046 In C, if the function is neither @code{extern} nor @code{static}, then
2047 the function is compiled as a standalone function, as well as being
2048 inlined where possible.
2050 This is how GCC traditionally handled functions declared
2051 @code{inline}. Since ISO C99 specifies a different semantics for
2052 @code{inline}, this function attribute is provided as a transition
2053 measure and as a useful feature in its own right. This attribute is
2054 available in GCC 4.1.3 and later. It is available if either of the
2055 preprocessor macros @code{__GNUC_GNU_INLINE__} or
2056 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
2057 Function is As Fast As a Macro}.
2059 In C++, this attribute does not depend on @code{extern} in any way,
2060 but it still requires the @code{inline} keyword to enable its special
2064 @cindex @code{artificial} function attribute
2065 This attribute is useful for small inline wrappers which if possible
2066 should appear during debugging as a unit, depending on the debug
2067 info format it will either mean marking the function as artificial
2068 or using the caller location for all instructions within the inlined
2072 @cindex interrupt handler functions
2073 When added to an interrupt handler with the M32C port, causes the
2074 prologue and epilogue to use bank switching to preserve the registers
2075 rather than saving them on the stack.
2078 @cindex @code{flatten} function attribute
2079 Generally, inlining into a function is limited. For a function marked with
2080 this attribute, every call inside this function will be inlined, if possible.
2081 Whether the function itself is considered for inlining depends on its size and
2082 the current inlining parameters.
2084 @item error ("@var{message}")
2085 @cindex @code{error} function attribute
2086 If this attribute is used on a function declaration and a call to such a function
2087 is not eliminated through dead code elimination or other optimizations, an error
2088 which will include @var{message} will be diagnosed. This is useful
2089 for compile time checking, especially together with @code{__builtin_constant_p}
2090 and inline functions where checking the inline function arguments is not
2091 possible through @code{extern char [(condition) ? 1 : -1];} tricks.
2092 While it is possible to leave the function undefined and thus invoke
2093 a link failure, when using this attribute the problem will be diagnosed
2094 earlier and with exact location of the call even in presence of inline
2095 functions or when not emitting debugging information.
2097 @item warning ("@var{message}")
2098 @cindex @code{warning} function attribute
2099 If this attribute is used on a function declaration and a call to such a function
2100 is not eliminated through dead code elimination or other optimizations, a warning
2101 which will include @var{message} will be diagnosed. This is useful
2102 for compile time checking, especially together with @code{__builtin_constant_p}
2103 and inline functions. While it is possible to define the function with
2104 a message in @code{.gnu.warning*} section, when using this attribute the problem
2105 will be diagnosed earlier and with exact location of the call even in presence
2106 of inline functions or when not emitting debugging information.
2109 @cindex functions that do pop the argument stack on the 386
2111 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2112 assume that the calling function will pop off the stack space used to
2113 pass arguments. This is
2114 useful to override the effects of the @option{-mrtd} switch.
2117 @cindex @code{const} function attribute
2118 Many functions do not examine any values except their arguments, and
2119 have no effects except the return value. Basically this is just slightly
2120 more strict class than the @code{pure} attribute below, since function is not
2121 allowed to read global memory.
2123 @cindex pointer arguments
2124 Note that a function that has pointer arguments and examines the data
2125 pointed to must @emph{not} be declared @code{const}. Likewise, a
2126 function that calls a non-@code{const} function usually must not be
2127 @code{const}. It does not make sense for a @code{const} function to
2130 The attribute @code{const} is not implemented in GCC versions earlier
2131 than 2.5. An alternative way to declare that a function has no side
2132 effects, which works in the current version and in some older versions,
2136 typedef int intfn ();
2138 extern const intfn square;
2141 This approach does not work in GNU C++ from 2.6.0 on, since the language
2142 specifies that the @samp{const} must be attached to the return value.
2146 @itemx constructor (@var{priority})
2147 @itemx destructor (@var{priority})
2148 @cindex @code{constructor} function attribute
2149 @cindex @code{destructor} function attribute
2150 The @code{constructor} attribute causes the function to be called
2151 automatically before execution enters @code{main ()}. Similarly, the
2152 @code{destructor} attribute causes the function to be called
2153 automatically after @code{main ()} has completed or @code{exit ()} has
2154 been called. Functions with these attributes are useful for
2155 initializing data that will be used implicitly during the execution of
2158 You may provide an optional integer priority to control the order in
2159 which constructor and destructor functions are run. A constructor
2160 with a smaller priority number runs before a constructor with a larger
2161 priority number; the opposite relationship holds for destructors. So,
2162 if you have a constructor that allocates a resource and a destructor
2163 that deallocates the same resource, both functions typically have the
2164 same priority. The priorities for constructor and destructor
2165 functions are the same as those specified for namespace-scope C++
2166 objects (@pxref{C++ Attributes}).
2168 These attributes are not currently implemented for Objective-C@.
2171 @itemx deprecated (@var{msg})
2172 @cindex @code{deprecated} attribute.
2173 The @code{deprecated} attribute results in a warning if the function
2174 is used anywhere in the source file. This is useful when identifying
2175 functions that are expected to be removed in a future version of a
2176 program. The warning also includes the location of the declaration
2177 of the deprecated function, to enable users to easily find further
2178 information about why the function is deprecated, or what they should
2179 do instead. Note that the warnings only occurs for uses:
2182 int old_fn () __attribute__ ((deprecated));
2184 int (*fn_ptr)() = old_fn;
2187 results in a warning on line 3 but not line 2. The optional msg
2188 argument, which must be a string, will be printed in the warning if
2191 The @code{deprecated} attribute can also be used for variables and
2192 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2195 @cindex @code{disinterrupt} attribute
2196 On Epiphany and MeP targets, this attribute causes the compiler to emit
2197 instructions to disable interrupts for the duration of the given
2201 @cindex @code{__declspec(dllexport)}
2202 On Microsoft Windows targets and Symbian OS targets the
2203 @code{dllexport} attribute causes the compiler to provide a global
2204 pointer to a pointer in a DLL, so that it can be referenced with the
2205 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
2206 name is formed by combining @code{_imp__} and the function or variable
2209 You can use @code{__declspec(dllexport)} as a synonym for
2210 @code{__attribute__ ((dllexport))} for compatibility with other
2213 On systems that support the @code{visibility} attribute, this
2214 attribute also implies ``default'' visibility. It is an error to
2215 explicitly specify any other visibility.
2217 In previous versions of GCC, the @code{dllexport} attribute was ignored
2218 for inlined functions, unless the @option{-fkeep-inline-functions} flag
2219 had been used. The default behaviour now is to emit all dllexported
2220 inline functions; however, this can cause object file-size bloat, in
2221 which case the old behaviour can be restored by using
2222 @option{-fno-keep-inline-dllexport}.
2224 The attribute is also ignored for undefined symbols.
2226 When applied to C++ classes, the attribute marks defined non-inlined
2227 member functions and static data members as exports. Static consts
2228 initialized in-class are not marked unless they are also defined
2231 For Microsoft Windows targets there are alternative methods for
2232 including the symbol in the DLL's export table such as using a
2233 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
2234 the @option{--export-all} linker flag.
2237 @cindex @code{__declspec(dllimport)}
2238 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
2239 attribute causes the compiler to reference a function or variable via
2240 a global pointer to a pointer that is set up by the DLL exporting the
2241 symbol. The attribute implies @code{extern}. On Microsoft Windows
2242 targets, the pointer name is formed by combining @code{_imp__} and the
2243 function or variable name.
2245 You can use @code{__declspec(dllimport)} as a synonym for
2246 @code{__attribute__ ((dllimport))} for compatibility with other
2249 On systems that support the @code{visibility} attribute, this
2250 attribute also implies ``default'' visibility. It is an error to
2251 explicitly specify any other visibility.
2253 Currently, the attribute is ignored for inlined functions. If the
2254 attribute is applied to a symbol @emph{definition}, an error is reported.
2255 If a symbol previously declared @code{dllimport} is later defined, the
2256 attribute is ignored in subsequent references, and a warning is emitted.
2257 The attribute is also overridden by a subsequent declaration as
2260 When applied to C++ classes, the attribute marks non-inlined
2261 member functions and static data members as imports. However, the
2262 attribute is ignored for virtual methods to allow creation of vtables
2265 On the SH Symbian OS target the @code{dllimport} attribute also has
2266 another affect---it can cause the vtable and run-time type information
2267 for a class to be exported. This happens when the class has a
2268 dllimport'ed constructor or a non-inline, non-pure virtual function
2269 and, for either of those two conditions, the class also has an inline
2270 constructor or destructor and has a key function that is defined in
2271 the current translation unit.
2273 For Microsoft Windows based targets the use of the @code{dllimport}
2274 attribute on functions is not necessary, but provides a small
2275 performance benefit by eliminating a thunk in the DLL@. The use of the
2276 @code{dllimport} attribute on imported variables was required on older
2277 versions of the GNU linker, but can now be avoided by passing the
2278 @option{--enable-auto-import} switch to the GNU linker. As with
2279 functions, using the attribute for a variable eliminates a thunk in
2282 One drawback to using this attribute is that a pointer to a
2283 @emph{variable} marked as @code{dllimport} cannot be used as a constant
2284 address. However, a pointer to a @emph{function} with the
2285 @code{dllimport} attribute can be used as a constant initializer; in
2286 this case, the address of a stub function in the import lib is
2287 referenced. On Microsoft Windows targets, the attribute can be disabled
2288 for functions by setting the @option{-mnop-fun-dllimport} flag.
2291 @cindex eight bit data on the H8/300, H8/300H, and H8S
2292 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2293 variable should be placed into the eight bit data section.
2294 The compiler will generate more efficient code for certain operations
2295 on data in the eight bit data area. Note the eight bit data area is limited to
2298 You must use GAS and GLD from GNU binutils version 2.7 or later for
2299 this attribute to work correctly.
2301 @item exception_handler
2302 @cindex exception handler functions on the Blackfin processor
2303 Use this attribute on the Blackfin to indicate that the specified function
2304 is an exception handler. The compiler will generate function entry and
2305 exit sequences suitable for use in an exception handler when this
2306 attribute is present.
2308 @item externally_visible
2309 @cindex @code{externally_visible} attribute.
2310 This attribute, attached to a global variable or function, nullifies
2311 the effect of the @option{-fwhole-program} command-line option, so the
2312 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.
2315 @cindex functions which handle memory bank switching
2316 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
2317 use a calling convention that takes care of switching memory banks when
2318 entering and leaving a function. This calling convention is also the
2319 default when using the @option{-mlong-calls} option.
2321 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
2322 to call and return from a function.
2324 On 68HC11 the compiler will generate a sequence of instructions
2325 to invoke a board-specific routine to switch the memory bank and call the
2326 real function. The board-specific routine simulates a @code{call}.
2327 At the end of a function, it will jump to a board-specific routine
2328 instead of using @code{rts}. The board-specific return routine simulates
2331 On MeP targets this causes the compiler to use a calling convention
2332 which assumes the called function is too far away for the built-in
2335 @item fast_interrupt
2336 @cindex interrupt handler functions
2337 Use this attribute on the M32C and RX ports to indicate that the specified
2338 function is a fast interrupt handler. This is just like the
2339 @code{interrupt} attribute, except that @code{freit} is used to return
2340 instead of @code{reit}.
2343 @cindex functions that pop the argument stack on the 386
2344 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2345 pass the first argument (if of integral type) in the register ECX and
2346 the second argument (if of integral type) in the register EDX@. Subsequent
2347 and other typed arguments are passed on the stack. The called function will
2348 pop the arguments off the stack. If the number of arguments is variable all
2349 arguments are pushed on the stack.
2352 @cindex functions that pop the argument stack on the 386
2353 On the Intel 386, the @code{thiscall} attribute causes the compiler to
2354 pass the first argument (if of integral type) in the register ECX.
2355 Subsequent and other typed arguments are passed on the stack. The called
2356 function will pop the arguments off the stack.
2357 If the number of arguments is variable all arguments are pushed on the
2359 The @code{thiscall} attribute is intended for C++ non-static member functions.
2360 As gcc extension this calling convention can be used for C-functions
2361 and for static member methods.
2363 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2364 @cindex @code{format} function attribute
2366 The @code{format} attribute specifies that a function takes @code{printf},
2367 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2368 should be type-checked against a format string. For example, the
2373 my_printf (void *my_object, const char *my_format, ...)
2374 __attribute__ ((format (printf, 2, 3)));
2378 causes the compiler to check the arguments in calls to @code{my_printf}
2379 for consistency with the @code{printf} style format string argument
2382 The parameter @var{archetype} determines how the format string is
2383 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime},
2384 @code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or
2385 @code{strfmon}. (You can also use @code{__printf__},
2386 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On
2387 MinGW targets, @code{ms_printf}, @code{ms_scanf}, and
2388 @code{ms_strftime} are also present.
2389 @var{archtype} values such as @code{printf} refer to the formats accepted
2390 by the system's C run-time library, while @code{gnu_} values always refer
2391 to the formats accepted by the GNU C Library. On Microsoft Windows
2392 targets, @code{ms_} values refer to the formats accepted by the
2393 @file{msvcrt.dll} library.
2394 The parameter @var{string-index}
2395 specifies which argument is the format string argument (starting
2396 from 1), while @var{first-to-check} is the number of the first
2397 argument to check against the format string. For functions
2398 where the arguments are not available to be checked (such as
2399 @code{vprintf}), specify the third parameter as zero. In this case the
2400 compiler only checks the format string for consistency. For
2401 @code{strftime} formats, the third parameter is required to be zero.
2402 Since non-static C++ methods have an implicit @code{this} argument, the
2403 arguments of such methods should be counted from two, not one, when
2404 giving values for @var{string-index} and @var{first-to-check}.
2406 In the example above, the format string (@code{my_format}) is the second
2407 argument of the function @code{my_print}, and the arguments to check
2408 start with the third argument, so the correct parameters for the format
2409 attribute are 2 and 3.
2411 @opindex ffreestanding
2412 @opindex fno-builtin
2413 The @code{format} attribute allows you to identify your own functions
2414 which take format strings as arguments, so that GCC can check the
2415 calls to these functions for errors. The compiler always (unless
2416 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2417 for the standard library functions @code{printf}, @code{fprintf},
2418 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2419 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2420 warnings are requested (using @option{-Wformat}), so there is no need to
2421 modify the header file @file{stdio.h}. In C99 mode, the functions
2422 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2423 @code{vsscanf} are also checked. Except in strictly conforming C
2424 standard modes, the X/Open function @code{strfmon} is also checked as
2425 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2426 @xref{C Dialect Options,,Options Controlling C Dialect}.
2428 For Objective-C dialects, @code{NSString} (or @code{__NSString__}) is
2429 recognized in the same context. Declarations including these format attributes
2430 will be parsed for correct syntax, however the result of checking of such format
2431 strings is not yet defined, and will not be carried out by this version of the
2434 The target may also provide additional types of format checks.
2435 @xref{Target Format Checks,,Format Checks Specific to Particular
2438 @item format_arg (@var{string-index})
2439 @cindex @code{format_arg} function attribute
2440 @opindex Wformat-nonliteral
2441 The @code{format_arg} attribute specifies that a function takes a format
2442 string for a @code{printf}, @code{scanf}, @code{strftime} or
2443 @code{strfmon} style function and modifies it (for example, to translate
2444 it into another language), so the result can be passed to a
2445 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2446 function (with the remaining arguments to the format function the same
2447 as they would have been for the unmodified string). For example, the
2452 my_dgettext (char *my_domain, const char *my_format)
2453 __attribute__ ((format_arg (2)));
2457 causes the compiler to check the arguments in calls to a @code{printf},
2458 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2459 format string argument is a call to the @code{my_dgettext} function, for
2460 consistency with the format string argument @code{my_format}. If the
2461 @code{format_arg} attribute had not been specified, all the compiler
2462 could tell in such calls to format functions would be that the format
2463 string argument is not constant; this would generate a warning when
2464 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2465 without the attribute.
2467 The parameter @var{string-index} specifies which argument is the format
2468 string argument (starting from one). Since non-static C++ methods have
2469 an implicit @code{this} argument, the arguments of such methods should
2470 be counted from two.
2472 The @code{format-arg} attribute allows you to identify your own
2473 functions which modify format strings, so that GCC can check the
2474 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2475 type function whose operands are a call to one of your own function.
2476 The compiler always treats @code{gettext}, @code{dgettext}, and
2477 @code{dcgettext} in this manner except when strict ISO C support is
2478 requested by @option{-ansi} or an appropriate @option{-std} option, or
2479 @option{-ffreestanding} or @option{-fno-builtin}
2480 is used. @xref{C Dialect Options,,Options
2481 Controlling C Dialect}.
2483 For Objective-C dialects, the @code{format-arg} attribute may refer to an
2484 @code{NSString} reference for compatibility with the @code{format} attribute
2487 The target may also allow additional types in @code{format-arg} attributes.
2488 @xref{Target Format Checks,,Format Checks Specific to Particular
2491 @item function_vector
2492 @cindex calling functions through the function vector on H8/300, M16C, M32C and SH2A processors
2493 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2494 function should be called through the function vector. Calling a
2495 function through the function vector will reduce code size, however;
2496 the function vector has a limited size (maximum 128 entries on the H8/300
2497 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
2499 In SH2A target, this attribute declares a function to be called using the
2500 TBR relative addressing mode. The argument to this attribute is the entry
2501 number of the same function in a vector table containing all the TBR
2502 relative addressable functions. For the successful jump, register TBR
2503 should contain the start address of this TBR relative vector table.
2504 In the startup routine of the user application, user needs to care of this
2505 TBR register initialization. The TBR relative vector table can have at
2506 max 256 function entries. The jumps to these functions will be generated
2507 using a SH2A specific, non delayed branch instruction JSR/N @@(disp8,TBR).
2508 You must use GAS and GLD from GNU binutils version 2.7 or later for
2509 this attribute to work correctly.
2511 Please refer the example of M16C target, to see the use of this
2512 attribute while declaring a function,
2514 In an application, for a function being called once, this attribute will
2515 save at least 8 bytes of code; and if other successive calls are being
2516 made to the same function, it will save 2 bytes of code per each of these
2519 On M16C/M32C targets, the @code{function_vector} attribute declares a
2520 special page subroutine call function. Use of this attribute reduces
2521 the code size by 2 bytes for each call generated to the
2522 subroutine. The argument to the attribute is the vector number entry
2523 from the special page vector table which contains the 16 low-order
2524 bits of the subroutine's entry address. Each vector table has special
2525 page number (18 to 255) which are used in @code{jsrs} instruction.
2526 Jump addresses of the routines are generated by adding 0x0F0000 (in
2527 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the 2
2528 byte addresses set in the vector table. Therefore you need to ensure
2529 that all the special page vector routines should get mapped within the
2530 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
2533 In the following example 2 bytes will be saved for each call to
2534 function @code{foo}.
2537 void foo (void) __attribute__((function_vector(0x18)));
2548 If functions are defined in one file and are called in another file,
2549 then be sure to write this declaration in both files.
2551 This attribute is ignored for R8C target.
2554 @cindex interrupt handler functions
2555 Use this attribute on the ARM, AVR, Epiphany, M32C, M32R/D, m68k, MeP, MIPS,
2556 RX and Xstormy16 ports to indicate that the specified function is an
2557 interrupt handler. The compiler will generate function entry and exit
2558 sequences suitable for use in an interrupt handler when this attribute
2561 Note, interrupt handlers for the Blackfin, H8/300, H8/300H, H8S, MicroBlaze,
2562 and SH processors can be specified via the @code{interrupt_handler} attribute.
2564 Note, on the AVR, interrupts will be enabled inside the function.
2566 Note, for the ARM, you can specify the kind of interrupt to be handled by
2567 adding an optional parameter to the interrupt attribute like this:
2570 void f () __attribute__ ((interrupt ("IRQ")));
2573 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2575 On ARMv7-M the interrupt type is ignored, and the attribute means the function
2576 may be called with a word aligned stack pointer.
2578 On MIPS targets, you can use the following attributes to modify the behavior
2579 of an interrupt handler:
2581 @item use_shadow_register_set
2582 @cindex @code{use_shadow_register_set} attribute
2583 Assume that the handler uses a shadow register set, instead of
2584 the main general-purpose registers.
2586 @item keep_interrupts_masked
2587 @cindex @code{keep_interrupts_masked} attribute
2588 Keep interrupts masked for the whole function. Without this attribute,
2589 GCC tries to reenable interrupts for as much of the function as it can.
2591 @item use_debug_exception_return
2592 @cindex @code{use_debug_exception_return} attribute
2593 Return using the @code{deret} instruction. Interrupt handlers that don't
2594 have this attribute return using @code{eret} instead.
2597 You can use any combination of these attributes, as shown below:
2599 void __attribute__ ((interrupt)) v0 ();
2600 void __attribute__ ((interrupt, use_shadow_register_set)) v1 ();
2601 void __attribute__ ((interrupt, keep_interrupts_masked)) v2 ();
2602 void __attribute__ ((interrupt, use_debug_exception_return)) v3 ();
2603 void __attribute__ ((interrupt, use_shadow_register_set,
2604 keep_interrupts_masked)) v4 ();
2605 void __attribute__ ((interrupt, use_shadow_register_set,
2606 use_debug_exception_return)) v5 ();
2607 void __attribute__ ((interrupt, keep_interrupts_masked,
2608 use_debug_exception_return)) v6 ();
2609 void __attribute__ ((interrupt, use_shadow_register_set,
2610 keep_interrupts_masked,
2611 use_debug_exception_return)) v7 ();
2614 @item ifunc ("@var{resolver}")
2615 @cindex @code{ifunc} attribute
2616 The @code{ifunc} attribute is used to mark a function as an indirect
2617 function using the STT_GNU_IFUNC symbol type extension to the ELF
2618 standard. This allows the resolution of the symbol value to be
2619 determined dynamically at load time, and an optimized version of the
2620 routine can be selected for the particular processor or other system
2621 characteristics determined then. To use this attribute, first define
2622 the implementation functions available, and a resolver function that
2623 returns a pointer to the selected implementation function. The
2624 implementation functions' declarations must match the API of the
2625 function being implemented, the resolver's declaration is be a
2626 function returning pointer to void function returning void:
2629 void *my_memcpy (void *dst, const void *src, size_t len)
2634 static void (*resolve_memcpy (void)) (void)
2636 return my_memcpy; // we'll just always select this routine
2640 The exported header file declaring the function the user calls would
2644 extern void *memcpy (void *, const void *, size_t);
2647 allowing the user to call this as a regular function, unaware of the
2648 implementation. Finally, the indirect function needs to be defined in
2649 the same translation unit as the resolver function:
2652 void *memcpy (void *, const void *, size_t)
2653 __attribute__ ((ifunc ("resolve_memcpy")));
2656 Indirect functions cannot be weak, and require a recent binutils (at
2657 least version 2.20.1), and GNU C library (at least version 2.11.1).
2659 @item interrupt_handler
2660 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
2661 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
2662 indicate that the specified function is an interrupt handler. The compiler
2663 will generate function entry and exit sequences suitable for use in an
2664 interrupt handler when this attribute is present.
2666 @item interrupt_thread
2667 @cindex interrupt thread functions on fido
2668 Use this attribute on fido, a subarchitecture of the m68k, to indicate
2669 that the specified function is an interrupt handler that is designed
2670 to run as a thread. The compiler omits generate prologue/epilogue
2671 sequences and replaces the return instruction with a @code{sleep}
2672 instruction. This attribute is available only on fido.
2675 @cindex interrupt service routines on ARM
2676 Use this attribute on ARM to write Interrupt Service Routines. This is an
2677 alias to the @code{interrupt} attribute above.
2680 @cindex User stack pointer in interrupts on the Blackfin
2681 When used together with @code{interrupt_handler}, @code{exception_handler}
2682 or @code{nmi_handler}, code will be generated to load the stack pointer
2683 from the USP register in the function prologue.
2686 @cindex @code{l1_text} function attribute
2687 This attribute specifies a function to be placed into L1 Instruction
2688 SRAM@. The function will be put into a specific section named @code{.l1.text}.
2689 With @option{-mfdpic}, function calls with a such function as the callee
2690 or caller will use inlined PLT.
2693 @cindex @code{l2} function attribute
2694 On the Blackfin, this attribute specifies a function to be placed into L2
2695 SRAM. The function will be put into a specific section named
2696 @code{.l1.text}. With @option{-mfdpic}, callers of such functions will use
2700 @cindex @code{leaf} function attribute
2701 Calls to external functions with this attribute must return to the current
2702 compilation unit only by return or by exception handling. In particular, leaf
2703 functions are not allowed to call callback function passed to it from the current
2704 compilation unit or directly call functions exported by the unit or longjmp
2705 into the unit. Leaf function might still call functions from other compilation
2706 units and thus they are not necessarily leaf in the sense that they contain no
2707 function calls at all.
2709 The attribute is intended for library functions to improve dataflow analysis.
2710 The compiler takes the hint that any data not escaping the current compilation unit can
2711 not be used or modified by the leaf function. For example, the @code{sin} function
2712 is a leaf function, but @code{qsort} is not.
2714 Note that leaf functions might invoke signals and signal handlers might be
2715 defined in the current compilation unit and use static variables. The only
2716 compliant way to write such a signal handler is to declare such variables
2719 The attribute has no effect on functions defined within the current compilation
2720 unit. This is to allow easy merging of multiple compilation units into one,
2721 for example, by using the link time optimization. For this reason the
2722 attribute is not allowed on types to annotate indirect calls.
2724 @item long_call/short_call
2725 @cindex indirect calls on ARM
2726 This attribute specifies how a particular function is called on
2727 ARM and Epiphany. Both attributes override the
2728 @option{-mlong-calls} (@pxref{ARM Options})
2729 command-line switch and @code{#pragma long_calls} settings. The
2730 @code{long_call} attribute indicates that the function might be far
2731 away from the call site and require a different (more expensive)
2732 calling sequence. The @code{short_call} attribute always places
2733 the offset to the function from the call site into the @samp{BL}
2734 instruction directly.
2736 @item longcall/shortcall
2737 @cindex functions called via pointer on the RS/6000 and PowerPC
2738 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute
2739 indicates that the function might be far away from the call site and
2740 require a different (more expensive) calling sequence. The
2741 @code{shortcall} attribute indicates that the function is always close
2742 enough for the shorter calling sequence to be used. These attributes
2743 override both the @option{-mlongcall} switch and, on the RS/6000 and
2744 PowerPC, the @code{#pragma longcall} setting.
2746 @xref{RS/6000 and PowerPC Options}, for more information on whether long
2747 calls are necessary.
2749 @item long_call/near/far
2750 @cindex indirect calls on MIPS
2751 These attributes specify how a particular function is called on MIPS@.
2752 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
2753 command-line switch. The @code{long_call} and @code{far} attributes are
2754 synonyms, and cause the compiler to always call
2755 the function by first loading its address into a register, and then using
2756 the contents of that register. The @code{near} attribute has the opposite
2757 effect; it specifies that non-PIC calls should be made using the more
2758 efficient @code{jal} instruction.
2761 @cindex @code{malloc} attribute
2762 The @code{malloc} attribute is used to tell the compiler that a function
2763 may be treated as if any non-@code{NULL} pointer it returns cannot
2764 alias any other pointer valid when the function returns.
2765 This will often improve optimization.
2766 Standard functions with this property include @code{malloc} and
2767 @code{calloc}. @code{realloc}-like functions have this property as
2768 long as the old pointer is never referred to (including comparing it
2769 to the new pointer) after the function returns a non-@code{NULL}
2772 @item mips16/nomips16
2773 @cindex @code{mips16} attribute
2774 @cindex @code{nomips16} attribute
2776 On MIPS targets, you can use the @code{mips16} and @code{nomips16}
2777 function attributes to locally select or turn off MIPS16 code generation.
2778 A function with the @code{mips16} attribute is emitted as MIPS16 code,
2779 while MIPS16 code generation is disabled for functions with the
2780 @code{nomips16} attribute. These attributes override the
2781 @option{-mips16} and @option{-mno-mips16} options on the command line
2782 (@pxref{MIPS Options}).
2784 When compiling files containing mixed MIPS16 and non-MIPS16 code, the
2785 preprocessor symbol @code{__mips16} reflects the setting on the command line,
2786 not that within individual functions. Mixed MIPS16 and non-MIPS16 code
2787 may interact badly with some GCC extensions such as @code{__builtin_apply}
2788 (@pxref{Constructing Calls}).
2790 @item model (@var{model-name})
2791 @cindex function addressability on the M32R/D
2792 @cindex variable addressability on the IA-64
2794 On the M32R/D, use this attribute to set the addressability of an
2795 object, and of the code generated for a function. The identifier
2796 @var{model-name} is one of @code{small}, @code{medium}, or
2797 @code{large}, representing each of the code models.
2799 Small model objects live in the lower 16MB of memory (so that their
2800 addresses can be loaded with the @code{ld24} instruction), and are
2801 callable with the @code{bl} instruction.
2803 Medium model objects may live anywhere in the 32-bit address space (the
2804 compiler will generate @code{seth/add3} instructions to load their addresses),
2805 and are callable with the @code{bl} instruction.
2807 Large model objects may live anywhere in the 32-bit address space (the
2808 compiler will generate @code{seth/add3} instructions to load their addresses),
2809 and may not be reachable with the @code{bl} instruction (the compiler will
2810 generate the much slower @code{seth/add3/jl} instruction sequence).
2812 On IA-64, use this attribute to set the addressability of an object.
2813 At present, the only supported identifier for @var{model-name} is
2814 @code{small}, indicating addressability via ``small'' (22-bit)
2815 addresses (so that their addresses can be loaded with the @code{addl}
2816 instruction). Caveat: such addressing is by definition not position
2817 independent and hence this attribute must not be used for objects
2818 defined by shared libraries.
2820 @item ms_abi/sysv_abi
2821 @cindex @code{ms_abi} attribute
2822 @cindex @code{sysv_abi} attribute
2824 On 32-bit and 64-bit (i?86|x86_64)-*-* targets, you can use an ABI attribute
2825 to indicate which calling convention should be used for a function. The
2826 @code{ms_abi} attribute tells the compiler to use the Microsoft ABI,
2827 while the @code{sysv_abi} attribute tells the compiler to use the ABI
2828 used on GNU/Linux and other systems. The default is to use the Microsoft ABI
2829 when targeting Windows. On all other systems, the default is the x86/AMD ABI.
2831 Note, the @code{ms_abi} attribute for Windows 64-bit targets currently
2832 requires the @option{-maccumulate-outgoing-args} option.
2834 @item callee_pop_aggregate_return (@var{number})
2835 @cindex @code{callee_pop_aggregate_return} attribute
2837 On 32-bit i?86-*-* targets, you can control by those attribute for
2838 aggregate return in memory, if the caller is responsible to pop the hidden
2839 pointer together with the rest of the arguments - @var{number} equal to
2840 zero -, or if the callee is responsible to pop hidden pointer - @var{number}
2841 equal to one. The default i386 ABI assumes that the callee pops the
2842 stack for hidden pointer.
2844 Note, that on 32-bit i386 Windows targets the compiler assumes that the
2845 caller pops the stack for hidden pointer.
2847 @item ms_hook_prologue
2848 @cindex @code{ms_hook_prologue} attribute
2850 On 32 bit i[34567]86-*-* targets and 64 bit x86_64-*-* targets, you can use
2851 this function attribute to make gcc generate the "hot-patching" function
2852 prologue used in Win32 API functions in Microsoft Windows XP Service Pack 2
2856 @cindex function without a prologue/epilogue code
2857 Use this attribute on the ARM, AVR, MCORE, RX and SPU ports to indicate that
2858 the specified function does not need prologue/epilogue sequences generated by
2859 the compiler. It is up to the programmer to provide these sequences. The
2860 only statements that can be safely included in naked functions are
2861 @code{asm} statements that do not have operands. All other statements,
2862 including declarations of local variables, @code{if} statements, and so
2863 forth, should be avoided. Naked functions should be used to implement the
2864 body of an assembly function, while allowing the compiler to construct
2865 the requisite function declaration for the assembler.
2868 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2869 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2870 use the normal calling convention based on @code{jsr} and @code{rts}.
2871 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2874 On MeP targets this attribute causes the compiler to assume the called
2875 function is close enough to use the normal calling convention,
2876 overriding the @code{-mtf} command line option.
2879 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
2880 Use this attribute together with @code{interrupt_handler},
2881 @code{exception_handler} or @code{nmi_handler} to indicate that the function
2882 entry code should enable nested interrupts or exceptions.
2885 @cindex NMI handler functions on the Blackfin processor
2886 Use this attribute on the Blackfin to indicate that the specified function
2887 is an NMI handler. The compiler will generate function entry and
2888 exit sequences suitable for use in an NMI handler when this
2889 attribute is present.
2891 @item no_instrument_function
2892 @cindex @code{no_instrument_function} function attribute
2893 @opindex finstrument-functions
2894 If @option{-finstrument-functions} is given, profiling function calls will
2895 be generated at entry and exit of most user-compiled functions.
2896 Functions with this attribute will not be so instrumented.
2898 @item no_split_stack
2899 @cindex @code{no_split_stack} function attribute
2900 @opindex fsplit-stack
2901 If @option{-fsplit-stack} is given, functions will have a small
2902 prologue which decides whether to split the stack. Functions with the
2903 @code{no_split_stack} attribute will not have that prologue, and thus
2904 may run with only a small amount of stack space available.
2907 @cindex @code{noinline} function attribute
2908 This function attribute prevents a function from being considered for
2910 @c Don't enumerate the optimizations by name here; we try to be
2911 @c future-compatible with this mechanism.
2912 If the function does not have side-effects, there are optimizations
2913 other than inlining that causes function calls to be optimized away,
2914 although the function call is live. To keep such calls from being
2919 (@pxref{Extended Asm}) in the called function, to serve as a special
2923 @cindex @code{noclone} function attribute
2924 This function attribute prevents a function from being considered for
2925 cloning - a mechanism which produces specialized copies of functions
2926 and which is (currently) performed by interprocedural constant
2929 @item nonnull (@var{arg-index}, @dots{})
2930 @cindex @code{nonnull} function attribute
2931 The @code{nonnull} attribute specifies that some function parameters should
2932 be non-null pointers. For instance, the declaration:
2936 my_memcpy (void *dest, const void *src, size_t len)
2937 __attribute__((nonnull (1, 2)));
2941 causes the compiler to check that, in calls to @code{my_memcpy},
2942 arguments @var{dest} and @var{src} are non-null. If the compiler
2943 determines that a null pointer is passed in an argument slot marked
2944 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2945 is issued. The compiler may also choose to make optimizations based
2946 on the knowledge that certain function arguments will not be null.
2948 If no argument index list is given to the @code{nonnull} attribute,
2949 all pointer arguments are marked as non-null. To illustrate, the
2950 following declaration is equivalent to the previous example:
2954 my_memcpy (void *dest, const void *src, size_t len)
2955 __attribute__((nonnull));
2959 @cindex @code{noreturn} function attribute
2960 A few standard library functions, such as @code{abort} and @code{exit},
2961 cannot return. GCC knows this automatically. Some programs define
2962 their own functions that never return. You can declare them
2963 @code{noreturn} to tell the compiler this fact. For example,
2967 void fatal () __attribute__ ((noreturn));
2970 fatal (/* @r{@dots{}} */)
2972 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2978 The @code{noreturn} keyword tells the compiler to assume that
2979 @code{fatal} cannot return. It can then optimize without regard to what
2980 would happen if @code{fatal} ever did return. This makes slightly
2981 better code. More importantly, it helps avoid spurious warnings of
2982 uninitialized variables.
2984 The @code{noreturn} keyword does not affect the exceptional path when that
2985 applies: a @code{noreturn}-marked function may still return to the caller
2986 by throwing an exception or calling @code{longjmp}.
2988 Do not assume that registers saved by the calling function are
2989 restored before calling the @code{noreturn} function.
2991 It does not make sense for a @code{noreturn} function to have a return
2992 type other than @code{void}.
2994 The attribute @code{noreturn} is not implemented in GCC versions
2995 earlier than 2.5. An alternative way to declare that a function does
2996 not return, which works in the current version and in some older
2997 versions, is as follows:
3000 typedef void voidfn ();
3002 volatile voidfn fatal;
3005 This approach does not work in GNU C++.
3008 @cindex @code{nothrow} function attribute
3009 The @code{nothrow} attribute is used to inform the compiler that a
3010 function cannot throw an exception. For example, most functions in
3011 the standard C library can be guaranteed not to throw an exception
3012 with the notable exceptions of @code{qsort} and @code{bsearch} that
3013 take function pointer arguments. The @code{nothrow} attribute is not
3014 implemented in GCC versions earlier than 3.3.
3017 @cindex @code{optimize} function attribute
3018 The @code{optimize} attribute is used to specify that a function is to
3019 be compiled with different optimization options than specified on the
3020 command line. Arguments can either be numbers or strings. Numbers
3021 are assumed to be an optimization level. Strings that begin with
3022 @code{O} are assumed to be an optimization option, while other options
3023 are assumed to be used with a @code{-f} prefix. You can also use the
3024 @samp{#pragma GCC optimize} pragma to set the optimization options
3025 that affect more than one function.
3026 @xref{Function Specific Option Pragmas}, for details about the
3027 @samp{#pragma GCC optimize} pragma.
3029 This can be used for instance to have frequently executed functions
3030 compiled with more aggressive optimization options that produce faster
3031 and larger code, while other functions can be called with less
3034 @item OS_main/OS_task
3035 @cindex @code{OS_main} AVR function attribute
3036 @cindex @code{OS_task} AVR function attribute
3037 On AVR, functions with the @code{OS_main} or @code{OS_task} attribute
3038 do not save/restore any call-saved register in their prologue/epilogue.
3040 The @code{OS_main} attribute can be used when there @emph{is
3041 guarantee} that interrupts are disabled at the time when the function
3042 is entered. This will save resources when the stack pointer has to be
3043 changed to set up a frame for local variables.
3045 The @code{OS_task} attribute can be used when there is @emph{no
3046 guarantee} that interrupts are disabled at that time when the function
3047 is entered like for, e@.g@. task functions in a multi-threading operating
3048 system. In that case, changing the stack pointer register will be
3049 guarded by save/clear/restore of the global interrupt enable flag.
3051 The differences to the @code{naked} function attrubute are:
3053 @item @code{naked} functions do not have a return instruction whereas
3054 @code{OS_main} and @code{OS_task} functions will have a @code{RET} or
3055 @code{RETI} return instruction.
3056 @item @code{naked} functions do not set up a frame for local variables
3057 or a frame pointer whereas @code{OS_main} and @code{OS_task} do this
3062 @cindex @code{pcs} function attribute
3064 The @code{pcs} attribute can be used to control the calling convention
3065 used for a function on ARM. The attribute takes an argument that specifies
3066 the calling convention to use.
3068 When compiling using the AAPCS ABI (or a variant of that) then valid
3069 values for the argument are @code{"aapcs"} and @code{"aapcs-vfp"}. In
3070 order to use a variant other than @code{"aapcs"} then the compiler must
3071 be permitted to use the appropriate co-processor registers (i.e., the
3072 VFP registers must be available in order to use @code{"aapcs-vfp"}).
3076 /* Argument passed in r0, and result returned in r0+r1. */
3077 double f2d (float) __attribute__((pcs("aapcs")));
3080 Variadic functions always use the @code{"aapcs"} calling convention and
3081 the compiler will reject attempts to specify an alternative.
3084 @cindex @code{pure} function attribute
3085 Many functions have no effects except the return value and their
3086 return value depends only on the parameters and/or global variables.
3087 Such a function can be subject
3088 to common subexpression elimination and loop optimization just as an
3089 arithmetic operator would be. These functions should be declared
3090 with the attribute @code{pure}. For example,
3093 int square (int) __attribute__ ((pure));
3097 says that the hypothetical function @code{square} is safe to call
3098 fewer times than the program says.
3100 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
3101 Interesting non-pure functions are functions with infinite loops or those
3102 depending on volatile memory or other system resource, that may change between
3103 two consecutive calls (such as @code{feof} in a multithreading environment).
3105 The attribute @code{pure} is not implemented in GCC versions earlier
3109 @cindex @code{hot} function attribute
3110 The @code{hot} attribute is used to inform the compiler that a function is a
3111 hot spot of the compiled program. The function is optimized more aggressively
3112 and on many target it is placed into special subsection of the text section so
3113 all hot functions appears close together improving locality.
3115 When profile feedback is available, via @option{-fprofile-use}, hot functions
3116 are automatically detected and this attribute is ignored.
3118 The @code{hot} attribute is not implemented in GCC versions earlier
3122 @cindex @code{cold} function attribute
3123 The @code{cold} attribute is used to inform the compiler that a function is
3124 unlikely executed. The function is optimized for size rather than speed and on
3125 many targets it is placed into special subsection of the text section so all
3126 cold functions appears close together improving code locality of non-cold parts
3127 of program. The paths leading to call of cold functions within code are marked
3128 as unlikely by the branch prediction mechanism. It is thus useful to mark
3129 functions used to handle unlikely conditions, such as @code{perror}, as cold to
3130 improve optimization of hot functions that do call marked functions in rare
3133 When profile feedback is available, via @option{-fprofile-use}, hot functions
3134 are automatically detected and this attribute is ignored.
3136 The @code{cold} attribute is not implemented in GCC versions earlier than 4.3.
3138 @item regparm (@var{number})
3139 @cindex @code{regparm} attribute
3140 @cindex functions that are passed arguments in registers on the 386
3141 On the Intel 386, the @code{regparm} attribute causes the compiler to
3142 pass arguments number one to @var{number} if they are of integral type
3143 in registers EAX, EDX, and ECX instead of on the stack. Functions that
3144 take a variable number of arguments will continue to be passed all of their
3145 arguments on the stack.
3147 Beware that on some ELF systems this attribute is unsuitable for
3148 global functions in shared libraries with lazy binding (which is the
3149 default). Lazy binding will send the first call via resolving code in
3150 the loader, which might assume EAX, EDX and ECX can be clobbered, as
3151 per the standard calling conventions. Solaris 8 is affected by this.
3152 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
3153 safe since the loaders there save EAX, EDX and ECX. (Lazy binding can be
3154 disabled with the linker or the loader if desired, to avoid the
3158 @cindex @code{sseregparm} attribute
3159 On the Intel 386 with SSE support, the @code{sseregparm} attribute
3160 causes the compiler to pass up to 3 floating point arguments in
3161 SSE registers instead of on the stack. Functions that take a
3162 variable number of arguments will continue to pass all of their
3163 floating point arguments on the stack.
3165 @item force_align_arg_pointer
3166 @cindex @code{force_align_arg_pointer} attribute
3167 On the Intel x86, the @code{force_align_arg_pointer} attribute may be
3168 applied to individual function definitions, generating an alternate
3169 prologue and epilogue that realigns the runtime stack if necessary.
3170 This supports mixing legacy codes that run with a 4-byte aligned stack
3171 with modern codes that keep a 16-byte stack for SSE compatibility.
3174 @cindex @code{resbank} attribute
3175 On the SH2A target, this attribute enables the high-speed register
3176 saving and restoration using a register bank for @code{interrupt_handler}
3177 routines. Saving to the bank is performed automatically after the CPU
3178 accepts an interrupt that uses a register bank.
3180 The nineteen 32-bit registers comprising general register R0 to R14,
3181 control register GBR, and system registers MACH, MACL, and PR and the
3182 vector table address offset are saved into a register bank. Register
3183 banks are stacked in first-in last-out (FILO) sequence. Restoration
3184 from the bank is executed by issuing a RESBANK instruction.
3187 @cindex @code{returns_twice} attribute
3188 The @code{returns_twice} attribute tells the compiler that a function may
3189 return more than one time. The compiler will ensure that all registers
3190 are dead before calling such a function and will emit a warning about
3191 the variables that may be clobbered after the second return from the
3192 function. Examples of such functions are @code{setjmp} and @code{vfork}.
3193 The @code{longjmp}-like counterpart of such function, if any, might need
3194 to be marked with the @code{noreturn} attribute.
3197 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
3198 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
3199 all registers except the stack pointer should be saved in the prologue
3200 regardless of whether they are used or not.
3202 @item save_volatiles
3203 @cindex save volatile registers on the MicroBlaze
3204 Use this attribute on the MicroBlaze to indicate that the function is
3205 an interrupt handler. All volatile registers (in addition to non-volatile
3206 registers) will be saved in the function prologue. If the function is a leaf
3207 function, only volatiles used by the function are saved. A normal function
3208 return is generated instead of a return from interrupt.
3210 @item section ("@var{section-name}")
3211 @cindex @code{section} function attribute
3212 Normally, the compiler places the code it generates in the @code{text} section.
3213 Sometimes, however, you need additional sections, or you need certain
3214 particular functions to appear in special sections. The @code{section}
3215 attribute specifies that a function lives in a particular section.
3216 For example, the declaration:
3219 extern void foobar (void) __attribute__ ((section ("bar")));
3223 puts the function @code{foobar} in the @code{bar} section.
3225 Some file formats do not support arbitrary sections so the @code{section}
3226 attribute is not available on all platforms.
3227 If you need to map the entire contents of a module to a particular
3228 section, consider using the facilities of the linker instead.
3231 @cindex @code{sentinel} function attribute
3232 This function attribute ensures that a parameter in a function call is
3233 an explicit @code{NULL}. The attribute is only valid on variadic
3234 functions. By default, the sentinel is located at position zero, the
3235 last parameter of the function call. If an optional integer position
3236 argument P is supplied to the attribute, the sentinel must be located at
3237 position P counting backwards from the end of the argument list.
3240 __attribute__ ((sentinel))
3242 __attribute__ ((sentinel(0)))
3245 The attribute is automatically set with a position of 0 for the built-in
3246 functions @code{execl} and @code{execlp}. The built-in function
3247 @code{execle} has the attribute set with a position of 1.
3249 A valid @code{NULL} in this context is defined as zero with any pointer
3250 type. If your system defines the @code{NULL} macro with an integer type
3251 then you need to add an explicit cast. GCC replaces @code{stddef.h}
3252 with a copy that redefines NULL appropriately.
3254 The warnings for missing or incorrect sentinels are enabled with
3258 See long_call/short_call.
3261 See longcall/shortcall.
3264 @cindex signal handler functions on the AVR processors
3265 Use this attribute on the AVR to indicate that the specified
3266 function is a signal handler. The compiler will generate function
3267 entry and exit sequences suitable for use in a signal handler when this
3268 attribute is present. Interrupts will be disabled inside the function.
3271 Use this attribute on the SH to indicate an @code{interrupt_handler}
3272 function should switch to an alternate stack. It expects a string
3273 argument that names a global variable holding the address of the
3278 void f () __attribute__ ((interrupt_handler,
3279 sp_switch ("alt_stack")));
3283 @cindex functions that pop the argument stack on the 386
3284 On the Intel 386, the @code{stdcall} attribute causes the compiler to
3285 assume that the called function will pop off the stack space used to
3286 pass arguments, unless it takes a variable number of arguments.
3288 @item syscall_linkage
3289 @cindex @code{syscall_linkage} attribute
3290 This attribute is used to modify the IA64 calling convention by marking
3291 all input registers as live at all function exits. This makes it possible
3292 to restart a system call after an interrupt without having to save/restore
3293 the input registers. This also prevents kernel data from leaking into
3297 @cindex @code{target} function attribute
3298 The @code{target} attribute is used to specify that a function is to
3299 be compiled with different target options than specified on the
3300 command line. This can be used for instance to have functions
3301 compiled with a different ISA (instruction set architecture) than the
3302 default. You can also use the @samp{#pragma GCC target} pragma to set
3303 more than one function to be compiled with specific target options.
3304 @xref{Function Specific Option Pragmas}, for details about the
3305 @samp{#pragma GCC target} pragma.
3307 For instance on a 386, you could compile one function with
3308 @code{target("sse4.1,arch=core2")} and another with
3309 @code{target("sse4a,arch=amdfam10")} that would be equivalent to
3310 compiling the first function with @option{-msse4.1} and
3311 @option{-march=core2} options, and the second function with
3312 @option{-msse4a} and @option{-march=amdfam10} options. It is up to the
3313 user to make sure that a function is only invoked on a machine that
3314 supports the particular ISA it was compiled for (for example by using
3315 @code{cpuid} on 386 to determine what feature bits and architecture
3319 int core2_func (void) __attribute__ ((__target__ ("arch=core2")));
3320 int sse3_func (void) __attribute__ ((__target__ ("sse3")));
3323 On the 386, the following options are allowed:
3328 @cindex @code{target("abm")} attribute
3329 Enable/disable the generation of the advanced bit instructions.
3333 @cindex @code{target("aes")} attribute
3334 Enable/disable the generation of the AES instructions.
3338 @cindex @code{target("mmx")} attribute
3339 Enable/disable the generation of the MMX instructions.
3343 @cindex @code{target("pclmul")} attribute
3344 Enable/disable the generation of the PCLMUL instructions.
3348 @cindex @code{target("popcnt")} attribute
3349 Enable/disable the generation of the POPCNT instruction.
3353 @cindex @code{target("sse")} attribute
3354 Enable/disable the generation of the SSE instructions.
3358 @cindex @code{target("sse2")} attribute
3359 Enable/disable the generation of the SSE2 instructions.
3363 @cindex @code{target("sse3")} attribute
3364 Enable/disable the generation of the SSE3 instructions.
3368 @cindex @code{target("sse4")} attribute
3369 Enable/disable the generation of the SSE4 instructions (both SSE4.1
3374 @cindex @code{target("sse4.1")} attribute
3375 Enable/disable the generation of the sse4.1 instructions.
3379 @cindex @code{target("sse4.2")} attribute
3380 Enable/disable the generation of the sse4.2 instructions.
3384 @cindex @code{target("sse4a")} attribute
3385 Enable/disable the generation of the SSE4A instructions.
3389 @cindex @code{target("fma4")} attribute
3390 Enable/disable the generation of the FMA4 instructions.
3394 @cindex @code{target("xop")} attribute
3395 Enable/disable the generation of the XOP instructions.
3399 @cindex @code{target("lwp")} attribute
3400 Enable/disable the generation of the LWP instructions.
3404 @cindex @code{target("ssse3")} attribute
3405 Enable/disable the generation of the SSSE3 instructions.
3409 @cindex @code{target("cld")} attribute
3410 Enable/disable the generation of the CLD before string moves.
3412 @item fancy-math-387
3413 @itemx no-fancy-math-387
3414 @cindex @code{target("fancy-math-387")} attribute
3415 Enable/disable the generation of the @code{sin}, @code{cos}, and
3416 @code{sqrt} instructions on the 387 floating point unit.
3419 @itemx no-fused-madd
3420 @cindex @code{target("fused-madd")} attribute
3421 Enable/disable the generation of the fused multiply/add instructions.
3425 @cindex @code{target("ieee-fp")} attribute
3426 Enable/disable the generation of floating point that depends on IEEE arithmetic.
3428 @item inline-all-stringops
3429 @itemx no-inline-all-stringops
3430 @cindex @code{target("inline-all-stringops")} attribute
3431 Enable/disable inlining of string operations.
3433 @item inline-stringops-dynamically
3434 @itemx no-inline-stringops-dynamically
3435 @cindex @code{target("inline-stringops-dynamically")} attribute
3436 Enable/disable the generation of the inline code to do small string
3437 operations and calling the library routines for large operations.
3439 @item align-stringops
3440 @itemx no-align-stringops
3441 @cindex @code{target("align-stringops")} attribute
3442 Do/do not align destination of inlined string operations.
3446 @cindex @code{target("recip")} attribute
3447 Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS
3448 instructions followed an additional Newton-Raphson step instead of
3449 doing a floating point division.
3451 @item arch=@var{ARCH}
3452 @cindex @code{target("arch=@var{ARCH}")} attribute
3453 Specify the architecture to generate code for in compiling the function.
3455 @item tune=@var{TUNE}
3456 @cindex @code{target("tune=@var{TUNE}")} attribute
3457 Specify the architecture to tune for in compiling the function.
3459 @item fpmath=@var{FPMATH}
3460 @cindex @code{target("fpmath=@var{FPMATH}")} attribute
3461 Specify which floating point unit to use. The
3462 @code{target("fpmath=sse,387")} option must be specified as
3463 @code{target("fpmath=sse+387")} because the comma would separate
3467 On the PowerPC, the following options are allowed:
3472 @cindex @code{target("altivec")} attribute
3473 Generate code that uses (does not use) AltiVec instructions. In
3474 32-bit code, you cannot enable Altivec instructions unless
3475 @option{-mabi=altivec} was used on the command line.
3479 @cindex @code{target("cmpb")} attribute
3480 Generate code that uses (does not use) the compare bytes instruction
3481 implemented on the POWER6 processor and other processors that support
3482 the PowerPC V2.05 architecture.
3486 @cindex @code{target("dlmzb")} attribute
3487 Generate code that uses (does not use) the string-search @samp{dlmzb}
3488 instruction on the IBM 405, 440, 464 and 476 processors. This instruction is
3489 generated by default when targetting those processors.
3493 @cindex @code{target("fprnd")} attribute
3494 Generate code that uses (does not use) the FP round to integer
3495 instructions implemented on the POWER5+ processor and other processors
3496 that support the PowerPC V2.03 architecture.
3500 @cindex @code{target("hard-dfp")} attribute
3501 Generate code that uses (does not use) the decimal floating point
3502 instructions implemented on some POWER processors.
3506 @cindex @code{target("isel")} attribute
3507 Generate code that uses (does not use) ISEL instruction.
3511 @cindex @code{target("mfcrf")} attribute
3512 Generate code that uses (does not use) the move from condition
3513 register field instruction implemented on the POWER4 processor and
3514 other processors that support the PowerPC V2.01 architecture.
3518 @cindex @code{target("mfpgpr")} attribute
3519 Generate code that uses (does not use) the FP move to/from general
3520 purpose register instructions implemented on the POWER6X processor and
3521 other processors that support the extended PowerPC V2.05 architecture.
3525 @cindex @code{target("mulhw")} attribute
3526 Generate code that uses (does not use) the half-word multiply and
3527 multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors.
3528 These instructions are generated by default when targetting those
3533 @cindex @code{target("multiple")} attribute
3534 Generate code that uses (does not use) the load multiple word
3535 instructions and the store multiple word instructions.
3539 @cindex @code{target("update")} attribute
3540 Generate code that uses (does not use) the load or store instructions
3541 that update the base register to the address of the calculated memory
3546 @cindex @code{target("popcntb")} attribute
3547 Generate code that uses (does not use) the popcount and double
3548 precision FP reciprocal estimate instruction implemented on the POWER5
3549 processor and other processors that support the PowerPC V2.02
3554 @cindex @code{target("popcntd")} attribute
3555 Generate code that uses (does not use) the popcount instruction
3556 implemented on the POWER7 processor and other processors that support
3557 the PowerPC V2.06 architecture.
3559 @item powerpc-gfxopt
3560 @itemx no-powerpc-gfxopt
3561 @cindex @code{target("powerpc-gfxopt")} attribute
3562 Generate code that uses (does not use) the optional PowerPC
3563 architecture instructions in the Graphics group, including
3564 floating-point select.
3567 @itemx no-powerpc-gpopt
3568 @cindex @code{target("powerpc-gpopt")} attribute
3569 Generate code that uses (does not use) the optional PowerPC
3570 architecture instructions in the General Purpose group, including
3571 floating-point square root.
3573 @item recip-precision
3574 @itemx no-recip-precision
3575 @cindex @code{target("recip-precision")} attribute
3576 Assume (do not assume) that the reciprocal estimate instructions
3577 provide higher precision estimates than is mandated by the powerpc
3582 @cindex @code{target("string")} attribute
3583 Generate code that uses (does not use) the load string instructions
3584 and the store string word instructions to save multiple registers and
3585 do small block moves.
3589 @cindex @code{target("vsx")} attribute
3590 Generate code that uses (does not use) vector/scalar (VSX)
3591 instructions, and also enable the use of built-in functions that allow
3592 more direct access to the VSX instruction set. In 32-bit code, you
3593 cannot enable VSX or Altivec instructions unless
3594 @option{-mabi=altivec} was used on the command line.
3598 @cindex @code{target("friz")} attribute
3599 Generate (do not generate) the @code{friz} instruction when the
3600 @option{-funsafe-math-optimizations} option is used to optimize
3601 rounding a floating point value to 64-bit integer and back to floating
3602 point. The @code{friz} instruction does not return the same value if
3603 the floating point number is too large to fit in an integer.
3605 @item avoid-indexed-addresses
3606 @itemx no-avoid-indexed-addresses
3607 @cindex @code{target("avoid-indexed-addresses")} attribute
3608 Generate code that tries to avoid (not avoid) the use of indexed load
3609 or store instructions.
3613 @cindex @code{target("paired")} attribute
3614 Generate code that uses (does not use) the generation of PAIRED simd
3619 @cindex @code{target("longcall")} attribute
3620 Generate code that assumes (does not assume) that all calls are far
3621 away so that a longer more expensive calling sequence is required.
3624 @cindex @code{target("cpu=@var{CPU}")} attribute
3625 Specify the architecture to generate code for when compiling the
3626 function. If you select the @code{target("cpu=power7")} attribute when
3627 generating 32-bit code, VSX and Altivec instructions are not generated
3628 unless you use the @option{-mabi=altivec} option on the command line.
3630 @item tune=@var{TUNE}
3631 @cindex @code{target("tune=@var{TUNE}")} attribute
3632 Specify the architecture to tune for when compiling the function. If
3633 you do not specify the @code{target("tune=@var{TUNE}")} attribute and
3634 you do specify the @code{target("cpu=@var{CPU}")} attribute,
3635 compilation will tune for the @var{CPU} architecture, and not the
3636 default tuning specified on the command line.
3639 On the 386/x86_64 and PowerPC backends, you can use either multiple
3640 strings to specify multiple options, or you can separate the option
3641 with a comma (@code{,}).
3643 On the 386/x86_64 and PowerPC backends, the inliner will not inline a
3644 function that has different target options than the caller, unless the
3645 callee has a subset of the target options of the caller. For example
3646 a function declared with @code{target("sse3")} can inline a function
3647 with @code{target("sse2")}, since @code{-msse3} implies @code{-msse2}.
3649 The @code{target} attribute is not implemented in GCC versions earlier
3650 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC backends. It is
3651 not currently implemented for other backends.
3654 @cindex tiny data section on the H8/300H and H8S
3655 Use this attribute on the H8/300H and H8S to indicate that the specified
3656 variable should be placed into the tiny data section.
3657 The compiler will generate more efficient code for loads and stores
3658 on data in the tiny data section. Note the tiny data area is limited to
3659 slightly under 32kbytes of data.
3662 Use this attribute on the SH for an @code{interrupt_handler} to return using
3663 @code{trapa} instead of @code{rte}. This attribute expects an integer
3664 argument specifying the trap number to be used.
3667 @cindex @code{unused} attribute.
3668 This attribute, attached to a function, means that the function is meant
3669 to be possibly unused. GCC will not produce a warning for this
3673 @cindex @code{used} attribute.
3674 This attribute, attached to a function, means that code must be emitted
3675 for the function even if it appears that the function is not referenced.
3676 This is useful, for example, when the function is referenced only in
3679 When applied to a member function of a C++ class template, the
3680 attribute also means that the function will be instantiated if the
3681 class itself is instantiated.
3684 @cindex @code{version_id} attribute
3685 This IA64 HP-UX attribute, attached to a global variable or function, renames a
3686 symbol to contain a version string, thus allowing for function level
3687 versioning. HP-UX system header files may use version level functioning
3688 for some system calls.
3691 extern int foo () __attribute__((version_id ("20040821")));
3694 Calls to @var{foo} will be mapped to calls to @var{foo@{20040821@}}.
3696 @item visibility ("@var{visibility_type}")
3697 @cindex @code{visibility} attribute
3698 This attribute affects the linkage of the declaration to which it is attached.
3699 There are four supported @var{visibility_type} values: default,
3700 hidden, protected or internal visibility.
3703 void __attribute__ ((visibility ("protected")))
3704 f () @{ /* @r{Do something.} */; @}
3705 int i __attribute__ ((visibility ("hidden")));
3708 The possible values of @var{visibility_type} correspond to the
3709 visibility settings in the ELF gABI.
3712 @c keep this list of visibilities in alphabetical order.
3715 Default visibility is the normal case for the object file format.
3716 This value is available for the visibility attribute to override other
3717 options that may change the assumed visibility of entities.
3719 On ELF, default visibility means that the declaration is visible to other
3720 modules and, in shared libraries, means that the declared entity may be
3723 On Darwin, default visibility means that the declaration is visible to
3726 Default visibility corresponds to ``external linkage'' in the language.
3729 Hidden visibility indicates that the entity declared will have a new
3730 form of linkage, which we'll call ``hidden linkage''. Two
3731 declarations of an object with hidden linkage refer to the same object
3732 if they are in the same shared object.
3735 Internal visibility is like hidden visibility, but with additional
3736 processor specific semantics. Unless otherwise specified by the
3737 psABI, GCC defines internal visibility to mean that a function is
3738 @emph{never} called from another module. Compare this with hidden
3739 functions which, while they cannot be referenced directly by other
3740 modules, can be referenced indirectly via function pointers. By
3741 indicating that a function cannot be called from outside the module,
3742 GCC may for instance omit the load of a PIC register since it is known
3743 that the calling function loaded the correct value.
3746 Protected visibility is like default visibility except that it
3747 indicates that references within the defining module will bind to the
3748 definition in that module. That is, the declared entity cannot be
3749 overridden by another module.
3753 All visibilities are supported on many, but not all, ELF targets
3754 (supported when the assembler supports the @samp{.visibility}
3755 pseudo-op). Default visibility is supported everywhere. Hidden
3756 visibility is supported on Darwin targets.
3758 The visibility attribute should be applied only to declarations which
3759 would otherwise have external linkage. The attribute should be applied
3760 consistently, so that the same entity should not be declared with
3761 different settings of the attribute.
3763 In C++, the visibility attribute applies to types as well as functions
3764 and objects, because in C++ types have linkage. A class must not have
3765 greater visibility than its non-static data member types and bases,
3766 and class members default to the visibility of their class. Also, a
3767 declaration without explicit visibility is limited to the visibility
3770 In C++, you can mark member functions and static member variables of a
3771 class with the visibility attribute. This is useful if you know a
3772 particular method or static member variable should only be used from
3773 one shared object; then you can mark it hidden while the rest of the
3774 class has default visibility. Care must be taken to avoid breaking
3775 the One Definition Rule; for example, it is usually not useful to mark
3776 an inline method as hidden without marking the whole class as hidden.
3778 A C++ namespace declaration can also have the visibility attribute.
3779 This attribute applies only to the particular namespace body, not to
3780 other definitions of the same namespace; it is equivalent to using
3781 @samp{#pragma GCC visibility} before and after the namespace
3782 definition (@pxref{Visibility Pragmas}).
3784 In C++, if a template argument has limited visibility, this
3785 restriction is implicitly propagated to the template instantiation.
3786 Otherwise, template instantiations and specializations default to the
3787 visibility of their template.
3789 If both the template and enclosing class have explicit visibility, the
3790 visibility from the template is used.
3793 @cindex @code{vliw} attribute
3794 On MeP, the @code{vliw} attribute tells the compiler to emit
3795 instructions in VLIW mode instead of core mode. Note that this
3796 attribute is not allowed unless a VLIW coprocessor has been configured
3797 and enabled through command line options.
3799 @item warn_unused_result
3800 @cindex @code{warn_unused_result} attribute
3801 The @code{warn_unused_result} attribute causes a warning to be emitted
3802 if a caller of the function with this attribute does not use its
3803 return value. This is useful for functions where not checking
3804 the result is either a security problem or always a bug, such as
3808 int fn () __attribute__ ((warn_unused_result));
3811 if (fn () < 0) return -1;
3817 results in warning on line 5.
3820 @cindex @code{weak} attribute
3821 The @code{weak} attribute causes the declaration to be emitted as a weak
3822 symbol rather than a global. This is primarily useful in defining
3823 library functions which can be overridden in user code, though it can
3824 also be used with non-function declarations. Weak symbols are supported
3825 for ELF targets, and also for a.out targets when using the GNU assembler
3829 @itemx weakref ("@var{target}")
3830 @cindex @code{weakref} attribute
3831 The @code{weakref} attribute marks a declaration as a weak reference.
3832 Without arguments, it should be accompanied by an @code{alias} attribute
3833 naming the target symbol. Optionally, the @var{target} may be given as
3834 an argument to @code{weakref} itself. In either case, @code{weakref}
3835 implicitly marks the declaration as @code{weak}. Without a
3836 @var{target}, given as an argument to @code{weakref} or to @code{alias},
3837 @code{weakref} is equivalent to @code{weak}.
3840 static int x() __attribute__ ((weakref ("y")));
3841 /* is equivalent to... */
3842 static int x() __attribute__ ((weak, weakref, alias ("y")));
3844 static int x() __attribute__ ((weakref));
3845 static int x() __attribute__ ((alias ("y")));
3848 A weak reference is an alias that does not by itself require a
3849 definition to be given for the target symbol. If the target symbol is
3850 only referenced through weak references, then it becomes a @code{weak}
3851 undefined symbol. If it is directly referenced, however, then such
3852 strong references prevail, and a definition will be required for the
3853 symbol, not necessarily in the same translation unit.
3855 The effect is equivalent to moving all references to the alias to a
3856 separate translation unit, renaming the alias to the aliased symbol,
3857 declaring it as weak, compiling the two separate translation units and
3858 performing a reloadable link on them.
3860 At present, a declaration to which @code{weakref} is attached can
3861 only be @code{static}.
3865 You can specify multiple attributes in a declaration by separating them
3866 by commas within the double parentheses or by immediately following an
3867 attribute declaration with another attribute declaration.
3869 @cindex @code{#pragma}, reason for not using
3870 @cindex pragma, reason for not using
3871 Some people object to the @code{__attribute__} feature, suggesting that
3872 ISO C's @code{#pragma} should be used instead. At the time
3873 @code{__attribute__} was designed, there were two reasons for not doing
3878 It is impossible to generate @code{#pragma} commands from a macro.
3881 There is no telling what the same @code{#pragma} might mean in another
3885 These two reasons applied to almost any application that might have been
3886 proposed for @code{#pragma}. It was basically a mistake to use
3887 @code{#pragma} for @emph{anything}.
3889 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
3890 to be generated from macros. In addition, a @code{#pragma GCC}
3891 namespace is now in use for GCC-specific pragmas. However, it has been
3892 found convenient to use @code{__attribute__} to achieve a natural
3893 attachment of attributes to their corresponding declarations, whereas
3894 @code{#pragma GCC} is of use for constructs that do not naturally form
3895 part of the grammar. @xref{Other Directives,,Miscellaneous
3896 Preprocessing Directives, cpp, The GNU C Preprocessor}.
3898 @node Attribute Syntax
3899 @section Attribute Syntax
3900 @cindex attribute syntax
3902 This section describes the syntax with which @code{__attribute__} may be
3903 used, and the constructs to which attribute specifiers bind, for the C
3904 language. Some details may vary for C++ and Objective-C@. Because of
3905 infelicities in the grammar for attributes, some forms described here
3906 may not be successfully parsed in all cases.
3908 There are some problems with the semantics of attributes in C++. For
3909 example, there are no manglings for attributes, although they may affect
3910 code generation, so problems may arise when attributed types are used in
3911 conjunction with templates or overloading. Similarly, @code{typeid}
3912 does not distinguish between types with different attributes. Support
3913 for attributes in C++ may be restricted in future to attributes on
3914 declarations only, but not on nested declarators.
3916 @xref{Function Attributes}, for details of the semantics of attributes
3917 applying to functions. @xref{Variable Attributes}, for details of the
3918 semantics of attributes applying to variables. @xref{Type Attributes},
3919 for details of the semantics of attributes applying to structure, union
3920 and enumerated types.
3922 An @dfn{attribute specifier} is of the form
3923 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
3924 is a possibly empty comma-separated sequence of @dfn{attributes}, where
3925 each attribute is one of the following:
3929 Empty. Empty attributes are ignored.
3932 A word (which may be an identifier such as @code{unused}, or a reserved
3933 word such as @code{const}).
3936 A word, followed by, in parentheses, parameters for the attribute.
3937 These parameters take one of the following forms:
3941 An identifier. For example, @code{mode} attributes use this form.
3944 An identifier followed by a comma and a non-empty comma-separated list
3945 of expressions. For example, @code{format} attributes use this form.
3948 A possibly empty comma-separated list of expressions. For example,
3949 @code{format_arg} attributes use this form with the list being a single
3950 integer constant expression, and @code{alias} attributes use this form
3951 with the list being a single string constant.
3955 An @dfn{attribute specifier list} is a sequence of one or more attribute
3956 specifiers, not separated by any other tokens.
3958 In GNU C, an attribute specifier list may appear after the colon following a
3959 label, other than a @code{case} or @code{default} label. The only
3960 attribute it makes sense to use after a label is @code{unused}. This
3961 feature is intended for code generated by programs which contains labels
3962 that may be unused but which is compiled with @option{-Wall}. It would
3963 not normally be appropriate to use in it human-written code, though it
3964 could be useful in cases where the code that jumps to the label is
3965 contained within an @code{#ifdef} conditional. GNU C++ only permits
3966 attributes on labels if the attribute specifier is immediately
3967 followed by a semicolon (i.e., the label applies to an empty
3968 statement). If the semicolon is missing, C++ label attributes are
3969 ambiguous, as it is permissible for a declaration, which could begin
3970 with an attribute list, to be labelled in C++. Declarations cannot be
3971 labelled in C90 or C99, so the ambiguity does not arise there.
3973 An attribute specifier list may appear as part of a @code{struct},
3974 @code{union} or @code{enum} specifier. It may go either immediately
3975 after the @code{struct}, @code{union} or @code{enum} keyword, or after
3976 the closing brace. The former syntax is preferred.
3977 Where attribute specifiers follow the closing brace, they are considered
3978 to relate to the structure, union or enumerated type defined, not to any
3979 enclosing declaration the type specifier appears in, and the type
3980 defined is not complete until after the attribute specifiers.
3981 @c Otherwise, there would be the following problems: a shift/reduce
3982 @c conflict between attributes binding the struct/union/enum and
3983 @c binding to the list of specifiers/qualifiers; and "aligned"
3984 @c attributes could use sizeof for the structure, but the size could be
3985 @c changed later by "packed" attributes.
3987 Otherwise, an attribute specifier appears as part of a declaration,
3988 counting declarations of unnamed parameters and type names, and relates
3989 to that declaration (which may be nested in another declaration, for
3990 example in the case of a parameter declaration), or to a particular declarator
3991 within a declaration. Where an
3992 attribute specifier is applied to a parameter declared as a function or
3993 an array, it should apply to the function or array rather than the
3994 pointer to which the parameter is implicitly converted, but this is not
3995 yet correctly implemented.
3997 Any list of specifiers and qualifiers at the start of a declaration may
3998 contain attribute specifiers, whether or not such a list may in that
3999 context contain storage class specifiers. (Some attributes, however,
4000 are essentially in the nature of storage class specifiers, and only make
4001 sense where storage class specifiers may be used; for example,
4002 @code{section}.) There is one necessary limitation to this syntax: the
4003 first old-style parameter declaration in a function definition cannot
4004 begin with an attribute specifier, because such an attribute applies to
4005 the function instead by syntax described below (which, however, is not
4006 yet implemented in this case). In some other cases, attribute
4007 specifiers are permitted by this grammar but not yet supported by the
4008 compiler. All attribute specifiers in this place relate to the
4009 declaration as a whole. In the obsolescent usage where a type of
4010 @code{int} is implied by the absence of type specifiers, such a list of
4011 specifiers and qualifiers may be an attribute specifier list with no
4012 other specifiers or qualifiers.
4014 At present, the first parameter in a function prototype must have some
4015 type specifier which is not an attribute specifier; this resolves an
4016 ambiguity in the interpretation of @code{void f(int
4017 (__attribute__((foo)) x))}, but is subject to change. At present, if
4018 the parentheses of a function declarator contain only attributes then
4019 those attributes are ignored, rather than yielding an error or warning
4020 or implying a single parameter of type int, but this is subject to
4023 An attribute specifier list may appear immediately before a declarator
4024 (other than the first) in a comma-separated list of declarators in a
4025 declaration of more than one identifier using a single list of
4026 specifiers and qualifiers. Such attribute specifiers apply
4027 only to the identifier before whose declarator they appear. For
4031 __attribute__((noreturn)) void d0 (void),
4032 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
4037 the @code{noreturn} attribute applies to all the functions
4038 declared; the @code{format} attribute only applies to @code{d1}.
4040 An attribute specifier list may appear immediately before the comma,
4041 @code{=} or semicolon terminating the declaration of an identifier other
4042 than a function definition. Such attribute specifiers apply
4043 to the declared object or function. Where an
4044 assembler name for an object or function is specified (@pxref{Asm
4045 Labels}), the attribute must follow the @code{asm}
4048 An attribute specifier list may, in future, be permitted to appear after
4049 the declarator in a function definition (before any old-style parameter
4050 declarations or the function body).
4052 Attribute specifiers may be mixed with type qualifiers appearing inside
4053 the @code{[]} of a parameter array declarator, in the C99 construct by
4054 which such qualifiers are applied to the pointer to which the array is
4055 implicitly converted. Such attribute specifiers apply to the pointer,
4056 not to the array, but at present this is not implemented and they are
4059 An attribute specifier list may appear at the start of a nested
4060 declarator. At present, there are some limitations in this usage: the
4061 attributes correctly apply to the declarator, but for most individual
4062 attributes the semantics this implies are not implemented.
4063 When attribute specifiers follow the @code{*} of a pointer
4064 declarator, they may be mixed with any type qualifiers present.
4065 The following describes the formal semantics of this syntax. It will make the
4066 most sense if you are familiar with the formal specification of
4067 declarators in the ISO C standard.
4069 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
4070 D1}, where @code{T} contains declaration specifiers that specify a type
4071 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
4072 contains an identifier @var{ident}. The type specified for @var{ident}
4073 for derived declarators whose type does not include an attribute
4074 specifier is as in the ISO C standard.
4076 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
4077 and the declaration @code{T D} specifies the type
4078 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4079 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4080 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
4082 If @code{D1} has the form @code{*
4083 @var{type-qualifier-and-attribute-specifier-list} D}, and the
4084 declaration @code{T D} specifies the type
4085 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4086 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4087 @var{type-qualifier-and-attribute-specifier-list} pointer to @var{Type}'' for
4093 void (__attribute__((noreturn)) ****f) (void);
4097 specifies the type ``pointer to pointer to pointer to pointer to
4098 non-returning function returning @code{void}''. As another example,
4101 char *__attribute__((aligned(8))) *f;
4105 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
4106 Note again that this does not work with most attributes; for example,
4107 the usage of @samp{aligned} and @samp{noreturn} attributes given above
4108 is not yet supported.
4110 For compatibility with existing code written for compiler versions that
4111 did not implement attributes on nested declarators, some laxity is
4112 allowed in the placing of attributes. If an attribute that only applies
4113 to types is applied to a declaration, it will be treated as applying to
4114 the type of that declaration. If an attribute that only applies to
4115 declarations is applied to the type of a declaration, it will be treated
4116 as applying to that declaration; and, for compatibility with code
4117 placing the attributes immediately before the identifier declared, such
4118 an attribute applied to a function return type will be treated as
4119 applying to the function type, and such an attribute applied to an array
4120 element type will be treated as applying to the array type. If an
4121 attribute that only applies to function types is applied to a
4122 pointer-to-function type, it will be treated as applying to the pointer
4123 target type; if such an attribute is applied to a function return type
4124 that is not a pointer-to-function type, it will be treated as applying
4125 to the function type.
4127 @node Function Prototypes
4128 @section Prototypes and Old-Style Function Definitions
4129 @cindex function prototype declarations
4130 @cindex old-style function definitions
4131 @cindex promotion of formal parameters
4133 GNU C extends ISO C to allow a function prototype to override a later
4134 old-style non-prototype definition. Consider the following example:
4137 /* @r{Use prototypes unless the compiler is old-fashioned.} */
4144 /* @r{Prototype function declaration.} */
4145 int isroot P((uid_t));
4147 /* @r{Old-style function definition.} */
4149 isroot (x) /* @r{??? lossage here ???} */
4156 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
4157 not allow this example, because subword arguments in old-style
4158 non-prototype definitions are promoted. Therefore in this example the
4159 function definition's argument is really an @code{int}, which does not
4160 match the prototype argument type of @code{short}.
4162 This restriction of ISO C makes it hard to write code that is portable
4163 to traditional C compilers, because the programmer does not know
4164 whether the @code{uid_t} type is @code{short}, @code{int}, or
4165 @code{long}. Therefore, in cases like these GNU C allows a prototype
4166 to override a later old-style definition. More precisely, in GNU C, a
4167 function prototype argument type overrides the argument type specified
4168 by a later old-style definition if the former type is the same as the
4169 latter type before promotion. Thus in GNU C the above example is
4170 equivalent to the following:
4183 GNU C++ does not support old-style function definitions, so this
4184 extension is irrelevant.
4187 @section C++ Style Comments
4189 @cindex C++ comments
4190 @cindex comments, C++ style
4192 In GNU C, you may use C++ style comments, which start with @samp{//} and
4193 continue until the end of the line. Many other C implementations allow
4194 such comments, and they are included in the 1999 C standard. However,
4195 C++ style comments are not recognized if you specify an @option{-std}
4196 option specifying a version of ISO C before C99, or @option{-ansi}
4197 (equivalent to @option{-std=c90}).
4200 @section Dollar Signs in Identifier Names
4202 @cindex dollar signs in identifier names
4203 @cindex identifier names, dollar signs in
4205 In GNU C, you may normally use dollar signs in identifier names.
4206 This is because many traditional C implementations allow such identifiers.
4207 However, dollar signs in identifiers are not supported on a few target
4208 machines, typically because the target assembler does not allow them.
4210 @node Character Escapes
4211 @section The Character @key{ESC} in Constants
4213 You can use the sequence @samp{\e} in a string or character constant to
4214 stand for the ASCII character @key{ESC}.
4216 @node Variable Attributes
4217 @section Specifying Attributes of Variables
4218 @cindex attribute of variables
4219 @cindex variable attributes
4221 The keyword @code{__attribute__} allows you to specify special
4222 attributes of variables or structure fields. This keyword is followed
4223 by an attribute specification inside double parentheses. Some
4224 attributes are currently defined generically for variables.
4225 Other attributes are defined for variables on particular target
4226 systems. Other attributes are available for functions
4227 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
4228 Other front ends might define more attributes
4229 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
4231 You may also specify attributes with @samp{__} preceding and following
4232 each keyword. This allows you to use them in header files without
4233 being concerned about a possible macro of the same name. For example,
4234 you may use @code{__aligned__} instead of @code{aligned}.
4236 @xref{Attribute Syntax}, for details of the exact syntax for using
4240 @cindex @code{aligned} attribute
4241 @item aligned (@var{alignment})
4242 This attribute specifies a minimum alignment for the variable or
4243 structure field, measured in bytes. For example, the declaration:
4246 int x __attribute__ ((aligned (16))) = 0;
4250 causes the compiler to allocate the global variable @code{x} on a
4251 16-byte boundary. On a 68040, this could be used in conjunction with
4252 an @code{asm} expression to access the @code{move16} instruction which
4253 requires 16-byte aligned operands.
4255 You can also specify the alignment of structure fields. For example, to
4256 create a double-word aligned @code{int} pair, you could write:
4259 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
4263 This is an alternative to creating a union with a @code{double} member
4264 that forces the union to be double-word aligned.
4266 As in the preceding examples, you can explicitly specify the alignment
4267 (in bytes) that you wish the compiler to use for a given variable or
4268 structure field. Alternatively, you can leave out the alignment factor
4269 and just ask the compiler to align a variable or field to the
4270 default alignment for the target architecture you are compiling for.
4271 The default alignment is sufficient for all scalar types, but may not be
4272 enough for all vector types on a target which supports vector operations.
4273 The default alignment is fixed for a particular target ABI.
4275 Gcc also provides a target specific macro @code{__BIGGEST_ALIGNMENT__},
4276 which is the largest alignment ever used for any data type on the
4277 target machine you are compiling for. For example, you could write:
4280 short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__)));
4283 The compiler automatically sets the alignment for the declared
4284 variable or field to @code{__BIGGEST_ALIGNMENT__}. Doing this can
4285 often make copy operations more efficient, because the compiler can
4286 use whatever instructions copy the biggest chunks of memory when
4287 performing copies to or from the variables or fields that you have
4288 aligned this way. Note that the value of @code{__BIGGEST_ALIGNMENT__}
4289 may change depending on command line options.
4291 When used on a struct, or struct member, the @code{aligned} attribute can
4292 only increase the alignment; in order to decrease it, the @code{packed}
4293 attribute must be specified as well. When used as part of a typedef, the
4294 @code{aligned} attribute can both increase and decrease alignment, and
4295 specifying the @code{packed} attribute will generate a warning.
4297 Note that the effectiveness of @code{aligned} attributes may be limited
4298 by inherent limitations in your linker. On many systems, the linker is
4299 only able to arrange for variables to be aligned up to a certain maximum
4300 alignment. (For some linkers, the maximum supported alignment may
4301 be very very small.) If your linker is only able to align variables
4302 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
4303 in an @code{__attribute__} will still only provide you with 8 byte
4304 alignment. See your linker documentation for further information.
4306 The @code{aligned} attribute can also be used for functions
4307 (@pxref{Function Attributes}.)
4309 @item cleanup (@var{cleanup_function})
4310 @cindex @code{cleanup} attribute
4311 The @code{cleanup} attribute runs a function when the variable goes
4312 out of scope. This attribute can only be applied to auto function
4313 scope variables; it may not be applied to parameters or variables
4314 with static storage duration. The function must take one parameter,
4315 a pointer to a type compatible with the variable. The return value
4316 of the function (if any) is ignored.
4318 If @option{-fexceptions} is enabled, then @var{cleanup_function}
4319 will be run during the stack unwinding that happens during the
4320 processing of the exception. Note that the @code{cleanup} attribute
4321 does not allow the exception to be caught, only to perform an action.
4322 It is undefined what happens if @var{cleanup_function} does not
4327 @cindex @code{common} attribute
4328 @cindex @code{nocommon} attribute
4331 The @code{common} attribute requests GCC to place a variable in
4332 ``common'' storage. The @code{nocommon} attribute requests the
4333 opposite---to allocate space for it directly.
4335 These attributes override the default chosen by the
4336 @option{-fno-common} and @option{-fcommon} flags respectively.
4339 @itemx deprecated (@var{msg})
4340 @cindex @code{deprecated} attribute
4341 The @code{deprecated} attribute results in a warning if the variable
4342 is used anywhere in the source file. This is useful when identifying
4343 variables that are expected to be removed in a future version of a
4344 program. The warning also includes the location of the declaration
4345 of the deprecated variable, to enable users to easily find further
4346 information about why the variable is deprecated, or what they should
4347 do instead. Note that the warning only occurs for uses:
4350 extern int old_var __attribute__ ((deprecated));
4352 int new_fn () @{ return old_var; @}
4355 results in a warning on line 3 but not line 2. The optional msg
4356 argument, which must be a string, will be printed in the warning if
4359 The @code{deprecated} attribute can also be used for functions and
4360 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
4362 @item mode (@var{mode})
4363 @cindex @code{mode} attribute
4364 This attribute specifies the data type for the declaration---whichever
4365 type corresponds to the mode @var{mode}. This in effect lets you
4366 request an integer or floating point type according to its width.
4368 You may also specify a mode of @samp{byte} or @samp{__byte__} to
4369 indicate the mode corresponding to a one-byte integer, @samp{word} or
4370 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
4371 or @samp{__pointer__} for the mode used to represent pointers.
4374 @cindex @code{packed} attribute
4375 The @code{packed} attribute specifies that a variable or structure field
4376 should have the smallest possible alignment---one byte for a variable,
4377 and one bit for a field, unless you specify a larger value with the
4378 @code{aligned} attribute.
4380 Here is a structure in which the field @code{x} is packed, so that it
4381 immediately follows @code{a}:
4387 int x[2] __attribute__ ((packed));
4391 @emph{Note:} The 4.1, 4.2 and 4.3 series of GCC ignore the
4392 @code{packed} attribute on bit-fields of type @code{char}. This has
4393 been fixed in GCC 4.4 but the change can lead to differences in the
4394 structure layout. See the documentation of
4395 @option{-Wpacked-bitfield-compat} for more information.
4397 @item section ("@var{section-name}")
4398 @cindex @code{section} variable attribute
4399 Normally, the compiler places the objects it generates in sections like
4400 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
4401 or you need certain particular variables to appear in special sections,
4402 for example to map to special hardware. The @code{section}
4403 attribute specifies that a variable (or function) lives in a particular
4404 section. For example, this small program uses several specific section names:
4407 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
4408 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
4409 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
4410 int init_data __attribute__ ((section ("INITDATA")));
4414 /* @r{Initialize stack pointer} */
4415 init_sp (stack + sizeof (stack));
4417 /* @r{Initialize initialized data} */
4418 memcpy (&init_data, &data, &edata - &data);
4420 /* @r{Turn on the serial ports} */
4427 Use the @code{section} attribute with
4428 @emph{global} variables and not @emph{local} variables,
4429 as shown in the example.
4431 You may use the @code{section} attribute with initialized or
4432 uninitialized global variables but the linker requires
4433 each object be defined once, with the exception that uninitialized
4434 variables tentatively go in the @code{common} (or @code{bss}) section
4435 and can be multiply ``defined''. Using the @code{section} attribute
4436 will change what section the variable goes into and may cause the
4437 linker to issue an error if an uninitialized variable has multiple
4438 definitions. You can force a variable to be initialized with the
4439 @option{-fno-common} flag or the @code{nocommon} attribute.
4441 Some file formats do not support arbitrary sections so the @code{section}
4442 attribute is not available on all platforms.
4443 If you need to map the entire contents of a module to a particular
4444 section, consider using the facilities of the linker instead.
4447 @cindex @code{shared} variable attribute
4448 On Microsoft Windows, in addition to putting variable definitions in a named
4449 section, the section can also be shared among all running copies of an
4450 executable or DLL@. For example, this small program defines shared data
4451 by putting it in a named section @code{shared} and marking the section
4455 int foo __attribute__((section ("shared"), shared)) = 0;
4460 /* @r{Read and write foo. All running
4461 copies see the same value.} */
4467 You may only use the @code{shared} attribute along with @code{section}
4468 attribute with a fully initialized global definition because of the way
4469 linkers work. See @code{section} attribute for more information.
4471 The @code{shared} attribute is only available on Microsoft Windows@.
4473 @item tls_model ("@var{tls_model}")
4474 @cindex @code{tls_model} attribute
4475 The @code{tls_model} attribute sets thread-local storage model
4476 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
4477 overriding @option{-ftls-model=} command-line switch on a per-variable
4479 The @var{tls_model} argument should be one of @code{global-dynamic},
4480 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
4482 Not all targets support this attribute.
4485 This attribute, attached to a variable, means that the variable is meant
4486 to be possibly unused. GCC will not produce a warning for this
4490 This attribute, attached to a variable, means that the variable must be
4491 emitted even if it appears that the variable is not referenced.
4493 When applied to a static data member of a C++ class template, the
4494 attribute also means that the member will be instantiated if the
4495 class itself is instantiated.
4497 @item vector_size (@var{bytes})
4498 This attribute specifies the vector size for the variable, measured in
4499 bytes. For example, the declaration:
4502 int foo __attribute__ ((vector_size (16)));
4506 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
4507 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
4508 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
4510 This attribute is only applicable to integral and float scalars,
4511 although arrays, pointers, and function return values are allowed in
4512 conjunction with this construct.
4514 Aggregates with this attribute are invalid, even if they are of the same
4515 size as a corresponding scalar. For example, the declaration:
4518 struct S @{ int a; @};
4519 struct S __attribute__ ((vector_size (16))) foo;
4523 is invalid even if the size of the structure is the same as the size of
4527 The @code{selectany} attribute causes an initialized global variable to
4528 have link-once semantics. When multiple definitions of the variable are
4529 encountered by the linker, the first is selected and the remainder are
4530 discarded. Following usage by the Microsoft compiler, the linker is told
4531 @emph{not} to warn about size or content differences of the multiple
4534 Although the primary usage of this attribute is for POD types, the
4535 attribute can also be applied to global C++ objects that are initialized
4536 by a constructor. In this case, the static initialization and destruction
4537 code for the object is emitted in each translation defining the object,
4538 but the calls to the constructor and destructor are protected by a
4539 link-once guard variable.
4541 The @code{selectany} attribute is only available on Microsoft Windows
4542 targets. You can use @code{__declspec (selectany)} as a synonym for
4543 @code{__attribute__ ((selectany))} for compatibility with other
4547 The @code{weak} attribute is described in @ref{Function Attributes}.
4550 The @code{dllimport} attribute is described in @ref{Function Attributes}.
4553 The @code{dllexport} attribute is described in @ref{Function Attributes}.
4557 @subsection AVR Variable Attributes
4561 @cindex @code{progmem} AVR variable attribute
4562 The @code{progmem} attribute is used on the AVR to place data in the program
4563 memory address space (flash). This is accomplished by putting
4564 respective variables into a section whose name starts with @code{.progmem}.
4566 AVR is a Harvard architecture processor and data and reas only data
4567 normally resides in the data memory address space (RAM).
4570 @subsection Blackfin Variable Attributes
4572 Three attributes are currently defined for the Blackfin.
4578 @cindex @code{l1_data} variable attribute
4579 @cindex @code{l1_data_A} variable attribute
4580 @cindex @code{l1_data_B} variable attribute
4581 Use these attributes on the Blackfin to place the variable into L1 Data SRAM.
4582 Variables with @code{l1_data} attribute will be put into the specific section
4583 named @code{.l1.data}. Those with @code{l1_data_A} attribute will be put into
4584 the specific section named @code{.l1.data.A}. Those with @code{l1_data_B}
4585 attribute will be put into the specific section named @code{.l1.data.B}.
4588 @cindex @code{l2} variable attribute
4589 Use this attribute on the Blackfin to place the variable into L2 SRAM.
4590 Variables with @code{l2} attribute will be put into the specific section
4591 named @code{.l2.data}.
4594 @subsection M32R/D Variable Attributes
4596 One attribute is currently defined for the M32R/D@.
4599 @item model (@var{model-name})
4600 @cindex variable addressability on the M32R/D
4601 Use this attribute on the M32R/D to set the addressability of an object.
4602 The identifier @var{model-name} is one of @code{small}, @code{medium},
4603 or @code{large}, representing each of the code models.
4605 Small model objects live in the lower 16MB of memory (so that their
4606 addresses can be loaded with the @code{ld24} instruction).
4608 Medium and large model objects may live anywhere in the 32-bit address space
4609 (the compiler will generate @code{seth/add3} instructions to load their
4613 @anchor{MeP Variable Attributes}
4614 @subsection MeP Variable Attributes
4616 The MeP target has a number of addressing modes and busses. The
4617 @code{near} space spans the standard memory space's first 16 megabytes
4618 (24 bits). The @code{far} space spans the entire 32-bit memory space.
4619 The @code{based} space is a 128 byte region in the memory space which
4620 is addressed relative to the @code{$tp} register. The @code{tiny}
4621 space is a 65536 byte region relative to the @code{$gp} register. In
4622 addition to these memory regions, the MeP target has a separate 16-bit
4623 control bus which is specified with @code{cb} attributes.
4628 Any variable with the @code{based} attribute will be assigned to the
4629 @code{.based} section, and will be accessed with relative to the
4630 @code{$tp} register.
4633 Likewise, the @code{tiny} attribute assigned variables to the
4634 @code{.tiny} section, relative to the @code{$gp} register.
4637 Variables with the @code{near} attribute are assumed to have addresses
4638 that fit in a 24-bit addressing mode. This is the default for large
4639 variables (@code{-mtiny=4} is the default) but this attribute can
4640 override @code{-mtiny=} for small variables, or override @code{-ml}.
4643 Variables with the @code{far} attribute are addressed using a full
4644 32-bit address. Since this covers the entire memory space, this
4645 allows modules to make no assumptions about where variables might be
4649 @itemx io (@var{addr})
4650 Variables with the @code{io} attribute are used to address
4651 memory-mapped peripherals. If an address is specified, the variable
4652 is assigned that address, else it is not assigned an address (it is
4653 assumed some other module will assign an address). Example:
4656 int timer_count __attribute__((io(0x123)));
4660 @itemx cb (@var{addr})
4661 Variables with the @code{cb} attribute are used to access the control
4662 bus, using special instructions. @code{addr} indicates the control bus
4666 int cpu_clock __attribute__((cb(0x123)));
4671 @anchor{i386 Variable Attributes}
4672 @subsection i386 Variable Attributes
4674 Two attributes are currently defined for i386 configurations:
4675 @code{ms_struct} and @code{gcc_struct}
4680 @cindex @code{ms_struct} attribute
4681 @cindex @code{gcc_struct} attribute
4683 If @code{packed} is used on a structure, or if bit-fields are used
4684 it may be that the Microsoft ABI packs them differently
4685 than GCC would normally pack them. Particularly when moving packed
4686 data between functions compiled with GCC and the native Microsoft compiler
4687 (either via function call or as data in a file), it may be necessary to access
4690 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
4691 compilers to match the native Microsoft compiler.
4693 The Microsoft structure layout algorithm is fairly simple with the exception
4694 of the bitfield packing:
4696 The padding and alignment of members of structures and whether a bit field
4697 can straddle a storage-unit boundary
4700 @item Structure members are stored sequentially in the order in which they are
4701 declared: the first member has the lowest memory address and the last member
4704 @item Every data object has an alignment-requirement. The alignment-requirement
4705 for all data except structures, unions, and arrays is either the size of the
4706 object or the current packing size (specified with either the aligned attribute
4707 or the pack pragma), whichever is less. For structures, unions, and arrays,
4708 the alignment-requirement is the largest alignment-requirement of its members.
4709 Every object is allocated an offset so that:
4711 offset % alignment-requirement == 0
4713 @item Adjacent bit fields are packed into the same 1-, 2-, or 4-byte allocation
4714 unit if the integral types are the same size and if the next bit field fits
4715 into the current allocation unit without crossing the boundary imposed by the
4716 common alignment requirements of the bit fields.
4719 Handling of zero-length bitfields:
4721 MSVC interprets zero-length bitfields in the following ways:
4724 @item If a zero-length bitfield is inserted between two bitfields that would
4725 normally be coalesced, the bitfields will not be coalesced.
4732 unsigned long bf_1 : 12;
4734 unsigned long bf_2 : 12;
4738 The size of @code{t1} would be 8 bytes with the zero-length bitfield. If the
4739 zero-length bitfield were removed, @code{t1}'s size would be 4 bytes.
4741 @item If a zero-length bitfield is inserted after a bitfield, @code{foo}, and the
4742 alignment of the zero-length bitfield is greater than the member that follows it,
4743 @code{bar}, @code{bar} will be aligned as the type of the zero-length bitfield.
4763 For @code{t2}, @code{bar} will be placed at offset 2, rather than offset 1.
4764 Accordingly, the size of @code{t2} will be 4. For @code{t3}, the zero-length
4765 bitfield will not affect the alignment of @code{bar} or, as a result, the size
4768 Taking this into account, it is important to note the following:
4771 @item If a zero-length bitfield follows a normal bitfield, the type of the
4772 zero-length bitfield may affect the alignment of the structure as whole. For
4773 example, @code{t2} has a size of 4 bytes, since the zero-length bitfield follows a
4774 normal bitfield, and is of type short.
4776 @item Even if a zero-length bitfield is not followed by a normal bitfield, it may
4777 still affect the alignment of the structure:
4787 Here, @code{t4} will take up 4 bytes.
4790 @item Zero-length bitfields following non-bitfield members are ignored:
4801 Here, @code{t5} will take up 2 bytes.
4805 @subsection PowerPC Variable Attributes
4807 Three attributes currently are defined for PowerPC configurations:
4808 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
4810 For full documentation of the struct attributes please see the
4811 documentation in @ref{i386 Variable Attributes}.
4813 For documentation of @code{altivec} attribute please see the
4814 documentation in @ref{PowerPC Type Attributes}.
4816 @subsection SPU Variable Attributes
4818 The SPU supports the @code{spu_vector} attribute for variables. For
4819 documentation of this attribute please see the documentation in
4820 @ref{SPU Type Attributes}.
4822 @subsection Xstormy16 Variable Attributes
4824 One attribute is currently defined for xstormy16 configurations:
4829 @cindex @code{below100} attribute
4831 If a variable has the @code{below100} attribute (@code{BELOW100} is
4832 allowed also), GCC will place the variable in the first 0x100 bytes of
4833 memory and use special opcodes to access it. Such variables will be
4834 placed in either the @code{.bss_below100} section or the
4835 @code{.data_below100} section.
4839 @node Type Attributes
4840 @section Specifying Attributes of Types
4841 @cindex attribute of types
4842 @cindex type attributes
4844 The keyword @code{__attribute__} allows you to specify special
4845 attributes of @code{struct} and @code{union} types when you define
4846 such types. This keyword is followed by an attribute specification
4847 inside double parentheses. Seven attributes are currently defined for
4848 types: @code{aligned}, @code{packed}, @code{transparent_union},
4849 @code{unused}, @code{deprecated}, @code{visibility}, and
4850 @code{may_alias}. Other attributes are defined for functions
4851 (@pxref{Function Attributes}) and for variables (@pxref{Variable
4854 You may also specify any one of these attributes with @samp{__}
4855 preceding and following its keyword. This allows you to use these
4856 attributes in header files without being concerned about a possible
4857 macro of the same name. For example, you may use @code{__aligned__}
4858 instead of @code{aligned}.
4860 You may specify type attributes in an enum, struct or union type
4861 declaration or definition, or for other types in a @code{typedef}
4864 For an enum, struct or union type, you may specify attributes either
4865 between the enum, struct or union tag and the name of the type, or
4866 just past the closing curly brace of the @emph{definition}. The
4867 former syntax is preferred.
4869 @xref{Attribute Syntax}, for details of the exact syntax for using
4873 @cindex @code{aligned} attribute
4874 @item aligned (@var{alignment})
4875 This attribute specifies a minimum alignment (in bytes) for variables
4876 of the specified type. For example, the declarations:
4879 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
4880 typedef int more_aligned_int __attribute__ ((aligned (8)));
4884 force the compiler to insure (as far as it can) that each variable whose
4885 type is @code{struct S} or @code{more_aligned_int} will be allocated and
4886 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
4887 variables of type @code{struct S} aligned to 8-byte boundaries allows
4888 the compiler to use the @code{ldd} and @code{std} (doubleword load and
4889 store) instructions when copying one variable of type @code{struct S} to
4890 another, thus improving run-time efficiency.
4892 Note that the alignment of any given @code{struct} or @code{union} type
4893 is required by the ISO C standard to be at least a perfect multiple of
4894 the lowest common multiple of the alignments of all of the members of
4895 the @code{struct} or @code{union} in question. This means that you @emph{can}
4896 effectively adjust the alignment of a @code{struct} or @code{union}
4897 type by attaching an @code{aligned} attribute to any one of the members
4898 of such a type, but the notation illustrated in the example above is a
4899 more obvious, intuitive, and readable way to request the compiler to
4900 adjust the alignment of an entire @code{struct} or @code{union} type.
4902 As in the preceding example, you can explicitly specify the alignment
4903 (in bytes) that you wish the compiler to use for a given @code{struct}
4904 or @code{union} type. Alternatively, you can leave out the alignment factor
4905 and just ask the compiler to align a type to the maximum
4906 useful alignment for the target machine you are compiling for. For
4907 example, you could write:
4910 struct S @{ short f[3]; @} __attribute__ ((aligned));
4913 Whenever you leave out the alignment factor in an @code{aligned}
4914 attribute specification, the compiler automatically sets the alignment
4915 for the type to the largest alignment which is ever used for any data
4916 type on the target machine you are compiling for. Doing this can often
4917 make copy operations more efficient, because the compiler can use
4918 whatever instructions copy the biggest chunks of memory when performing
4919 copies to or from the variables which have types that you have aligned
4922 In the example above, if the size of each @code{short} is 2 bytes, then
4923 the size of the entire @code{struct S} type is 6 bytes. The smallest
4924 power of two which is greater than or equal to that is 8, so the
4925 compiler sets the alignment for the entire @code{struct S} type to 8
4928 Note that although you can ask the compiler to select a time-efficient
4929 alignment for a given type and then declare only individual stand-alone
4930 objects of that type, the compiler's ability to select a time-efficient
4931 alignment is primarily useful only when you plan to create arrays of
4932 variables having the relevant (efficiently aligned) type. If you
4933 declare or use arrays of variables of an efficiently-aligned type, then
4934 it is likely that your program will also be doing pointer arithmetic (or
4935 subscripting, which amounts to the same thing) on pointers to the
4936 relevant type, and the code that the compiler generates for these
4937 pointer arithmetic operations will often be more efficient for
4938 efficiently-aligned types than for other types.
4940 The @code{aligned} attribute can only increase the alignment; but you
4941 can decrease it by specifying @code{packed} as well. See below.
4943 Note that the effectiveness of @code{aligned} attributes may be limited
4944 by inherent limitations in your linker. On many systems, the linker is
4945 only able to arrange for variables to be aligned up to a certain maximum
4946 alignment. (For some linkers, the maximum supported alignment may
4947 be very very small.) If your linker is only able to align variables
4948 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
4949 in an @code{__attribute__} will still only provide you with 8 byte
4950 alignment. See your linker documentation for further information.
4953 This attribute, attached to @code{struct} or @code{union} type
4954 definition, specifies that each member (other than zero-width bitfields)
4955 of the structure or union is placed to minimize the memory required. When
4956 attached to an @code{enum} definition, it indicates that the smallest
4957 integral type should be used.
4959 @opindex fshort-enums
4960 Specifying this attribute for @code{struct} and @code{union} types is
4961 equivalent to specifying the @code{packed} attribute on each of the
4962 structure or union members. Specifying the @option{-fshort-enums}
4963 flag on the line is equivalent to specifying the @code{packed}
4964 attribute on all @code{enum} definitions.
4966 In the following example @code{struct my_packed_struct}'s members are
4967 packed closely together, but the internal layout of its @code{s} member
4968 is not packed---to do that, @code{struct my_unpacked_struct} would need to
4972 struct my_unpacked_struct
4978 struct __attribute__ ((__packed__)) my_packed_struct
4982 struct my_unpacked_struct s;
4986 You may only specify this attribute on the definition of an @code{enum},
4987 @code{struct} or @code{union}, not on a @code{typedef} which does not
4988 also define the enumerated type, structure or union.
4990 @item transparent_union
4991 This attribute, attached to a @code{union} type definition, indicates
4992 that any function parameter having that union type causes calls to that
4993 function to be treated in a special way.
4995 First, the argument corresponding to a transparent union type can be of
4996 any type in the union; no cast is required. Also, if the union contains
4997 a pointer type, the corresponding argument can be a null pointer
4998 constant or a void pointer expression; and if the union contains a void
4999 pointer type, the corresponding argument can be any pointer expression.
5000 If the union member type is a pointer, qualifiers like @code{const} on
5001 the referenced type must be respected, just as with normal pointer
5004 Second, the argument is passed to the function using the calling
5005 conventions of the first member of the transparent union, not the calling
5006 conventions of the union itself. All members of the union must have the
5007 same machine representation; this is necessary for this argument passing
5010 Transparent unions are designed for library functions that have multiple
5011 interfaces for compatibility reasons. For example, suppose the
5012 @code{wait} function must accept either a value of type @code{int *} to
5013 comply with Posix, or a value of type @code{union wait *} to comply with
5014 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
5015 @code{wait} would accept both kinds of arguments, but it would also
5016 accept any other pointer type and this would make argument type checking
5017 less useful. Instead, @code{<sys/wait.h>} might define the interface
5021 typedef union __attribute__ ((__transparent_union__))
5025 @} wait_status_ptr_t;
5027 pid_t wait (wait_status_ptr_t);
5030 This interface allows either @code{int *} or @code{union wait *}
5031 arguments to be passed, using the @code{int *} calling convention.
5032 The program can call @code{wait} with arguments of either type:
5035 int w1 () @{ int w; return wait (&w); @}
5036 int w2 () @{ union wait w; return wait (&w); @}
5039 With this interface, @code{wait}'s implementation might look like this:
5042 pid_t wait (wait_status_ptr_t p)
5044 return waitpid (-1, p.__ip, 0);
5049 When attached to a type (including a @code{union} or a @code{struct}),
5050 this attribute means that variables of that type are meant to appear
5051 possibly unused. GCC will not produce a warning for any variables of
5052 that type, even if the variable appears to do nothing. This is often
5053 the case with lock or thread classes, which are usually defined and then
5054 not referenced, but contain constructors and destructors that have
5055 nontrivial bookkeeping functions.
5058 @itemx deprecated (@var{msg})
5059 The @code{deprecated} attribute results in a warning if the type
5060 is used anywhere in the source file. This is useful when identifying
5061 types that are expected to be removed in a future version of a program.
5062 If possible, the warning also includes the location of the declaration
5063 of the deprecated type, to enable users to easily find further
5064 information about why the type is deprecated, or what they should do
5065 instead. Note that the warnings only occur for uses and then only
5066 if the type is being applied to an identifier that itself is not being
5067 declared as deprecated.
5070 typedef int T1 __attribute__ ((deprecated));
5074 typedef T1 T3 __attribute__ ((deprecated));
5075 T3 z __attribute__ ((deprecated));
5078 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
5079 warning is issued for line 4 because T2 is not explicitly
5080 deprecated. Line 5 has no warning because T3 is explicitly
5081 deprecated. Similarly for line 6. The optional msg
5082 argument, which must be a string, will be printed in the warning if
5085 The @code{deprecated} attribute can also be used for functions and
5086 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
5089 Accesses through pointers to types with this attribute are not subject
5090 to type-based alias analysis, but are instead assumed to be able to alias
5091 any other type of objects. In the context of 6.5/7 an lvalue expression
5092 dereferencing such a pointer is treated like having a character type.
5093 See @option{-fstrict-aliasing} for more information on aliasing issues.
5094 This extension exists to support some vector APIs, in which pointers to
5095 one vector type are permitted to alias pointers to a different vector type.
5097 Note that an object of a type with this attribute does not have any
5103 typedef short __attribute__((__may_alias__)) short_a;
5109 short_a *b = (short_a *) &a;
5113 if (a == 0x12345678)
5120 If you replaced @code{short_a} with @code{short} in the variable
5121 declaration, the above program would abort when compiled with
5122 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
5123 above in recent GCC versions.
5126 In C++, attribute visibility (@pxref{Function Attributes}) can also be
5127 applied to class, struct, union and enum types. Unlike other type
5128 attributes, the attribute must appear between the initial keyword and
5129 the name of the type; it cannot appear after the body of the type.
5131 Note that the type visibility is applied to vague linkage entities
5132 associated with the class (vtable, typeinfo node, etc.). In
5133 particular, if a class is thrown as an exception in one shared object
5134 and caught in another, the class must have default visibility.
5135 Otherwise the two shared objects will be unable to use the same
5136 typeinfo node and exception handling will break.
5140 @subsection ARM Type Attributes
5142 On those ARM targets that support @code{dllimport} (such as Symbian
5143 OS), you can use the @code{notshared} attribute to indicate that the
5144 virtual table and other similar data for a class should not be
5145 exported from a DLL@. For example:
5148 class __declspec(notshared) C @{
5150 __declspec(dllimport) C();
5154 __declspec(dllexport)
5158 In this code, @code{C::C} is exported from the current DLL, but the
5159 virtual table for @code{C} is not exported. (You can use
5160 @code{__attribute__} instead of @code{__declspec} if you prefer, but
5161 most Symbian OS code uses @code{__declspec}.)
5163 @anchor{MeP Type Attributes}
5164 @subsection MeP Type Attributes
5166 Many of the MeP variable attributes may be applied to types as well.
5167 Specifically, the @code{based}, @code{tiny}, @code{near}, and
5168 @code{far} attributes may be applied to either. The @code{io} and
5169 @code{cb} attributes may not be applied to types.
5171 @anchor{i386 Type Attributes}
5172 @subsection i386 Type Attributes
5174 Two attributes are currently defined for i386 configurations:
5175 @code{ms_struct} and @code{gcc_struct}.
5181 @cindex @code{ms_struct}
5182 @cindex @code{gcc_struct}
5184 If @code{packed} is used on a structure, or if bit-fields are used
5185 it may be that the Microsoft ABI packs them differently
5186 than GCC would normally pack them. Particularly when moving packed
5187 data between functions compiled with GCC and the native Microsoft compiler
5188 (either via function call or as data in a file), it may be necessary to access
5191 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
5192 compilers to match the native Microsoft compiler.
5195 To specify multiple attributes, separate them by commas within the
5196 double parentheses: for example, @samp{__attribute__ ((aligned (16),
5199 @anchor{PowerPC Type Attributes}
5200 @subsection PowerPC Type Attributes
5202 Three attributes currently are defined for PowerPC configurations:
5203 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
5205 For full documentation of the @code{ms_struct} and @code{gcc_struct}
5206 attributes please see the documentation in @ref{i386 Type Attributes}.
5208 The @code{altivec} attribute allows one to declare AltiVec vector data
5209 types supported by the AltiVec Programming Interface Manual. The
5210 attribute requires an argument to specify one of three vector types:
5211 @code{vector__}, @code{pixel__} (always followed by unsigned short),
5212 and @code{bool__} (always followed by unsigned).
5215 __attribute__((altivec(vector__)))
5216 __attribute__((altivec(pixel__))) unsigned short
5217 __attribute__((altivec(bool__))) unsigned
5220 These attributes mainly are intended to support the @code{__vector},
5221 @code{__pixel}, and @code{__bool} AltiVec keywords.
5223 @anchor{SPU Type Attributes}
5224 @subsection SPU Type Attributes
5226 The SPU supports the @code{spu_vector} attribute for types. This attribute
5227 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
5228 Language Extensions Specification. It is intended to support the
5229 @code{__vector} keyword.
5232 @section Inquiring on Alignment of Types or Variables
5234 @cindex type alignment
5235 @cindex variable alignment
5237 The keyword @code{__alignof__} allows you to inquire about how an object
5238 is aligned, or the minimum alignment usually required by a type. Its
5239 syntax is just like @code{sizeof}.
5241 For example, if the target machine requires a @code{double} value to be
5242 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
5243 This is true on many RISC machines. On more traditional machine
5244 designs, @code{__alignof__ (double)} is 4 or even 2.
5246 Some machines never actually require alignment; they allow reference to any
5247 data type even at an odd address. For these machines, @code{__alignof__}
5248 reports the smallest alignment that GCC will give the data type, usually as
5249 mandated by the target ABI.
5251 If the operand of @code{__alignof__} is an lvalue rather than a type,
5252 its value is the required alignment for its type, taking into account
5253 any minimum alignment specified with GCC's @code{__attribute__}
5254 extension (@pxref{Variable Attributes}). For example, after this
5258 struct foo @{ int x; char y; @} foo1;
5262 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
5263 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
5265 It is an error to ask for the alignment of an incomplete type.
5269 @section An Inline Function is As Fast As a Macro
5270 @cindex inline functions
5271 @cindex integrating function code
5273 @cindex macros, inline alternative
5275 By declaring a function inline, you can direct GCC to make
5276 calls to that function faster. One way GCC can achieve this is to
5277 integrate that function's code into the code for its callers. This
5278 makes execution faster by eliminating the function-call overhead; in
5279 addition, if any of the actual argument values are constant, their
5280 known values may permit simplifications at compile time so that not
5281 all of the inline function's code needs to be included. The effect on
5282 code size is less predictable; object code may be larger or smaller
5283 with function inlining, depending on the particular case. You can
5284 also direct GCC to try to integrate all ``simple enough'' functions
5285 into their callers with the option @option{-finline-functions}.
5287 GCC implements three different semantics of declaring a function
5288 inline. One is available with @option{-std=gnu89} or
5289 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
5290 on all inline declarations, another when
5291 @option{-std=c99}, @option{-std=c1x},
5292 @option{-std=gnu99} or @option{-std=gnu1x}
5293 (without @option{-fgnu89-inline}), and the third
5294 is used when compiling C++.
5296 To declare a function inline, use the @code{inline} keyword in its
5297 declaration, like this:
5307 If you are writing a header file to be included in ISO C90 programs, write
5308 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
5310 The three types of inlining behave similarly in two important cases:
5311 when the @code{inline} keyword is used on a @code{static} function,
5312 like the example above, and when a function is first declared without
5313 using the @code{inline} keyword and then is defined with
5314 @code{inline}, like this:
5317 extern int inc (int *a);
5325 In both of these common cases, the program behaves the same as if you
5326 had not used the @code{inline} keyword, except for its speed.
5328 @cindex inline functions, omission of
5329 @opindex fkeep-inline-functions
5330 When a function is both inline and @code{static}, if all calls to the
5331 function are integrated into the caller, and the function's address is
5332 never used, then the function's own assembler code is never referenced.
5333 In this case, GCC does not actually output assembler code for the
5334 function, unless you specify the option @option{-fkeep-inline-functions}.
5335 Some calls cannot be integrated for various reasons (in particular,
5336 calls that precede the function's definition cannot be integrated, and
5337 neither can recursive calls within the definition). If there is a
5338 nonintegrated call, then the function is compiled to assembler code as
5339 usual. The function must also be compiled as usual if the program
5340 refers to its address, because that can't be inlined.
5343 Note that certain usages in a function definition can make it unsuitable
5344 for inline substitution. Among these usages are: use of varargs, use of
5345 alloca, use of variable sized data types (@pxref{Variable Length}),
5346 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
5347 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
5348 will warn when a function marked @code{inline} could not be substituted,
5349 and will give the reason for the failure.
5351 @cindex automatic @code{inline} for C++ member fns
5352 @cindex @code{inline} automatic for C++ member fns
5353 @cindex member fns, automatically @code{inline}
5354 @cindex C++ member fns, automatically @code{inline}
5355 @opindex fno-default-inline
5356 As required by ISO C++, GCC considers member functions defined within
5357 the body of a class to be marked inline even if they are
5358 not explicitly declared with the @code{inline} keyword. You can
5359 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
5360 Options,,Options Controlling C++ Dialect}.
5362 GCC does not inline any functions when not optimizing unless you specify
5363 the @samp{always_inline} attribute for the function, like this:
5366 /* @r{Prototype.} */
5367 inline void foo (const char) __attribute__((always_inline));
5370 The remainder of this section is specific to GNU C90 inlining.
5372 @cindex non-static inline function
5373 When an inline function is not @code{static}, then the compiler must assume
5374 that there may be calls from other source files; since a global symbol can
5375 be defined only once in any program, the function must not be defined in
5376 the other source files, so the calls therein cannot be integrated.
5377 Therefore, a non-@code{static} inline function is always compiled on its
5378 own in the usual fashion.
5380 If you specify both @code{inline} and @code{extern} in the function
5381 definition, then the definition is used only for inlining. In no case
5382 is the function compiled on its own, not even if you refer to its
5383 address explicitly. Such an address becomes an external reference, as
5384 if you had only declared the function, and had not defined it.
5386 This combination of @code{inline} and @code{extern} has almost the
5387 effect of a macro. The way to use it is to put a function definition in
5388 a header file with these keywords, and put another copy of the
5389 definition (lacking @code{inline} and @code{extern}) in a library file.
5390 The definition in the header file will cause most calls to the function
5391 to be inlined. If any uses of the function remain, they will refer to
5392 the single copy in the library.
5395 @section When is a Volatile Object Accessed?
5396 @cindex accessing volatiles
5397 @cindex volatile read
5398 @cindex volatile write
5399 @cindex volatile access
5401 C has the concept of volatile objects. These are normally accessed by
5402 pointers and used for accessing hardware or inter-thread
5403 communication. The standard encourages compilers to refrain from
5404 optimizations concerning accesses to volatile objects, but leaves it
5405 implementation defined as to what constitutes a volatile access. The
5406 minimum requirement is that at a sequence point all previous accesses
5407 to volatile objects have stabilized and no subsequent accesses have
5408 occurred. Thus an implementation is free to reorder and combine
5409 volatile accesses which occur between sequence points, but cannot do
5410 so for accesses across a sequence point. The use of volatile does
5411 not allow you to violate the restriction on updating objects multiple
5412 times between two sequence points.
5414 Accesses to non-volatile objects are not ordered with respect to
5415 volatile accesses. You cannot use a volatile object as a memory
5416 barrier to order a sequence of writes to non-volatile memory. For
5420 int *ptr = @var{something};
5422 *ptr = @var{something};
5426 Unless @var{*ptr} and @var{vobj} can be aliased, it is not guaranteed
5427 that the write to @var{*ptr} will have occurred by the time the update
5428 of @var{vobj} has happened. If you need this guarantee, you must use
5429 a stronger memory barrier such as:
5432 int *ptr = @var{something};
5434 *ptr = @var{something};
5435 asm volatile ("" : : : "memory");
5439 A scalar volatile object is read when it is accessed in a void context:
5442 volatile int *src = @var{somevalue};
5446 Such expressions are rvalues, and GCC implements this as a
5447 read of the volatile object being pointed to.
5449 Assignments are also expressions and have an rvalue. However when
5450 assigning to a scalar volatile, the volatile object is not reread,
5451 regardless of whether the assignment expression's rvalue is used or
5452 not. If the assignment's rvalue is used, the value is that assigned
5453 to the volatile object. For instance, there is no read of @var{vobj}
5454 in all the following cases:
5459 vobj = @var{something};
5460 obj = vobj = @var{something};
5461 obj ? vobj = @var{onething} : vobj = @var{anotherthing};
5462 obj = (@var{something}, vobj = @var{anotherthing});
5465 If you need to read the volatile object after an assignment has
5466 occurred, you must use a separate expression with an intervening
5469 As bitfields are not individually addressable, volatile bitfields may
5470 be implicitly read when written to, or when adjacent bitfields are
5471 accessed. Bitfield operations may be optimized such that adjacent
5472 bitfields are only partially accessed, if they straddle a storage unit
5473 boundary. For these reasons it is unwise to use volatile bitfields to
5477 @section Assembler Instructions with C Expression Operands
5478 @cindex extended @code{asm}
5479 @cindex @code{asm} expressions
5480 @cindex assembler instructions
5483 In an assembler instruction using @code{asm}, you can specify the
5484 operands of the instruction using C expressions. This means you need not
5485 guess which registers or memory locations will contain the data you want
5488 You must specify an assembler instruction template much like what
5489 appears in a machine description, plus an operand constraint string for
5492 For example, here is how to use the 68881's @code{fsinx} instruction:
5495 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
5499 Here @code{angle} is the C expression for the input operand while
5500 @code{result} is that of the output operand. Each has @samp{"f"} as its
5501 operand constraint, saying that a floating point register is required.
5502 The @samp{=} in @samp{=f} indicates that the operand is an output; all
5503 output operands' constraints must use @samp{=}. The constraints use the
5504 same language used in the machine description (@pxref{Constraints}).
5506 Each operand is described by an operand-constraint string followed by
5507 the C expression in parentheses. A colon separates the assembler
5508 template from the first output operand and another separates the last
5509 output operand from the first input, if any. Commas separate the
5510 operands within each group. The total number of operands is currently
5511 limited to 30; this limitation may be lifted in some future version of
5514 If there are no output operands but there are input operands, you must
5515 place two consecutive colons surrounding the place where the output
5518 As of GCC version 3.1, it is also possible to specify input and output
5519 operands using symbolic names which can be referenced within the
5520 assembler code. These names are specified inside square brackets
5521 preceding the constraint string, and can be referenced inside the
5522 assembler code using @code{%[@var{name}]} instead of a percentage sign
5523 followed by the operand number. Using named operands the above example
5527 asm ("fsinx %[angle],%[output]"
5528 : [output] "=f" (result)
5529 : [angle] "f" (angle));
5533 Note that the symbolic operand names have no relation whatsoever to
5534 other C identifiers. You may use any name you like, even those of
5535 existing C symbols, but you must ensure that no two operands within the same
5536 assembler construct use the same symbolic name.
5538 Output operand expressions must be lvalues; the compiler can check this.
5539 The input operands need not be lvalues. The compiler cannot check
5540 whether the operands have data types that are reasonable for the
5541 instruction being executed. It does not parse the assembler instruction
5542 template and does not know what it means or even whether it is valid
5543 assembler input. The extended @code{asm} feature is most often used for
5544 machine instructions the compiler itself does not know exist. If
5545 the output expression cannot be directly addressed (for example, it is a
5546 bit-field), your constraint must allow a register. In that case, GCC
5547 will use the register as the output of the @code{asm}, and then store
5548 that register into the output.
5550 The ordinary output operands must be write-only; GCC will assume that
5551 the values in these operands before the instruction are dead and need
5552 not be generated. Extended asm supports input-output or read-write
5553 operands. Use the constraint character @samp{+} to indicate such an
5554 operand and list it with the output operands. You should only use
5555 read-write operands when the constraints for the operand (or the
5556 operand in which only some of the bits are to be changed) allow a
5559 You may, as an alternative, logically split its function into two
5560 separate operands, one input operand and one write-only output
5561 operand. The connection between them is expressed by constraints
5562 which say they need to be in the same location when the instruction
5563 executes. You can use the same C expression for both operands, or
5564 different expressions. For example, here we write the (fictitious)
5565 @samp{combine} instruction with @code{bar} as its read-only source
5566 operand and @code{foo} as its read-write destination:
5569 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
5573 The constraint @samp{"0"} for operand 1 says that it must occupy the
5574 same location as operand 0. A number in constraint is allowed only in
5575 an input operand and it must refer to an output operand.
5577 Only a number in the constraint can guarantee that one operand will be in
5578 the same place as another. The mere fact that @code{foo} is the value
5579 of both operands is not enough to guarantee that they will be in the
5580 same place in the generated assembler code. The following would not
5584 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
5587 Various optimizations or reloading could cause operands 0 and 1 to be in
5588 different registers; GCC knows no reason not to do so. For example, the
5589 compiler might find a copy of the value of @code{foo} in one register and
5590 use it for operand 1, but generate the output operand 0 in a different
5591 register (copying it afterward to @code{foo}'s own address). Of course,
5592 since the register for operand 1 is not even mentioned in the assembler
5593 code, the result will not work, but GCC can't tell that.
5595 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
5596 the operand number for a matching constraint. For example:
5599 asm ("cmoveq %1,%2,%[result]"
5600 : [result] "=r"(result)
5601 : "r" (test), "r"(new), "[result]"(old));
5604 Sometimes you need to make an @code{asm} operand be a specific register,
5605 but there's no matching constraint letter for that register @emph{by
5606 itself}. To force the operand into that register, use a local variable
5607 for the operand and specify the register in the variable declaration.
5608 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
5609 register constraint letter that matches the register:
5612 register int *p1 asm ("r0") = @dots{};
5613 register int *p2 asm ("r1") = @dots{};
5614 register int *result asm ("r0");
5615 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
5618 @anchor{Example of asm with clobbered asm reg}
5619 In the above example, beware that a register that is call-clobbered by
5620 the target ABI will be overwritten by any function call in the
5621 assignment, including library calls for arithmetic operators.
5622 Also a register may be clobbered when generating some operations,
5623 like variable shift, memory copy or memory move on x86.
5624 Assuming it is a call-clobbered register, this may happen to @code{r0}
5625 above by the assignment to @code{p2}. If you have to use such a
5626 register, use temporary variables for expressions between the register
5631 register int *p1 asm ("r0") = @dots{};
5632 register int *p2 asm ("r1") = t1;
5633 register int *result asm ("r0");
5634 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
5637 Some instructions clobber specific hard registers. To describe this,
5638 write a third colon after the input operands, followed by the names of
5639 the clobbered hard registers (given as strings). Here is a realistic
5640 example for the VAX:
5643 asm volatile ("movc3 %0,%1,%2"
5644 : /* @r{no outputs} */
5645 : "g" (from), "g" (to), "g" (count)
5646 : "r0", "r1", "r2", "r3", "r4", "r5");
5649 You may not write a clobber description in a way that overlaps with an
5650 input or output operand. For example, you may not have an operand
5651 describing a register class with one member if you mention that register
5652 in the clobber list. Variables declared to live in specific registers
5653 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
5654 have no part mentioned in the clobber description.
5655 There is no way for you to specify that an input
5656 operand is modified without also specifying it as an output
5657 operand. Note that if all the output operands you specify are for this
5658 purpose (and hence unused), you will then also need to specify
5659 @code{volatile} for the @code{asm} construct, as described below, to
5660 prevent GCC from deleting the @code{asm} statement as unused.
5662 If you refer to a particular hardware register from the assembler code,
5663 you will probably have to list the register after the third colon to
5664 tell the compiler the register's value is modified. In some assemblers,
5665 the register names begin with @samp{%}; to produce one @samp{%} in the
5666 assembler code, you must write @samp{%%} in the input.
5668 If your assembler instruction can alter the condition code register, add
5669 @samp{cc} to the list of clobbered registers. GCC on some machines
5670 represents the condition codes as a specific hardware register;
5671 @samp{cc} serves to name this register. On other machines, the
5672 condition code is handled differently, and specifying @samp{cc} has no
5673 effect. But it is valid no matter what the machine.
5675 If your assembler instructions access memory in an unpredictable
5676 fashion, add @samp{memory} to the list of clobbered registers. This
5677 will cause GCC to not keep memory values cached in registers across the
5678 assembler instruction and not optimize stores or loads to that memory.
5679 You will also want to add the @code{volatile} keyword if the memory
5680 affected is not listed in the inputs or outputs of the @code{asm}, as
5681 the @samp{memory} clobber does not count as a side-effect of the
5682 @code{asm}. If you know how large the accessed memory is, you can add
5683 it as input or output but if this is not known, you should add
5684 @samp{memory}. As an example, if you access ten bytes of a string, you
5685 can use a memory input like:
5688 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
5691 Note that in the following example the memory input is necessary,
5692 otherwise GCC might optimize the store to @code{x} away:
5699 asm ("magic stuff accessing an 'int' pointed to by '%1'"
5700 "=&d" (r) : "a" (y), "m" (*y));
5705 You can put multiple assembler instructions together in a single
5706 @code{asm} template, separated by the characters normally used in assembly
5707 code for the system. A combination that works in most places is a newline
5708 to break the line, plus a tab character to move to the instruction field
5709 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
5710 assembler allows semicolons as a line-breaking character. Note that some
5711 assembler dialects use semicolons to start a comment.
5712 The input operands are guaranteed not to use any of the clobbered
5713 registers, and neither will the output operands' addresses, so you can
5714 read and write the clobbered registers as many times as you like. Here
5715 is an example of multiple instructions in a template; it assumes the
5716 subroutine @code{_foo} accepts arguments in registers 9 and 10:
5719 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
5721 : "g" (from), "g" (to)
5725 Unless an output operand has the @samp{&} constraint modifier, GCC
5726 may allocate it in the same register as an unrelated input operand, on
5727 the assumption the inputs are consumed before the outputs are produced.
5728 This assumption may be false if the assembler code actually consists of
5729 more than one instruction. In such a case, use @samp{&} for each output
5730 operand that may not overlap an input. @xref{Modifiers}.
5732 If you want to test the condition code produced by an assembler
5733 instruction, you must include a branch and a label in the @code{asm}
5734 construct, as follows:
5737 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
5743 This assumes your assembler supports local labels, as the GNU assembler
5744 and most Unix assemblers do.
5746 Speaking of labels, jumps from one @code{asm} to another are not
5747 supported. The compiler's optimizers do not know about these jumps, and
5748 therefore they cannot take account of them when deciding how to
5749 optimize. @xref{Extended asm with goto}.
5751 @cindex macros containing @code{asm}
5752 Usually the most convenient way to use these @code{asm} instructions is to
5753 encapsulate them in macros that look like functions. For example,
5757 (@{ double __value, __arg = (x); \
5758 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
5763 Here the variable @code{__arg} is used to make sure that the instruction
5764 operates on a proper @code{double} value, and to accept only those
5765 arguments @code{x} which can convert automatically to a @code{double}.
5767 Another way to make sure the instruction operates on the correct data
5768 type is to use a cast in the @code{asm}. This is different from using a
5769 variable @code{__arg} in that it converts more different types. For
5770 example, if the desired type were @code{int}, casting the argument to
5771 @code{int} would accept a pointer with no complaint, while assigning the
5772 argument to an @code{int} variable named @code{__arg} would warn about
5773 using a pointer unless the caller explicitly casts it.
5775 If an @code{asm} has output operands, GCC assumes for optimization
5776 purposes the instruction has no side effects except to change the output
5777 operands. This does not mean instructions with a side effect cannot be
5778 used, but you must be careful, because the compiler may eliminate them
5779 if the output operands aren't used, or move them out of loops, or
5780 replace two with one if they constitute a common subexpression. Also,
5781 if your instruction does have a side effect on a variable that otherwise
5782 appears not to change, the old value of the variable may be reused later
5783 if it happens to be found in a register.
5785 You can prevent an @code{asm} instruction from being deleted
5786 by writing the keyword @code{volatile} after
5787 the @code{asm}. For example:
5790 #define get_and_set_priority(new) \
5792 asm volatile ("get_and_set_priority %0, %1" \
5793 : "=g" (__old) : "g" (new)); \
5798 The @code{volatile} keyword indicates that the instruction has
5799 important side-effects. GCC will not delete a volatile @code{asm} if
5800 it is reachable. (The instruction can still be deleted if GCC can
5801 prove that control-flow will never reach the location of the
5802 instruction.) Note that even a volatile @code{asm} instruction
5803 can be moved relative to other code, including across jump
5804 instructions. For example, on many targets there is a system
5805 register which can be set to control the rounding mode of
5806 floating point operations. You might try
5807 setting it with a volatile @code{asm}, like this PowerPC example:
5810 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
5815 This will not work reliably, as the compiler may move the addition back
5816 before the volatile @code{asm}. To make it work you need to add an
5817 artificial dependency to the @code{asm} referencing a variable in the code
5818 you don't want moved, for example:
5821 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
5825 Similarly, you can't expect a
5826 sequence of volatile @code{asm} instructions to remain perfectly
5827 consecutive. If you want consecutive output, use a single @code{asm}.
5828 Also, GCC will perform some optimizations across a volatile @code{asm}
5829 instruction; GCC does not ``forget everything'' when it encounters
5830 a volatile @code{asm} instruction the way some other compilers do.
5832 An @code{asm} instruction without any output operands will be treated
5833 identically to a volatile @code{asm} instruction.
5835 It is a natural idea to look for a way to give access to the condition
5836 code left by the assembler instruction. However, when we attempted to
5837 implement this, we found no way to make it work reliably. The problem
5838 is that output operands might need reloading, which would result in
5839 additional following ``store'' instructions. On most machines, these
5840 instructions would alter the condition code before there was time to
5841 test it. This problem doesn't arise for ordinary ``test'' and
5842 ``compare'' instructions because they don't have any output operands.
5844 For reasons similar to those described above, it is not possible to give
5845 an assembler instruction access to the condition code left by previous
5848 @anchor{Extended asm with goto}
5849 As of GCC version 4.5, @code{asm goto} may be used to have the assembly
5850 jump to one or more C labels. In this form, a fifth section after the
5851 clobber list contains a list of all C labels to which the assembly may jump.
5852 Each label operand is implicitly self-named. The @code{asm} is also assumed
5853 to fall through to the next statement.
5855 This form of @code{asm} is restricted to not have outputs. This is due
5856 to a internal restriction in the compiler that control transfer instructions
5857 cannot have outputs. This restriction on @code{asm goto} may be lifted
5858 in some future version of the compiler. In the mean time, @code{asm goto}
5859 may include a memory clobber, and so leave outputs in memory.
5865 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
5866 : : "r"(x), "r"(&y) : "r5", "memory" : error);
5873 In this (inefficient) example, the @code{frob} instruction sets the
5874 carry bit to indicate an error. The @code{jc} instruction detects
5875 this and branches to the @code{error} label. Finally, the output
5876 of the @code{frob} instruction (@code{%r5}) is stored into the memory
5877 for variable @code{y}, which is later read by the @code{return} statement.
5883 asm goto ("mfsr %%r1, 123; jmp %%r1;"
5884 ".pushsection doit_table;"
5885 ".long %l0, %l1, %l2, %l3;"
5887 : : : "r1" : label1, label2, label3, label4);
5888 __builtin_unreachable ();
5903 In this (also inefficient) example, the @code{mfsr} instruction reads
5904 an address from some out-of-band machine register, and the following
5905 @code{jmp} instruction branches to that address. The address read by
5906 the @code{mfsr} instruction is assumed to have been previously set via
5907 some application-specific mechanism to be one of the four values stored
5908 in the @code{doit_table} section. Finally, the @code{asm} is followed
5909 by a call to @code{__builtin_unreachable} to indicate that the @code{asm}
5910 does not in fact fall through.
5913 #define TRACE1(NUM) \
5915 asm goto ("0: nop;" \
5916 ".pushsection trace_table;" \
5919 : : : : trace#NUM); \
5920 if (0) @{ trace#NUM: trace(); @} \
5922 #define TRACE TRACE1(__COUNTER__)
5925 In this example (which in fact inspired the @code{asm goto} feature)
5926 we want on rare occasions to call the @code{trace} function; on other
5927 occasions we'd like to keep the overhead to the absolute minimum.
5928 The normal code path consists of a single @code{nop} instruction.
5929 However, we record the address of this @code{nop} together with the
5930 address of a label that calls the @code{trace} function. This allows
5931 the @code{nop} instruction to be patched at runtime to be an
5932 unconditional branch to the stored label. It is assumed that an
5933 optimizing compiler will move the labeled block out of line, to
5934 optimize the fall through path from the @code{asm}.
5936 If you are writing a header file that should be includable in ISO C
5937 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
5940 @subsection Size of an @code{asm}
5942 Some targets require that GCC track the size of each instruction used in
5943 order to generate correct code. Because the final length of an
5944 @code{asm} is only known by the assembler, GCC must make an estimate as
5945 to how big it will be. The estimate is formed by counting the number of
5946 statements in the pattern of the @code{asm} and multiplying that by the
5947 length of the longest instruction on that processor. Statements in the
5948 @code{asm} are identified by newline characters and whatever statement
5949 separator characters are supported by the assembler; on most processors
5950 this is the `@code{;}' character.
5952 Normally, GCC's estimate is perfectly adequate to ensure that correct
5953 code is generated, but it is possible to confuse the compiler if you use
5954 pseudo instructions or assembler macros that expand into multiple real
5955 instructions or if you use assembler directives that expand to more
5956 space in the object file than would be needed for a single instruction.
5957 If this happens then the assembler will produce a diagnostic saying that
5958 a label is unreachable.
5960 @subsection i386 floating point asm operands
5962 There are several rules on the usage of stack-like regs in
5963 asm_operands insns. These rules apply only to the operands that are
5968 Given a set of input regs that die in an asm_operands, it is
5969 necessary to know which are implicitly popped by the asm, and
5970 which must be explicitly popped by gcc.
5972 An input reg that is implicitly popped by the asm must be
5973 explicitly clobbered, unless it is constrained to match an
5977 For any input reg that is implicitly popped by an asm, it is
5978 necessary to know how to adjust the stack to compensate for the pop.
5979 If any non-popped input is closer to the top of the reg-stack than
5980 the implicitly popped reg, it would not be possible to know what the
5981 stack looked like---it's not clear how the rest of the stack ``slides
5984 All implicitly popped input regs must be closer to the top of
5985 the reg-stack than any input that is not implicitly popped.
5987 It is possible that if an input dies in an insn, reload might
5988 use the input reg for an output reload. Consider this example:
5991 asm ("foo" : "=t" (a) : "f" (b));
5994 This asm says that input B is not popped by the asm, and that
5995 the asm pushes a result onto the reg-stack, i.e., the stack is one
5996 deeper after the asm than it was before. But, it is possible that
5997 reload will think that it can use the same reg for both the input and
5998 the output, if input B dies in this insn.
6000 If any input operand uses the @code{f} constraint, all output reg
6001 constraints must use the @code{&} earlyclobber.
6003 The asm above would be written as
6006 asm ("foo" : "=&t" (a) : "f" (b));
6010 Some operands need to be in particular places on the stack. All
6011 output operands fall in this category---there is no other way to
6012 know which regs the outputs appear in unless the user indicates
6013 this in the constraints.
6015 Output operands must specifically indicate which reg an output
6016 appears in after an asm. @code{=f} is not allowed: the operand
6017 constraints must select a class with a single reg.
6020 Output operands may not be ``inserted'' between existing stack regs.
6021 Since no 387 opcode uses a read/write operand, all output operands
6022 are dead before the asm_operands, and are pushed by the asm_operands.
6023 It makes no sense to push anywhere but the top of the reg-stack.
6025 Output operands must start at the top of the reg-stack: output
6026 operands may not ``skip'' a reg.
6029 Some asm statements may need extra stack space for internal
6030 calculations. This can be guaranteed by clobbering stack registers
6031 unrelated to the inputs and outputs.
6035 Here are a couple of reasonable asms to want to write. This asm
6036 takes one input, which is internally popped, and produces two outputs.
6039 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
6042 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
6043 and replaces them with one output. The user must code the @code{st(1)}
6044 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
6047 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
6053 @section Controlling Names Used in Assembler Code
6054 @cindex assembler names for identifiers
6055 @cindex names used in assembler code
6056 @cindex identifiers, names in assembler code
6058 You can specify the name to be used in the assembler code for a C
6059 function or variable by writing the @code{asm} (or @code{__asm__})
6060 keyword after the declarator as follows:
6063 int foo asm ("myfoo") = 2;
6067 This specifies that the name to be used for the variable @code{foo} in
6068 the assembler code should be @samp{myfoo} rather than the usual
6071 On systems where an underscore is normally prepended to the name of a C
6072 function or variable, this feature allows you to define names for the
6073 linker that do not start with an underscore.
6075 It does not make sense to use this feature with a non-static local
6076 variable since such variables do not have assembler names. If you are
6077 trying to put the variable in a particular register, see @ref{Explicit
6078 Reg Vars}. GCC presently accepts such code with a warning, but will
6079 probably be changed to issue an error, rather than a warning, in the
6082 You cannot use @code{asm} in this way in a function @emph{definition}; but
6083 you can get the same effect by writing a declaration for the function
6084 before its definition and putting @code{asm} there, like this:
6087 extern func () asm ("FUNC");
6094 It is up to you to make sure that the assembler names you choose do not
6095 conflict with any other assembler symbols. Also, you must not use a
6096 register name; that would produce completely invalid assembler code. GCC
6097 does not as yet have the ability to store static variables in registers.
6098 Perhaps that will be added.
6100 @node Explicit Reg Vars
6101 @section Variables in Specified Registers
6102 @cindex explicit register variables
6103 @cindex variables in specified registers
6104 @cindex specified registers
6105 @cindex registers, global allocation
6107 GNU C allows you to put a few global variables into specified hardware
6108 registers. You can also specify the register in which an ordinary
6109 register variable should be allocated.
6113 Global register variables reserve registers throughout the program.
6114 This may be useful in programs such as programming language
6115 interpreters which have a couple of global variables that are accessed
6119 Local register variables in specific registers do not reserve the
6120 registers, except at the point where they are used as input or output
6121 operands in an @code{asm} statement and the @code{asm} statement itself is
6122 not deleted. The compiler's data flow analysis is capable of determining
6123 where the specified registers contain live values, and where they are
6124 available for other uses. Stores into local register variables may be deleted
6125 when they appear to be dead according to dataflow analysis. References
6126 to local register variables may be deleted or moved or simplified.
6128 These local variables are sometimes convenient for use with the extended
6129 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
6130 output of the assembler instruction directly into a particular register.
6131 (This will work provided the register you specify fits the constraints
6132 specified for that operand in the @code{asm}.)
6140 @node Global Reg Vars
6141 @subsection Defining Global Register Variables
6142 @cindex global register variables
6143 @cindex registers, global variables in
6145 You can define a global register variable in GNU C like this:
6148 register int *foo asm ("a5");
6152 Here @code{a5} is the name of the register which should be used. Choose a
6153 register which is normally saved and restored by function calls on your
6154 machine, so that library routines will not clobber it.
6156 Naturally the register name is cpu-dependent, so you would need to
6157 conditionalize your program according to cpu type. The register
6158 @code{a5} would be a good choice on a 68000 for a variable of pointer
6159 type. On machines with register windows, be sure to choose a ``global''
6160 register that is not affected magically by the function call mechanism.
6162 In addition, operating systems on one type of cpu may differ in how they
6163 name the registers; then you would need additional conditionals. For
6164 example, some 68000 operating systems call this register @code{%a5}.
6166 Eventually there may be a way of asking the compiler to choose a register
6167 automatically, but first we need to figure out how it should choose and
6168 how to enable you to guide the choice. No solution is evident.
6170 Defining a global register variable in a certain register reserves that
6171 register entirely for this use, at least within the current compilation.
6172 The register will not be allocated for any other purpose in the functions
6173 in the current compilation. The register will not be saved and restored by
6174 these functions. Stores into this register are never deleted even if they
6175 would appear to be dead, but references may be deleted or moved or
6178 It is not safe to access the global register variables from signal
6179 handlers, or from more than one thread of control, because the system
6180 library routines may temporarily use the register for other things (unless
6181 you recompile them specially for the task at hand).
6183 @cindex @code{qsort}, and global register variables
6184 It is not safe for one function that uses a global register variable to
6185 call another such function @code{foo} by way of a third function
6186 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
6187 different source file in which the variable wasn't declared). This is
6188 because @code{lose} might save the register and put some other value there.
6189 For example, you can't expect a global register variable to be available in
6190 the comparison-function that you pass to @code{qsort}, since @code{qsort}
6191 might have put something else in that register. (If you are prepared to
6192 recompile @code{qsort} with the same global register variable, you can
6193 solve this problem.)
6195 If you want to recompile @code{qsort} or other source files which do not
6196 actually use your global register variable, so that they will not use that
6197 register for any other purpose, then it suffices to specify the compiler
6198 option @option{-ffixed-@var{reg}}. You need not actually add a global
6199 register declaration to their source code.
6201 A function which can alter the value of a global register variable cannot
6202 safely be called from a function compiled without this variable, because it
6203 could clobber the value the caller expects to find there on return.
6204 Therefore, the function which is the entry point into the part of the
6205 program that uses the global register variable must explicitly save and
6206 restore the value which belongs to its caller.
6208 @cindex register variable after @code{longjmp}
6209 @cindex global register after @code{longjmp}
6210 @cindex value after @code{longjmp}
6213 On most machines, @code{longjmp} will restore to each global register
6214 variable the value it had at the time of the @code{setjmp}. On some
6215 machines, however, @code{longjmp} will not change the value of global
6216 register variables. To be portable, the function that called @code{setjmp}
6217 should make other arrangements to save the values of the global register
6218 variables, and to restore them in a @code{longjmp}. This way, the same
6219 thing will happen regardless of what @code{longjmp} does.
6221 All global register variable declarations must precede all function
6222 definitions. If such a declaration could appear after function
6223 definitions, the declaration would be too late to prevent the register from
6224 being used for other purposes in the preceding functions.
6226 Global register variables may not have initial values, because an
6227 executable file has no means to supply initial contents for a register.
6229 On the SPARC, there are reports that g3 @dots{} g7 are suitable
6230 registers, but certain library functions, such as @code{getwd}, as well
6231 as the subroutines for division and remainder, modify g3 and g4. g1 and
6232 g2 are local temporaries.
6234 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
6235 Of course, it will not do to use more than a few of those.
6237 @node Local Reg Vars
6238 @subsection Specifying Registers for Local Variables
6239 @cindex local variables, specifying registers
6240 @cindex specifying registers for local variables
6241 @cindex registers for local variables
6243 You can define a local register variable with a specified register
6247 register int *foo asm ("a5");
6251 Here @code{a5} is the name of the register which should be used. Note
6252 that this is the same syntax used for defining global register
6253 variables, but for a local variable it would appear within a function.
6255 Naturally the register name is cpu-dependent, but this is not a
6256 problem, since specific registers are most often useful with explicit
6257 assembler instructions (@pxref{Extended Asm}). Both of these things
6258 generally require that you conditionalize your program according to
6261 In addition, operating systems on one type of cpu may differ in how they
6262 name the registers; then you would need additional conditionals. For
6263 example, some 68000 operating systems call this register @code{%a5}.
6265 Defining such a register variable does not reserve the register; it
6266 remains available for other uses in places where flow control determines
6267 the variable's value is not live.
6269 This option does not guarantee that GCC will generate code that has
6270 this variable in the register you specify at all times. You may not
6271 code an explicit reference to this register in the @emph{assembler
6272 instruction template} part of an @code{asm} statement and assume it will
6273 always refer to this variable. However, using the variable as an
6274 @code{asm} @emph{operand} guarantees that the specified register is used
6277 Stores into local register variables may be deleted when they appear to be dead
6278 according to dataflow analysis. References to local register variables may
6279 be deleted or moved or simplified.
6281 As for global register variables, it's recommended that you choose a
6282 register which is normally saved and restored by function calls on
6283 your machine, so that library routines will not clobber it. A common
6284 pitfall is to initialize multiple call-clobbered registers with
6285 arbitrary expressions, where a function call or library call for an
6286 arithmetic operator will overwrite a register value from a previous
6287 assignment, for example @code{r0} below:
6289 register int *p1 asm ("r0") = @dots{};
6290 register int *p2 asm ("r1") = @dots{};
6292 In those cases, a solution is to use a temporary variable for
6293 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
6295 @node Alternate Keywords
6296 @section Alternate Keywords
6297 @cindex alternate keywords
6298 @cindex keywords, alternate
6300 @option{-ansi} and the various @option{-std} options disable certain
6301 keywords. This causes trouble when you want to use GNU C extensions, or
6302 a general-purpose header file that should be usable by all programs,
6303 including ISO C programs. The keywords @code{asm}, @code{typeof} and
6304 @code{inline} are not available in programs compiled with
6305 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
6306 program compiled with @option{-std=c99} or @option{-std=c1x}). The
6308 @code{restrict} is only available when @option{-std=gnu99} (which will
6309 eventually be the default) or @option{-std=c99} (or the equivalent
6310 @option{-std=iso9899:1999}), or an option for a later standard
6313 The way to solve these problems is to put @samp{__} at the beginning and
6314 end of each problematical keyword. For example, use @code{__asm__}
6315 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
6317 Other C compilers won't accept these alternative keywords; if you want to
6318 compile with another compiler, you can define the alternate keywords as
6319 macros to replace them with the customary keywords. It looks like this:
6327 @findex __extension__
6329 @option{-pedantic} and other options cause warnings for many GNU C extensions.
6331 prevent such warnings within one expression by writing
6332 @code{__extension__} before the expression. @code{__extension__} has no
6333 effect aside from this.
6335 @node Incomplete Enums
6336 @section Incomplete @code{enum} Types
6338 You can define an @code{enum} tag without specifying its possible values.
6339 This results in an incomplete type, much like what you get if you write
6340 @code{struct foo} without describing the elements. A later declaration
6341 which does specify the possible values completes the type.
6343 You can't allocate variables or storage using the type while it is
6344 incomplete. However, you can work with pointers to that type.
6346 This extension may not be very useful, but it makes the handling of
6347 @code{enum} more consistent with the way @code{struct} and @code{union}
6350 This extension is not supported by GNU C++.
6352 @node Function Names
6353 @section Function Names as Strings
6354 @cindex @code{__func__} identifier
6355 @cindex @code{__FUNCTION__} identifier
6356 @cindex @code{__PRETTY_FUNCTION__} identifier
6358 GCC provides three magic variables which hold the name of the current
6359 function, as a string. The first of these is @code{__func__}, which
6360 is part of the C99 standard:
6362 The identifier @code{__func__} is implicitly declared by the translator
6363 as if, immediately following the opening brace of each function
6364 definition, the declaration
6367 static const char __func__[] = "function-name";
6371 appeared, where function-name is the name of the lexically-enclosing
6372 function. This name is the unadorned name of the function.
6374 @code{__FUNCTION__} is another name for @code{__func__}. Older
6375 versions of GCC recognize only this name. However, it is not
6376 standardized. For maximum portability, we recommend you use
6377 @code{__func__}, but provide a fallback definition with the
6381 #if __STDC_VERSION__ < 199901L
6383 # define __func__ __FUNCTION__
6385 # define __func__ "<unknown>"
6390 In C, @code{__PRETTY_FUNCTION__} is yet another name for
6391 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
6392 the type signature of the function as well as its bare name. For
6393 example, this program:
6397 extern int printf (char *, ...);
6404 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
6405 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
6423 __PRETTY_FUNCTION__ = void a::sub(int)
6426 These identifiers are not preprocessor macros. In GCC 3.3 and
6427 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
6428 were treated as string literals; they could be used to initialize
6429 @code{char} arrays, and they could be concatenated with other string
6430 literals. GCC 3.4 and later treat them as variables, like
6431 @code{__func__}. In C++, @code{__FUNCTION__} and
6432 @code{__PRETTY_FUNCTION__} have always been variables.
6434 @node Return Address
6435 @section Getting the Return or Frame Address of a Function
6437 These functions may be used to get information about the callers of a
6440 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
6441 This function returns the return address of the current function, or of
6442 one of its callers. The @var{level} argument is number of frames to
6443 scan up the call stack. A value of @code{0} yields the return address
6444 of the current function, a value of @code{1} yields the return address
6445 of the caller of the current function, and so forth. When inlining
6446 the expected behavior is that the function will return the address of
6447 the function that will be returned to. To work around this behavior use
6448 the @code{noinline} function attribute.
6450 The @var{level} argument must be a constant integer.
6452 On some machines it may be impossible to determine the return address of
6453 any function other than the current one; in such cases, or when the top
6454 of the stack has been reached, this function will return @code{0} or a
6455 random value. In addition, @code{__builtin_frame_address} may be used
6456 to determine if the top of the stack has been reached.
6458 Additional post-processing of the returned value may be needed, see
6459 @code{__builtin_extract_return_address}.
6461 This function should only be used with a nonzero argument for debugging
6465 @deftypefn {Built-in Function} {void *} __builtin_extract_return_address (void *@var{addr})
6466 The address as returned by @code{__builtin_return_address} may have to be fed
6467 through this function to get the actual encoded address. For example, on the
6468 31-bit S/390 platform the highest bit has to be masked out, or on SPARC
6469 platforms an offset has to be added for the true next instruction to be
6472 If no fixup is needed, this function simply passes through @var{addr}.
6475 @deftypefn {Built-in Function} {void *} __builtin_frob_return_address (void *@var{addr})
6476 This function does the reverse of @code{__builtin_extract_return_address}.
6479 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
6480 This function is similar to @code{__builtin_return_address}, but it
6481 returns the address of the function frame rather than the return address
6482 of the function. Calling @code{__builtin_frame_address} with a value of
6483 @code{0} yields the frame address of the current function, a value of
6484 @code{1} yields the frame address of the caller of the current function,
6487 The frame is the area on the stack which holds local variables and saved
6488 registers. The frame address is normally the address of the first word
6489 pushed on to the stack by the function. However, the exact definition
6490 depends upon the processor and the calling convention. If the processor
6491 has a dedicated frame pointer register, and the function has a frame,
6492 then @code{__builtin_frame_address} will return the value of the frame
6495 On some machines it may be impossible to determine the frame address of
6496 any function other than the current one; in such cases, or when the top
6497 of the stack has been reached, this function will return @code{0} if
6498 the first frame pointer is properly initialized by the startup code.
6500 This function should only be used with a nonzero argument for debugging
6504 @node Vector Extensions
6505 @section Using vector instructions through built-in functions
6507 On some targets, the instruction set contains SIMD vector instructions that
6508 operate on multiple values contained in one large register at the same time.
6509 For example, on the i386 the MMX, 3DNow!@: and SSE extensions can be used
6512 The first step in using these extensions is to provide the necessary data
6513 types. This should be done using an appropriate @code{typedef}:
6516 typedef int v4si __attribute__ ((vector_size (16)));
6519 The @code{int} type specifies the base type, while the attribute specifies
6520 the vector size for the variable, measured in bytes. For example, the
6521 declaration above causes the compiler to set the mode for the @code{v4si}
6522 type to be 16 bytes wide and divided into @code{int} sized units. For
6523 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
6524 corresponding mode of @code{foo} will be @acronym{V4SI}.
6526 The @code{vector_size} attribute is only applicable to integral and
6527 float scalars, although arrays, pointers, and function return values
6528 are allowed in conjunction with this construct.
6530 All the basic integer types can be used as base types, both as signed
6531 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
6532 @code{long long}. In addition, @code{float} and @code{double} can be
6533 used to build floating-point vector types.
6535 Specifying a combination that is not valid for the current architecture
6536 will cause GCC to synthesize the instructions using a narrower mode.
6537 For example, if you specify a variable of type @code{V4SI} and your
6538 architecture does not allow for this specific SIMD type, GCC will
6539 produce code that uses 4 @code{SIs}.
6541 The types defined in this manner can be used with a subset of normal C
6542 operations. Currently, GCC will allow using the following operators
6543 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~, %}@.
6545 The operations behave like C++ @code{valarrays}. Addition is defined as
6546 the addition of the corresponding elements of the operands. For
6547 example, in the code below, each of the 4 elements in @var{a} will be
6548 added to the corresponding 4 elements in @var{b} and the resulting
6549 vector will be stored in @var{c}.
6552 typedef int v4si __attribute__ ((vector_size (16)));
6559 Subtraction, multiplication, division, and the logical operations
6560 operate in a similar manner. Likewise, the result of using the unary
6561 minus or complement operators on a vector type is a vector whose
6562 elements are the negative or complemented values of the corresponding
6563 elements in the operand.
6565 In C it is possible to use shifting operators @code{<<}, @code{>>} on
6566 integer-type vectors. The operation is defined as following: @code{@{a0,
6567 a1, @dots{}, an@} >> @{b0, b1, @dots{}, bn@} == @{a0 >> b0, a1 >> b1,
6568 @dots{}, an >> bn@}}@. Vector operands must have the same number of
6571 For the convenience in C it is allowed to use a binary vector operation
6572 where one operand is a scalar. In that case the compiler will transform
6573 the scalar operand into a vector where each element is the scalar from
6574 the operation. The transformation will happen only if the scalar could be
6575 safely converted to the vector-element type.
6576 Consider the following code.
6579 typedef int v4si __attribute__ ((vector_size (16)));
6584 a = b + 1; /* a = b + @{1,1,1,1@}; */
6585 a = 2 * b; /* a = @{2,2,2,2@} * b; */
6587 a = l + a; /* Error, cannot convert long to int. */
6590 In C vectors can be subscripted as if the vector were an array with
6591 the same number of elements and base type. Out of bound accesses
6592 invoke undefined behavior at runtime. Warnings for out of bound
6593 accesses for vector subscription can be enabled with
6594 @option{-Warray-bounds}.
6596 In GNU C vector comparison is supported within standard comparison
6597 operators: @code{==, !=, <, <=, >, >=}. Comparison operands can be
6598 vector expressions of integer-type or real-type. Comparison between
6599 integer-type vectors and real-type vectors are not supported. The
6600 result of the comparison is a vector of the same width and number of
6601 elements as the comparison operands with a signed integral element
6604 Vectors are compared element-wise producing 0 when comparison is false
6605 and -1 (constant of the appropriate type where all bits are set)
6606 otherwise. Consider the following example.
6609 typedef int v4si __attribute__ ((vector_size (16)));
6611 v4si a = @{1,2,3,4@};
6612 v4si b = @{3,2,1,4@};
6615 c = a > b; /* The result would be @{0, 0,-1, 0@} */
6616 c = a == b; /* The result would be @{0,-1, 0,-1@} */
6619 Vector shuffling is available using functions
6620 @code{__builtin_shuffle (vec, mask)} and
6621 @code{__builtin_shuffle (vec0, vec1, mask)}.
6622 Both functions construct a permutation of elements from one or two
6623 vectors and return a vector of the same type as the input vector(s).
6624 The @var{mask} is an integral vector with the same width (@var{W})
6625 and element count (@var{N}) as the output vector.
6627 The elements of the input vectors are numbered in memory ordering of
6628 @var{vec0} beginning at 0 and @var{vec1} beginning at @var{N}. The
6629 elements of @var{mask} are considered modulo @var{N} in the single-operand
6630 case and modulo @math{2*@var{N}} in the two-operand case.
6632 Consider the following example,
6635 typedef int v4si __attribute__ ((vector_size (16)));
6637 v4si a = @{1,2,3,4@};
6638 v4si b = @{5,6,7,8@};
6639 v4si mask1 = @{0,1,1,3@};
6640 v4si mask2 = @{0,4,2,5@};
6643 res = __builtin_shuffle (a, mask1); /* res is @{1,2,2,4@} */
6644 res = __builtin_shuffle (a, b, mask2); /* res is @{1,5,3,6@} */
6647 Note that @code{__builtin_shuffle} is intentionally semantically
6648 compatible with the OpenCL @code{shuffle} and @code{shuffle2} functions.
6650 You can declare variables and use them in function calls and returns, as
6651 well as in assignments and some casts. You can specify a vector type as
6652 a return type for a function. Vector types can also be used as function
6653 arguments. It is possible to cast from one vector type to another,
6654 provided they are of the same size (in fact, you can also cast vectors
6655 to and from other datatypes of the same size).
6657 You cannot operate between vectors of different lengths or different
6658 signedness without a cast.
6662 @findex __builtin_offsetof
6664 GCC implements for both C and C++ a syntactic extension to implement
6665 the @code{offsetof} macro.
6669 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
6671 offsetof_member_designator:
6673 | offsetof_member_designator "." @code{identifier}
6674 | offsetof_member_designator "[" @code{expr} "]"
6677 This extension is sufficient such that
6680 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
6683 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
6684 may be dependent. In either case, @var{member} may consist of a single
6685 identifier, or a sequence of member accesses and array references.
6687 @node __sync Builtins
6688 @section Legacy __sync built-in functions for atomic memory access
6690 The following builtins are intended to be compatible with those described
6691 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
6692 section 7.4. As such, they depart from the normal GCC practice of using
6693 the ``__builtin_'' prefix, and further that they are overloaded such that
6694 they work on multiple types.
6696 The definition given in the Intel documentation allows only for the use of
6697 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
6698 counterparts. GCC will allow any integral scalar or pointer type that is
6699 1, 2, 4 or 8 bytes in length.
6701 Not all operations are supported by all target processors. If a particular
6702 operation cannot be implemented on the target processor, a warning will be
6703 generated and a call an external function will be generated. The external
6704 function will carry the same name as the builtin, with an additional suffix
6705 @samp{_@var{n}} where @var{n} is the size of the data type.
6707 @c ??? Should we have a mechanism to suppress this warning? This is almost
6708 @c useful for implementing the operation under the control of an external
6711 In most cases, these builtins are considered a @dfn{full barrier}. That is,
6712 no memory operand will be moved across the operation, either forward or
6713 backward. Further, instructions will be issued as necessary to prevent the
6714 processor from speculating loads across the operation and from queuing stores
6715 after the operation.
6717 All of the routines are described in the Intel documentation to take
6718 ``an optional list of variables protected by the memory barrier''. It's
6719 not clear what is meant by that; it could mean that @emph{only} the
6720 following variables are protected, or it could mean that these variables
6721 should in addition be protected. At present GCC ignores this list and
6722 protects all variables which are globally accessible. If in the future
6723 we make some use of this list, an empty list will continue to mean all
6724 globally accessible variables.
6727 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
6728 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
6729 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
6730 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
6731 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
6732 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
6733 @findex __sync_fetch_and_add
6734 @findex __sync_fetch_and_sub
6735 @findex __sync_fetch_and_or
6736 @findex __sync_fetch_and_and
6737 @findex __sync_fetch_and_xor
6738 @findex __sync_fetch_and_nand
6739 These builtins perform the operation suggested by the name, and
6740 returns the value that had previously been in memory. That is,
6743 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
6744 @{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; @} // nand
6747 @emph{Note:} GCC 4.4 and later implement @code{__sync_fetch_and_nand}
6748 builtin as @code{*ptr = ~(tmp & value)} instead of @code{*ptr = ~tmp & value}.
6750 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
6751 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
6752 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
6753 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
6754 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
6755 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
6756 @findex __sync_add_and_fetch
6757 @findex __sync_sub_and_fetch
6758 @findex __sync_or_and_fetch
6759 @findex __sync_and_and_fetch
6760 @findex __sync_xor_and_fetch
6761 @findex __sync_nand_and_fetch
6762 These builtins perform the operation suggested by the name, and
6763 return the new value. That is,
6766 @{ *ptr @var{op}= value; return *ptr; @}
6767 @{ *ptr = ~(*ptr & value); return *ptr; @} // nand
6770 @emph{Note:} GCC 4.4 and later implement @code{__sync_nand_and_fetch}
6771 builtin as @code{*ptr = ~(*ptr & value)} instead of
6772 @code{*ptr = ~*ptr & value}.
6774 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
6775 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
6776 @findex __sync_bool_compare_and_swap
6777 @findex __sync_val_compare_and_swap
6778 These builtins perform an atomic compare and swap. That is, if the current
6779 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
6782 The ``bool'' version returns true if the comparison is successful and
6783 @var{newval} was written. The ``val'' version returns the contents
6784 of @code{*@var{ptr}} before the operation.
6786 @item __sync_synchronize (...)
6787 @findex __sync_synchronize
6788 This builtin issues a full memory barrier.
6790 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
6791 @findex __sync_lock_test_and_set
6792 This builtin, as described by Intel, is not a traditional test-and-set
6793 operation, but rather an atomic exchange operation. It writes @var{value}
6794 into @code{*@var{ptr}}, and returns the previous contents of
6797 Many targets have only minimal support for such locks, and do not support
6798 a full exchange operation. In this case, a target may support reduced
6799 functionality here by which the @emph{only} valid value to store is the
6800 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
6801 is implementation defined.
6803 This builtin is not a full barrier, but rather an @dfn{acquire barrier}.
6804 This means that references after the builtin cannot move to (or be
6805 speculated to) before the builtin, but previous memory stores may not
6806 be globally visible yet, and previous memory loads may not yet be
6809 @item void __sync_lock_release (@var{type} *ptr, ...)
6810 @findex __sync_lock_release
6811 This builtin releases the lock acquired by @code{__sync_lock_test_and_set}.
6812 Normally this means writing the constant 0 to @code{*@var{ptr}}.
6814 This builtin is not a full barrier, but rather a @dfn{release barrier}.
6815 This means that all previous memory stores are globally visible, and all
6816 previous memory loads have been satisfied, but following memory reads
6817 are not prevented from being speculated to before the barrier.
6820 @node __atomic Builtins
6821 @section Built-in functions for memory model aware atomic operations
6823 The following built-in functions approximately match the requirements for
6824 C++11 memory model. Many are similar to the @samp{__sync} prefixed built-in
6825 functions, but all also have a memory model parameter. These are all
6826 identified by being prefixed with @samp{__atomic}, and most are overloaded
6827 such that they work with multiple types.
6829 GCC will allow any integral scalar or pointer type that is 1, 2, 4, or 8
6830 bytes in length. 16-byte integral types are also allowed if
6831 @samp{__int128} (@pxref{__int128}) is supported by the architecture.
6833 Target architectures are encouraged to provide their own patterns for
6834 each of these built-in functions. If no target is provided, the original
6835 non-memory model set of @samp{__sync} atomic built-in functions will be
6836 utilized, along with any required synchronization fences surrounding it in
6837 order to achieve the proper behaviour. Execution in this case is subject
6838 to the same restrictions as those built-in functions.
6840 If there is no pattern or mechanism to provide a lock free instruction
6841 sequence, a call is made to an external routine with the same parameters
6842 to be resolved at runtime.
6844 The four non-arithmetic functions (load, store, exchange, and
6845 compare_exchange) all have a generic version as well. This generic
6846 version will work on any data type. If the data type size maps to one
6847 of the integral sizes which may have lock free support, the generic
6848 version will utilize the lock free built-in function. Otherwise an
6849 external call is left to be resolved at runtime. This external call will
6850 be the same format with the addition of a @samp{size_t} parameter inserted
6851 as the first parameter indicating the size of the object being pointed to.
6852 All objects must be the same size.
6854 There are 6 different memory models which can be specified. These map
6855 to the same names in the C++11 standard. Refer there or to the
6856 @uref{http://gcc.gnu.org/wiki/Atomic/GCCMM/AtomicSync,GCC wiki on
6857 atomic synchronization} for more detailed definitions. These memory
6858 models integrate both barriers to code motion as well as synchronization
6859 requirements with other threads. These are listed in approximately
6860 ascending order of strength.
6863 @item __ATOMIC_RELAXED
6864 No barriers or synchronization.
6865 @item __ATOMIC_CONSUME
6866 Data dependency only for both barrier and synchronization with another
6868 @item __ATOMIC_ACQUIRE
6869 Barrier to hoisting of code and synchronizes with release (or stronger)
6870 semantic stores from another thread.
6871 @item __ATOMIC_RELEASE
6872 Barrier to sinking of code and synchronizes with acquire (or stronger)
6873 semantic loads from another thread.
6874 @item __ATOMIC_ACQ_REL
6875 Full barrier in both directions and synchronizes with acquire loads and
6876 release stores in another thread.
6877 @item __ATOMIC_SEQ_CST
6878 Full barrier in both directions and synchronizes with acquire loads and
6879 release stores in all threads.
6882 When implementing patterns for these built-in functions , the memory model
6883 parameter can be ignored as long as the pattern implements the most
6884 restrictive @code{__ATOMIC_SEQ_CST} model. Any of the other memory models
6885 will execute correctly with this memory model but they may not execute as
6886 efficiently as they could with a more appropriate implemention of the
6887 relaxed requirements.
6889 Note that the C++11 standard allows for the memory model parameter to be
6890 determined at runtime rather than at compile time. These built-in
6891 functions will map any runtime value to @code{__ATOMIC_SEQ_CST} rather
6892 than invoke a runtime library call or inline a switch statement. This is
6893 standard compliant, safe, and the simplest approach for now.
6895 @deftypefn {Built-in Function} @var{type} __atomic_load_n (@var{type} *ptr, int memmodel)
6896 This built-in function implements an atomic load operation. It returns the
6897 contents of @code{*@var{ptr}}.
6899 The valid memory model variants are
6900 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
6901 and @code{__ATOMIC_CONSUME}.
6905 @deftypefn {Built-in Function} void __atomic_load (@var{type} *ptr, @var{type} *ret, int memmodel)
6906 This is the generic version of an atomic load. It will return the
6907 contents of @code{*@var{ptr}} in @code{*@var{ret}}.
6911 @deftypefn {Built-in Function} void __atomic_store_n (@var{type} *ptr, @var{type} val, int memmodel)
6912 This built-in function implements an atomic store operation. It writes
6913 @code{@var{val}} into @code{*@var{ptr}}.
6915 The valid memory model variants are
6916 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and @code{__ATOMIC_RELEASE}.
6920 @deftypefn {Built-in Function} void __atomic_store (@var{type} *ptr, @var{type} *val, int memmodel)
6921 This is the generic version of an atomic store. It will store the value
6922 of @code{*@var{val}} into @code{*@var{ptr}}.
6926 @deftypefn {Built-in Function} @var{type} __atomic_exchange_n (@var{type} *ptr, @var{type} val, int memmodel)
6927 This built-in function implements an atomic exchange operation. It writes
6928 @var{val} into @code{*@var{ptr}}, and returns the previous contents of
6931 The valid memory model variants are
6932 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
6933 @code{__ATOMIC_RELEASE}, and @code{__ATOMIC_ACQ_REL}.
6937 @deftypefn {Built-in Function} void __atomic_exchange (@var{type} *ptr, @var{type} *val, @var{type} *ret, int memmodel)
6938 This is the generic version of an atomic exchange. It will store the
6939 contents of @code{*@var{val}} into @code{*@var{ptr}}. The original value
6940 of @code{*@var{ptr}} will be copied into @code{*@var{ret}}.
6944 @deftypefn {Built-in Function} bool __atomic_compare_exchange_n (@var{type} *ptr, @var{type} *expected, @var{type} desired, bool weak, int success_memmodel, int failure_memmodel)
6945 This built-in function implements an atomic compare and exchange operation.
6946 This compares the contents of @code{*@var{ptr}} with the contents of
6947 @code{*@var{expected}} and if equal, writes @var{desired} into
6948 @code{*@var{ptr}}. If they are not equal, the current contents of
6949 @code{*@var{ptr}} is written into @code{*@var{expected}}.
6951 True is returned if @code{*@var{desired}} is written into
6952 @code{*@var{ptr}} and the execution is considered to conform to the
6953 memory model specified by @var{success_memmodel}. There are no
6954 restrictions on what memory model can be used here.
6956 False is returned otherwise, and the execution is considered to conform
6957 to @var{failure_memmodel}. This memory model cannot be
6958 @code{__ATOMIC_RELEASE} nor @code{__ATOMIC_ACQ_REL}. It also cannot be a
6959 stronger model than that specified by @var{success_memmodel}.
6963 @deftypefn {Built-in Function} bool __atomic_compare_exchange (@var{type} *ptr, @var{type} *expected, @var{type} *desired, bool weak, int success_memmodel, int failure_memmodel)
6964 This built-in function implements the generic version of
6965 @code{__atomic_compare_exchange}. The function is virtually identical to
6966 @code{__atomic_compare_exchange_n}, except the desired value is also a
6971 @deftypefn {Built-in Function} @var{type} __atomic_add_fetch (@var{type} *ptr, @var{type} val, int memmodel)
6972 @deftypefnx {Built-in Function} @var{type} __atomic_sub_fetch (@var{type} *ptr, @var{type} val, int memmodel)
6973 @deftypefnx {Built-in Function} @var{type} __atomic_and_fetch (@var{type} *ptr, @var{type} val, int memmodel)
6974 @deftypefnx {Built-in Function} @var{type} __atomic_xor_fetch (@var{type} *ptr, @var{type} val, int memmodel)
6975 @deftypefnx {Built-in Function} @var{type} __atomic_or_fetch (@var{type} *ptr, @var{type} val, int memmodel)
6976 @deftypefnx {Built-in Function} @var{type} __atomic_nand_fetch (@var{type} *ptr, @var{type} val, int memmodel)
6977 These built-in functions perform the operation suggested by the name, and
6978 return the result of the operation. That is,
6981 @{ *ptr @var{op}= val; return *ptr; @}
6984 All memory models are valid.
6988 @deftypefn {Built-in Function} @var{type} __atomic_fetch_add (@var{type} *ptr, @var{type} val, int memmodel)
6989 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_sub (@var{type} *ptr, @var{type} val, int memmodel)
6990 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_and (@var{type} *ptr, @var{type} val, int memmodel)
6991 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_xor (@var{type} *ptr, @var{type} val, int memmodel)
6992 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_or (@var{type} *ptr, @var{type} val, int memmodel)
6993 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_nand (@var{type} *ptr, @var{type} val, int memmodel)
6994 These built-in functions perform the operation suggested by the name, and
6995 return the value that had previously been in @code{*@var{ptr}}. That is,
6998 @{ tmp = *ptr; *ptr @var{op}= val; return tmp; @}
7001 All memory models are valid.
7005 @deftypefn {Built-in Function} bool __atomic_test_and_set (bool *ptr, int memmodel)
7007 This built-in function performs an atomic test-and-set operation on
7008 @code{*@var{ptr}}. @code{*@var{ptr}} is set to the value 1 and
7009 the previous contents are returned.
7011 All memory models are valid.
7015 @deftypefn {Built-in Function} void __atomic_clear (bool *ptr, int memmodel)
7017 This built-in function performs an atomic clear operation on
7018 @code{*@var{ptr}}. After the operation, @code{*@var{ptr}} will contain 0.
7020 The valid memory model variants are
7021 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and
7022 @code{__ATOMIC_RELEASE}.
7026 @deftypefn {Built-in Function} void __atomic_thread_fence (int memmodel)
7028 This built-in function acts as a synchronization fence between threads
7029 based on the specified memory model.
7031 All memory orders are valid.
7035 @deftypefn {Built-in Function} void __atomic_signal_fence (int memmodel)
7037 This built-in function acts as a synchronization fence between a thread
7038 and signal handlers based in the same thread.
7040 All memory orders are valid.
7044 @deftypefn {Built-in Function} bool __atomic_always_lock_free (size_t size)
7046 This built-in function returns true if objects of size bytes will always
7047 generate lock free atomic instructions for the target architecture.
7048 Otherwise false is returned.
7050 size must resolve to a compile time constant.
7053 if (_atomic_always_lock_free (sizeof (long long)))
7058 @deftypefn {Built-in Function} bool __atomic_is_lock_free (size_t size)
7060 This built-in function returns true if objects of size bytes will always
7061 generate lock free atomic instructions for the target architecture. If
7062 it is not known to be lock free a call is made to a runtime routine named
7063 @code{__atomic_is_lock_free}.
7067 @node Object Size Checking
7068 @section Object Size Checking Builtins
7069 @findex __builtin_object_size
7070 @findex __builtin___memcpy_chk
7071 @findex __builtin___mempcpy_chk
7072 @findex __builtin___memmove_chk
7073 @findex __builtin___memset_chk
7074 @findex __builtin___strcpy_chk
7075 @findex __builtin___stpcpy_chk
7076 @findex __builtin___strncpy_chk
7077 @findex __builtin___strcat_chk
7078 @findex __builtin___strncat_chk
7079 @findex __builtin___sprintf_chk
7080 @findex __builtin___snprintf_chk
7081 @findex __builtin___vsprintf_chk
7082 @findex __builtin___vsnprintf_chk
7083 @findex __builtin___printf_chk
7084 @findex __builtin___vprintf_chk
7085 @findex __builtin___fprintf_chk
7086 @findex __builtin___vfprintf_chk
7088 GCC implements a limited buffer overflow protection mechanism
7089 that can prevent some buffer overflow attacks.
7091 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
7092 is a built-in construct that returns a constant number of bytes from
7093 @var{ptr} to the end of the object @var{ptr} pointer points to
7094 (if known at compile time). @code{__builtin_object_size} never evaluates
7095 its arguments for side-effects. If there are any side-effects in them, it
7096 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
7097 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
7098 point to and all of them are known at compile time, the returned number
7099 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
7100 0 and minimum if nonzero. If it is not possible to determine which objects
7101 @var{ptr} points to at compile time, @code{__builtin_object_size} should
7102 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
7103 for @var{type} 2 or 3.
7105 @var{type} is an integer constant from 0 to 3. If the least significant
7106 bit is clear, objects are whole variables, if it is set, a closest
7107 surrounding subobject is considered the object a pointer points to.
7108 The second bit determines if maximum or minimum of remaining bytes
7112 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
7113 char *p = &var.buf1[1], *q = &var.b;
7115 /* Here the object p points to is var. */
7116 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
7117 /* The subobject p points to is var.buf1. */
7118 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
7119 /* The object q points to is var. */
7120 assert (__builtin_object_size (q, 0)
7121 == (char *) (&var + 1) - (char *) &var.b);
7122 /* The subobject q points to is var.b. */
7123 assert (__builtin_object_size (q, 1) == sizeof (var.b));
7127 There are built-in functions added for many common string operation
7128 functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk}
7129 built-in is provided. This built-in has an additional last argument,
7130 which is the number of bytes remaining in object the @var{dest}
7131 argument points to or @code{(size_t) -1} if the size is not known.
7133 The built-in functions are optimized into the normal string functions
7134 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
7135 it is known at compile time that the destination object will not
7136 be overflown. If the compiler can determine at compile time the
7137 object will be always overflown, it issues a warning.
7139 The intended use can be e.g.
7143 #define bos0(dest) __builtin_object_size (dest, 0)
7144 #define memcpy(dest, src, n) \
7145 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
7149 /* It is unknown what object p points to, so this is optimized
7150 into plain memcpy - no checking is possible. */
7151 memcpy (p, "abcde", n);
7152 /* Destination is known and length too. It is known at compile
7153 time there will be no overflow. */
7154 memcpy (&buf[5], "abcde", 5);
7155 /* Destination is known, but the length is not known at compile time.
7156 This will result in __memcpy_chk call that can check for overflow
7158 memcpy (&buf[5], "abcde", n);
7159 /* Destination is known and it is known at compile time there will
7160 be overflow. There will be a warning and __memcpy_chk call that
7161 will abort the program at runtime. */
7162 memcpy (&buf[6], "abcde", 5);
7165 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
7166 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
7167 @code{strcat} and @code{strncat}.
7169 There are also checking built-in functions for formatted output functions.
7171 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
7172 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
7173 const char *fmt, ...);
7174 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
7176 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
7177 const char *fmt, va_list ap);
7180 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
7181 etc.@: functions and can contain implementation specific flags on what
7182 additional security measures the checking function might take, such as
7183 handling @code{%n} differently.
7185 The @var{os} argument is the object size @var{s} points to, like in the
7186 other built-in functions. There is a small difference in the behavior
7187 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
7188 optimized into the non-checking functions only if @var{flag} is 0, otherwise
7189 the checking function is called with @var{os} argument set to
7192 In addition to this, there are checking built-in functions
7193 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
7194 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
7195 These have just one additional argument, @var{flag}, right before
7196 format string @var{fmt}. If the compiler is able to optimize them to
7197 @code{fputc} etc.@: functions, it will, otherwise the checking function
7198 should be called and the @var{flag} argument passed to it.
7200 @node Other Builtins
7201 @section Other built-in functions provided by GCC
7202 @cindex built-in functions
7203 @findex __builtin_fpclassify
7204 @findex __builtin_isfinite
7205 @findex __builtin_isnormal
7206 @findex __builtin_isgreater
7207 @findex __builtin_isgreaterequal
7208 @findex __builtin_isinf_sign
7209 @findex __builtin_isless
7210 @findex __builtin_islessequal
7211 @findex __builtin_islessgreater
7212 @findex __builtin_isunordered
7213 @findex __builtin_powi
7214 @findex __builtin_powif
7215 @findex __builtin_powil
7373 @findex fprintf_unlocked
7375 @findex fputs_unlocked
7492 @findex printf_unlocked
7524 @findex significandf
7525 @findex significandl
7596 GCC provides a large number of built-in functions other than the ones
7597 mentioned above. Some of these are for internal use in the processing
7598 of exceptions or variable-length argument lists and will not be
7599 documented here because they may change from time to time; we do not
7600 recommend general use of these functions.
7602 The remaining functions are provided for optimization purposes.
7604 @opindex fno-builtin
7605 GCC includes built-in versions of many of the functions in the standard
7606 C library. The versions prefixed with @code{__builtin_} will always be
7607 treated as having the same meaning as the C library function even if you
7608 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
7609 Many of these functions are only optimized in certain cases; if they are
7610 not optimized in a particular case, a call to the library function will
7615 Outside strict ISO C mode (@option{-ansi}, @option{-std=c90},
7616 @option{-std=c99} or @option{-std=c1x}), the functions
7617 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
7618 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
7619 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
7620 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
7621 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
7622 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
7623 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
7624 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
7625 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
7626 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
7627 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
7628 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
7629 @code{signbitd64}, @code{signbitd128}, @code{significandf},
7630 @code{significandl}, @code{significand}, @code{sincosf},
7631 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
7632 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
7633 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
7634 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
7636 may be handled as built-in functions.
7637 All these functions have corresponding versions
7638 prefixed with @code{__builtin_}, which may be used even in strict C90
7641 The ISO C99 functions
7642 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
7643 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
7644 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
7645 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
7646 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
7647 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
7648 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
7649 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
7650 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
7651 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
7652 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
7653 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
7654 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
7655 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
7656 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
7657 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
7658 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
7659 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
7660 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
7661 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
7662 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
7663 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
7664 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
7665 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
7666 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
7667 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
7668 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
7669 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
7670 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
7671 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
7672 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
7673 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
7674 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
7675 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
7676 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
7677 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
7678 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
7679 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
7680 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
7681 are handled as built-in functions
7682 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
7684 There are also built-in versions of the ISO C99 functions
7685 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
7686 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
7687 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
7688 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
7689 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
7690 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
7691 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
7692 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
7693 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
7694 that are recognized in any mode since ISO C90 reserves these names for
7695 the purpose to which ISO C99 puts them. All these functions have
7696 corresponding versions prefixed with @code{__builtin_}.
7698 The ISO C94 functions
7699 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
7700 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
7701 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
7703 are handled as built-in functions
7704 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
7706 The ISO C90 functions
7707 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
7708 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
7709 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
7710 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
7711 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
7712 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
7713 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
7714 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
7715 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
7716 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
7717 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
7718 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
7719 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
7720 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
7721 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
7722 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
7723 are all recognized as built-in functions unless
7724 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
7725 is specified for an individual function). All of these functions have
7726 corresponding versions prefixed with @code{__builtin_}.
7728 GCC provides built-in versions of the ISO C99 floating point comparison
7729 macros that avoid raising exceptions for unordered operands. They have
7730 the same names as the standard macros ( @code{isgreater},
7731 @code{isgreaterequal}, @code{isless}, @code{islessequal},
7732 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
7733 prefixed. We intend for a library implementor to be able to simply
7734 @code{#define} each standard macro to its built-in equivalent.
7735 In the same fashion, GCC provides @code{fpclassify}, @code{isfinite},
7736 @code{isinf_sign} and @code{isnormal} built-ins used with
7737 @code{__builtin_} prefixed. The @code{isinf} and @code{isnan}
7738 builtins appear both with and without the @code{__builtin_} prefix.
7740 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
7742 You can use the built-in function @code{__builtin_types_compatible_p} to
7743 determine whether two types are the same.
7745 This built-in function returns 1 if the unqualified versions of the
7746 types @var{type1} and @var{type2} (which are types, not expressions) are
7747 compatible, 0 otherwise. The result of this built-in function can be
7748 used in integer constant expressions.
7750 This built-in function ignores top level qualifiers (e.g., @code{const},
7751 @code{volatile}). For example, @code{int} is equivalent to @code{const
7754 The type @code{int[]} and @code{int[5]} are compatible. On the other
7755 hand, @code{int} and @code{char *} are not compatible, even if the size
7756 of their types, on the particular architecture are the same. Also, the
7757 amount of pointer indirection is taken into account when determining
7758 similarity. Consequently, @code{short *} is not similar to
7759 @code{short **}. Furthermore, two types that are typedefed are
7760 considered compatible if their underlying types are compatible.
7762 An @code{enum} type is not considered to be compatible with another
7763 @code{enum} type even if both are compatible with the same integer
7764 type; this is what the C standard specifies.
7765 For example, @code{enum @{foo, bar@}} is not similar to
7766 @code{enum @{hot, dog@}}.
7768 You would typically use this function in code whose execution varies
7769 depending on the arguments' types. For example:
7774 typeof (x) tmp = (x); \
7775 if (__builtin_types_compatible_p (typeof (x), long double)) \
7776 tmp = foo_long_double (tmp); \
7777 else if (__builtin_types_compatible_p (typeof (x), double)) \
7778 tmp = foo_double (tmp); \
7779 else if (__builtin_types_compatible_p (typeof (x), float)) \
7780 tmp = foo_float (tmp); \
7787 @emph{Note:} This construct is only available for C@.
7791 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
7793 You can use the built-in function @code{__builtin_choose_expr} to
7794 evaluate code depending on the value of a constant expression. This
7795 built-in function returns @var{exp1} if @var{const_exp}, which is an
7796 integer constant expression, is nonzero. Otherwise it returns @var{exp2}.
7798 This built-in function is analogous to the @samp{? :} operator in C,
7799 except that the expression returned has its type unaltered by promotion
7800 rules. Also, the built-in function does not evaluate the expression
7801 that was not chosen. For example, if @var{const_exp} evaluates to true,
7802 @var{exp2} is not evaluated even if it has side-effects.
7804 This built-in function can return an lvalue if the chosen argument is an
7807 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
7808 type. Similarly, if @var{exp2} is returned, its return type is the same
7815 __builtin_choose_expr ( \
7816 __builtin_types_compatible_p (typeof (x), double), \
7818 __builtin_choose_expr ( \
7819 __builtin_types_compatible_p (typeof (x), float), \
7821 /* @r{The void expression results in a compile-time error} \
7822 @r{when assigning the result to something.} */ \
7826 @emph{Note:} This construct is only available for C@. Furthermore, the
7827 unused expression (@var{exp1} or @var{exp2} depending on the value of
7828 @var{const_exp}) may still generate syntax errors. This may change in
7833 @deftypefn {Built-in Function} @var{type} __builtin_complex (@var{real}, @var{imag})
7835 The built-in function @code{__builtin_complex} is provided for use in
7836 implementing the ISO C1X macros @code{CMPLXF}, @code{CMPLX} and
7837 @code{CMPLXL}. @var{real} and @var{imag} must have the same type, a
7838 real binary floating-point type, and the result has the corresponding
7839 complex type with real and imaginary parts @var{real} and @var{imag}.
7840 Unlike @samp{@var{real} + I * @var{imag}}, this works even when
7841 infinities, NaNs and negative zeros are involved.
7845 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
7846 You can use the built-in function @code{__builtin_constant_p} to
7847 determine if a value is known to be constant at compile-time and hence
7848 that GCC can perform constant-folding on expressions involving that
7849 value. The argument of the function is the value to test. The function
7850 returns the integer 1 if the argument is known to be a compile-time
7851 constant and 0 if it is not known to be a compile-time constant. A
7852 return of 0 does not indicate that the value is @emph{not} a constant,
7853 but merely that GCC cannot prove it is a constant with the specified
7854 value of the @option{-O} option.
7856 You would typically use this function in an embedded application where
7857 memory was a critical resource. If you have some complex calculation,
7858 you may want it to be folded if it involves constants, but need to call
7859 a function if it does not. For example:
7862 #define Scale_Value(X) \
7863 (__builtin_constant_p (X) \
7864 ? ((X) * SCALE + OFFSET) : Scale (X))
7867 You may use this built-in function in either a macro or an inline
7868 function. However, if you use it in an inlined function and pass an
7869 argument of the function as the argument to the built-in, GCC will
7870 never return 1 when you call the inline function with a string constant
7871 or compound literal (@pxref{Compound Literals}) and will not return 1
7872 when you pass a constant numeric value to the inline function unless you
7873 specify the @option{-O} option.
7875 You may also use @code{__builtin_constant_p} in initializers for static
7876 data. For instance, you can write
7879 static const int table[] = @{
7880 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
7886 This is an acceptable initializer even if @var{EXPRESSION} is not a
7887 constant expression, including the case where
7888 @code{__builtin_constant_p} returns 1 because @var{EXPRESSION} can be
7889 folded to a constant but @var{EXPRESSION} contains operands that would
7890 not otherwise be permitted in a static initializer (for example,
7891 @code{0 && foo ()}). GCC must be more conservative about evaluating the
7892 built-in in this case, because it has no opportunity to perform
7895 Previous versions of GCC did not accept this built-in in data
7896 initializers. The earliest version where it is completely safe is
7900 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
7901 @opindex fprofile-arcs
7902 You may use @code{__builtin_expect} to provide the compiler with
7903 branch prediction information. In general, you should prefer to
7904 use actual profile feedback for this (@option{-fprofile-arcs}), as
7905 programmers are notoriously bad at predicting how their programs
7906 actually perform. However, there are applications in which this
7907 data is hard to collect.
7909 The return value is the value of @var{exp}, which should be an integral
7910 expression. The semantics of the built-in are that it is expected that
7911 @var{exp} == @var{c}. For example:
7914 if (__builtin_expect (x, 0))
7919 would indicate that we do not expect to call @code{foo}, since
7920 we expect @code{x} to be zero. Since you are limited to integral
7921 expressions for @var{exp}, you should use constructions such as
7924 if (__builtin_expect (ptr != NULL, 1))
7929 when testing pointer or floating-point values.
7932 @deftypefn {Built-in Function} void __builtin_trap (void)
7933 This function causes the program to exit abnormally. GCC implements
7934 this function by using a target-dependent mechanism (such as
7935 intentionally executing an illegal instruction) or by calling
7936 @code{abort}. The mechanism used may vary from release to release so
7937 you should not rely on any particular implementation.
7940 @deftypefn {Built-in Function} void __builtin_unreachable (void)
7941 If control flow reaches the point of the @code{__builtin_unreachable},
7942 the program is undefined. It is useful in situations where the
7943 compiler cannot deduce the unreachability of the code.
7945 One such case is immediately following an @code{asm} statement that
7946 will either never terminate, or one that transfers control elsewhere
7947 and never returns. In this example, without the
7948 @code{__builtin_unreachable}, GCC would issue a warning that control
7949 reaches the end of a non-void function. It would also generate code
7950 to return after the @code{asm}.
7953 int f (int c, int v)
7961 asm("jmp error_handler");
7962 __builtin_unreachable ();
7967 Because the @code{asm} statement unconditionally transfers control out
7968 of the function, control will never reach the end of the function
7969 body. The @code{__builtin_unreachable} is in fact unreachable and
7970 communicates this fact to the compiler.
7972 Another use for @code{__builtin_unreachable} is following a call a
7973 function that never returns but that is not declared
7974 @code{__attribute__((noreturn))}, as in this example:
7977 void function_that_never_returns (void);
7987 function_that_never_returns ();
7988 __builtin_unreachable ();
7995 @deftypefn {Built-in Function} void *__builtin_assume_aligned (const void *@var{exp}, size_t @var{align}, ...)
7996 This function returns its first argument, and allows the compiler
7997 to assume that the returned pointer is at least @var{align} bytes
7998 aligned. This built-in can have either two or three arguments,
7999 if it has three, the third argument should have integer type, and
8000 if it is non-zero means misalignment offset. For example:
8003 void *x = __builtin_assume_aligned (arg, 16);
8006 means that the compiler can assume x, set to arg, is at least
8007 16 byte aligned, while:
8010 void *x = __builtin_assume_aligned (arg, 32, 8);
8013 means that the compiler can assume for x, set to arg, that
8014 (char *) x - 8 is 32 byte aligned.
8017 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
8018 This function is used to flush the processor's instruction cache for
8019 the region of memory between @var{begin} inclusive and @var{end}
8020 exclusive. Some targets require that the instruction cache be
8021 flushed, after modifying memory containing code, in order to obtain
8022 deterministic behavior.
8024 If the target does not require instruction cache flushes,
8025 @code{__builtin___clear_cache} has no effect. Otherwise either
8026 instructions are emitted in-line to clear the instruction cache or a
8027 call to the @code{__clear_cache} function in libgcc is made.
8030 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
8031 This function is used to minimize cache-miss latency by moving data into
8032 a cache before it is accessed.
8033 You can insert calls to @code{__builtin_prefetch} into code for which
8034 you know addresses of data in memory that is likely to be accessed soon.
8035 If the target supports them, data prefetch instructions will be generated.
8036 If the prefetch is done early enough before the access then the data will
8037 be in the cache by the time it is accessed.
8039 The value of @var{addr} is the address of the memory to prefetch.
8040 There are two optional arguments, @var{rw} and @var{locality}.
8041 The value of @var{rw} is a compile-time constant one or zero; one
8042 means that the prefetch is preparing for a write to the memory address
8043 and zero, the default, means that the prefetch is preparing for a read.
8044 The value @var{locality} must be a compile-time constant integer between
8045 zero and three. A value of zero means that the data has no temporal
8046 locality, so it need not be left in the cache after the access. A value
8047 of three means that the data has a high degree of temporal locality and
8048 should be left in all levels of cache possible. Values of one and two
8049 mean, respectively, a low or moderate degree of temporal locality. The
8053 for (i = 0; i < n; i++)
8056 __builtin_prefetch (&a[i+j], 1, 1);
8057 __builtin_prefetch (&b[i+j], 0, 1);
8062 Data prefetch does not generate faults if @var{addr} is invalid, but
8063 the address expression itself must be valid. For example, a prefetch
8064 of @code{p->next} will not fault if @code{p->next} is not a valid
8065 address, but evaluation will fault if @code{p} is not a valid address.
8067 If the target does not support data prefetch, the address expression
8068 is evaluated if it includes side effects but no other code is generated
8069 and GCC does not issue a warning.
8072 @deftypefn {Built-in Function} double __builtin_huge_val (void)
8073 Returns a positive infinity, if supported by the floating-point format,
8074 else @code{DBL_MAX}. This function is suitable for implementing the
8075 ISO C macro @code{HUGE_VAL}.
8078 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
8079 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
8082 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
8083 Similar to @code{__builtin_huge_val}, except the return
8084 type is @code{long double}.
8087 @deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...)
8088 This built-in implements the C99 fpclassify functionality. The first
8089 five int arguments should be the target library's notion of the
8090 possible FP classes and are used for return values. They must be
8091 constant values and they must appear in this order: @code{FP_NAN},
8092 @code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and
8093 @code{FP_ZERO}. The ellipsis is for exactly one floating point value
8094 to classify. GCC treats the last argument as type-generic, which
8095 means it does not do default promotion from float to double.
8098 @deftypefn {Built-in Function} double __builtin_inf (void)
8099 Similar to @code{__builtin_huge_val}, except a warning is generated
8100 if the target floating-point format does not support infinities.
8103 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
8104 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
8107 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
8108 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
8111 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
8112 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
8115 @deftypefn {Built-in Function} float __builtin_inff (void)
8116 Similar to @code{__builtin_inf}, except the return type is @code{float}.
8117 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
8120 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
8121 Similar to @code{__builtin_inf}, except the return
8122 type is @code{long double}.
8125 @deftypefn {Built-in Function} int __builtin_isinf_sign (...)
8126 Similar to @code{isinf}, except the return value will be negative for
8127 an argument of @code{-Inf}. Note while the parameter list is an
8128 ellipsis, this function only accepts exactly one floating point
8129 argument. GCC treats this parameter as type-generic, which means it
8130 does not do default promotion from float to double.
8133 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
8134 This is an implementation of the ISO C99 function @code{nan}.
8136 Since ISO C99 defines this function in terms of @code{strtod}, which we
8137 do not implement, a description of the parsing is in order. The string
8138 is parsed as by @code{strtol}; that is, the base is recognized by
8139 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
8140 in the significand such that the least significant bit of the number
8141 is at the least significant bit of the significand. The number is
8142 truncated to fit the significand field provided. The significand is
8143 forced to be a quiet NaN@.
8145 This function, if given a string literal all of which would have been
8146 consumed by strtol, is evaluated early enough that it is considered a
8147 compile-time constant.
8150 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
8151 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
8154 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
8155 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
8158 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
8159 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
8162 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
8163 Similar to @code{__builtin_nan}, except the return type is @code{float}.
8166 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
8167 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
8170 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
8171 Similar to @code{__builtin_nan}, except the significand is forced
8172 to be a signaling NaN@. The @code{nans} function is proposed by
8173 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
8176 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
8177 Similar to @code{__builtin_nans}, except the return type is @code{float}.
8180 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
8181 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
8184 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
8185 Returns one plus the index of the least significant 1-bit of @var{x}, or
8186 if @var{x} is zero, returns zero.
8189 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
8190 Returns the number of leading 0-bits in @var{x}, starting at the most
8191 significant bit position. If @var{x} is 0, the result is undefined.
8194 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
8195 Returns the number of trailing 0-bits in @var{x}, starting at the least
8196 significant bit position. If @var{x} is 0, the result is undefined.
8199 @deftypefn {Built-in Function} int __builtin_clrsb (int x)
8200 Returns the number of leading redundant sign bits in @var{x}, i.e. the
8201 number of bits following the most significant bit which are identical
8202 to it. There are no special cases for 0 or other values.
8205 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
8206 Returns the number of 1-bits in @var{x}.
8209 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
8210 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
8214 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
8215 Similar to @code{__builtin_ffs}, except the argument type is
8216 @code{unsigned long}.
8219 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
8220 Similar to @code{__builtin_clz}, except the argument type is
8221 @code{unsigned long}.
8224 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
8225 Similar to @code{__builtin_ctz}, except the argument type is
8226 @code{unsigned long}.
8229 @deftypefn {Built-in Function} int __builtin_clrsbl (long)
8230 Similar to @code{__builtin_clrsb}, except the argument type is
8234 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
8235 Similar to @code{__builtin_popcount}, except the argument type is
8236 @code{unsigned long}.
8239 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
8240 Similar to @code{__builtin_parity}, except the argument type is
8241 @code{unsigned long}.
8244 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
8245 Similar to @code{__builtin_ffs}, except the argument type is
8246 @code{unsigned long long}.
8249 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
8250 Similar to @code{__builtin_clz}, except the argument type is
8251 @code{unsigned long long}.
8254 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
8255 Similar to @code{__builtin_ctz}, except the argument type is
8256 @code{unsigned long long}.
8259 @deftypefn {Built-in Function} int __builtin_clrsbll (long long)
8260 Similar to @code{__builtin_clrsb}, except the argument type is
8264 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
8265 Similar to @code{__builtin_popcount}, except the argument type is
8266 @code{unsigned long long}.
8269 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
8270 Similar to @code{__builtin_parity}, except the argument type is
8271 @code{unsigned long long}.
8274 @deftypefn {Built-in Function} double __builtin_powi (double, int)
8275 Returns the first argument raised to the power of the second. Unlike the
8276 @code{pow} function no guarantees about precision and rounding are made.
8279 @deftypefn {Built-in Function} float __builtin_powif (float, int)
8280 Similar to @code{__builtin_powi}, except the argument and return types
8284 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
8285 Similar to @code{__builtin_powi}, except the argument and return types
8286 are @code{long double}.
8289 @deftypefn {Built-in Function} int32_t __builtin_bswap32 (int32_t x)
8290 Returns @var{x} with the order of the bytes reversed; for example,
8291 @code{0xaabbccdd} becomes @code{0xddccbbaa}. Byte here always means
8295 @deftypefn {Built-in Function} int64_t __builtin_bswap64 (int64_t x)
8296 Similar to @code{__builtin_bswap32}, except the argument and return types
8300 @node Target Builtins
8301 @section Built-in Functions Specific to Particular Target Machines
8303 On some target machines, GCC supports many built-in functions specific
8304 to those machines. Generally these generate calls to specific machine
8305 instructions, but allow the compiler to schedule those calls.
8308 * Alpha Built-in Functions::
8309 * ARM iWMMXt Built-in Functions::
8310 * ARM NEON Intrinsics::
8311 * AVR Built-in Functions::
8312 * Blackfin Built-in Functions::
8313 * FR-V Built-in Functions::
8314 * X86 Built-in Functions::
8315 * MIPS DSP Built-in Functions::
8316 * MIPS Paired-Single Support::
8317 * MIPS Loongson Built-in Functions::
8318 * Other MIPS Built-in Functions::
8319 * picoChip Built-in Functions::
8320 * PowerPC AltiVec/VSX Built-in Functions::
8321 * RX Built-in Functions::
8322 * SPARC VIS Built-in Functions::
8323 * SPU Built-in Functions::
8324 * TI C6X Built-in Functions::
8327 @node Alpha Built-in Functions
8328 @subsection Alpha Built-in Functions
8330 These built-in functions are available for the Alpha family of
8331 processors, depending on the command-line switches used.
8333 The following built-in functions are always available. They
8334 all generate the machine instruction that is part of the name.
8337 long __builtin_alpha_implver (void)
8338 long __builtin_alpha_rpcc (void)
8339 long __builtin_alpha_amask (long)
8340 long __builtin_alpha_cmpbge (long, long)
8341 long __builtin_alpha_extbl (long, long)
8342 long __builtin_alpha_extwl (long, long)
8343 long __builtin_alpha_extll (long, long)
8344 long __builtin_alpha_extql (long, long)
8345 long __builtin_alpha_extwh (long, long)
8346 long __builtin_alpha_extlh (long, long)
8347 long __builtin_alpha_extqh (long, long)
8348 long __builtin_alpha_insbl (long, long)
8349 long __builtin_alpha_inswl (long, long)
8350 long __builtin_alpha_insll (long, long)
8351 long __builtin_alpha_insql (long, long)
8352 long __builtin_alpha_inswh (long, long)
8353 long __builtin_alpha_inslh (long, long)
8354 long __builtin_alpha_insqh (long, long)
8355 long __builtin_alpha_mskbl (long, long)
8356 long __builtin_alpha_mskwl (long, long)
8357 long __builtin_alpha_mskll (long, long)
8358 long __builtin_alpha_mskql (long, long)
8359 long __builtin_alpha_mskwh (long, long)
8360 long __builtin_alpha_msklh (long, long)
8361 long __builtin_alpha_mskqh (long, long)
8362 long __builtin_alpha_umulh (long, long)
8363 long __builtin_alpha_zap (long, long)
8364 long __builtin_alpha_zapnot (long, long)
8367 The following built-in functions are always with @option{-mmax}
8368 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
8369 later. They all generate the machine instruction that is part
8373 long __builtin_alpha_pklb (long)
8374 long __builtin_alpha_pkwb (long)
8375 long __builtin_alpha_unpkbl (long)
8376 long __builtin_alpha_unpkbw (long)
8377 long __builtin_alpha_minub8 (long, long)
8378 long __builtin_alpha_minsb8 (long, long)
8379 long __builtin_alpha_minuw4 (long, long)
8380 long __builtin_alpha_minsw4 (long, long)
8381 long __builtin_alpha_maxub8 (long, long)
8382 long __builtin_alpha_maxsb8 (long, long)
8383 long __builtin_alpha_maxuw4 (long, long)
8384 long __builtin_alpha_maxsw4 (long, long)
8385 long __builtin_alpha_perr (long, long)
8388 The following built-in functions are always with @option{-mcix}
8389 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
8390 later. They all generate the machine instruction that is part
8394 long __builtin_alpha_cttz (long)
8395 long __builtin_alpha_ctlz (long)
8396 long __builtin_alpha_ctpop (long)
8399 The following builtins are available on systems that use the OSF/1
8400 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
8401 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
8402 @code{rdval} and @code{wrval}.
8405 void *__builtin_thread_pointer (void)
8406 void __builtin_set_thread_pointer (void *)
8409 @node ARM iWMMXt Built-in Functions
8410 @subsection ARM iWMMXt Built-in Functions
8412 These built-in functions are available for the ARM family of
8413 processors when the @option{-mcpu=iwmmxt} switch is used:
8416 typedef int v2si __attribute__ ((vector_size (8)));
8417 typedef short v4hi __attribute__ ((vector_size (8)));
8418 typedef char v8qi __attribute__ ((vector_size (8)));
8420 int __builtin_arm_getwcx (int)
8421 void __builtin_arm_setwcx (int, int)
8422 int __builtin_arm_textrmsb (v8qi, int)
8423 int __builtin_arm_textrmsh (v4hi, int)
8424 int __builtin_arm_textrmsw (v2si, int)
8425 int __builtin_arm_textrmub (v8qi, int)
8426 int __builtin_arm_textrmuh (v4hi, int)
8427 int __builtin_arm_textrmuw (v2si, int)
8428 v8qi __builtin_arm_tinsrb (v8qi, int)
8429 v4hi __builtin_arm_tinsrh (v4hi, int)
8430 v2si __builtin_arm_tinsrw (v2si, int)
8431 long long __builtin_arm_tmia (long long, int, int)
8432 long long __builtin_arm_tmiabb (long long, int, int)
8433 long long __builtin_arm_tmiabt (long long, int, int)
8434 long long __builtin_arm_tmiaph (long long, int, int)
8435 long long __builtin_arm_tmiatb (long long, int, int)
8436 long long __builtin_arm_tmiatt (long long, int, int)
8437 int __builtin_arm_tmovmskb (v8qi)
8438 int __builtin_arm_tmovmskh (v4hi)
8439 int __builtin_arm_tmovmskw (v2si)
8440 long long __builtin_arm_waccb (v8qi)
8441 long long __builtin_arm_wacch (v4hi)
8442 long long __builtin_arm_waccw (v2si)
8443 v8qi __builtin_arm_waddb (v8qi, v8qi)
8444 v8qi __builtin_arm_waddbss (v8qi, v8qi)
8445 v8qi __builtin_arm_waddbus (v8qi, v8qi)
8446 v4hi __builtin_arm_waddh (v4hi, v4hi)
8447 v4hi __builtin_arm_waddhss (v4hi, v4hi)
8448 v4hi __builtin_arm_waddhus (v4hi, v4hi)
8449 v2si __builtin_arm_waddw (v2si, v2si)
8450 v2si __builtin_arm_waddwss (v2si, v2si)
8451 v2si __builtin_arm_waddwus (v2si, v2si)
8452 v8qi __builtin_arm_walign (v8qi, v8qi, int)
8453 long long __builtin_arm_wand(long long, long long)
8454 long long __builtin_arm_wandn (long long, long long)
8455 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
8456 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
8457 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
8458 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
8459 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
8460 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
8461 v2si __builtin_arm_wcmpeqw (v2si, v2si)
8462 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
8463 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
8464 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
8465 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
8466 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
8467 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
8468 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
8469 long long __builtin_arm_wmacsz (v4hi, v4hi)
8470 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
8471 long long __builtin_arm_wmacuz (v4hi, v4hi)
8472 v4hi __builtin_arm_wmadds (v4hi, v4hi)
8473 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
8474 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
8475 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
8476 v2si __builtin_arm_wmaxsw (v2si, v2si)
8477 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
8478 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
8479 v2si __builtin_arm_wmaxuw (v2si, v2si)
8480 v8qi __builtin_arm_wminsb (v8qi, v8qi)
8481 v4hi __builtin_arm_wminsh (v4hi, v4hi)
8482 v2si __builtin_arm_wminsw (v2si, v2si)
8483 v8qi __builtin_arm_wminub (v8qi, v8qi)
8484 v4hi __builtin_arm_wminuh (v4hi, v4hi)
8485 v2si __builtin_arm_wminuw (v2si, v2si)
8486 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
8487 v4hi __builtin_arm_wmulul (v4hi, v4hi)
8488 v4hi __builtin_arm_wmulum (v4hi, v4hi)
8489 long long __builtin_arm_wor (long long, long long)
8490 v2si __builtin_arm_wpackdss (long long, long long)
8491 v2si __builtin_arm_wpackdus (long long, long long)
8492 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
8493 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
8494 v4hi __builtin_arm_wpackwss (v2si, v2si)
8495 v4hi __builtin_arm_wpackwus (v2si, v2si)
8496 long long __builtin_arm_wrord (long long, long long)
8497 long long __builtin_arm_wrordi (long long, int)
8498 v4hi __builtin_arm_wrorh (v4hi, long long)
8499 v4hi __builtin_arm_wrorhi (v4hi, int)
8500 v2si __builtin_arm_wrorw (v2si, long long)
8501 v2si __builtin_arm_wrorwi (v2si, int)
8502 v2si __builtin_arm_wsadb (v8qi, v8qi)
8503 v2si __builtin_arm_wsadbz (v8qi, v8qi)
8504 v2si __builtin_arm_wsadh (v4hi, v4hi)
8505 v2si __builtin_arm_wsadhz (v4hi, v4hi)
8506 v4hi __builtin_arm_wshufh (v4hi, int)
8507 long long __builtin_arm_wslld (long long, long long)
8508 long long __builtin_arm_wslldi (long long, int)
8509 v4hi __builtin_arm_wsllh (v4hi, long long)
8510 v4hi __builtin_arm_wsllhi (v4hi, int)
8511 v2si __builtin_arm_wsllw (v2si, long long)
8512 v2si __builtin_arm_wsllwi (v2si, int)
8513 long long __builtin_arm_wsrad (long long, long long)
8514 long long __builtin_arm_wsradi (long long, int)
8515 v4hi __builtin_arm_wsrah (v4hi, long long)
8516 v4hi __builtin_arm_wsrahi (v4hi, int)
8517 v2si __builtin_arm_wsraw (v2si, long long)
8518 v2si __builtin_arm_wsrawi (v2si, int)
8519 long long __builtin_arm_wsrld (long long, long long)
8520 long long __builtin_arm_wsrldi (long long, int)
8521 v4hi __builtin_arm_wsrlh (v4hi, long long)
8522 v4hi __builtin_arm_wsrlhi (v4hi, int)
8523 v2si __builtin_arm_wsrlw (v2si, long long)
8524 v2si __builtin_arm_wsrlwi (v2si, int)
8525 v8qi __builtin_arm_wsubb (v8qi, v8qi)
8526 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
8527 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
8528 v4hi __builtin_arm_wsubh (v4hi, v4hi)
8529 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
8530 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
8531 v2si __builtin_arm_wsubw (v2si, v2si)
8532 v2si __builtin_arm_wsubwss (v2si, v2si)
8533 v2si __builtin_arm_wsubwus (v2si, v2si)
8534 v4hi __builtin_arm_wunpckehsb (v8qi)
8535 v2si __builtin_arm_wunpckehsh (v4hi)
8536 long long __builtin_arm_wunpckehsw (v2si)
8537 v4hi __builtin_arm_wunpckehub (v8qi)
8538 v2si __builtin_arm_wunpckehuh (v4hi)
8539 long long __builtin_arm_wunpckehuw (v2si)
8540 v4hi __builtin_arm_wunpckelsb (v8qi)
8541 v2si __builtin_arm_wunpckelsh (v4hi)
8542 long long __builtin_arm_wunpckelsw (v2si)
8543 v4hi __builtin_arm_wunpckelub (v8qi)
8544 v2si __builtin_arm_wunpckeluh (v4hi)
8545 long long __builtin_arm_wunpckeluw (v2si)
8546 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
8547 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
8548 v2si __builtin_arm_wunpckihw (v2si, v2si)
8549 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
8550 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
8551 v2si __builtin_arm_wunpckilw (v2si, v2si)
8552 long long __builtin_arm_wxor (long long, long long)
8553 long long __builtin_arm_wzero ()
8556 @node ARM NEON Intrinsics
8557 @subsection ARM NEON Intrinsics
8559 These built-in intrinsics for the ARM Advanced SIMD extension are available
8560 when the @option{-mfpu=neon} switch is used:
8562 @include arm-neon-intrinsics.texi
8564 @node AVR Built-in Functions
8565 @subsection AVR Built-in Functions
8567 For each built-in function for AVR, there is an equally named,
8568 uppercase built-in macro defined. That way users can easily query if
8569 or if not a specific built-in is implemented or not. For example, if
8570 @code{__builtin_avr_nop} is available the macro
8571 @code{__BUILTIN_AVR_NOP} is defined to @code{1} and undefined otherwise.
8573 The following built-in functions map to the respective machine
8574 instruction, i.e. @code{nop}, @code{sei}, @code{cli}, @code{sleep},
8575 @code{wdr}, @code{swap}, @code{fmul}, @code{fmuls}
8576 resp. @code{fmulsu}. The three @code{fmul*} built-ins are implemented
8577 as library call if no hardware multiplier is available.
8580 void __builtin_avr_nop (void)
8581 void __builtin_avr_sei (void)
8582 void __builtin_avr_cli (void)
8583 void __builtin_avr_sleep (void)
8584 void __builtin_avr_wdr (void)
8585 unsigned char __builtin_avr_swap (unsigned char)
8586 unsigned int __builtin_avr_fmul (unsigned char, unsigned char)
8587 int __builtin_avr_fmuls (char, char)
8588 int __builtin_avr_fmulsu (char, unsigned char)
8591 In order to delay execution for a specific number of cycles, GCC
8594 void __builtin_avr_delay_cycles (unsigned long ticks)
8598 @code{ticks} is the number of ticks to delay execution. Note that this
8599 built-in does not take into account the effect of interrupts which
8600 might increase delay time. @code{ticks} must be a compile time
8601 integer constant; delays with a variable number of cycles are not supported.
8604 unsigned char __builtin_avr_map8 (unsigned long map, unsigned char val)
8608 Each bit of the result is copied from a specific bit of @code{val}.
8609 @code{map} is a compile time constant that represents a map composed
8610 of 8 nibbles (4-bit groups):
8611 The @var{n}-th nibble of @code{map} specifies which bit of @code{val}
8612 is to be moved to the @var{n}-th bit of the result.
8613 For example, @code{map = 0x76543210} represents identity: The MSB of
8614 the result is read from the 7-th bit of @code{val}, the LSB is
8615 read from the 0-th bit to @code{val}, etc.
8616 Two more examples: @code{0x01234567} reverses the bit order and
8617 @code{0x32107654} is equivalent to a @code{swap} instruction.
8620 One typical use case for this and the following built-in is adjusting input and
8621 output values to non-contiguous port layouts.
8624 unsigned int __builtin_avr_map16 (unsigned long long map, unsigned int val)
8628 Similar to the previous built-in except that it operates on @code{int}
8629 and thus 16 bits are involved. Again, @code{map} must be a compile
8632 @node Blackfin Built-in Functions
8633 @subsection Blackfin Built-in Functions
8635 Currently, there are two Blackfin-specific built-in functions. These are
8636 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
8637 using inline assembly; by using these built-in functions the compiler can
8638 automatically add workarounds for hardware errata involving these
8639 instructions. These functions are named as follows:
8642 void __builtin_bfin_csync (void)
8643 void __builtin_bfin_ssync (void)
8646 @node FR-V Built-in Functions
8647 @subsection FR-V Built-in Functions
8649 GCC provides many FR-V-specific built-in functions. In general,
8650 these functions are intended to be compatible with those described
8651 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
8652 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
8653 @code{__MBTOHE}, the gcc forms of which pass 128-bit values by
8654 pointer rather than by value.
8656 Most of the functions are named after specific FR-V instructions.
8657 Such functions are said to be ``directly mapped'' and are summarized
8658 here in tabular form.
8662 * Directly-mapped Integer Functions::
8663 * Directly-mapped Media Functions::
8664 * Raw read/write Functions::
8665 * Other Built-in Functions::
8668 @node Argument Types
8669 @subsubsection Argument Types
8671 The arguments to the built-in functions can be divided into three groups:
8672 register numbers, compile-time constants and run-time values. In order
8673 to make this classification clear at a glance, the arguments and return
8674 values are given the following pseudo types:
8676 @multitable @columnfractions .20 .30 .15 .35
8677 @item Pseudo type @tab Real C type @tab Constant? @tab Description
8678 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
8679 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
8680 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
8681 @item @code{uw2} @tab @code{unsigned long long} @tab No
8682 @tab an unsigned doubleword
8683 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
8684 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
8685 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
8686 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
8689 These pseudo types are not defined by GCC, they are simply a notational
8690 convenience used in this manual.
8692 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
8693 and @code{sw2} are evaluated at run time. They correspond to
8694 register operands in the underlying FR-V instructions.
8696 @code{const} arguments represent immediate operands in the underlying
8697 FR-V instructions. They must be compile-time constants.
8699 @code{acc} arguments are evaluated at compile time and specify the number
8700 of an accumulator register. For example, an @code{acc} argument of 2
8701 will select the ACC2 register.
8703 @code{iacc} arguments are similar to @code{acc} arguments but specify the
8704 number of an IACC register. See @pxref{Other Built-in Functions}
8707 @node Directly-mapped Integer Functions
8708 @subsubsection Directly-mapped Integer Functions
8710 The functions listed below map directly to FR-V I-type instructions.
8712 @multitable @columnfractions .45 .32 .23
8713 @item Function prototype @tab Example usage @tab Assembly output
8714 @item @code{sw1 __ADDSS (sw1, sw1)}
8715 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
8716 @tab @code{ADDSS @var{a},@var{b},@var{c}}
8717 @item @code{sw1 __SCAN (sw1, sw1)}
8718 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
8719 @tab @code{SCAN @var{a},@var{b},@var{c}}
8720 @item @code{sw1 __SCUTSS (sw1)}
8721 @tab @code{@var{b} = __SCUTSS (@var{a})}
8722 @tab @code{SCUTSS @var{a},@var{b}}
8723 @item @code{sw1 __SLASS (sw1, sw1)}
8724 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
8725 @tab @code{SLASS @var{a},@var{b},@var{c}}
8726 @item @code{void __SMASS (sw1, sw1)}
8727 @tab @code{__SMASS (@var{a}, @var{b})}
8728 @tab @code{SMASS @var{a},@var{b}}
8729 @item @code{void __SMSSS (sw1, sw1)}
8730 @tab @code{__SMSSS (@var{a}, @var{b})}
8731 @tab @code{SMSSS @var{a},@var{b}}
8732 @item @code{void __SMU (sw1, sw1)}
8733 @tab @code{__SMU (@var{a}, @var{b})}
8734 @tab @code{SMU @var{a},@var{b}}
8735 @item @code{sw2 __SMUL (sw1, sw1)}
8736 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
8737 @tab @code{SMUL @var{a},@var{b},@var{c}}
8738 @item @code{sw1 __SUBSS (sw1, sw1)}
8739 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
8740 @tab @code{SUBSS @var{a},@var{b},@var{c}}
8741 @item @code{uw2 __UMUL (uw1, uw1)}
8742 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
8743 @tab @code{UMUL @var{a},@var{b},@var{c}}
8746 @node Directly-mapped Media Functions
8747 @subsubsection Directly-mapped Media Functions
8749 The functions listed below map directly to FR-V M-type instructions.
8751 @multitable @columnfractions .45 .32 .23
8752 @item Function prototype @tab Example usage @tab Assembly output
8753 @item @code{uw1 __MABSHS (sw1)}
8754 @tab @code{@var{b} = __MABSHS (@var{a})}
8755 @tab @code{MABSHS @var{a},@var{b}}
8756 @item @code{void __MADDACCS (acc, acc)}
8757 @tab @code{__MADDACCS (@var{b}, @var{a})}
8758 @tab @code{MADDACCS @var{a},@var{b}}
8759 @item @code{sw1 __MADDHSS (sw1, sw1)}
8760 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
8761 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
8762 @item @code{uw1 __MADDHUS (uw1, uw1)}
8763 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
8764 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
8765 @item @code{uw1 __MAND (uw1, uw1)}
8766 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
8767 @tab @code{MAND @var{a},@var{b},@var{c}}
8768 @item @code{void __MASACCS (acc, acc)}
8769 @tab @code{__MASACCS (@var{b}, @var{a})}
8770 @tab @code{MASACCS @var{a},@var{b}}
8771 @item @code{uw1 __MAVEH (uw1, uw1)}
8772 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
8773 @tab @code{MAVEH @var{a},@var{b},@var{c}}
8774 @item @code{uw2 __MBTOH (uw1)}
8775 @tab @code{@var{b} = __MBTOH (@var{a})}
8776 @tab @code{MBTOH @var{a},@var{b}}
8777 @item @code{void __MBTOHE (uw1 *, uw1)}
8778 @tab @code{__MBTOHE (&@var{b}, @var{a})}
8779 @tab @code{MBTOHE @var{a},@var{b}}
8780 @item @code{void __MCLRACC (acc)}
8781 @tab @code{__MCLRACC (@var{a})}
8782 @tab @code{MCLRACC @var{a}}
8783 @item @code{void __MCLRACCA (void)}
8784 @tab @code{__MCLRACCA ()}
8785 @tab @code{MCLRACCA}
8786 @item @code{uw1 __Mcop1 (uw1, uw1)}
8787 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
8788 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
8789 @item @code{uw1 __Mcop2 (uw1, uw1)}
8790 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
8791 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
8792 @item @code{uw1 __MCPLHI (uw2, const)}
8793 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
8794 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
8795 @item @code{uw1 __MCPLI (uw2, const)}
8796 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
8797 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
8798 @item @code{void __MCPXIS (acc, sw1, sw1)}
8799 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
8800 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
8801 @item @code{void __MCPXIU (acc, uw1, uw1)}
8802 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
8803 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
8804 @item @code{void __MCPXRS (acc, sw1, sw1)}
8805 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
8806 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
8807 @item @code{void __MCPXRU (acc, uw1, uw1)}
8808 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
8809 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
8810 @item @code{uw1 __MCUT (acc, uw1)}
8811 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
8812 @tab @code{MCUT @var{a},@var{b},@var{c}}
8813 @item @code{uw1 __MCUTSS (acc, sw1)}
8814 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
8815 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
8816 @item @code{void __MDADDACCS (acc, acc)}
8817 @tab @code{__MDADDACCS (@var{b}, @var{a})}
8818 @tab @code{MDADDACCS @var{a},@var{b}}
8819 @item @code{void __MDASACCS (acc, acc)}
8820 @tab @code{__MDASACCS (@var{b}, @var{a})}
8821 @tab @code{MDASACCS @var{a},@var{b}}
8822 @item @code{uw2 __MDCUTSSI (acc, const)}
8823 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
8824 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
8825 @item @code{uw2 __MDPACKH (uw2, uw2)}
8826 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
8827 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
8828 @item @code{uw2 __MDROTLI (uw2, const)}
8829 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
8830 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
8831 @item @code{void __MDSUBACCS (acc, acc)}
8832 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
8833 @tab @code{MDSUBACCS @var{a},@var{b}}
8834 @item @code{void __MDUNPACKH (uw1 *, uw2)}
8835 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
8836 @tab @code{MDUNPACKH @var{a},@var{b}}
8837 @item @code{uw2 __MEXPDHD (uw1, const)}
8838 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
8839 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
8840 @item @code{uw1 __MEXPDHW (uw1, const)}
8841 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
8842 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
8843 @item @code{uw1 __MHDSETH (uw1, const)}
8844 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
8845 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
8846 @item @code{sw1 __MHDSETS (const)}
8847 @tab @code{@var{b} = __MHDSETS (@var{a})}
8848 @tab @code{MHDSETS #@var{a},@var{b}}
8849 @item @code{uw1 __MHSETHIH (uw1, const)}
8850 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
8851 @tab @code{MHSETHIH #@var{a},@var{b}}
8852 @item @code{sw1 __MHSETHIS (sw1, const)}
8853 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
8854 @tab @code{MHSETHIS #@var{a},@var{b}}
8855 @item @code{uw1 __MHSETLOH (uw1, const)}
8856 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
8857 @tab @code{MHSETLOH #@var{a},@var{b}}
8858 @item @code{sw1 __MHSETLOS (sw1, const)}
8859 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
8860 @tab @code{MHSETLOS #@var{a},@var{b}}
8861 @item @code{uw1 __MHTOB (uw2)}
8862 @tab @code{@var{b} = __MHTOB (@var{a})}
8863 @tab @code{MHTOB @var{a},@var{b}}
8864 @item @code{void __MMACHS (acc, sw1, sw1)}
8865 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
8866 @tab @code{MMACHS @var{a},@var{b},@var{c}}
8867 @item @code{void __MMACHU (acc, uw1, uw1)}
8868 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
8869 @tab @code{MMACHU @var{a},@var{b},@var{c}}
8870 @item @code{void __MMRDHS (acc, sw1, sw1)}
8871 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
8872 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
8873 @item @code{void __MMRDHU (acc, uw1, uw1)}
8874 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
8875 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
8876 @item @code{void __MMULHS (acc, sw1, sw1)}
8877 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
8878 @tab @code{MMULHS @var{a},@var{b},@var{c}}
8879 @item @code{void __MMULHU (acc, uw1, uw1)}
8880 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
8881 @tab @code{MMULHU @var{a},@var{b},@var{c}}
8882 @item @code{void __MMULXHS (acc, sw1, sw1)}
8883 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
8884 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
8885 @item @code{void __MMULXHU (acc, uw1, uw1)}
8886 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
8887 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
8888 @item @code{uw1 __MNOT (uw1)}
8889 @tab @code{@var{b} = __MNOT (@var{a})}
8890 @tab @code{MNOT @var{a},@var{b}}
8891 @item @code{uw1 __MOR (uw1, uw1)}
8892 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
8893 @tab @code{MOR @var{a},@var{b},@var{c}}
8894 @item @code{uw1 __MPACKH (uh, uh)}
8895 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
8896 @tab @code{MPACKH @var{a},@var{b},@var{c}}
8897 @item @code{sw2 __MQADDHSS (sw2, sw2)}
8898 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
8899 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
8900 @item @code{uw2 __MQADDHUS (uw2, uw2)}
8901 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
8902 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
8903 @item @code{void __MQCPXIS (acc, sw2, sw2)}
8904 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
8905 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
8906 @item @code{void __MQCPXIU (acc, uw2, uw2)}
8907 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
8908 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
8909 @item @code{void __MQCPXRS (acc, sw2, sw2)}
8910 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
8911 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
8912 @item @code{void __MQCPXRU (acc, uw2, uw2)}
8913 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
8914 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
8915 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
8916 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
8917 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
8918 @item @code{sw2 __MQLMTHS (sw2, sw2)}
8919 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
8920 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
8921 @item @code{void __MQMACHS (acc, sw2, sw2)}
8922 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
8923 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
8924 @item @code{void __MQMACHU (acc, uw2, uw2)}
8925 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
8926 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
8927 @item @code{void __MQMACXHS (acc, sw2, sw2)}
8928 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
8929 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
8930 @item @code{void __MQMULHS (acc, sw2, sw2)}
8931 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
8932 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
8933 @item @code{void __MQMULHU (acc, uw2, uw2)}
8934 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
8935 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
8936 @item @code{void __MQMULXHS (acc, sw2, sw2)}
8937 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
8938 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
8939 @item @code{void __MQMULXHU (acc, uw2, uw2)}
8940 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
8941 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
8942 @item @code{sw2 __MQSATHS (sw2, sw2)}
8943 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
8944 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
8945 @item @code{uw2 __MQSLLHI (uw2, int)}
8946 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
8947 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
8948 @item @code{sw2 __MQSRAHI (sw2, int)}
8949 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
8950 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
8951 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
8952 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
8953 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
8954 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
8955 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
8956 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
8957 @item @code{void __MQXMACHS (acc, sw2, sw2)}
8958 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
8959 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
8960 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
8961 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
8962 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
8963 @item @code{uw1 __MRDACC (acc)}
8964 @tab @code{@var{b} = __MRDACC (@var{a})}
8965 @tab @code{MRDACC @var{a},@var{b}}
8966 @item @code{uw1 __MRDACCG (acc)}
8967 @tab @code{@var{b} = __MRDACCG (@var{a})}
8968 @tab @code{MRDACCG @var{a},@var{b}}
8969 @item @code{uw1 __MROTLI (uw1, const)}
8970 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
8971 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
8972 @item @code{uw1 __MROTRI (uw1, const)}
8973 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
8974 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
8975 @item @code{sw1 __MSATHS (sw1, sw1)}
8976 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
8977 @tab @code{MSATHS @var{a},@var{b},@var{c}}
8978 @item @code{uw1 __MSATHU (uw1, uw1)}
8979 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
8980 @tab @code{MSATHU @var{a},@var{b},@var{c}}
8981 @item @code{uw1 __MSLLHI (uw1, const)}
8982 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
8983 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
8984 @item @code{sw1 __MSRAHI (sw1, const)}
8985 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
8986 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
8987 @item @code{uw1 __MSRLHI (uw1, const)}
8988 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
8989 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
8990 @item @code{void __MSUBACCS (acc, acc)}
8991 @tab @code{__MSUBACCS (@var{b}, @var{a})}
8992 @tab @code{MSUBACCS @var{a},@var{b}}
8993 @item @code{sw1 __MSUBHSS (sw1, sw1)}
8994 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
8995 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
8996 @item @code{uw1 __MSUBHUS (uw1, uw1)}
8997 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
8998 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
8999 @item @code{void __MTRAP (void)}
9000 @tab @code{__MTRAP ()}
9002 @item @code{uw2 __MUNPACKH (uw1)}
9003 @tab @code{@var{b} = __MUNPACKH (@var{a})}
9004 @tab @code{MUNPACKH @var{a},@var{b}}
9005 @item @code{uw1 __MWCUT (uw2, uw1)}
9006 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
9007 @tab @code{MWCUT @var{a},@var{b},@var{c}}
9008 @item @code{void __MWTACC (acc, uw1)}
9009 @tab @code{__MWTACC (@var{b}, @var{a})}
9010 @tab @code{MWTACC @var{a},@var{b}}
9011 @item @code{void __MWTACCG (acc, uw1)}
9012 @tab @code{__MWTACCG (@var{b}, @var{a})}
9013 @tab @code{MWTACCG @var{a},@var{b}}
9014 @item @code{uw1 __MXOR (uw1, uw1)}
9015 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
9016 @tab @code{MXOR @var{a},@var{b},@var{c}}
9019 @node Raw read/write Functions
9020 @subsubsection Raw read/write Functions
9022 This sections describes built-in functions related to read and write
9023 instructions to access memory. These functions generate
9024 @code{membar} instructions to flush the I/O load and stores where
9025 appropriate, as described in Fujitsu's manual described above.
9029 @item unsigned char __builtin_read8 (void *@var{data})
9030 @item unsigned short __builtin_read16 (void *@var{data})
9031 @item unsigned long __builtin_read32 (void *@var{data})
9032 @item unsigned long long __builtin_read64 (void *@var{data})
9034 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
9035 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
9036 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
9037 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
9040 @node Other Built-in Functions
9041 @subsubsection Other Built-in Functions
9043 This section describes built-in functions that are not named after
9044 a specific FR-V instruction.
9047 @item sw2 __IACCreadll (iacc @var{reg})
9048 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
9049 for future expansion and must be 0.
9051 @item sw1 __IACCreadl (iacc @var{reg})
9052 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
9053 Other values of @var{reg} are rejected as invalid.
9055 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
9056 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
9057 is reserved for future expansion and must be 0.
9059 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
9060 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
9061 is 1. Other values of @var{reg} are rejected as invalid.
9063 @item void __data_prefetch0 (const void *@var{x})
9064 Use the @code{dcpl} instruction to load the contents of address @var{x}
9065 into the data cache.
9067 @item void __data_prefetch (const void *@var{x})
9068 Use the @code{nldub} instruction to load the contents of address @var{x}
9069 into the data cache. The instruction will be issued in slot I1@.
9072 @node X86 Built-in Functions
9073 @subsection X86 Built-in Functions
9075 These built-in functions are available for the i386 and x86-64 family
9076 of computers, depending on the command-line switches used.
9078 Note that, if you specify command-line switches such as @option{-msse},
9079 the compiler could use the extended instruction sets even if the built-ins
9080 are not used explicitly in the program. For this reason, applications
9081 which perform runtime CPU detection must compile separate files for each
9082 supported architecture, using the appropriate flags. In particular,
9083 the file containing the CPU detection code should be compiled without
9086 The following machine modes are available for use with MMX built-in functions
9087 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
9088 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
9089 vector of eight 8-bit integers. Some of the built-in functions operate on
9090 MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode.
9092 If 3DNow!@: extensions are enabled, @code{V2SF} is used as a mode for a vector
9093 of two 32-bit floating point values.
9095 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
9096 floating point values. Some instructions use a vector of four 32-bit
9097 integers, these use @code{V4SI}. Finally, some instructions operate on an
9098 entire vector register, interpreting it as a 128-bit integer, these use mode
9101 In 64-bit mode, the x86-64 family of processors uses additional built-in
9102 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
9103 floating point and @code{TC} 128-bit complex floating point values.
9105 The following floating point built-in functions are available in 64-bit
9106 mode. All of them implement the function that is part of the name.
9109 __float128 __builtin_fabsq (__float128)
9110 __float128 __builtin_copysignq (__float128, __float128)
9113 The following built-in function is always available.
9116 @item void __builtin_ia32_pause (void)
9117 Generates the @code{pause} machine instruction with a compiler memory
9121 The following floating point built-in functions are made available in the
9125 @item __float128 __builtin_infq (void)
9126 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
9127 @findex __builtin_infq
9129 @item __float128 __builtin_huge_valq (void)
9130 Similar to @code{__builtin_huge_val}, except the return type is @code{__float128}.
9131 @findex __builtin_huge_valq
9134 The following built-in functions are made available by @option{-mmmx}.
9135 All of them generate the machine instruction that is part of the name.
9138 v8qi __builtin_ia32_paddb (v8qi, v8qi)
9139 v4hi __builtin_ia32_paddw (v4hi, v4hi)
9140 v2si __builtin_ia32_paddd (v2si, v2si)
9141 v8qi __builtin_ia32_psubb (v8qi, v8qi)
9142 v4hi __builtin_ia32_psubw (v4hi, v4hi)
9143 v2si __builtin_ia32_psubd (v2si, v2si)
9144 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
9145 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
9146 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
9147 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
9148 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
9149 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
9150 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
9151 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
9152 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
9153 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
9154 di __builtin_ia32_pand (di, di)
9155 di __builtin_ia32_pandn (di,di)
9156 di __builtin_ia32_por (di, di)
9157 di __builtin_ia32_pxor (di, di)
9158 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
9159 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
9160 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
9161 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
9162 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
9163 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
9164 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
9165 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
9166 v2si __builtin_ia32_punpckhdq (v2si, v2si)
9167 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
9168 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
9169 v2si __builtin_ia32_punpckldq (v2si, v2si)
9170 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
9171 v4hi __builtin_ia32_packssdw (v2si, v2si)
9172 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
9174 v4hi __builtin_ia32_psllw (v4hi, v4hi)
9175 v2si __builtin_ia32_pslld (v2si, v2si)
9176 v1di __builtin_ia32_psllq (v1di, v1di)
9177 v4hi __builtin_ia32_psrlw (v4hi, v4hi)
9178 v2si __builtin_ia32_psrld (v2si, v2si)
9179 v1di __builtin_ia32_psrlq (v1di, v1di)
9180 v4hi __builtin_ia32_psraw (v4hi, v4hi)
9181 v2si __builtin_ia32_psrad (v2si, v2si)
9182 v4hi __builtin_ia32_psllwi (v4hi, int)
9183 v2si __builtin_ia32_pslldi (v2si, int)
9184 v1di __builtin_ia32_psllqi (v1di, int)
9185 v4hi __builtin_ia32_psrlwi (v4hi, int)
9186 v2si __builtin_ia32_psrldi (v2si, int)
9187 v1di __builtin_ia32_psrlqi (v1di, int)
9188 v4hi __builtin_ia32_psrawi (v4hi, int)
9189 v2si __builtin_ia32_psradi (v2si, int)
9193 The following built-in functions are made available either with
9194 @option{-msse}, or with a combination of @option{-m3dnow} and
9195 @option{-march=athlon}. All of them generate the machine
9196 instruction that is part of the name.
9199 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
9200 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
9201 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
9202 v1di __builtin_ia32_psadbw (v8qi, v8qi)
9203 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
9204 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
9205 v8qi __builtin_ia32_pminub (v8qi, v8qi)
9206 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
9207 int __builtin_ia32_pextrw (v4hi, int)
9208 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
9209 int __builtin_ia32_pmovmskb (v8qi)
9210 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
9211 void __builtin_ia32_movntq (di *, di)
9212 void __builtin_ia32_sfence (void)
9215 The following built-in functions are available when @option{-msse} is used.
9216 All of them generate the machine instruction that is part of the name.
9219 int __builtin_ia32_comieq (v4sf, v4sf)
9220 int __builtin_ia32_comineq (v4sf, v4sf)
9221 int __builtin_ia32_comilt (v4sf, v4sf)
9222 int __builtin_ia32_comile (v4sf, v4sf)
9223 int __builtin_ia32_comigt (v4sf, v4sf)
9224 int __builtin_ia32_comige (v4sf, v4sf)
9225 int __builtin_ia32_ucomieq (v4sf, v4sf)
9226 int __builtin_ia32_ucomineq (v4sf, v4sf)
9227 int __builtin_ia32_ucomilt (v4sf, v4sf)
9228 int __builtin_ia32_ucomile (v4sf, v4sf)
9229 int __builtin_ia32_ucomigt (v4sf, v4sf)
9230 int __builtin_ia32_ucomige (v4sf, v4sf)
9231 v4sf __builtin_ia32_addps (v4sf, v4sf)
9232 v4sf __builtin_ia32_subps (v4sf, v4sf)
9233 v4sf __builtin_ia32_mulps (v4sf, v4sf)
9234 v4sf __builtin_ia32_divps (v4sf, v4sf)
9235 v4sf __builtin_ia32_addss (v4sf, v4sf)
9236 v4sf __builtin_ia32_subss (v4sf, v4sf)
9237 v4sf __builtin_ia32_mulss (v4sf, v4sf)
9238 v4sf __builtin_ia32_divss (v4sf, v4sf)
9239 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
9240 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
9241 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
9242 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
9243 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
9244 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
9245 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
9246 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
9247 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
9248 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
9249 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
9250 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
9251 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
9252 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
9253 v4si __builtin_ia32_cmpless (v4sf, v4sf)
9254 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
9255 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
9256 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
9257 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
9258 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
9259 v4sf __builtin_ia32_maxps (v4sf, v4sf)
9260 v4sf __builtin_ia32_maxss (v4sf, v4sf)
9261 v4sf __builtin_ia32_minps (v4sf, v4sf)
9262 v4sf __builtin_ia32_minss (v4sf, v4sf)
9263 v4sf __builtin_ia32_andps (v4sf, v4sf)
9264 v4sf __builtin_ia32_andnps (v4sf, v4sf)
9265 v4sf __builtin_ia32_orps (v4sf, v4sf)
9266 v4sf __builtin_ia32_xorps (v4sf, v4sf)
9267 v4sf __builtin_ia32_movss (v4sf, v4sf)
9268 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
9269 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
9270 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
9271 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
9272 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
9273 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
9274 v2si __builtin_ia32_cvtps2pi (v4sf)
9275 int __builtin_ia32_cvtss2si (v4sf)
9276 v2si __builtin_ia32_cvttps2pi (v4sf)
9277 int __builtin_ia32_cvttss2si (v4sf)
9278 v4sf __builtin_ia32_rcpps (v4sf)
9279 v4sf __builtin_ia32_rsqrtps (v4sf)
9280 v4sf __builtin_ia32_sqrtps (v4sf)
9281 v4sf __builtin_ia32_rcpss (v4sf)
9282 v4sf __builtin_ia32_rsqrtss (v4sf)
9283 v4sf __builtin_ia32_sqrtss (v4sf)
9284 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
9285 void __builtin_ia32_movntps (float *, v4sf)
9286 int __builtin_ia32_movmskps (v4sf)
9289 The following built-in functions are available when @option{-msse} is used.
9292 @item v4sf __builtin_ia32_loadaps (float *)
9293 Generates the @code{movaps} machine instruction as a load from memory.
9294 @item void __builtin_ia32_storeaps (float *, v4sf)
9295 Generates the @code{movaps} machine instruction as a store to memory.
9296 @item v4sf __builtin_ia32_loadups (float *)
9297 Generates the @code{movups} machine instruction as a load from memory.
9298 @item void __builtin_ia32_storeups (float *, v4sf)
9299 Generates the @code{movups} machine instruction as a store to memory.
9300 @item v4sf __builtin_ia32_loadsss (float *)
9301 Generates the @code{movss} machine instruction as a load from memory.
9302 @item void __builtin_ia32_storess (float *, v4sf)
9303 Generates the @code{movss} machine instruction as a store to memory.
9304 @item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)
9305 Generates the @code{movhps} machine instruction as a load from memory.
9306 @item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)
9307 Generates the @code{movlps} machine instruction as a load from memory
9308 @item void __builtin_ia32_storehps (v2sf *, v4sf)
9309 Generates the @code{movhps} machine instruction as a store to memory.
9310 @item void __builtin_ia32_storelps (v2sf *, v4sf)
9311 Generates the @code{movlps} machine instruction as a store to memory.
9314 The following built-in functions are available when @option{-msse2} is used.
9315 All of them generate the machine instruction that is part of the name.
9318 int __builtin_ia32_comisdeq (v2df, v2df)
9319 int __builtin_ia32_comisdlt (v2df, v2df)
9320 int __builtin_ia32_comisdle (v2df, v2df)
9321 int __builtin_ia32_comisdgt (v2df, v2df)
9322 int __builtin_ia32_comisdge (v2df, v2df)
9323 int __builtin_ia32_comisdneq (v2df, v2df)
9324 int __builtin_ia32_ucomisdeq (v2df, v2df)
9325 int __builtin_ia32_ucomisdlt (v2df, v2df)
9326 int __builtin_ia32_ucomisdle (v2df, v2df)
9327 int __builtin_ia32_ucomisdgt (v2df, v2df)
9328 int __builtin_ia32_ucomisdge (v2df, v2df)
9329 int __builtin_ia32_ucomisdneq (v2df, v2df)
9330 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
9331 v2df __builtin_ia32_cmpltpd (v2df, v2df)
9332 v2df __builtin_ia32_cmplepd (v2df, v2df)
9333 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
9334 v2df __builtin_ia32_cmpgepd (v2df, v2df)
9335 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
9336 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
9337 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
9338 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
9339 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
9340 v2df __builtin_ia32_cmpngepd (v2df, v2df)
9341 v2df __builtin_ia32_cmpordpd (v2df, v2df)
9342 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
9343 v2df __builtin_ia32_cmpltsd (v2df, v2df)
9344 v2df __builtin_ia32_cmplesd (v2df, v2df)
9345 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
9346 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
9347 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
9348 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
9349 v2df __builtin_ia32_cmpordsd (v2df, v2df)
9350 v2di __builtin_ia32_paddq (v2di, v2di)
9351 v2di __builtin_ia32_psubq (v2di, v2di)
9352 v2df __builtin_ia32_addpd (v2df, v2df)
9353 v2df __builtin_ia32_subpd (v2df, v2df)
9354 v2df __builtin_ia32_mulpd (v2df, v2df)
9355 v2df __builtin_ia32_divpd (v2df, v2df)
9356 v2df __builtin_ia32_addsd (v2df, v2df)
9357 v2df __builtin_ia32_subsd (v2df, v2df)
9358 v2df __builtin_ia32_mulsd (v2df, v2df)
9359 v2df __builtin_ia32_divsd (v2df, v2df)
9360 v2df __builtin_ia32_minpd (v2df, v2df)
9361 v2df __builtin_ia32_maxpd (v2df, v2df)
9362 v2df __builtin_ia32_minsd (v2df, v2df)
9363 v2df __builtin_ia32_maxsd (v2df, v2df)
9364 v2df __builtin_ia32_andpd (v2df, v2df)
9365 v2df __builtin_ia32_andnpd (v2df, v2df)
9366 v2df __builtin_ia32_orpd (v2df, v2df)
9367 v2df __builtin_ia32_xorpd (v2df, v2df)
9368 v2df __builtin_ia32_movsd (v2df, v2df)
9369 v2df __builtin_ia32_unpckhpd (v2df, v2df)
9370 v2df __builtin_ia32_unpcklpd (v2df, v2df)
9371 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
9372 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
9373 v4si __builtin_ia32_paddd128 (v4si, v4si)
9374 v2di __builtin_ia32_paddq128 (v2di, v2di)
9375 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
9376 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
9377 v4si __builtin_ia32_psubd128 (v4si, v4si)
9378 v2di __builtin_ia32_psubq128 (v2di, v2di)
9379 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
9380 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
9381 v2di __builtin_ia32_pand128 (v2di, v2di)
9382 v2di __builtin_ia32_pandn128 (v2di, v2di)
9383 v2di __builtin_ia32_por128 (v2di, v2di)
9384 v2di __builtin_ia32_pxor128 (v2di, v2di)
9385 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
9386 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
9387 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
9388 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
9389 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
9390 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
9391 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
9392 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
9393 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
9394 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
9395 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
9396 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
9397 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
9398 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
9399 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
9400 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
9401 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
9402 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
9403 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
9404 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
9405 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi)
9406 v8hi __builtin_ia32_packssdw128 (v4si, v4si)
9407 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi)
9408 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
9409 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
9410 v2df __builtin_ia32_loadupd (double *)
9411 void __builtin_ia32_storeupd (double *, v2df)
9412 v2df __builtin_ia32_loadhpd (v2df, double const *)
9413 v2df __builtin_ia32_loadlpd (v2df, double const *)
9414 int __builtin_ia32_movmskpd (v2df)
9415 int __builtin_ia32_pmovmskb128 (v16qi)
9416 void __builtin_ia32_movnti (int *, int)
9417 void __builtin_ia32_movnti64 (long long int *, long long int)
9418 void __builtin_ia32_movntpd (double *, v2df)
9419 void __builtin_ia32_movntdq (v2df *, v2df)
9420 v4si __builtin_ia32_pshufd (v4si, int)
9421 v8hi __builtin_ia32_pshuflw (v8hi, int)
9422 v8hi __builtin_ia32_pshufhw (v8hi, int)
9423 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
9424 v2df __builtin_ia32_sqrtpd (v2df)
9425 v2df __builtin_ia32_sqrtsd (v2df)
9426 v2df __builtin_ia32_shufpd (v2df, v2df, int)
9427 v2df __builtin_ia32_cvtdq2pd (v4si)
9428 v4sf __builtin_ia32_cvtdq2ps (v4si)
9429 v4si __builtin_ia32_cvtpd2dq (v2df)
9430 v2si __builtin_ia32_cvtpd2pi (v2df)
9431 v4sf __builtin_ia32_cvtpd2ps (v2df)
9432 v4si __builtin_ia32_cvttpd2dq (v2df)
9433 v2si __builtin_ia32_cvttpd2pi (v2df)
9434 v2df __builtin_ia32_cvtpi2pd (v2si)
9435 int __builtin_ia32_cvtsd2si (v2df)
9436 int __builtin_ia32_cvttsd2si (v2df)
9437 long long __builtin_ia32_cvtsd2si64 (v2df)
9438 long long __builtin_ia32_cvttsd2si64 (v2df)
9439 v4si __builtin_ia32_cvtps2dq (v4sf)
9440 v2df __builtin_ia32_cvtps2pd (v4sf)
9441 v4si __builtin_ia32_cvttps2dq (v4sf)
9442 v2df __builtin_ia32_cvtsi2sd (v2df, int)
9443 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
9444 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
9445 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
9446 void __builtin_ia32_clflush (const void *)
9447 void __builtin_ia32_lfence (void)
9448 void __builtin_ia32_mfence (void)
9449 v16qi __builtin_ia32_loaddqu (const char *)
9450 void __builtin_ia32_storedqu (char *, v16qi)
9451 v1di __builtin_ia32_pmuludq (v2si, v2si)
9452 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
9453 v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
9454 v4si __builtin_ia32_pslld128 (v4si, v4si)
9455 v2di __builtin_ia32_psllq128 (v2di, v2di)
9456 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
9457 v4si __builtin_ia32_psrld128 (v4si, v4si)
9458 v2di __builtin_ia32_psrlq128 (v2di, v2di)
9459 v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
9460 v4si __builtin_ia32_psrad128 (v4si, v4si)
9461 v2di __builtin_ia32_pslldqi128 (v2di, int)
9462 v8hi __builtin_ia32_psllwi128 (v8hi, int)
9463 v4si __builtin_ia32_pslldi128 (v4si, int)
9464 v2di __builtin_ia32_psllqi128 (v2di, int)
9465 v2di __builtin_ia32_psrldqi128 (v2di, int)
9466 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
9467 v4si __builtin_ia32_psrldi128 (v4si, int)
9468 v2di __builtin_ia32_psrlqi128 (v2di, int)
9469 v8hi __builtin_ia32_psrawi128 (v8hi, int)
9470 v4si __builtin_ia32_psradi128 (v4si, int)
9471 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
9472 v2di __builtin_ia32_movq128 (v2di)
9475 The following built-in functions are available when @option{-msse3} is used.
9476 All of them generate the machine instruction that is part of the name.
9479 v2df __builtin_ia32_addsubpd (v2df, v2df)
9480 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
9481 v2df __builtin_ia32_haddpd (v2df, v2df)
9482 v4sf __builtin_ia32_haddps (v4sf, v4sf)
9483 v2df __builtin_ia32_hsubpd (v2df, v2df)
9484 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
9485 v16qi __builtin_ia32_lddqu (char const *)
9486 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
9487 v2df __builtin_ia32_movddup (v2df)
9488 v4sf __builtin_ia32_movshdup (v4sf)
9489 v4sf __builtin_ia32_movsldup (v4sf)
9490 void __builtin_ia32_mwait (unsigned int, unsigned int)
9493 The following built-in functions are available when @option{-msse3} is used.
9496 @item v2df __builtin_ia32_loadddup (double const *)
9497 Generates the @code{movddup} machine instruction as a load from memory.
9500 The following built-in functions are available when @option{-mssse3} is used.
9501 All of them generate the machine instruction that is part of the name
9505 v2si __builtin_ia32_phaddd (v2si, v2si)
9506 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
9507 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
9508 v2si __builtin_ia32_phsubd (v2si, v2si)
9509 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
9510 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
9511 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi)
9512 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
9513 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
9514 v8qi __builtin_ia32_psignb (v8qi, v8qi)
9515 v2si __builtin_ia32_psignd (v2si, v2si)
9516 v4hi __builtin_ia32_psignw (v4hi, v4hi)
9517 v1di __builtin_ia32_palignr (v1di, v1di, int)
9518 v8qi __builtin_ia32_pabsb (v8qi)
9519 v2si __builtin_ia32_pabsd (v2si)
9520 v4hi __builtin_ia32_pabsw (v4hi)
9523 The following built-in functions are available when @option{-mssse3} is used.
9524 All of them generate the machine instruction that is part of the name
9528 v4si __builtin_ia32_phaddd128 (v4si, v4si)
9529 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
9530 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
9531 v4si __builtin_ia32_phsubd128 (v4si, v4si)
9532 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
9533 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
9534 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
9535 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
9536 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
9537 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
9538 v4si __builtin_ia32_psignd128 (v4si, v4si)
9539 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
9540 v2di __builtin_ia32_palignr128 (v2di, v2di, int)
9541 v16qi __builtin_ia32_pabsb128 (v16qi)
9542 v4si __builtin_ia32_pabsd128 (v4si)
9543 v8hi __builtin_ia32_pabsw128 (v8hi)
9546 The following built-in functions are available when @option{-msse4.1} is
9547 used. All of them generate the machine instruction that is part of the
9551 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
9552 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
9553 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
9554 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
9555 v2df __builtin_ia32_dppd (v2df, v2df, const int)
9556 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
9557 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
9558 v2di __builtin_ia32_movntdqa (v2di *);
9559 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
9560 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
9561 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
9562 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
9563 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
9564 v8hi __builtin_ia32_phminposuw128 (v8hi)
9565 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
9566 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
9567 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
9568 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
9569 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
9570 v4si __builtin_ia32_pminsd128 (v4si, v4si)
9571 v4si __builtin_ia32_pminud128 (v4si, v4si)
9572 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
9573 v4si __builtin_ia32_pmovsxbd128 (v16qi)
9574 v2di __builtin_ia32_pmovsxbq128 (v16qi)
9575 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
9576 v2di __builtin_ia32_pmovsxdq128 (v4si)
9577 v4si __builtin_ia32_pmovsxwd128 (v8hi)
9578 v2di __builtin_ia32_pmovsxwq128 (v8hi)
9579 v4si __builtin_ia32_pmovzxbd128 (v16qi)
9580 v2di __builtin_ia32_pmovzxbq128 (v16qi)
9581 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
9582 v2di __builtin_ia32_pmovzxdq128 (v4si)
9583 v4si __builtin_ia32_pmovzxwd128 (v8hi)
9584 v2di __builtin_ia32_pmovzxwq128 (v8hi)
9585 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
9586 v4si __builtin_ia32_pmulld128 (v4si, v4si)
9587 int __builtin_ia32_ptestc128 (v2di, v2di)
9588 int __builtin_ia32_ptestnzc128 (v2di, v2di)
9589 int __builtin_ia32_ptestz128 (v2di, v2di)
9590 v2df __builtin_ia32_roundpd (v2df, const int)
9591 v4sf __builtin_ia32_roundps (v4sf, const int)
9592 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
9593 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
9596 The following built-in functions are available when @option{-msse4.1} is
9600 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
9601 Generates the @code{insertps} machine instruction.
9602 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
9603 Generates the @code{pextrb} machine instruction.
9604 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
9605 Generates the @code{pinsrb} machine instruction.
9606 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
9607 Generates the @code{pinsrd} machine instruction.
9608 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
9609 Generates the @code{pinsrq} machine instruction in 64bit mode.
9612 The following built-in functions are changed to generate new SSE4.1
9613 instructions when @option{-msse4.1} is used.
9616 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
9617 Generates the @code{extractps} machine instruction.
9618 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
9619 Generates the @code{pextrd} machine instruction.
9620 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
9621 Generates the @code{pextrq} machine instruction in 64bit mode.
9624 The following built-in functions are available when @option{-msse4.2} is
9625 used. All of them generate the machine instruction that is part of the
9629 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
9630 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
9631 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
9632 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
9633 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
9634 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
9635 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
9636 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
9637 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
9638 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
9639 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
9640 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
9641 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
9642 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
9643 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
9646 The following built-in functions are available when @option{-msse4.2} is
9650 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
9651 Generates the @code{crc32b} machine instruction.
9652 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
9653 Generates the @code{crc32w} machine instruction.
9654 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
9655 Generates the @code{crc32l} machine instruction.
9656 @item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)
9657 Generates the @code{crc32q} machine instruction.
9660 The following built-in functions are changed to generate new SSE4.2
9661 instructions when @option{-msse4.2} is used.
9664 @item int __builtin_popcount (unsigned int)
9665 Generates the @code{popcntl} machine instruction.
9666 @item int __builtin_popcountl (unsigned long)
9667 Generates the @code{popcntl} or @code{popcntq} machine instruction,
9668 depending on the size of @code{unsigned long}.
9669 @item int __builtin_popcountll (unsigned long long)
9670 Generates the @code{popcntq} machine instruction.
9673 The following built-in functions are available when @option{-mavx} is
9674 used. All of them generate the machine instruction that is part of the
9678 v4df __builtin_ia32_addpd256 (v4df,v4df)
9679 v8sf __builtin_ia32_addps256 (v8sf,v8sf)
9680 v4df __builtin_ia32_addsubpd256 (v4df,v4df)
9681 v8sf __builtin_ia32_addsubps256 (v8sf,v8sf)
9682 v4df __builtin_ia32_andnpd256 (v4df,v4df)
9683 v8sf __builtin_ia32_andnps256 (v8sf,v8sf)
9684 v4df __builtin_ia32_andpd256 (v4df,v4df)
9685 v8sf __builtin_ia32_andps256 (v8sf,v8sf)
9686 v4df __builtin_ia32_blendpd256 (v4df,v4df,int)
9687 v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int)
9688 v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df)
9689 v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf)
9690 v2df __builtin_ia32_cmppd (v2df,v2df,int)
9691 v4df __builtin_ia32_cmppd256 (v4df,v4df,int)
9692 v4sf __builtin_ia32_cmpps (v4sf,v4sf,int)
9693 v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int)
9694 v2df __builtin_ia32_cmpsd (v2df,v2df,int)
9695 v4sf __builtin_ia32_cmpss (v4sf,v4sf,int)
9696 v4df __builtin_ia32_cvtdq2pd256 (v4si)
9697 v8sf __builtin_ia32_cvtdq2ps256 (v8si)
9698 v4si __builtin_ia32_cvtpd2dq256 (v4df)
9699 v4sf __builtin_ia32_cvtpd2ps256 (v4df)
9700 v8si __builtin_ia32_cvtps2dq256 (v8sf)
9701 v4df __builtin_ia32_cvtps2pd256 (v4sf)
9702 v4si __builtin_ia32_cvttpd2dq256 (v4df)
9703 v8si __builtin_ia32_cvttps2dq256 (v8sf)
9704 v4df __builtin_ia32_divpd256 (v4df,v4df)
9705 v8sf __builtin_ia32_divps256 (v8sf,v8sf)
9706 v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int)
9707 v4df __builtin_ia32_haddpd256 (v4df,v4df)
9708 v8sf __builtin_ia32_haddps256 (v8sf,v8sf)
9709 v4df __builtin_ia32_hsubpd256 (v4df,v4df)
9710 v8sf __builtin_ia32_hsubps256 (v8sf,v8sf)
9711 v32qi __builtin_ia32_lddqu256 (pcchar)
9712 v32qi __builtin_ia32_loaddqu256 (pcchar)
9713 v4df __builtin_ia32_loadupd256 (pcdouble)
9714 v8sf __builtin_ia32_loadups256 (pcfloat)
9715 v2df __builtin_ia32_maskloadpd (pcv2df,v2df)
9716 v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df)
9717 v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf)
9718 v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf)
9719 void __builtin_ia32_maskstorepd (pv2df,v2df,v2df)
9720 void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df)
9721 void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf)
9722 void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf)
9723 v4df __builtin_ia32_maxpd256 (v4df,v4df)
9724 v8sf __builtin_ia32_maxps256 (v8sf,v8sf)
9725 v4df __builtin_ia32_minpd256 (v4df,v4df)
9726 v8sf __builtin_ia32_minps256 (v8sf,v8sf)
9727 v4df __builtin_ia32_movddup256 (v4df)
9728 int __builtin_ia32_movmskpd256 (v4df)
9729 int __builtin_ia32_movmskps256 (v8sf)
9730 v8sf __builtin_ia32_movshdup256 (v8sf)
9731 v8sf __builtin_ia32_movsldup256 (v8sf)
9732 v4df __builtin_ia32_mulpd256 (v4df,v4df)
9733 v8sf __builtin_ia32_mulps256 (v8sf,v8sf)
9734 v4df __builtin_ia32_orpd256 (v4df,v4df)
9735 v8sf __builtin_ia32_orps256 (v8sf,v8sf)
9736 v2df __builtin_ia32_pd_pd256 (v4df)
9737 v4df __builtin_ia32_pd256_pd (v2df)
9738 v4sf __builtin_ia32_ps_ps256 (v8sf)
9739 v8sf __builtin_ia32_ps256_ps (v4sf)
9740 int __builtin_ia32_ptestc256 (v4di,v4di,ptest)
9741 int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest)
9742 int __builtin_ia32_ptestz256 (v4di,v4di,ptest)
9743 v8sf __builtin_ia32_rcpps256 (v8sf)
9744 v4df __builtin_ia32_roundpd256 (v4df,int)
9745 v8sf __builtin_ia32_roundps256 (v8sf,int)
9746 v8sf __builtin_ia32_rsqrtps_nr256 (v8sf)
9747 v8sf __builtin_ia32_rsqrtps256 (v8sf)
9748 v4df __builtin_ia32_shufpd256 (v4df,v4df,int)
9749 v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int)
9750 v4si __builtin_ia32_si_si256 (v8si)
9751 v8si __builtin_ia32_si256_si (v4si)
9752 v4df __builtin_ia32_sqrtpd256 (v4df)
9753 v8sf __builtin_ia32_sqrtps_nr256 (v8sf)
9754 v8sf __builtin_ia32_sqrtps256 (v8sf)
9755 void __builtin_ia32_storedqu256 (pchar,v32qi)
9756 void __builtin_ia32_storeupd256 (pdouble,v4df)
9757 void __builtin_ia32_storeups256 (pfloat,v8sf)
9758 v4df __builtin_ia32_subpd256 (v4df,v4df)
9759 v8sf __builtin_ia32_subps256 (v8sf,v8sf)
9760 v4df __builtin_ia32_unpckhpd256 (v4df,v4df)
9761 v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf)
9762 v4df __builtin_ia32_unpcklpd256 (v4df,v4df)
9763 v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf)
9764 v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df)
9765 v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf)
9766 v4df __builtin_ia32_vbroadcastsd256 (pcdouble)
9767 v4sf __builtin_ia32_vbroadcastss (pcfloat)
9768 v8sf __builtin_ia32_vbroadcastss256 (pcfloat)
9769 v2df __builtin_ia32_vextractf128_pd256 (v4df,int)
9770 v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int)
9771 v4si __builtin_ia32_vextractf128_si256 (v8si,int)
9772 v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int)
9773 v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int)
9774 v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int)
9775 v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int)
9776 v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int)
9777 v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int)
9778 v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int)
9779 v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int)
9780 v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int)
9781 v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int)
9782 v2df __builtin_ia32_vpermilpd (v2df,int)
9783 v4df __builtin_ia32_vpermilpd256 (v4df,int)
9784 v4sf __builtin_ia32_vpermilps (v4sf,int)
9785 v8sf __builtin_ia32_vpermilps256 (v8sf,int)
9786 v2df __builtin_ia32_vpermilvarpd (v2df,v2di)
9787 v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di)
9788 v4sf __builtin_ia32_vpermilvarps (v4sf,v4si)
9789 v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si)
9790 int __builtin_ia32_vtestcpd (v2df,v2df,ptest)
9791 int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest)
9792 int __builtin_ia32_vtestcps (v4sf,v4sf,ptest)
9793 int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest)
9794 int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest)
9795 int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest)
9796 int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest)
9797 int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest)
9798 int __builtin_ia32_vtestzpd (v2df,v2df,ptest)
9799 int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest)
9800 int __builtin_ia32_vtestzps (v4sf,v4sf,ptest)
9801 int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest)
9802 void __builtin_ia32_vzeroall (void)
9803 void __builtin_ia32_vzeroupper (void)
9804 v4df __builtin_ia32_xorpd256 (v4df,v4df)
9805 v8sf __builtin_ia32_xorps256 (v8sf,v8sf)
9808 The following built-in functions are available when @option{-mavx2} is
9809 used. All of them generate the machine instruction that is part of the
9813 v32qi __builtin_ia32_mpsadbw256 (v32qi,v32qi,v32qi,int)
9814 v32qi __builtin_ia32_pabsb256 (v32qi)
9815 v16hi __builtin_ia32_pabsw256 (v16hi)
9816 v8si __builtin_ia32_pabsd256 (v8si)
9817 v16hi builtin_ia32_packssdw256 (v8si,v8si)
9818 v32qi __builtin_ia32_packsswb256 (v16hi,v16hi)
9819 v16hi __builtin_ia32_packusdw256 (v8si,v8si)
9820 v32qi __builtin_ia32_packuswb256 (v16hi,v16hi)
9821 v32qi__builtin_ia32_paddb256 (v32qi,v32qi)
9822 v16hi __builtin_ia32_paddw256 (v16hi,v16hi)
9823 v8si __builtin_ia32_paddd256 (v8si,v8si)
9824 v4di __builtin_ia32_paddq256 (v4di,v4di)
9825 v32qi __builtin_ia32_paddsb256 (v32qi,v32qi)
9826 v16hi __builtin_ia32_paddsw256 (v16hi,v16hi)
9827 v32qi __builtin_ia32_paddusb256 (v32qi,v32qi)
9828 v16hi __builtin_ia32_paddusw256 (v16hi,v16hi)
9829 v4di __builtin_ia32_palignr256 (v4di,v4di,int)
9830 v4di __builtin_ia32_andsi256 (v4di,v4di)
9831 v4di __builtin_ia32_andnotsi256 (v4di,v4di)
9832 v32qi__builtin_ia32_pavgb256 (v32qi,v32qi)
9833 v16hi __builtin_ia32_pavgw256 (v16hi,v16hi)
9834 v32qi __builtin_ia32_pblendvb256 (v32qi,v32qi,v32qi)
9835 v16hi __builtin_ia32_pblendw256 (v16hi,v16hi,int)
9836 v32qi __builtin_ia32_pcmpeqb256 (v32qi,v32qi)
9837 v16hi __builtin_ia32_pcmpeqw256 (v16hi,v16hi)
9838 v8si __builtin_ia32_pcmpeqd256 (c8si,v8si)
9839 v4di __builtin_ia32_pcmpeqq256 (v4di,v4di)
9840 v32qi __builtin_ia32_pcmpgtb256 (v32qi,v32qi)
9841 v16hi __builtin_ia32_pcmpgtw256 (16hi,v16hi)
9842 v8si __builtin_ia32_pcmpgtd256 (v8si,v8si)
9843 v4di __builtin_ia32_pcmpgtq256 (v4di,v4di)
9844 v16hi __builtin_ia32_phaddw256 (v16hi,v16hi)
9845 v8si __builtin_ia32_phaddd256 (v8si,v8si)
9846 v16hi __builtin_ia32_phaddsw256 (v16hi,v16hi)
9847 v16hi __builtin_ia32_phsubw256 (v16hi,v16hi)
9848 v8si __builtin_ia32_phsubd256 (v8si,v8si)
9849 v16hi __builtin_ia32_phsubsw256 (v16hi,v16hi)
9850 v32qi __builtin_ia32_pmaddubsw256 (v32qi,v32qi)
9851 v16hi __builtin_ia32_pmaddwd256 (v16hi,v16hi)
9852 v32qi __builtin_ia32_pmaxsb256 (v32qi,v32qi)
9853 v16hi __builtin_ia32_pmaxsw256 (v16hi,v16hi)
9854 v8si __builtin_ia32_pmaxsd256 (v8si,v8si)
9855 v32qi __builtin_ia32_pmaxub256 (v32qi,v32qi)
9856 v16hi __builtin_ia32_pmaxuw256 (v16hi,v16hi)
9857 v8si __builtin_ia32_pmaxud256 (v8si,v8si)
9858 v32qi __builtin_ia32_pminsb256 (v32qi,v32qi)
9859 v16hi __builtin_ia32_pminsw256 (v16hi,v16hi)
9860 v8si __builtin_ia32_pminsd256 (v8si,v8si)
9861 v32qi __builtin_ia32_pminub256 (v32qi,v32qi)
9862 v16hi __builtin_ia32_pminuw256 (v16hi,v16hi)
9863 v8si __builtin_ia32_pminud256 (v8si,v8si)
9864 int __builtin_ia32_pmovmskb256 (v32qi)
9865 v16hi __builtin_ia32_pmovsxbw256 (v16qi)
9866 v8si __builtin_ia32_pmovsxbd256 (v16qi)
9867 v4di __builtin_ia32_pmovsxbq256 (v16qi)
9868 v8si __builtin_ia32_pmovsxwd256 (v8hi)
9869 v4di __builtin_ia32_pmovsxwq256 (v8hi)
9870 v4di __builtin_ia32_pmovsxdq256 (v4si)
9871 v16hi __builtin_ia32_pmovzxbw256 (v16qi)
9872 v8si __builtin_ia32_pmovzxbd256 (v16qi)
9873 v4di __builtin_ia32_pmovzxbq256 (v16qi)
9874 v8si __builtin_ia32_pmovzxwd256 (v8hi)
9875 v4di __builtin_ia32_pmovzxwq256 (v8hi)
9876 v4di __builtin_ia32_pmovzxdq256 (v4si)
9877 v4di __builtin_ia32_pmuldq256 (v8si,v8si)
9878 v16hi __builtin_ia32_pmulhrsw256 (v16hi, v16hi)
9879 v16hi __builtin_ia32_pmulhuw256 (v16hi,v16hi)
9880 v16hi __builtin_ia32_pmulhw256 (v16hi,v16hi)
9881 v16hi __builtin_ia32_pmullw256 (v16hi,v16hi)
9882 v8si __builtin_ia32_pmulld256 (v8si,v8si)
9883 v4di __builtin_ia32_pmuludq256 (v8si,v8si)
9884 v4di __builtin_ia32_por256 (v4di,v4di)
9885 v16hi __builtin_ia32_psadbw256 (v32qi,v32qi)
9886 v32qi __builtin_ia32_pshufb256 (v32qi,v32qi)
9887 v8si __builtin_ia32_pshufd256 (v8si,int)
9888 v16hi __builtin_ia32_pshufhw256 (v16hi,int)
9889 v16hi __builtin_ia32_pshuflw256 (v16hi,int)
9890 v32qi __builtin_ia32_psignb256 (v32qi,v32qi)
9891 v16hi __builtin_ia32_psignw256 (v16hi,v16hi)
9892 v8si __builtin_ia32_psignd256 (v8si,v8si)
9893 v4di __builtin_ia32_pslldqi256 (v4di,int)
9894 v16hi __builtin_ia32_psllwi256 (16hi,int)
9895 v16hi __builtin_ia32_psllw256(v16hi,v8hi)
9896 v8si __builtin_ia32_pslldi256 (v8si,int)
9897 v8si __builtin_ia32_pslld256(v8si,v4si)
9898 v4di __builtin_ia32_psllqi256 (v4di,int)
9899 v4di __builtin_ia32_psllq256(v4di,v2di)
9900 v16hi __builtin_ia32_psrawi256 (v16hi,int)
9901 v16hi __builtin_ia32_psraw256 (v16hi,v8hi)
9902 v8si __builtin_ia32_psradi256 (v8si,int)
9903 v8si __builtin_ia32_psrad256 (v8si,v4si)
9904 v4di __builtin_ia32_psrldqi256 (v4di, int)
9905 v16hi __builtin_ia32_psrlwi256 (v16hi,int)
9906 v16hi __builtin_ia32_psrlw256 (v16hi,v8hi)
9907 v8si __builtin_ia32_psrldi256 (v8si,int)
9908 v8si __builtin_ia32_psrld256 (v8si,v4si)
9909 v4di __builtin_ia32_psrlqi256 (v4di,int)
9910 v4di __builtin_ia32_psrlq256(v4di,v2di)
9911 v32qi __builtin_ia32_psubb256 (v32qi,v32qi)
9912 v32hi __builtin_ia32_psubw256 (v16hi,v16hi)
9913 v8si __builtin_ia32_psubd256 (v8si,v8si)
9914 v4di __builtin_ia32_psubq256 (v4di,v4di)
9915 v32qi __builtin_ia32_psubsb256 (v32qi,v32qi)
9916 v16hi __builtin_ia32_psubsw256 (v16hi,v16hi)
9917 v32qi __builtin_ia32_psubusb256 (v32qi,v32qi)
9918 v16hi __builtin_ia32_psubusw256 (v16hi,v16hi)
9919 v32qi __builtin_ia32_punpckhbw256 (v32qi,v32qi)
9920 v16hi __builtin_ia32_punpckhwd256 (v16hi,v16hi)
9921 v8si __builtin_ia32_punpckhdq256 (v8si,v8si)
9922 v4di __builtin_ia32_punpckhqdq256 (v4di,v4di)
9923 v32qi __builtin_ia32_punpcklbw256 (v32qi,v32qi)
9924 v16hi __builtin_ia32_punpcklwd256 (v16hi,v16hi)
9925 v8si __builtin_ia32_punpckldq256 (v8si,v8si)
9926 v4di __builtin_ia32_punpcklqdq256 (v4di,v4di)
9927 v4di __builtin_ia32_pxor256 (v4di,v4di)
9928 v4di __builtin_ia32_movntdqa256 (pv4di)
9929 v4sf __builtin_ia32_vbroadcastss_ps (v4sf)
9930 v8sf __builtin_ia32_vbroadcastss_ps256 (v4sf)
9931 v4df __builtin_ia32_vbroadcastsd_pd256 (v2df)
9932 v4di __builtin_ia32_vbroadcastsi256 (v2di)
9933 v4si __builtin_ia32_pblendd128 (v4si,v4si)
9934 v8si __builtin_ia32_pblendd256 (v8si,v8si)
9935 v32qi __builtin_ia32_pbroadcastb256 (v16qi)
9936 v16hi __builtin_ia32_pbroadcastw256 (v8hi)
9937 v8si __builtin_ia32_pbroadcastd256 (v4si)
9938 v4di __builtin_ia32_pbroadcastq256 (v2di)
9939 v16qi __builtin_ia32_pbroadcastb128 (v16qi)
9940 v8hi __builtin_ia32_pbroadcastw128 (v8hi)
9941 v4si __builtin_ia32_pbroadcastd128 (v4si)
9942 v2di __builtin_ia32_pbroadcastq128 (v2di)
9943 v8si __builtin_ia32_permvarsi256 (v8si,v8si)
9944 v4df __builtin_ia32_permdf256 (v4df,int)
9945 v8sf __builtin_ia32_permvarsf256 (v8sf,v8sf)
9946 v4di __builtin_ia32_permdi256 (v4di,int)
9947 v4di __builtin_ia32_permti256 (v4di,v4di,int)
9948 v4di __builtin_ia32_extract128i256 (v4di,int)
9949 v4di __builtin_ia32_insert128i256 (v4di,v2di,int)
9950 v8si __builtin_ia32_maskloadd256 (pcv8si,v8si)
9951 v4di __builtin_ia32_maskloadq256 (pcv4di,v4di)
9952 v4si __builtin_ia32_maskloadd (pcv4si,v4si)
9953 v2di __builtin_ia32_maskloadq (pcv2di,v2di)
9954 void __builtin_ia32_maskstored256 (pv8si,v8si,v8si)
9955 void __builtin_ia32_maskstoreq256 (pv4di,v4di,v4di)
9956 void __builtin_ia32_maskstored (pv4si,v4si,v4si)
9957 void __builtin_ia32_maskstoreq (pv2di,v2di,v2di)
9958 v8si __builtin_ia32_psllv8si (v8si,v8si)
9959 v4si __builtin_ia32_psllv4si (v4si,v4si)
9960 v4di __builtin_ia32_psllv4di (v4di,v4di)
9961 v2di __builtin_ia32_psllv2di (v2di,v2di)
9962 v8si __builtin_ia32_psrav8si (v8si,v8si)
9963 v4si __builtin_ia32_psrav4si (v4si,v4si)
9964 v8si __builtin_ia32_psrlv8si (v8si,v8si)
9965 v4si __builtin_ia32_psrlv4si (v4si,v4si)
9966 v4di __builtin_ia32_psrlv4di (v4di,v4di)
9967 v2di __builtin_ia32_psrlv2di (v2di,v2di)
9968 v2df __builtin_ia32_gathersiv2df (v2df, pcdouble,v4si,v2df,int)
9969 v4df __builtin_ia32_gathersiv4df (v4df, pcdouble,v4si,v4df,int)
9970 v2df __builtin_ia32_gatherdiv2df (v2df, pcdouble,v2di,v2df,int)
9971 v4df __builtin_ia32_gatherdiv4df (v4df, pcdouble,v4di,v4df,int)
9972 v4sf __builtin_ia32_gathersiv4sf (v4sf, pcfloat,v4si,v4sf,int)
9973 v8sf __builtin_ia32_gathersiv8sf (v8sf, pcfloat,v8si,v8sf,int)
9974 v4sf __builtin_ia32_gatherdiv4sf (v4sf, pcfloat,v2di,v4sf,int)
9975 v4sf __builtin_ia32_gatherdiv4sf256 (v4sf, pcfloat,v4di,v4sf,int)
9976 v2di __builtin_ia32_gathersiv2di (v2di, pcint64,v4si,v2di,int)
9977 v4di __builtin_ia32_gathersiv4di (v4di, pcint64,v4si,v4di,int)
9978 v2di __builtin_ia32_gatherdiv2di (v2di, pcint64,v2di,v2di,int)
9979 v4di __builtin_ia32_gatherdiv4di (v4di, pcint64,v4di,v4di,int)
9980 v4si __builtin_ia32_gathersiv4si (v4si, pcint,v4si,v4si,int)
9981 v8si __builtin_ia32_gathersiv8si (v8si, pcint,v8si,v8si,int)
9982 v4si __builtin_ia32_gatherdiv4si (v4si, pcint,v2di,v4si,int)
9983 v4si __builtin_ia32_gatherdiv4si256 (v4si, pcint,v4di,v4si,int)
9986 The following built-in functions are available when @option{-maes} is
9987 used. All of them generate the machine instruction that is part of the
9991 v2di __builtin_ia32_aesenc128 (v2di, v2di)
9992 v2di __builtin_ia32_aesenclast128 (v2di, v2di)
9993 v2di __builtin_ia32_aesdec128 (v2di, v2di)
9994 v2di __builtin_ia32_aesdeclast128 (v2di, v2di)
9995 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int)
9996 v2di __builtin_ia32_aesimc128 (v2di)
9999 The following built-in function is available when @option{-mpclmul} is
10003 @item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)
10004 Generates the @code{pclmulqdq} machine instruction.
10007 The following built-in function is available when @option{-mfsgsbase} is
10008 used. All of them generate the machine instruction that is part of the
10012 unsigned int __builtin_ia32_rdfsbase32 (void)
10013 unsigned long long __builtin_ia32_rdfsbase64 (void)
10014 unsigned int __builtin_ia32_rdgsbase32 (void)
10015 unsigned long long __builtin_ia32_rdgsbase64 (void)
10016 void _writefsbase_u32 (unsigned int)
10017 void _writefsbase_u64 (unsigned long long)
10018 void _writegsbase_u32 (unsigned int)
10019 void _writegsbase_u64 (unsigned long long)
10022 The following built-in function is available when @option{-mrdrnd} is
10023 used. All of them generate the machine instruction that is part of the
10027 unsigned int __builtin_ia32_rdrand16_step (unsigned short *)
10028 unsigned int __builtin_ia32_rdrand32_step (unsigned int *)
10029 unsigned int __builtin_ia32_rdrand64_step (unsigned long long *)
10032 The following built-in functions are available when @option{-msse4a} is used.
10033 All of them generate the machine instruction that is part of the name.
10036 void __builtin_ia32_movntsd (double *, v2df)
10037 void __builtin_ia32_movntss (float *, v4sf)
10038 v2di __builtin_ia32_extrq (v2di, v16qi)
10039 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
10040 v2di __builtin_ia32_insertq (v2di, v2di)
10041 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
10044 The following built-in functions are available when @option{-mxop} is used.
10046 v2df __builtin_ia32_vfrczpd (v2df)
10047 v4sf __builtin_ia32_vfrczps (v4sf)
10048 v2df __builtin_ia32_vfrczsd (v2df, v2df)
10049 v4sf __builtin_ia32_vfrczss (v4sf, v4sf)
10050 v4df __builtin_ia32_vfrczpd256 (v4df)
10051 v8sf __builtin_ia32_vfrczps256 (v8sf)
10052 v2di __builtin_ia32_vpcmov (v2di, v2di, v2di)
10053 v2di __builtin_ia32_vpcmov_v2di (v2di, v2di, v2di)
10054 v4si __builtin_ia32_vpcmov_v4si (v4si, v4si, v4si)
10055 v8hi __builtin_ia32_vpcmov_v8hi (v8hi, v8hi, v8hi)
10056 v16qi __builtin_ia32_vpcmov_v16qi (v16qi, v16qi, v16qi)
10057 v2df __builtin_ia32_vpcmov_v2df (v2df, v2df, v2df)
10058 v4sf __builtin_ia32_vpcmov_v4sf (v4sf, v4sf, v4sf)
10059 v4di __builtin_ia32_vpcmov_v4di256 (v4di, v4di, v4di)
10060 v8si __builtin_ia32_vpcmov_v8si256 (v8si, v8si, v8si)
10061 v16hi __builtin_ia32_vpcmov_v16hi256 (v16hi, v16hi, v16hi)
10062 v32qi __builtin_ia32_vpcmov_v32qi256 (v32qi, v32qi, v32qi)
10063 v4df __builtin_ia32_vpcmov_v4df256 (v4df, v4df, v4df)
10064 v8sf __builtin_ia32_vpcmov_v8sf256 (v8sf, v8sf, v8sf)
10065 v16qi __builtin_ia32_vpcomeqb (v16qi, v16qi)
10066 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
10067 v4si __builtin_ia32_vpcomeqd (v4si, v4si)
10068 v2di __builtin_ia32_vpcomeqq (v2di, v2di)
10069 v16qi __builtin_ia32_vpcomequb (v16qi, v16qi)
10070 v4si __builtin_ia32_vpcomequd (v4si, v4si)
10071 v2di __builtin_ia32_vpcomequq (v2di, v2di)
10072 v8hi __builtin_ia32_vpcomequw (v8hi, v8hi)
10073 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
10074 v16qi __builtin_ia32_vpcomfalseb (v16qi, v16qi)
10075 v4si __builtin_ia32_vpcomfalsed (v4si, v4si)
10076 v2di __builtin_ia32_vpcomfalseq (v2di, v2di)
10077 v16qi __builtin_ia32_vpcomfalseub (v16qi, v16qi)
10078 v4si __builtin_ia32_vpcomfalseud (v4si, v4si)
10079 v2di __builtin_ia32_vpcomfalseuq (v2di, v2di)
10080 v8hi __builtin_ia32_vpcomfalseuw (v8hi, v8hi)
10081 v8hi __builtin_ia32_vpcomfalsew (v8hi, v8hi)
10082 v16qi __builtin_ia32_vpcomgeb (v16qi, v16qi)
10083 v4si __builtin_ia32_vpcomged (v4si, v4si)
10084 v2di __builtin_ia32_vpcomgeq (v2di, v2di)
10085 v16qi __builtin_ia32_vpcomgeub (v16qi, v16qi)
10086 v4si __builtin_ia32_vpcomgeud (v4si, v4si)
10087 v2di __builtin_ia32_vpcomgeuq (v2di, v2di)
10088 v8hi __builtin_ia32_vpcomgeuw (v8hi, v8hi)
10089 v8hi __builtin_ia32_vpcomgew (v8hi, v8hi)
10090 v16qi __builtin_ia32_vpcomgtb (v16qi, v16qi)
10091 v4si __builtin_ia32_vpcomgtd (v4si, v4si)
10092 v2di __builtin_ia32_vpcomgtq (v2di, v2di)
10093 v16qi __builtin_ia32_vpcomgtub (v16qi, v16qi)
10094 v4si __builtin_ia32_vpcomgtud (v4si, v4si)
10095 v2di __builtin_ia32_vpcomgtuq (v2di, v2di)
10096 v8hi __builtin_ia32_vpcomgtuw (v8hi, v8hi)
10097 v8hi __builtin_ia32_vpcomgtw (v8hi, v8hi)
10098 v16qi __builtin_ia32_vpcomleb (v16qi, v16qi)
10099 v4si __builtin_ia32_vpcomled (v4si, v4si)
10100 v2di __builtin_ia32_vpcomleq (v2di, v2di)
10101 v16qi __builtin_ia32_vpcomleub (v16qi, v16qi)
10102 v4si __builtin_ia32_vpcomleud (v4si, v4si)
10103 v2di __builtin_ia32_vpcomleuq (v2di, v2di)
10104 v8hi __builtin_ia32_vpcomleuw (v8hi, v8hi)
10105 v8hi __builtin_ia32_vpcomlew (v8hi, v8hi)
10106 v16qi __builtin_ia32_vpcomltb (v16qi, v16qi)
10107 v4si __builtin_ia32_vpcomltd (v4si, v4si)
10108 v2di __builtin_ia32_vpcomltq (v2di, v2di)
10109 v16qi __builtin_ia32_vpcomltub (v16qi, v16qi)
10110 v4si __builtin_ia32_vpcomltud (v4si, v4si)
10111 v2di __builtin_ia32_vpcomltuq (v2di, v2di)
10112 v8hi __builtin_ia32_vpcomltuw (v8hi, v8hi)
10113 v8hi __builtin_ia32_vpcomltw (v8hi, v8hi)
10114 v16qi __builtin_ia32_vpcomneb (v16qi, v16qi)
10115 v4si __builtin_ia32_vpcomned (v4si, v4si)
10116 v2di __builtin_ia32_vpcomneq (v2di, v2di)
10117 v16qi __builtin_ia32_vpcomneub (v16qi, v16qi)
10118 v4si __builtin_ia32_vpcomneud (v4si, v4si)
10119 v2di __builtin_ia32_vpcomneuq (v2di, v2di)
10120 v8hi __builtin_ia32_vpcomneuw (v8hi, v8hi)
10121 v8hi __builtin_ia32_vpcomnew (v8hi, v8hi)
10122 v16qi __builtin_ia32_vpcomtrueb (v16qi, v16qi)
10123 v4si __builtin_ia32_vpcomtrued (v4si, v4si)
10124 v2di __builtin_ia32_vpcomtrueq (v2di, v2di)
10125 v16qi __builtin_ia32_vpcomtrueub (v16qi, v16qi)
10126 v4si __builtin_ia32_vpcomtrueud (v4si, v4si)
10127 v2di __builtin_ia32_vpcomtrueuq (v2di, v2di)
10128 v8hi __builtin_ia32_vpcomtrueuw (v8hi, v8hi)
10129 v8hi __builtin_ia32_vpcomtruew (v8hi, v8hi)
10130 v4si __builtin_ia32_vphaddbd (v16qi)
10131 v2di __builtin_ia32_vphaddbq (v16qi)
10132 v8hi __builtin_ia32_vphaddbw (v16qi)
10133 v2di __builtin_ia32_vphadddq (v4si)
10134 v4si __builtin_ia32_vphaddubd (v16qi)
10135 v2di __builtin_ia32_vphaddubq (v16qi)
10136 v8hi __builtin_ia32_vphaddubw (v16qi)
10137 v2di __builtin_ia32_vphaddudq (v4si)
10138 v4si __builtin_ia32_vphadduwd (v8hi)
10139 v2di __builtin_ia32_vphadduwq (v8hi)
10140 v4si __builtin_ia32_vphaddwd (v8hi)
10141 v2di __builtin_ia32_vphaddwq (v8hi)
10142 v8hi __builtin_ia32_vphsubbw (v16qi)
10143 v2di __builtin_ia32_vphsubdq (v4si)
10144 v4si __builtin_ia32_vphsubwd (v8hi)
10145 v4si __builtin_ia32_vpmacsdd (v4si, v4si, v4si)
10146 v2di __builtin_ia32_vpmacsdqh (v4si, v4si, v2di)
10147 v2di __builtin_ia32_vpmacsdql (v4si, v4si, v2di)
10148 v4si __builtin_ia32_vpmacssdd (v4si, v4si, v4si)
10149 v2di __builtin_ia32_vpmacssdqh (v4si, v4si, v2di)
10150 v2di __builtin_ia32_vpmacssdql (v4si, v4si, v2di)
10151 v4si __builtin_ia32_vpmacsswd (v8hi, v8hi, v4si)
10152 v8hi __builtin_ia32_vpmacssww (v8hi, v8hi, v8hi)
10153 v4si __builtin_ia32_vpmacswd (v8hi, v8hi, v4si)
10154 v8hi __builtin_ia32_vpmacsww (v8hi, v8hi, v8hi)
10155 v4si __builtin_ia32_vpmadcsswd (v8hi, v8hi, v4si)
10156 v4si __builtin_ia32_vpmadcswd (v8hi, v8hi, v4si)
10157 v16qi __builtin_ia32_vpperm (v16qi, v16qi, v16qi)
10158 v16qi __builtin_ia32_vprotb (v16qi, v16qi)
10159 v4si __builtin_ia32_vprotd (v4si, v4si)
10160 v2di __builtin_ia32_vprotq (v2di, v2di)
10161 v8hi __builtin_ia32_vprotw (v8hi, v8hi)
10162 v16qi __builtin_ia32_vpshab (v16qi, v16qi)
10163 v4si __builtin_ia32_vpshad (v4si, v4si)
10164 v2di __builtin_ia32_vpshaq (v2di, v2di)
10165 v8hi __builtin_ia32_vpshaw (v8hi, v8hi)
10166 v16qi __builtin_ia32_vpshlb (v16qi, v16qi)
10167 v4si __builtin_ia32_vpshld (v4si, v4si)
10168 v2di __builtin_ia32_vpshlq (v2di, v2di)
10169 v8hi __builtin_ia32_vpshlw (v8hi, v8hi)
10172 The following built-in functions are available when @option{-mfma4} is used.
10173 All of them generate the machine instruction that is part of the name
10174 with MMX registers.
10177 v2df __builtin_ia32_fmaddpd (v2df, v2df, v2df)
10178 v4sf __builtin_ia32_fmaddps (v4sf, v4sf, v4sf)
10179 v2df __builtin_ia32_fmaddsd (v2df, v2df, v2df)
10180 v4sf __builtin_ia32_fmaddss (v4sf, v4sf, v4sf)
10181 v2df __builtin_ia32_fmsubpd (v2df, v2df, v2df)
10182 v4sf __builtin_ia32_fmsubps (v4sf, v4sf, v4sf)
10183 v2df __builtin_ia32_fmsubsd (v2df, v2df, v2df)
10184 v4sf __builtin_ia32_fmsubss (v4sf, v4sf, v4sf)
10185 v2df __builtin_ia32_fnmaddpd (v2df, v2df, v2df)
10186 v4sf __builtin_ia32_fnmaddps (v4sf, v4sf, v4sf)
10187 v2df __builtin_ia32_fnmaddsd (v2df, v2df, v2df)
10188 v4sf __builtin_ia32_fnmaddss (v4sf, v4sf, v4sf)
10189 v2df __builtin_ia32_fnmsubpd (v2df, v2df, v2df)
10190 v4sf __builtin_ia32_fnmsubps (v4sf, v4sf, v4sf)
10191 v2df __builtin_ia32_fnmsubsd (v2df, v2df, v2df)
10192 v4sf __builtin_ia32_fnmsubss (v4sf, v4sf, v4sf)
10193 v2df __builtin_ia32_fmaddsubpd (v2df, v2df, v2df)
10194 v4sf __builtin_ia32_fmaddsubps (v4sf, v4sf, v4sf)
10195 v2df __builtin_ia32_fmsubaddpd (v2df, v2df, v2df)
10196 v4sf __builtin_ia32_fmsubaddps (v4sf, v4sf, v4sf)
10197 v4df __builtin_ia32_fmaddpd256 (v4df, v4df, v4df)
10198 v8sf __builtin_ia32_fmaddps256 (v8sf, v8sf, v8sf)
10199 v4df __builtin_ia32_fmsubpd256 (v4df, v4df, v4df)
10200 v8sf __builtin_ia32_fmsubps256 (v8sf, v8sf, v8sf)
10201 v4df __builtin_ia32_fnmaddpd256 (v4df, v4df, v4df)
10202 v8sf __builtin_ia32_fnmaddps256 (v8sf, v8sf, v8sf)
10203 v4df __builtin_ia32_fnmsubpd256 (v4df, v4df, v4df)
10204 v8sf __builtin_ia32_fnmsubps256 (v8sf, v8sf, v8sf)
10205 v4df __builtin_ia32_fmaddsubpd256 (v4df, v4df, v4df)
10206 v8sf __builtin_ia32_fmaddsubps256 (v8sf, v8sf, v8sf)
10207 v4df __builtin_ia32_fmsubaddpd256 (v4df, v4df, v4df)
10208 v8sf __builtin_ia32_fmsubaddps256 (v8sf, v8sf, v8sf)
10212 The following built-in functions are available when @option{-mlwp} is used.
10215 void __builtin_ia32_llwpcb16 (void *);
10216 void __builtin_ia32_llwpcb32 (void *);
10217 void __builtin_ia32_llwpcb64 (void *);
10218 void * __builtin_ia32_llwpcb16 (void);
10219 void * __builtin_ia32_llwpcb32 (void);
10220 void * __builtin_ia32_llwpcb64 (void);
10221 void __builtin_ia32_lwpval16 (unsigned short, unsigned int, unsigned short)
10222 void __builtin_ia32_lwpval32 (unsigned int, unsigned int, unsigned int)
10223 void __builtin_ia32_lwpval64 (unsigned __int64, unsigned int, unsigned int)
10224 unsigned char __builtin_ia32_lwpins16 (unsigned short, unsigned int, unsigned short)
10225 unsigned char __builtin_ia32_lwpins32 (unsigned int, unsigned int, unsigned int)
10226 unsigned char __builtin_ia32_lwpins64 (unsigned __int64, unsigned int, unsigned int)
10229 The following built-in functions are available when @option{-mbmi} is used.
10230 All of them generate the machine instruction that is part of the name.
10232 unsigned int __builtin_ia32_bextr_u32(unsigned int, unsigned int);
10233 unsigned long long __builtin_ia32_bextr_u64 (unsigned long long, unsigned long long);
10236 The following built-in functions are available when @option{-mbmi2} is used.
10237 All of them generate the machine instruction that is part of the name.
10239 unsigned int _bzhi_u32 (unsigned int, unsigned int)
10240 unsigned int _pdep_u32 (unsigned int, unsigned int)
10241 unsigned int _pext_u32 (unsigned int, unsigned int)
10242 unsigned long long _bzhi_u64 (unsigned long long, unsigned long long)
10243 unsigned long long _pdep_u64 (unsigned long long, unsigned long long)
10244 unsigned long long _pext_u64 (unsigned long long, unsigned long long)
10247 The following built-in functions are available when @option{-mlzcnt} is used.
10248 All of them generate the machine instruction that is part of the name.
10250 unsigned short __builtin_ia32_lzcnt_16(unsigned short);
10251 unsigned int __builtin_ia32_lzcnt_u32(unsigned int);
10252 unsigned long long __builtin_ia32_lzcnt_u64 (unsigned long long);
10255 The following built-in functions are available when @option{-mtbm} is used.
10256 Both of them generate the immediate form of the bextr machine instruction.
10258 unsigned int __builtin_ia32_bextri_u32 (unsigned int, const unsigned int);
10259 unsigned long long __builtin_ia32_bextri_u64 (unsigned long long, const unsigned long long);
10263 The following built-in functions are available when @option{-m3dnow} is used.
10264 All of them generate the machine instruction that is part of the name.
10267 void __builtin_ia32_femms (void)
10268 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
10269 v2si __builtin_ia32_pf2id (v2sf)
10270 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
10271 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
10272 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
10273 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
10274 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
10275 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
10276 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
10277 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
10278 v2sf __builtin_ia32_pfrcp (v2sf)
10279 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
10280 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
10281 v2sf __builtin_ia32_pfrsqrt (v2sf)
10282 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
10283 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
10284 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
10285 v2sf __builtin_ia32_pi2fd (v2si)
10286 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
10289 The following built-in functions are available when both @option{-m3dnow}
10290 and @option{-march=athlon} are used. All of them generate the machine
10291 instruction that is part of the name.
10294 v2si __builtin_ia32_pf2iw (v2sf)
10295 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
10296 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
10297 v2sf __builtin_ia32_pi2fw (v2si)
10298 v2sf __builtin_ia32_pswapdsf (v2sf)
10299 v2si __builtin_ia32_pswapdsi (v2si)
10302 @node MIPS DSP Built-in Functions
10303 @subsection MIPS DSP Built-in Functions
10305 The MIPS DSP Application-Specific Extension (ASE) includes new
10306 instructions that are designed to improve the performance of DSP and
10307 media applications. It provides instructions that operate on packed
10308 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
10310 GCC supports MIPS DSP operations using both the generic
10311 vector extensions (@pxref{Vector Extensions}) and a collection of
10312 MIPS-specific built-in functions. Both kinds of support are
10313 enabled by the @option{-mdsp} command-line option.
10315 Revision 2 of the ASE was introduced in the second half of 2006.
10316 This revision adds extra instructions to the original ASE, but is
10317 otherwise backwards-compatible with it. You can select revision 2
10318 using the command-line option @option{-mdspr2}; this option implies
10321 The SCOUNT and POS bits of the DSP control register are global. The
10322 WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and
10323 POS bits. During optimization, the compiler will not delete these
10324 instructions and it will not delete calls to functions containing
10325 these instructions.
10327 At present, GCC only provides support for operations on 32-bit
10328 vectors. The vector type associated with 8-bit integer data is
10329 usually called @code{v4i8}, the vector type associated with Q7
10330 is usually called @code{v4q7}, the vector type associated with 16-bit
10331 integer data is usually called @code{v2i16}, and the vector type
10332 associated with Q15 is usually called @code{v2q15}. They can be
10333 defined in C as follows:
10336 typedef signed char v4i8 __attribute__ ((vector_size(4)));
10337 typedef signed char v4q7 __attribute__ ((vector_size(4)));
10338 typedef short v2i16 __attribute__ ((vector_size(4)));
10339 typedef short v2q15 __attribute__ ((vector_size(4)));
10342 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
10343 initialized in the same way as aggregates. For example:
10346 v4i8 a = @{1, 2, 3, 4@};
10348 b = (v4i8) @{5, 6, 7, 8@};
10350 v2q15 c = @{0x0fcb, 0x3a75@};
10352 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
10355 @emph{Note:} The CPU's endianness determines the order in which values
10356 are packed. On little-endian targets, the first value is the least
10357 significant and the last value is the most significant. The opposite
10358 order applies to big-endian targets. For example, the code above will
10359 set the lowest byte of @code{a} to @code{1} on little-endian targets
10360 and @code{4} on big-endian targets.
10362 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
10363 representation. As shown in this example, the integer representation
10364 of a Q7 value can be obtained by multiplying the fractional value by
10365 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
10366 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
10369 The table below lists the @code{v4i8} and @code{v2q15} operations for which
10370 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
10371 and @code{c} and @code{d} are @code{v2q15} values.
10373 @multitable @columnfractions .50 .50
10374 @item C code @tab MIPS instruction
10375 @item @code{a + b} @tab @code{addu.qb}
10376 @item @code{c + d} @tab @code{addq.ph}
10377 @item @code{a - b} @tab @code{subu.qb}
10378 @item @code{c - d} @tab @code{subq.ph}
10381 The table below lists the @code{v2i16} operation for which
10382 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
10383 @code{v2i16} values.
10385 @multitable @columnfractions .50 .50
10386 @item C code @tab MIPS instruction
10387 @item @code{e * f} @tab @code{mul.ph}
10390 It is easier to describe the DSP built-in functions if we first define
10391 the following types:
10396 typedef unsigned int ui32;
10397 typedef long long a64;
10400 @code{q31} and @code{i32} are actually the same as @code{int}, but we
10401 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
10402 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
10403 @code{long long}, but we use @code{a64} to indicate values that will
10404 be placed in one of the four DSP accumulators (@code{$ac0},
10405 @code{$ac1}, @code{$ac2} or @code{$ac3}).
10407 Also, some built-in functions prefer or require immediate numbers as
10408 parameters, because the corresponding DSP instructions accept both immediate
10409 numbers and register operands, or accept immediate numbers only. The
10410 immediate parameters are listed as follows.
10418 imm0_255: 0 to 255.
10419 imm_n32_31: -32 to 31.
10420 imm_n512_511: -512 to 511.
10423 The following built-in functions map directly to a particular MIPS DSP
10424 instruction. Please refer to the architecture specification
10425 for details on what each instruction does.
10428 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
10429 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
10430 q31 __builtin_mips_addq_s_w (q31, q31)
10431 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
10432 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
10433 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
10434 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
10435 q31 __builtin_mips_subq_s_w (q31, q31)
10436 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
10437 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
10438 i32 __builtin_mips_addsc (i32, i32)
10439 i32 __builtin_mips_addwc (i32, i32)
10440 i32 __builtin_mips_modsub (i32, i32)
10441 i32 __builtin_mips_raddu_w_qb (v4i8)
10442 v2q15 __builtin_mips_absq_s_ph (v2q15)
10443 q31 __builtin_mips_absq_s_w (q31)
10444 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
10445 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
10446 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
10447 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
10448 q31 __builtin_mips_preceq_w_phl (v2q15)
10449 q31 __builtin_mips_preceq_w_phr (v2q15)
10450 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
10451 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
10452 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
10453 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
10454 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
10455 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
10456 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
10457 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
10458 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
10459 v4i8 __builtin_mips_shll_qb (v4i8, i32)
10460 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
10461 v2q15 __builtin_mips_shll_ph (v2q15, i32)
10462 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
10463 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
10464 q31 __builtin_mips_shll_s_w (q31, imm0_31)
10465 q31 __builtin_mips_shll_s_w (q31, i32)
10466 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
10467 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
10468 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
10469 v2q15 __builtin_mips_shra_ph (v2q15, i32)
10470 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
10471 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
10472 q31 __builtin_mips_shra_r_w (q31, imm0_31)
10473 q31 __builtin_mips_shra_r_w (q31, i32)
10474 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
10475 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
10476 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
10477 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
10478 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
10479 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
10480 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
10481 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
10482 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
10483 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
10484 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
10485 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
10486 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
10487 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
10488 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
10489 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
10490 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
10491 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
10492 i32 __builtin_mips_bitrev (i32)
10493 i32 __builtin_mips_insv (i32, i32)
10494 v4i8 __builtin_mips_repl_qb (imm0_255)
10495 v4i8 __builtin_mips_repl_qb (i32)
10496 v2q15 __builtin_mips_repl_ph (imm_n512_511)
10497 v2q15 __builtin_mips_repl_ph (i32)
10498 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
10499 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
10500 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
10501 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
10502 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
10503 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
10504 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
10505 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
10506 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
10507 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
10508 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
10509 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
10510 i32 __builtin_mips_extr_w (a64, imm0_31)
10511 i32 __builtin_mips_extr_w (a64, i32)
10512 i32 __builtin_mips_extr_r_w (a64, imm0_31)
10513 i32 __builtin_mips_extr_s_h (a64, i32)
10514 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
10515 i32 __builtin_mips_extr_rs_w (a64, i32)
10516 i32 __builtin_mips_extr_s_h (a64, imm0_31)
10517 i32 __builtin_mips_extr_r_w (a64, i32)
10518 i32 __builtin_mips_extp (a64, imm0_31)
10519 i32 __builtin_mips_extp (a64, i32)
10520 i32 __builtin_mips_extpdp (a64, imm0_31)
10521 i32 __builtin_mips_extpdp (a64, i32)
10522 a64 __builtin_mips_shilo (a64, imm_n32_31)
10523 a64 __builtin_mips_shilo (a64, i32)
10524 a64 __builtin_mips_mthlip (a64, i32)
10525 void __builtin_mips_wrdsp (i32, imm0_63)
10526 i32 __builtin_mips_rddsp (imm0_63)
10527 i32 __builtin_mips_lbux (void *, i32)
10528 i32 __builtin_mips_lhx (void *, i32)
10529 i32 __builtin_mips_lwx (void *, i32)
10530 i32 __builtin_mips_bposge32 (void)
10531 a64 __builtin_mips_madd (a64, i32, i32);
10532 a64 __builtin_mips_maddu (a64, ui32, ui32);
10533 a64 __builtin_mips_msub (a64, i32, i32);
10534 a64 __builtin_mips_msubu (a64, ui32, ui32);
10535 a64 __builtin_mips_mult (i32, i32);
10536 a64 __builtin_mips_multu (ui32, ui32);
10539 The following built-in functions map directly to a particular MIPS DSP REV 2
10540 instruction. Please refer to the architecture specification
10541 for details on what each instruction does.
10544 v4q7 __builtin_mips_absq_s_qb (v4q7);
10545 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
10546 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
10547 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
10548 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
10549 i32 __builtin_mips_append (i32, i32, imm0_31);
10550 i32 __builtin_mips_balign (i32, i32, imm0_3);
10551 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
10552 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
10553 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
10554 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
10555 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
10556 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
10557 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
10558 q31 __builtin_mips_mulq_rs_w (q31, q31);
10559 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
10560 q31 __builtin_mips_mulq_s_w (q31, q31);
10561 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
10562 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
10563 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
10564 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
10565 i32 __builtin_mips_prepend (i32, i32, imm0_31);
10566 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
10567 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
10568 v4i8 __builtin_mips_shra_qb (v4i8, i32);
10569 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
10570 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
10571 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
10572 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
10573 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
10574 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
10575 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
10576 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
10577 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
10578 q31 __builtin_mips_addqh_w (q31, q31);
10579 q31 __builtin_mips_addqh_r_w (q31, q31);
10580 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
10581 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
10582 q31 __builtin_mips_subqh_w (q31, q31);
10583 q31 __builtin_mips_subqh_r_w (q31, q31);
10584 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
10585 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
10586 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
10587 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
10588 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
10589 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
10593 @node MIPS Paired-Single Support
10594 @subsection MIPS Paired-Single Support
10596 The MIPS64 architecture includes a number of instructions that
10597 operate on pairs of single-precision floating-point values.
10598 Each pair is packed into a 64-bit floating-point register,
10599 with one element being designated the ``upper half'' and
10600 the other being designated the ``lower half''.
10602 GCC supports paired-single operations using both the generic
10603 vector extensions (@pxref{Vector Extensions}) and a collection of
10604 MIPS-specific built-in functions. Both kinds of support are
10605 enabled by the @option{-mpaired-single} command-line option.
10607 The vector type associated with paired-single values is usually
10608 called @code{v2sf}. It can be defined in C as follows:
10611 typedef float v2sf __attribute__ ((vector_size (8)));
10614 @code{v2sf} values are initialized in the same way as aggregates.
10618 v2sf a = @{1.5, 9.1@};
10621 b = (v2sf) @{e, f@};
10624 @emph{Note:} The CPU's endianness determines which value is stored in
10625 the upper half of a register and which value is stored in the lower half.
10626 On little-endian targets, the first value is the lower one and the second
10627 value is the upper one. The opposite order applies to big-endian targets.
10628 For example, the code above will set the lower half of @code{a} to
10629 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
10631 @node MIPS Loongson Built-in Functions
10632 @subsection MIPS Loongson Built-in Functions
10634 GCC provides intrinsics to access the SIMD instructions provided by the
10635 ST Microelectronics Loongson-2E and -2F processors. These intrinsics,
10636 available after inclusion of the @code{loongson.h} header file,
10637 operate on the following 64-bit vector types:
10640 @item @code{uint8x8_t}, a vector of eight unsigned 8-bit integers;
10641 @item @code{uint16x4_t}, a vector of four unsigned 16-bit integers;
10642 @item @code{uint32x2_t}, a vector of two unsigned 32-bit integers;
10643 @item @code{int8x8_t}, a vector of eight signed 8-bit integers;
10644 @item @code{int16x4_t}, a vector of four signed 16-bit integers;
10645 @item @code{int32x2_t}, a vector of two signed 32-bit integers.
10648 The intrinsics provided are listed below; each is named after the
10649 machine instruction to which it corresponds, with suffixes added as
10650 appropriate to distinguish intrinsics that expand to the same machine
10651 instruction yet have different argument types. Refer to the architecture
10652 documentation for a description of the functionality of each
10656 int16x4_t packsswh (int32x2_t s, int32x2_t t);
10657 int8x8_t packsshb (int16x4_t s, int16x4_t t);
10658 uint8x8_t packushb (uint16x4_t s, uint16x4_t t);
10659 uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t);
10660 uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t);
10661 uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t);
10662 int32x2_t paddw_s (int32x2_t s, int32x2_t t);
10663 int16x4_t paddh_s (int16x4_t s, int16x4_t t);
10664 int8x8_t paddb_s (int8x8_t s, int8x8_t t);
10665 uint64_t paddd_u (uint64_t s, uint64_t t);
10666 int64_t paddd_s (int64_t s, int64_t t);
10667 int16x4_t paddsh (int16x4_t s, int16x4_t t);
10668 int8x8_t paddsb (int8x8_t s, int8x8_t t);
10669 uint16x4_t paddush (uint16x4_t s, uint16x4_t t);
10670 uint8x8_t paddusb (uint8x8_t s, uint8x8_t t);
10671 uint64_t pandn_ud (uint64_t s, uint64_t t);
10672 uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t);
10673 uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t);
10674 uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t);
10675 int64_t pandn_sd (int64_t s, int64_t t);
10676 int32x2_t pandn_sw (int32x2_t s, int32x2_t t);
10677 int16x4_t pandn_sh (int16x4_t s, int16x4_t t);
10678 int8x8_t pandn_sb (int8x8_t s, int8x8_t t);
10679 uint16x4_t pavgh (uint16x4_t s, uint16x4_t t);
10680 uint8x8_t pavgb (uint8x8_t s, uint8x8_t t);
10681 uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t);
10682 uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t);
10683 uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t);
10684 int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t);
10685 int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t);
10686 int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t);
10687 uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t);
10688 uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t);
10689 uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t);
10690 int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t);
10691 int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t);
10692 int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t);
10693 uint16x4_t pextrh_u (uint16x4_t s, int field);
10694 int16x4_t pextrh_s (int16x4_t s, int field);
10695 uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t);
10696 uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t);
10697 uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t);
10698 uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t);
10699 int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t);
10700 int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t);
10701 int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t);
10702 int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t);
10703 int32x2_t pmaddhw (int16x4_t s, int16x4_t t);
10704 int16x4_t pmaxsh (int16x4_t s, int16x4_t t);
10705 uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t);
10706 int16x4_t pminsh (int16x4_t s, int16x4_t t);
10707 uint8x8_t pminub (uint8x8_t s, uint8x8_t t);
10708 uint8x8_t pmovmskb_u (uint8x8_t s);
10709 int8x8_t pmovmskb_s (int8x8_t s);
10710 uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t);
10711 int16x4_t pmulhh (int16x4_t s, int16x4_t t);
10712 int16x4_t pmullh (int16x4_t s, int16x4_t t);
10713 int64_t pmuluw (uint32x2_t s, uint32x2_t t);
10714 uint8x8_t pasubub (uint8x8_t s, uint8x8_t t);
10715 uint16x4_t biadd (uint8x8_t s);
10716 uint16x4_t psadbh (uint8x8_t s, uint8x8_t t);
10717 uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order);
10718 int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order);
10719 uint16x4_t psllh_u (uint16x4_t s, uint8_t amount);
10720 int16x4_t psllh_s (int16x4_t s, uint8_t amount);
10721 uint32x2_t psllw_u (uint32x2_t s, uint8_t amount);
10722 int32x2_t psllw_s (int32x2_t s, uint8_t amount);
10723 uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount);
10724 int16x4_t psrlh_s (int16x4_t s, uint8_t amount);
10725 uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount);
10726 int32x2_t psrlw_s (int32x2_t s, uint8_t amount);
10727 uint16x4_t psrah_u (uint16x4_t s, uint8_t amount);
10728 int16x4_t psrah_s (int16x4_t s, uint8_t amount);
10729 uint32x2_t psraw_u (uint32x2_t s, uint8_t amount);
10730 int32x2_t psraw_s (int32x2_t s, uint8_t amount);
10731 uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t);
10732 uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t);
10733 uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t);
10734 int32x2_t psubw_s (int32x2_t s, int32x2_t t);
10735 int16x4_t psubh_s (int16x4_t s, int16x4_t t);
10736 int8x8_t psubb_s (int8x8_t s, int8x8_t t);
10737 uint64_t psubd_u (uint64_t s, uint64_t t);
10738 int64_t psubd_s (int64_t s, int64_t t);
10739 int16x4_t psubsh (int16x4_t s, int16x4_t t);
10740 int8x8_t psubsb (int8x8_t s, int8x8_t t);
10741 uint16x4_t psubush (uint16x4_t s, uint16x4_t t);
10742 uint8x8_t psubusb (uint8x8_t s, uint8x8_t t);
10743 uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t);
10744 uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t);
10745 uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t);
10746 int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t);
10747 int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t);
10748 int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t);
10749 uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t);
10750 uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t);
10751 uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t);
10752 int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t);
10753 int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t);
10754 int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t);
10758 * Paired-Single Arithmetic::
10759 * Paired-Single Built-in Functions::
10760 * MIPS-3D Built-in Functions::
10763 @node Paired-Single Arithmetic
10764 @subsubsection Paired-Single Arithmetic
10766 The table below lists the @code{v2sf} operations for which hardware
10767 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
10768 values and @code{x} is an integral value.
10770 @multitable @columnfractions .50 .50
10771 @item C code @tab MIPS instruction
10772 @item @code{a + b} @tab @code{add.ps}
10773 @item @code{a - b} @tab @code{sub.ps}
10774 @item @code{-a} @tab @code{neg.ps}
10775 @item @code{a * b} @tab @code{mul.ps}
10776 @item @code{a * b + c} @tab @code{madd.ps}
10777 @item @code{a * b - c} @tab @code{msub.ps}
10778 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
10779 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
10780 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
10783 Note that the multiply-accumulate instructions can be disabled
10784 using the command-line option @code{-mno-fused-madd}.
10786 @node Paired-Single Built-in Functions
10787 @subsubsection Paired-Single Built-in Functions
10789 The following paired-single functions map directly to a particular
10790 MIPS instruction. Please refer to the architecture specification
10791 for details on what each instruction does.
10794 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
10795 Pair lower lower (@code{pll.ps}).
10797 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
10798 Pair upper lower (@code{pul.ps}).
10800 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
10801 Pair lower upper (@code{plu.ps}).
10803 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
10804 Pair upper upper (@code{puu.ps}).
10806 @item v2sf __builtin_mips_cvt_ps_s (float, float)
10807 Convert pair to paired single (@code{cvt.ps.s}).
10809 @item float __builtin_mips_cvt_s_pl (v2sf)
10810 Convert pair lower to single (@code{cvt.s.pl}).
10812 @item float __builtin_mips_cvt_s_pu (v2sf)
10813 Convert pair upper to single (@code{cvt.s.pu}).
10815 @item v2sf __builtin_mips_abs_ps (v2sf)
10816 Absolute value (@code{abs.ps}).
10818 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
10819 Align variable (@code{alnv.ps}).
10821 @emph{Note:} The value of the third parameter must be 0 or 4
10822 modulo 8, otherwise the result will be unpredictable. Please read the
10823 instruction description for details.
10826 The following multi-instruction functions are also available.
10827 In each case, @var{cond} can be any of the 16 floating-point conditions:
10828 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
10829 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
10830 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
10833 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10834 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10835 Conditional move based on floating point comparison (@code{c.@var{cond}.ps},
10836 @code{movt.ps}/@code{movf.ps}).
10838 The @code{movt} functions return the value @var{x} computed by:
10841 c.@var{cond}.ps @var{cc},@var{a},@var{b}
10842 mov.ps @var{x},@var{c}
10843 movt.ps @var{x},@var{d},@var{cc}
10846 The @code{movf} functions are similar but use @code{movf.ps} instead
10849 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10850 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10851 Comparison of two paired-single values (@code{c.@var{cond}.ps},
10852 @code{bc1t}/@code{bc1f}).
10854 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
10855 and return either the upper or lower half of the result. For example:
10859 if (__builtin_mips_upper_c_eq_ps (a, b))
10860 upper_halves_are_equal ();
10862 upper_halves_are_unequal ();
10864 if (__builtin_mips_lower_c_eq_ps (a, b))
10865 lower_halves_are_equal ();
10867 lower_halves_are_unequal ();
10871 @node MIPS-3D Built-in Functions
10872 @subsubsection MIPS-3D Built-in Functions
10874 The MIPS-3D Application-Specific Extension (ASE) includes additional
10875 paired-single instructions that are designed to improve the performance
10876 of 3D graphics operations. Support for these instructions is controlled
10877 by the @option{-mips3d} command-line option.
10879 The functions listed below map directly to a particular MIPS-3D
10880 instruction. Please refer to the architecture specification for
10881 more details on what each instruction does.
10884 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
10885 Reduction add (@code{addr.ps}).
10887 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
10888 Reduction multiply (@code{mulr.ps}).
10890 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
10891 Convert paired single to paired word (@code{cvt.pw.ps}).
10893 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
10894 Convert paired word to paired single (@code{cvt.ps.pw}).
10896 @item float __builtin_mips_recip1_s (float)
10897 @itemx double __builtin_mips_recip1_d (double)
10898 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
10899 Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
10901 @item float __builtin_mips_recip2_s (float, float)
10902 @itemx double __builtin_mips_recip2_d (double, double)
10903 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
10904 Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
10906 @item float __builtin_mips_rsqrt1_s (float)
10907 @itemx double __builtin_mips_rsqrt1_d (double)
10908 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
10909 Reduced precision reciprocal square root (sequence step 1)
10910 (@code{rsqrt1.@var{fmt}}).
10912 @item float __builtin_mips_rsqrt2_s (float, float)
10913 @itemx double __builtin_mips_rsqrt2_d (double, double)
10914 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
10915 Reduced precision reciprocal square root (sequence step 2)
10916 (@code{rsqrt2.@var{fmt}}).
10919 The following multi-instruction functions are also available.
10920 In each case, @var{cond} can be any of the 16 floating-point conditions:
10921 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
10922 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
10923 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
10926 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
10927 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
10928 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
10929 @code{bc1t}/@code{bc1f}).
10931 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
10932 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
10937 if (__builtin_mips_cabs_eq_s (a, b))
10943 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10944 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10945 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
10946 @code{bc1t}/@code{bc1f}).
10948 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
10949 and return either the upper or lower half of the result. For example:
10953 if (__builtin_mips_upper_cabs_eq_ps (a, b))
10954 upper_halves_are_equal ();
10956 upper_halves_are_unequal ();
10958 if (__builtin_mips_lower_cabs_eq_ps (a, b))
10959 lower_halves_are_equal ();
10961 lower_halves_are_unequal ();
10964 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10965 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
10966 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
10967 @code{movt.ps}/@code{movf.ps}).
10969 The @code{movt} functions return the value @var{x} computed by:
10972 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
10973 mov.ps @var{x},@var{c}
10974 movt.ps @var{x},@var{d},@var{cc}
10977 The @code{movf} functions are similar but use @code{movf.ps} instead
10980 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10981 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10982 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10983 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
10984 Comparison of two paired-single values
10985 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
10986 @code{bc1any2t}/@code{bc1any2f}).
10988 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
10989 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
10990 result is true and the @code{all} forms return true if both results are true.
10995 if (__builtin_mips_any_c_eq_ps (a, b))
11000 if (__builtin_mips_all_c_eq_ps (a, b))
11006 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11007 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11008 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11009 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11010 Comparison of four paired-single values
11011 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
11012 @code{bc1any4t}/@code{bc1any4f}).
11014 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
11015 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
11016 The @code{any} forms return true if any of the four results are true
11017 and the @code{all} forms return true if all four results are true.
11022 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
11027 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
11034 @node picoChip Built-in Functions
11035 @subsection picoChip Built-in Functions
11037 GCC provides an interface to selected machine instructions from the
11038 picoChip instruction set.
11041 @item int __builtin_sbc (int @var{value})
11042 Sign bit count. Return the number of consecutive bits in @var{value}
11043 which have the same value as the sign-bit. The result is the number of
11044 leading sign bits minus one, giving the number of redundant sign bits in
11047 @item int __builtin_byteswap (int @var{value})
11048 Byte swap. Return the result of swapping the upper and lower bytes of
11051 @item int __builtin_brev (int @var{value})
11052 Bit reversal. Return the result of reversing the bits in
11053 @var{value}. Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1,
11056 @item int __builtin_adds (int @var{x}, int @var{y})
11057 Saturating addition. Return the result of adding @var{x} and @var{y},
11058 storing the value 32767 if the result overflows.
11060 @item int __builtin_subs (int @var{x}, int @var{y})
11061 Saturating subtraction. Return the result of subtracting @var{y} from
11062 @var{x}, storing the value @minus{}32768 if the result overflows.
11064 @item void __builtin_halt (void)
11065 Halt. The processor will stop execution. This built-in is useful for
11066 implementing assertions.
11070 @node Other MIPS Built-in Functions
11071 @subsection Other MIPS Built-in Functions
11073 GCC provides other MIPS-specific built-in functions:
11076 @item void __builtin_mips_cache (int @var{op}, const volatile void *@var{addr})
11077 Insert a @samp{cache} instruction with operands @var{op} and @var{addr}.
11078 GCC defines the preprocessor macro @code{___GCC_HAVE_BUILTIN_MIPS_CACHE}
11079 when this function is available.
11082 @node PowerPC AltiVec/VSX Built-in Functions
11083 @subsection PowerPC AltiVec Built-in Functions
11085 GCC provides an interface for the PowerPC family of processors to access
11086 the AltiVec operations described in Motorola's AltiVec Programming
11087 Interface Manual. The interface is made available by including
11088 @code{<altivec.h>} and using @option{-maltivec} and
11089 @option{-mabi=altivec}. The interface supports the following vector
11093 vector unsigned char
11097 vector unsigned short
11098 vector signed short
11102 vector unsigned int
11108 If @option{-mvsx} is used the following additional vector types are
11112 vector unsigned long
11117 The long types are only implemented for 64-bit code generation, and
11118 the long type is only used in the floating point/integer conversion
11121 GCC's implementation of the high-level language interface available from
11122 C and C++ code differs from Motorola's documentation in several ways.
11127 A vector constant is a list of constant expressions within curly braces.
11130 A vector initializer requires no cast if the vector constant is of the
11131 same type as the variable it is initializing.
11134 If @code{signed} or @code{unsigned} is omitted, the signedness of the
11135 vector type is the default signedness of the base type. The default
11136 varies depending on the operating system, so a portable program should
11137 always specify the signedness.
11140 Compiling with @option{-maltivec} adds keywords @code{__vector},
11141 @code{vector}, @code{__pixel}, @code{pixel}, @code{__bool} and
11142 @code{bool}. When compiling ISO C, the context-sensitive substitution
11143 of the keywords @code{vector}, @code{pixel} and @code{bool} is
11144 disabled. To use them, you must include @code{<altivec.h>} instead.
11147 GCC allows using a @code{typedef} name as the type specifier for a
11151 For C, overloaded functions are implemented with macros so the following
11155 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
11158 Since @code{vec_add} is a macro, the vector constant in the example
11159 is treated as four separate arguments. Wrap the entire argument in
11160 parentheses for this to work.
11163 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
11164 Internally, GCC uses built-in functions to achieve the functionality in
11165 the aforementioned header file, but they are not supported and are
11166 subject to change without notice.
11168 The following interfaces are supported for the generic and specific
11169 AltiVec operations and the AltiVec predicates. In cases where there
11170 is a direct mapping between generic and specific operations, only the
11171 generic names are shown here, although the specific operations can also
11174 Arguments that are documented as @code{const int} require literal
11175 integral values within the range required for that operation.
11178 vector signed char vec_abs (vector signed char);
11179 vector signed short vec_abs (vector signed short);
11180 vector signed int vec_abs (vector signed int);
11181 vector float vec_abs (vector float);
11183 vector signed char vec_abss (vector signed char);
11184 vector signed short vec_abss (vector signed short);
11185 vector signed int vec_abss (vector signed int);
11187 vector signed char vec_add (vector bool char, vector signed char);
11188 vector signed char vec_add (vector signed char, vector bool char);
11189 vector signed char vec_add (vector signed char, vector signed char);
11190 vector unsigned char vec_add (vector bool char, vector unsigned char);
11191 vector unsigned char vec_add (vector unsigned char, vector bool char);
11192 vector unsigned char vec_add (vector unsigned char,
11193 vector unsigned char);
11194 vector signed short vec_add (vector bool short, vector signed short);
11195 vector signed short vec_add (vector signed short, vector bool short);
11196 vector signed short vec_add (vector signed short, vector signed short);
11197 vector unsigned short vec_add (vector bool short,
11198 vector unsigned short);
11199 vector unsigned short vec_add (vector unsigned short,
11200 vector bool short);
11201 vector unsigned short vec_add (vector unsigned short,
11202 vector unsigned short);
11203 vector signed int vec_add (vector bool int, vector signed int);
11204 vector signed int vec_add (vector signed int, vector bool int);
11205 vector signed int vec_add (vector signed int, vector signed int);
11206 vector unsigned int vec_add (vector bool int, vector unsigned int);
11207 vector unsigned int vec_add (vector unsigned int, vector bool int);
11208 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
11209 vector float vec_add (vector float, vector float);
11211 vector float vec_vaddfp (vector float, vector float);
11213 vector signed int vec_vadduwm (vector bool int, vector signed int);
11214 vector signed int vec_vadduwm (vector signed int, vector bool int);
11215 vector signed int vec_vadduwm (vector signed int, vector signed int);
11216 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
11217 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
11218 vector unsigned int vec_vadduwm (vector unsigned int,
11219 vector unsigned int);
11221 vector signed short vec_vadduhm (vector bool short,
11222 vector signed short);
11223 vector signed short vec_vadduhm (vector signed short,
11224 vector bool short);
11225 vector signed short vec_vadduhm (vector signed short,
11226 vector signed short);
11227 vector unsigned short vec_vadduhm (vector bool short,
11228 vector unsigned short);
11229 vector unsigned short vec_vadduhm (vector unsigned short,
11230 vector bool short);
11231 vector unsigned short vec_vadduhm (vector unsigned short,
11232 vector unsigned short);
11234 vector signed char vec_vaddubm (vector bool char, vector signed char);
11235 vector signed char vec_vaddubm (vector signed char, vector bool char);
11236 vector signed char vec_vaddubm (vector signed char, vector signed char);
11237 vector unsigned char vec_vaddubm (vector bool char,
11238 vector unsigned char);
11239 vector unsigned char vec_vaddubm (vector unsigned char,
11241 vector unsigned char vec_vaddubm (vector unsigned char,
11242 vector unsigned char);
11244 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
11246 vector unsigned char vec_adds (vector bool char, vector unsigned char);
11247 vector unsigned char vec_adds (vector unsigned char, vector bool char);
11248 vector unsigned char vec_adds (vector unsigned char,
11249 vector unsigned char);
11250 vector signed char vec_adds (vector bool char, vector signed char);
11251 vector signed char vec_adds (vector signed char, vector bool char);
11252 vector signed char vec_adds (vector signed char, vector signed char);
11253 vector unsigned short vec_adds (vector bool short,
11254 vector unsigned short);
11255 vector unsigned short vec_adds (vector unsigned short,
11256 vector bool short);
11257 vector unsigned short vec_adds (vector unsigned short,
11258 vector unsigned short);
11259 vector signed short vec_adds (vector bool short, vector signed short);
11260 vector signed short vec_adds (vector signed short, vector bool short);
11261 vector signed short vec_adds (vector signed short, vector signed short);
11262 vector unsigned int vec_adds (vector bool int, vector unsigned int);
11263 vector unsigned int vec_adds (vector unsigned int, vector bool int);
11264 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
11265 vector signed int vec_adds (vector bool int, vector signed int);
11266 vector signed int vec_adds (vector signed int, vector bool int);
11267 vector signed int vec_adds (vector signed int, vector signed int);
11269 vector signed int vec_vaddsws (vector bool int, vector signed int);
11270 vector signed int vec_vaddsws (vector signed int, vector bool int);
11271 vector signed int vec_vaddsws (vector signed int, vector signed int);
11273 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
11274 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
11275 vector unsigned int vec_vadduws (vector unsigned int,
11276 vector unsigned int);
11278 vector signed short vec_vaddshs (vector bool short,
11279 vector signed short);
11280 vector signed short vec_vaddshs (vector signed short,
11281 vector bool short);
11282 vector signed short vec_vaddshs (vector signed short,
11283 vector signed short);
11285 vector unsigned short vec_vadduhs (vector bool short,
11286 vector unsigned short);
11287 vector unsigned short vec_vadduhs (vector unsigned short,
11288 vector bool short);
11289 vector unsigned short vec_vadduhs (vector unsigned short,
11290 vector unsigned short);
11292 vector signed char vec_vaddsbs (vector bool char, vector signed char);
11293 vector signed char vec_vaddsbs (vector signed char, vector bool char);
11294 vector signed char vec_vaddsbs (vector signed char, vector signed char);
11296 vector unsigned char vec_vaddubs (vector bool char,
11297 vector unsigned char);
11298 vector unsigned char vec_vaddubs (vector unsigned char,
11300 vector unsigned char vec_vaddubs (vector unsigned char,
11301 vector unsigned char);
11303 vector float vec_and (vector float, vector float);
11304 vector float vec_and (vector float, vector bool int);
11305 vector float vec_and (vector bool int, vector float);
11306 vector bool int vec_and (vector bool int, vector bool int);
11307 vector signed int vec_and (vector bool int, vector signed int);
11308 vector signed int vec_and (vector signed int, vector bool int);
11309 vector signed int vec_and (vector signed int, vector signed int);
11310 vector unsigned int vec_and (vector bool int, vector unsigned int);
11311 vector unsigned int vec_and (vector unsigned int, vector bool int);
11312 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
11313 vector bool short vec_and (vector bool short, vector bool short);
11314 vector signed short vec_and (vector bool short, vector signed short);
11315 vector signed short vec_and (vector signed short, vector bool short);
11316 vector signed short vec_and (vector signed short, vector signed short);
11317 vector unsigned short vec_and (vector bool short,
11318 vector unsigned short);
11319 vector unsigned short vec_and (vector unsigned short,
11320 vector bool short);
11321 vector unsigned short vec_and (vector unsigned short,
11322 vector unsigned short);
11323 vector signed char vec_and (vector bool char, vector signed char);
11324 vector bool char vec_and (vector bool char, vector bool char);
11325 vector signed char vec_and (vector signed char, vector bool char);
11326 vector signed char vec_and (vector signed char, vector signed char);
11327 vector unsigned char vec_and (vector bool char, vector unsigned char);
11328 vector unsigned char vec_and (vector unsigned char, vector bool char);
11329 vector unsigned char vec_and (vector unsigned char,
11330 vector unsigned char);
11332 vector float vec_andc (vector float, vector float);
11333 vector float vec_andc (vector float, vector bool int);
11334 vector float vec_andc (vector bool int, vector float);
11335 vector bool int vec_andc (vector bool int, vector bool int);
11336 vector signed int vec_andc (vector bool int, vector signed int);
11337 vector signed int vec_andc (vector signed int, vector bool int);
11338 vector signed int vec_andc (vector signed int, vector signed int);
11339 vector unsigned int vec_andc (vector bool int, vector unsigned int);
11340 vector unsigned int vec_andc (vector unsigned int, vector bool int);
11341 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
11342 vector bool short vec_andc (vector bool short, vector bool short);
11343 vector signed short vec_andc (vector bool short, vector signed short);
11344 vector signed short vec_andc (vector signed short, vector bool short);
11345 vector signed short vec_andc (vector signed short, vector signed short);
11346 vector unsigned short vec_andc (vector bool short,
11347 vector unsigned short);
11348 vector unsigned short vec_andc (vector unsigned short,
11349 vector bool short);
11350 vector unsigned short vec_andc (vector unsigned short,
11351 vector unsigned short);
11352 vector signed char vec_andc (vector bool char, vector signed char);
11353 vector bool char vec_andc (vector bool char, vector bool char);
11354 vector signed char vec_andc (vector signed char, vector bool char);
11355 vector signed char vec_andc (vector signed char, vector signed char);
11356 vector unsigned char vec_andc (vector bool char, vector unsigned char);
11357 vector unsigned char vec_andc (vector unsigned char, vector bool char);
11358 vector unsigned char vec_andc (vector unsigned char,
11359 vector unsigned char);
11361 vector unsigned char vec_avg (vector unsigned char,
11362 vector unsigned char);
11363 vector signed char vec_avg (vector signed char, vector signed char);
11364 vector unsigned short vec_avg (vector unsigned short,
11365 vector unsigned short);
11366 vector signed short vec_avg (vector signed short, vector signed short);
11367 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
11368 vector signed int vec_avg (vector signed int, vector signed int);
11370 vector signed int vec_vavgsw (vector signed int, vector signed int);
11372 vector unsigned int vec_vavguw (vector unsigned int,
11373 vector unsigned int);
11375 vector signed short vec_vavgsh (vector signed short,
11376 vector signed short);
11378 vector unsigned short vec_vavguh (vector unsigned short,
11379 vector unsigned short);
11381 vector signed char vec_vavgsb (vector signed char, vector signed char);
11383 vector unsigned char vec_vavgub (vector unsigned char,
11384 vector unsigned char);
11386 vector float vec_copysign (vector float);
11388 vector float vec_ceil (vector float);
11390 vector signed int vec_cmpb (vector float, vector float);
11392 vector bool char vec_cmpeq (vector signed char, vector signed char);
11393 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
11394 vector bool short vec_cmpeq (vector signed short, vector signed short);
11395 vector bool short vec_cmpeq (vector unsigned short,
11396 vector unsigned short);
11397 vector bool int vec_cmpeq (vector signed int, vector signed int);
11398 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
11399 vector bool int vec_cmpeq (vector float, vector float);
11401 vector bool int vec_vcmpeqfp (vector float, vector float);
11403 vector bool int vec_vcmpequw (vector signed int, vector signed int);
11404 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
11406 vector bool short vec_vcmpequh (vector signed short,
11407 vector signed short);
11408 vector bool short vec_vcmpequh (vector unsigned short,
11409 vector unsigned short);
11411 vector bool char vec_vcmpequb (vector signed char, vector signed char);
11412 vector bool char vec_vcmpequb (vector unsigned char,
11413 vector unsigned char);
11415 vector bool int vec_cmpge (vector float, vector float);
11417 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
11418 vector bool char vec_cmpgt (vector signed char, vector signed char);
11419 vector bool short vec_cmpgt (vector unsigned short,
11420 vector unsigned short);
11421 vector bool short vec_cmpgt (vector signed short, vector signed short);
11422 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
11423 vector bool int vec_cmpgt (vector signed int, vector signed int);
11424 vector bool int vec_cmpgt (vector float, vector float);
11426 vector bool int vec_vcmpgtfp (vector float, vector float);
11428 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
11430 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
11432 vector bool short vec_vcmpgtsh (vector signed short,
11433 vector signed short);
11435 vector bool short vec_vcmpgtuh (vector unsigned short,
11436 vector unsigned short);
11438 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
11440 vector bool char vec_vcmpgtub (vector unsigned char,
11441 vector unsigned char);
11443 vector bool int vec_cmple (vector float, vector float);
11445 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
11446 vector bool char vec_cmplt (vector signed char, vector signed char);
11447 vector bool short vec_cmplt (vector unsigned short,
11448 vector unsigned short);
11449 vector bool short vec_cmplt (vector signed short, vector signed short);
11450 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
11451 vector bool int vec_cmplt (vector signed int, vector signed int);
11452 vector bool int vec_cmplt (vector float, vector float);
11454 vector float vec_ctf (vector unsigned int, const int);
11455 vector float vec_ctf (vector signed int, const int);
11457 vector float vec_vcfsx (vector signed int, const int);
11459 vector float vec_vcfux (vector unsigned int, const int);
11461 vector signed int vec_cts (vector float, const int);
11463 vector unsigned int vec_ctu (vector float, const int);
11465 void vec_dss (const int);
11467 void vec_dssall (void);
11469 void vec_dst (const vector unsigned char *, int, const int);
11470 void vec_dst (const vector signed char *, int, const int);
11471 void vec_dst (const vector bool char *, int, const int);
11472 void vec_dst (const vector unsigned short *, int, const int);
11473 void vec_dst (const vector signed short *, int, const int);
11474 void vec_dst (const vector bool short *, int, const int);
11475 void vec_dst (const vector pixel *, int, const int);
11476 void vec_dst (const vector unsigned int *, int, const int);
11477 void vec_dst (const vector signed int *, int, const int);
11478 void vec_dst (const vector bool int *, int, const int);
11479 void vec_dst (const vector float *, int, const int);
11480 void vec_dst (const unsigned char *, int, const int);
11481 void vec_dst (const signed char *, int, const int);
11482 void vec_dst (const unsigned short *, int, const int);
11483 void vec_dst (const short *, int, const int);
11484 void vec_dst (const unsigned int *, int, const int);
11485 void vec_dst (const int *, int, const int);
11486 void vec_dst (const unsigned long *, int, const int);
11487 void vec_dst (const long *, int, const int);
11488 void vec_dst (const float *, int, const int);
11490 void vec_dstst (const vector unsigned char *, int, const int);
11491 void vec_dstst (const vector signed char *, int, const int);
11492 void vec_dstst (const vector bool char *, int, const int);
11493 void vec_dstst (const vector unsigned short *, int, const int);
11494 void vec_dstst (const vector signed short *, int, const int);
11495 void vec_dstst (const vector bool short *, int, const int);
11496 void vec_dstst (const vector pixel *, int, const int);
11497 void vec_dstst (const vector unsigned int *, int, const int);
11498 void vec_dstst (const vector signed int *, int, const int);
11499 void vec_dstst (const vector bool int *, int, const int);
11500 void vec_dstst (const vector float *, int, const int);
11501 void vec_dstst (const unsigned char *, int, const int);
11502 void vec_dstst (const signed char *, int, const int);
11503 void vec_dstst (const unsigned short *, int, const int);
11504 void vec_dstst (const short *, int, const int);
11505 void vec_dstst (const unsigned int *, int, const int);
11506 void vec_dstst (const int *, int, const int);
11507 void vec_dstst (const unsigned long *, int, const int);
11508 void vec_dstst (const long *, int, const int);
11509 void vec_dstst (const float *, int, const int);
11511 void vec_dststt (const vector unsigned char *, int, const int);
11512 void vec_dststt (const vector signed char *, int, const int);
11513 void vec_dststt (const vector bool char *, int, const int);
11514 void vec_dststt (const vector unsigned short *, int, const int);
11515 void vec_dststt (const vector signed short *, int, const int);
11516 void vec_dststt (const vector bool short *, int, const int);
11517 void vec_dststt (const vector pixel *, int, const int);
11518 void vec_dststt (const vector unsigned int *, int, const int);
11519 void vec_dststt (const vector signed int *, int, const int);
11520 void vec_dststt (const vector bool int *, int, const int);
11521 void vec_dststt (const vector float *, int, const int);
11522 void vec_dststt (const unsigned char *, int, const int);
11523 void vec_dststt (const signed char *, int, const int);
11524 void vec_dststt (const unsigned short *, int, const int);
11525 void vec_dststt (const short *, int, const int);
11526 void vec_dststt (const unsigned int *, int, const int);
11527 void vec_dststt (const int *, int, const int);
11528 void vec_dststt (const unsigned long *, int, const int);
11529 void vec_dststt (const long *, int, const int);
11530 void vec_dststt (const float *, int, const int);
11532 void vec_dstt (const vector unsigned char *, int, const int);
11533 void vec_dstt (const vector signed char *, int, const int);
11534 void vec_dstt (const vector bool char *, int, const int);
11535 void vec_dstt (const vector unsigned short *, int, const int);
11536 void vec_dstt (const vector signed short *, int, const int);
11537 void vec_dstt (const vector bool short *, int, const int);
11538 void vec_dstt (const vector pixel *, int, const int);
11539 void vec_dstt (const vector unsigned int *, int, const int);
11540 void vec_dstt (const vector signed int *, int, const int);
11541 void vec_dstt (const vector bool int *, int, const int);
11542 void vec_dstt (const vector float *, int, const int);
11543 void vec_dstt (const unsigned char *, int, const int);
11544 void vec_dstt (const signed char *, int, const int);
11545 void vec_dstt (const unsigned short *, int, const int);
11546 void vec_dstt (const short *, int, const int);
11547 void vec_dstt (const unsigned int *, int, const int);
11548 void vec_dstt (const int *, int, const int);
11549 void vec_dstt (const unsigned long *, int, const int);
11550 void vec_dstt (const long *, int, const int);
11551 void vec_dstt (const float *, int, const int);
11553 vector float vec_expte (vector float);
11555 vector float vec_floor (vector float);
11557 vector float vec_ld (int, const vector float *);
11558 vector float vec_ld (int, const float *);
11559 vector bool int vec_ld (int, const vector bool int *);
11560 vector signed int vec_ld (int, const vector signed int *);
11561 vector signed int vec_ld (int, const int *);
11562 vector signed int vec_ld (int, const long *);
11563 vector unsigned int vec_ld (int, const vector unsigned int *);
11564 vector unsigned int vec_ld (int, const unsigned int *);
11565 vector unsigned int vec_ld (int, const unsigned long *);
11566 vector bool short vec_ld (int, const vector bool short *);
11567 vector pixel vec_ld (int, const vector pixel *);
11568 vector signed short vec_ld (int, const vector signed short *);
11569 vector signed short vec_ld (int, const short *);
11570 vector unsigned short vec_ld (int, const vector unsigned short *);
11571 vector unsigned short vec_ld (int, const unsigned short *);
11572 vector bool char vec_ld (int, const vector bool char *);
11573 vector signed char vec_ld (int, const vector signed char *);
11574 vector signed char vec_ld (int, const signed char *);
11575 vector unsigned char vec_ld (int, const vector unsigned char *);
11576 vector unsigned char vec_ld (int, const unsigned char *);
11578 vector signed char vec_lde (int, const signed char *);
11579 vector unsigned char vec_lde (int, const unsigned char *);
11580 vector signed short vec_lde (int, const short *);
11581 vector unsigned short vec_lde (int, const unsigned short *);
11582 vector float vec_lde (int, const float *);
11583 vector signed int vec_lde (int, const int *);
11584 vector unsigned int vec_lde (int, const unsigned int *);
11585 vector signed int vec_lde (int, const long *);
11586 vector unsigned int vec_lde (int, const unsigned long *);
11588 vector float vec_lvewx (int, float *);
11589 vector signed int vec_lvewx (int, int *);
11590 vector unsigned int vec_lvewx (int, unsigned int *);
11591 vector signed int vec_lvewx (int, long *);
11592 vector unsigned int vec_lvewx (int, unsigned long *);
11594 vector signed short vec_lvehx (int, short *);
11595 vector unsigned short vec_lvehx (int, unsigned short *);
11597 vector signed char vec_lvebx (int, char *);
11598 vector unsigned char vec_lvebx (int, unsigned char *);
11600 vector float vec_ldl (int, const vector float *);
11601 vector float vec_ldl (int, const float *);
11602 vector bool int vec_ldl (int, const vector bool int *);
11603 vector signed int vec_ldl (int, const vector signed int *);
11604 vector signed int vec_ldl (int, const int *);
11605 vector signed int vec_ldl (int, const long *);
11606 vector unsigned int vec_ldl (int, const vector unsigned int *);
11607 vector unsigned int vec_ldl (int, const unsigned int *);
11608 vector unsigned int vec_ldl (int, const unsigned long *);
11609 vector bool short vec_ldl (int, const vector bool short *);
11610 vector pixel vec_ldl (int, const vector pixel *);
11611 vector signed short vec_ldl (int, const vector signed short *);
11612 vector signed short vec_ldl (int, const short *);
11613 vector unsigned short vec_ldl (int, const vector unsigned short *);
11614 vector unsigned short vec_ldl (int, const unsigned short *);
11615 vector bool char vec_ldl (int, const vector bool char *);
11616 vector signed char vec_ldl (int, const vector signed char *);
11617 vector signed char vec_ldl (int, const signed char *);
11618 vector unsigned char vec_ldl (int, const vector unsigned char *);
11619 vector unsigned char vec_ldl (int, const unsigned char *);
11621 vector float vec_loge (vector float);
11623 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
11624 vector unsigned char vec_lvsl (int, const volatile signed char *);
11625 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
11626 vector unsigned char vec_lvsl (int, const volatile short *);
11627 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
11628 vector unsigned char vec_lvsl (int, const volatile int *);
11629 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
11630 vector unsigned char vec_lvsl (int, const volatile long *);
11631 vector unsigned char vec_lvsl (int, const volatile float *);
11633 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
11634 vector unsigned char vec_lvsr (int, const volatile signed char *);
11635 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
11636 vector unsigned char vec_lvsr (int, const volatile short *);
11637 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
11638 vector unsigned char vec_lvsr (int, const volatile int *);
11639 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
11640 vector unsigned char vec_lvsr (int, const volatile long *);
11641 vector unsigned char vec_lvsr (int, const volatile float *);
11643 vector float vec_madd (vector float, vector float, vector float);
11645 vector signed short vec_madds (vector signed short,
11646 vector signed short,
11647 vector signed short);
11649 vector unsigned char vec_max (vector bool char, vector unsigned char);
11650 vector unsigned char vec_max (vector unsigned char, vector bool char);
11651 vector unsigned char vec_max (vector unsigned char,
11652 vector unsigned char);
11653 vector signed char vec_max (vector bool char, vector signed char);
11654 vector signed char vec_max (vector signed char, vector bool char);
11655 vector signed char vec_max (vector signed char, vector signed char);
11656 vector unsigned short vec_max (vector bool short,
11657 vector unsigned short);
11658 vector unsigned short vec_max (vector unsigned short,
11659 vector bool short);
11660 vector unsigned short vec_max (vector unsigned short,
11661 vector unsigned short);
11662 vector signed short vec_max (vector bool short, vector signed short);
11663 vector signed short vec_max (vector signed short, vector bool short);
11664 vector signed short vec_max (vector signed short, vector signed short);
11665 vector unsigned int vec_max (vector bool int, vector unsigned int);
11666 vector unsigned int vec_max (vector unsigned int, vector bool int);
11667 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
11668 vector signed int vec_max (vector bool int, vector signed int);
11669 vector signed int vec_max (vector signed int, vector bool int);
11670 vector signed int vec_max (vector signed int, vector signed int);
11671 vector float vec_max (vector float, vector float);
11673 vector float vec_vmaxfp (vector float, vector float);
11675 vector signed int vec_vmaxsw (vector bool int, vector signed int);
11676 vector signed int vec_vmaxsw (vector signed int, vector bool int);
11677 vector signed int vec_vmaxsw (vector signed int, vector signed int);
11679 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
11680 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
11681 vector unsigned int vec_vmaxuw (vector unsigned int,
11682 vector unsigned int);
11684 vector signed short vec_vmaxsh (vector bool short, vector signed short);
11685 vector signed short vec_vmaxsh (vector signed short, vector bool short);
11686 vector signed short vec_vmaxsh (vector signed short,
11687 vector signed short);
11689 vector unsigned short vec_vmaxuh (vector bool short,
11690 vector unsigned short);
11691 vector unsigned short vec_vmaxuh (vector unsigned short,
11692 vector bool short);
11693 vector unsigned short vec_vmaxuh (vector unsigned short,
11694 vector unsigned short);
11696 vector signed char vec_vmaxsb (vector bool char, vector signed char);
11697 vector signed char vec_vmaxsb (vector signed char, vector bool char);
11698 vector signed char vec_vmaxsb (vector signed char, vector signed char);
11700 vector unsigned char vec_vmaxub (vector bool char,
11701 vector unsigned char);
11702 vector unsigned char vec_vmaxub (vector unsigned char,
11704 vector unsigned char vec_vmaxub (vector unsigned char,
11705 vector unsigned char);
11707 vector bool char vec_mergeh (vector bool char, vector bool char);
11708 vector signed char vec_mergeh (vector signed char, vector signed char);
11709 vector unsigned char vec_mergeh (vector unsigned char,
11710 vector unsigned char);
11711 vector bool short vec_mergeh (vector bool short, vector bool short);
11712 vector pixel vec_mergeh (vector pixel, vector pixel);
11713 vector signed short vec_mergeh (vector signed short,
11714 vector signed short);
11715 vector unsigned short vec_mergeh (vector unsigned short,
11716 vector unsigned short);
11717 vector float vec_mergeh (vector float, vector float);
11718 vector bool int vec_mergeh (vector bool int, vector bool int);
11719 vector signed int vec_mergeh (vector signed int, vector signed int);
11720 vector unsigned int vec_mergeh (vector unsigned int,
11721 vector unsigned int);
11723 vector float vec_vmrghw (vector float, vector float);
11724 vector bool int vec_vmrghw (vector bool int, vector bool int);
11725 vector signed int vec_vmrghw (vector signed int, vector signed int);
11726 vector unsigned int vec_vmrghw (vector unsigned int,
11727 vector unsigned int);
11729 vector bool short vec_vmrghh (vector bool short, vector bool short);
11730 vector signed short vec_vmrghh (vector signed short,
11731 vector signed short);
11732 vector unsigned short vec_vmrghh (vector unsigned short,
11733 vector unsigned short);
11734 vector pixel vec_vmrghh (vector pixel, vector pixel);
11736 vector bool char vec_vmrghb (vector bool char, vector bool char);
11737 vector signed char vec_vmrghb (vector signed char, vector signed char);
11738 vector unsigned char vec_vmrghb (vector unsigned char,
11739 vector unsigned char);
11741 vector bool char vec_mergel (vector bool char, vector bool char);
11742 vector signed char vec_mergel (vector signed char, vector signed char);
11743 vector unsigned char vec_mergel (vector unsigned char,
11744 vector unsigned char);
11745 vector bool short vec_mergel (vector bool short, vector bool short);
11746 vector pixel vec_mergel (vector pixel, vector pixel);
11747 vector signed short vec_mergel (vector signed short,
11748 vector signed short);
11749 vector unsigned short vec_mergel (vector unsigned short,
11750 vector unsigned short);
11751 vector float vec_mergel (vector float, vector float);
11752 vector bool int vec_mergel (vector bool int, vector bool int);
11753 vector signed int vec_mergel (vector signed int, vector signed int);
11754 vector unsigned int vec_mergel (vector unsigned int,
11755 vector unsigned int);
11757 vector float vec_vmrglw (vector float, vector float);
11758 vector signed int vec_vmrglw (vector signed int, vector signed int);
11759 vector unsigned int vec_vmrglw (vector unsigned int,
11760 vector unsigned int);
11761 vector bool int vec_vmrglw (vector bool int, vector bool int);
11763 vector bool short vec_vmrglh (vector bool short, vector bool short);
11764 vector signed short vec_vmrglh (vector signed short,
11765 vector signed short);
11766 vector unsigned short vec_vmrglh (vector unsigned short,
11767 vector unsigned short);
11768 vector pixel vec_vmrglh (vector pixel, vector pixel);
11770 vector bool char vec_vmrglb (vector bool char, vector bool char);
11771 vector signed char vec_vmrglb (vector signed char, vector signed char);
11772 vector unsigned char vec_vmrglb (vector unsigned char,
11773 vector unsigned char);
11775 vector unsigned short vec_mfvscr (void);
11777 vector unsigned char vec_min (vector bool char, vector unsigned char);
11778 vector unsigned char vec_min (vector unsigned char, vector bool char);
11779 vector unsigned char vec_min (vector unsigned char,
11780 vector unsigned char);
11781 vector signed char vec_min (vector bool char, vector signed char);
11782 vector signed char vec_min (vector signed char, vector bool char);
11783 vector signed char vec_min (vector signed char, vector signed char);
11784 vector unsigned short vec_min (vector bool short,
11785 vector unsigned short);
11786 vector unsigned short vec_min (vector unsigned short,
11787 vector bool short);
11788 vector unsigned short vec_min (vector unsigned short,
11789 vector unsigned short);
11790 vector signed short vec_min (vector bool short, vector signed short);
11791 vector signed short vec_min (vector signed short, vector bool short);
11792 vector signed short vec_min (vector signed short, vector signed short);
11793 vector unsigned int vec_min (vector bool int, vector unsigned int);
11794 vector unsigned int vec_min (vector unsigned int, vector bool int);
11795 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
11796 vector signed int vec_min (vector bool int, vector signed int);
11797 vector signed int vec_min (vector signed int, vector bool int);
11798 vector signed int vec_min (vector signed int, vector signed int);
11799 vector float vec_min (vector float, vector float);
11801 vector float vec_vminfp (vector float, vector float);
11803 vector signed int vec_vminsw (vector bool int, vector signed int);
11804 vector signed int vec_vminsw (vector signed int, vector bool int);
11805 vector signed int vec_vminsw (vector signed int, vector signed int);
11807 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
11808 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
11809 vector unsigned int vec_vminuw (vector unsigned int,
11810 vector unsigned int);
11812 vector signed short vec_vminsh (vector bool short, vector signed short);
11813 vector signed short vec_vminsh (vector signed short, vector bool short);
11814 vector signed short vec_vminsh (vector signed short,
11815 vector signed short);
11817 vector unsigned short vec_vminuh (vector bool short,
11818 vector unsigned short);
11819 vector unsigned short vec_vminuh (vector unsigned short,
11820 vector bool short);
11821 vector unsigned short vec_vminuh (vector unsigned short,
11822 vector unsigned short);
11824 vector signed char vec_vminsb (vector bool char, vector signed char);
11825 vector signed char vec_vminsb (vector signed char, vector bool char);
11826 vector signed char vec_vminsb (vector signed char, vector signed char);
11828 vector unsigned char vec_vminub (vector bool char,
11829 vector unsigned char);
11830 vector unsigned char vec_vminub (vector unsigned char,
11832 vector unsigned char vec_vminub (vector unsigned char,
11833 vector unsigned char);
11835 vector signed short vec_mladd (vector signed short,
11836 vector signed short,
11837 vector signed short);
11838 vector signed short vec_mladd (vector signed short,
11839 vector unsigned short,
11840 vector unsigned short);
11841 vector signed short vec_mladd (vector unsigned short,
11842 vector signed short,
11843 vector signed short);
11844 vector unsigned short vec_mladd (vector unsigned short,
11845 vector unsigned short,
11846 vector unsigned short);
11848 vector signed short vec_mradds (vector signed short,
11849 vector signed short,
11850 vector signed short);
11852 vector unsigned int vec_msum (vector unsigned char,
11853 vector unsigned char,
11854 vector unsigned int);
11855 vector signed int vec_msum (vector signed char,
11856 vector unsigned char,
11857 vector signed int);
11858 vector unsigned int vec_msum (vector unsigned short,
11859 vector unsigned short,
11860 vector unsigned int);
11861 vector signed int vec_msum (vector signed short,
11862 vector signed short,
11863 vector signed int);
11865 vector signed int vec_vmsumshm (vector signed short,
11866 vector signed short,
11867 vector signed int);
11869 vector unsigned int vec_vmsumuhm (vector unsigned short,
11870 vector unsigned short,
11871 vector unsigned int);
11873 vector signed int vec_vmsummbm (vector signed char,
11874 vector unsigned char,
11875 vector signed int);
11877 vector unsigned int vec_vmsumubm (vector unsigned char,
11878 vector unsigned char,
11879 vector unsigned int);
11881 vector unsigned int vec_msums (vector unsigned short,
11882 vector unsigned short,
11883 vector unsigned int);
11884 vector signed int vec_msums (vector signed short,
11885 vector signed short,
11886 vector signed int);
11888 vector signed int vec_vmsumshs (vector signed short,
11889 vector signed short,
11890 vector signed int);
11892 vector unsigned int vec_vmsumuhs (vector unsigned short,
11893 vector unsigned short,
11894 vector unsigned int);
11896 void vec_mtvscr (vector signed int);
11897 void vec_mtvscr (vector unsigned int);
11898 void vec_mtvscr (vector bool int);
11899 void vec_mtvscr (vector signed short);
11900 void vec_mtvscr (vector unsigned short);
11901 void vec_mtvscr (vector bool short);
11902 void vec_mtvscr (vector pixel);
11903 void vec_mtvscr (vector signed char);
11904 void vec_mtvscr (vector unsigned char);
11905 void vec_mtvscr (vector bool char);
11907 vector unsigned short vec_mule (vector unsigned char,
11908 vector unsigned char);
11909 vector signed short vec_mule (vector signed char,
11910 vector signed char);
11911 vector unsigned int vec_mule (vector unsigned short,
11912 vector unsigned short);
11913 vector signed int vec_mule (vector signed short, vector signed short);
11915 vector signed int vec_vmulesh (vector signed short,
11916 vector signed short);
11918 vector unsigned int vec_vmuleuh (vector unsigned short,
11919 vector unsigned short);
11921 vector signed short vec_vmulesb (vector signed char,
11922 vector signed char);
11924 vector unsigned short vec_vmuleub (vector unsigned char,
11925 vector unsigned char);
11927 vector unsigned short vec_mulo (vector unsigned char,
11928 vector unsigned char);
11929 vector signed short vec_mulo (vector signed char, vector signed char);
11930 vector unsigned int vec_mulo (vector unsigned short,
11931 vector unsigned short);
11932 vector signed int vec_mulo (vector signed short, vector signed short);
11934 vector signed int vec_vmulosh (vector signed short,
11935 vector signed short);
11937 vector unsigned int vec_vmulouh (vector unsigned short,
11938 vector unsigned short);
11940 vector signed short vec_vmulosb (vector signed char,
11941 vector signed char);
11943 vector unsigned short vec_vmuloub (vector unsigned char,
11944 vector unsigned char);
11946 vector float vec_nmsub (vector float, vector float, vector float);
11948 vector float vec_nor (vector float, vector float);
11949 vector signed int vec_nor (vector signed int, vector signed int);
11950 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
11951 vector bool int vec_nor (vector bool int, vector bool int);
11952 vector signed short vec_nor (vector signed short, vector signed short);
11953 vector unsigned short vec_nor (vector unsigned short,
11954 vector unsigned short);
11955 vector bool short vec_nor (vector bool short, vector bool short);
11956 vector signed char vec_nor (vector signed char, vector signed char);
11957 vector unsigned char vec_nor (vector unsigned char,
11958 vector unsigned char);
11959 vector bool char vec_nor (vector bool char, vector bool char);
11961 vector float vec_or (vector float, vector float);
11962 vector float vec_or (vector float, vector bool int);
11963 vector float vec_or (vector bool int, vector float);
11964 vector bool int vec_or (vector bool int, vector bool int);
11965 vector signed int vec_or (vector bool int, vector signed int);
11966 vector signed int vec_or (vector signed int, vector bool int);
11967 vector signed int vec_or (vector signed int, vector signed int);
11968 vector unsigned int vec_or (vector bool int, vector unsigned int);
11969 vector unsigned int vec_or (vector unsigned int, vector bool int);
11970 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
11971 vector bool short vec_or (vector bool short, vector bool short);
11972 vector signed short vec_or (vector bool short, vector signed short);
11973 vector signed short vec_or (vector signed short, vector bool short);
11974 vector signed short vec_or (vector signed short, vector signed short);
11975 vector unsigned short vec_or (vector bool short, vector unsigned short);
11976 vector unsigned short vec_or (vector unsigned short, vector bool short);
11977 vector unsigned short vec_or (vector unsigned short,
11978 vector unsigned short);
11979 vector signed char vec_or (vector bool char, vector signed char);
11980 vector bool char vec_or (vector bool char, vector bool char);
11981 vector signed char vec_or (vector signed char, vector bool char);
11982 vector signed char vec_or (vector signed char, vector signed char);
11983 vector unsigned char vec_or (vector bool char, vector unsigned char);
11984 vector unsigned char vec_or (vector unsigned char, vector bool char);
11985 vector unsigned char vec_or (vector unsigned char,
11986 vector unsigned char);
11988 vector signed char vec_pack (vector signed short, vector signed short);
11989 vector unsigned char vec_pack (vector unsigned short,
11990 vector unsigned short);
11991 vector bool char vec_pack (vector bool short, vector bool short);
11992 vector signed short vec_pack (vector signed int, vector signed int);
11993 vector unsigned short vec_pack (vector unsigned int,
11994 vector unsigned int);
11995 vector bool short vec_pack (vector bool int, vector bool int);
11997 vector bool short vec_vpkuwum (vector bool int, vector bool int);
11998 vector signed short vec_vpkuwum (vector signed int, vector signed int);
11999 vector unsigned short vec_vpkuwum (vector unsigned int,
12000 vector unsigned int);
12002 vector bool char vec_vpkuhum (vector bool short, vector bool short);
12003 vector signed char vec_vpkuhum (vector signed short,
12004 vector signed short);
12005 vector unsigned char vec_vpkuhum (vector unsigned short,
12006 vector unsigned short);
12008 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
12010 vector unsigned char vec_packs (vector unsigned short,
12011 vector unsigned short);
12012 vector signed char vec_packs (vector signed short, vector signed short);
12013 vector unsigned short vec_packs (vector unsigned int,
12014 vector unsigned int);
12015 vector signed short vec_packs (vector signed int, vector signed int);
12017 vector signed short vec_vpkswss (vector signed int, vector signed int);
12019 vector unsigned short vec_vpkuwus (vector unsigned int,
12020 vector unsigned int);
12022 vector signed char vec_vpkshss (vector signed short,
12023 vector signed short);
12025 vector unsigned char vec_vpkuhus (vector unsigned short,
12026 vector unsigned short);
12028 vector unsigned char vec_packsu (vector unsigned short,
12029 vector unsigned short);
12030 vector unsigned char vec_packsu (vector signed short,
12031 vector signed short);
12032 vector unsigned short vec_packsu (vector unsigned int,
12033 vector unsigned int);
12034 vector unsigned short vec_packsu (vector signed int, vector signed int);
12036 vector unsigned short vec_vpkswus (vector signed int,
12037 vector signed int);
12039 vector unsigned char vec_vpkshus (vector signed short,
12040 vector signed short);
12042 vector float vec_perm (vector float,
12044 vector unsigned char);
12045 vector signed int vec_perm (vector signed int,
12047 vector unsigned char);
12048 vector unsigned int vec_perm (vector unsigned int,
12049 vector unsigned int,
12050 vector unsigned char);
12051 vector bool int vec_perm (vector bool int,
12053 vector unsigned char);
12054 vector signed short vec_perm (vector signed short,
12055 vector signed short,
12056 vector unsigned char);
12057 vector unsigned short vec_perm (vector unsigned short,
12058 vector unsigned short,
12059 vector unsigned char);
12060 vector bool short vec_perm (vector bool short,
12062 vector unsigned char);
12063 vector pixel vec_perm (vector pixel,
12065 vector unsigned char);
12066 vector signed char vec_perm (vector signed char,
12067 vector signed char,
12068 vector unsigned char);
12069 vector unsigned char vec_perm (vector unsigned char,
12070 vector unsigned char,
12071 vector unsigned char);
12072 vector bool char vec_perm (vector bool char,
12074 vector unsigned char);
12076 vector float vec_re (vector float);
12078 vector signed char vec_rl (vector signed char,
12079 vector unsigned char);
12080 vector unsigned char vec_rl (vector unsigned char,
12081 vector unsigned char);
12082 vector signed short vec_rl (vector signed short, vector unsigned short);
12083 vector unsigned short vec_rl (vector unsigned short,
12084 vector unsigned short);
12085 vector signed int vec_rl (vector signed int, vector unsigned int);
12086 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
12088 vector signed int vec_vrlw (vector signed int, vector unsigned int);
12089 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
12091 vector signed short vec_vrlh (vector signed short,
12092 vector unsigned short);
12093 vector unsigned short vec_vrlh (vector unsigned short,
12094 vector unsigned short);
12096 vector signed char vec_vrlb (vector signed char, vector unsigned char);
12097 vector unsigned char vec_vrlb (vector unsigned char,
12098 vector unsigned char);
12100 vector float vec_round (vector float);
12102 vector float vec_recip (vector float, vector float);
12104 vector float vec_rsqrt (vector float);
12106 vector float vec_rsqrte (vector float);
12108 vector float vec_sel (vector float, vector float, vector bool int);
12109 vector float vec_sel (vector float, vector float, vector unsigned int);
12110 vector signed int vec_sel (vector signed int,
12113 vector signed int vec_sel (vector signed int,
12115 vector unsigned int);
12116 vector unsigned int vec_sel (vector unsigned int,
12117 vector unsigned int,
12119 vector unsigned int vec_sel (vector unsigned int,
12120 vector unsigned int,
12121 vector unsigned int);
12122 vector bool int vec_sel (vector bool int,
12125 vector bool int vec_sel (vector bool int,
12127 vector unsigned int);
12128 vector signed short vec_sel (vector signed short,
12129 vector signed short,
12130 vector bool short);
12131 vector signed short vec_sel (vector signed short,
12132 vector signed short,
12133 vector unsigned short);
12134 vector unsigned short vec_sel (vector unsigned short,
12135 vector unsigned short,
12136 vector bool short);
12137 vector unsigned short vec_sel (vector unsigned short,
12138 vector unsigned short,
12139 vector unsigned short);
12140 vector bool short vec_sel (vector bool short,
12142 vector bool short);
12143 vector bool short vec_sel (vector bool short,
12145 vector unsigned short);
12146 vector signed char vec_sel (vector signed char,
12147 vector signed char,
12149 vector signed char vec_sel (vector signed char,
12150 vector signed char,
12151 vector unsigned char);
12152 vector unsigned char vec_sel (vector unsigned char,
12153 vector unsigned char,
12155 vector unsigned char vec_sel (vector unsigned char,
12156 vector unsigned char,
12157 vector unsigned char);
12158 vector bool char vec_sel (vector bool char,
12161 vector bool char vec_sel (vector bool char,
12163 vector unsigned char);
12165 vector signed char vec_sl (vector signed char,
12166 vector unsigned char);
12167 vector unsigned char vec_sl (vector unsigned char,
12168 vector unsigned char);
12169 vector signed short vec_sl (vector signed short, vector unsigned short);
12170 vector unsigned short vec_sl (vector unsigned short,
12171 vector unsigned short);
12172 vector signed int vec_sl (vector signed int, vector unsigned int);
12173 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
12175 vector signed int vec_vslw (vector signed int, vector unsigned int);
12176 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
12178 vector signed short vec_vslh (vector signed short,
12179 vector unsigned short);
12180 vector unsigned short vec_vslh (vector unsigned short,
12181 vector unsigned short);
12183 vector signed char vec_vslb (vector signed char, vector unsigned char);
12184 vector unsigned char vec_vslb (vector unsigned char,
12185 vector unsigned char);
12187 vector float vec_sld (vector float, vector float, const int);
12188 vector signed int vec_sld (vector signed int,
12191 vector unsigned int vec_sld (vector unsigned int,
12192 vector unsigned int,
12194 vector bool int vec_sld (vector bool int,
12197 vector signed short vec_sld (vector signed short,
12198 vector signed short,
12200 vector unsigned short vec_sld (vector unsigned short,
12201 vector unsigned short,
12203 vector bool short vec_sld (vector bool short,
12206 vector pixel vec_sld (vector pixel,
12209 vector signed char vec_sld (vector signed char,
12210 vector signed char,
12212 vector unsigned char vec_sld (vector unsigned char,
12213 vector unsigned char,
12215 vector bool char vec_sld (vector bool char,
12219 vector signed int vec_sll (vector signed int,
12220 vector unsigned int);
12221 vector signed int vec_sll (vector signed int,
12222 vector unsigned short);
12223 vector signed int vec_sll (vector signed int,
12224 vector unsigned char);
12225 vector unsigned int vec_sll (vector unsigned int,
12226 vector unsigned int);
12227 vector unsigned int vec_sll (vector unsigned int,
12228 vector unsigned short);
12229 vector unsigned int vec_sll (vector unsigned int,
12230 vector unsigned char);
12231 vector bool int vec_sll (vector bool int,
12232 vector unsigned int);
12233 vector bool int vec_sll (vector bool int,
12234 vector unsigned short);
12235 vector bool int vec_sll (vector bool int,
12236 vector unsigned char);
12237 vector signed short vec_sll (vector signed short,
12238 vector unsigned int);
12239 vector signed short vec_sll (vector signed short,
12240 vector unsigned short);
12241 vector signed short vec_sll (vector signed short,
12242 vector unsigned char);
12243 vector unsigned short vec_sll (vector unsigned short,
12244 vector unsigned int);
12245 vector unsigned short vec_sll (vector unsigned short,
12246 vector unsigned short);
12247 vector unsigned short vec_sll (vector unsigned short,
12248 vector unsigned char);
12249 vector bool short vec_sll (vector bool short, vector unsigned int);
12250 vector bool short vec_sll (vector bool short, vector unsigned short);
12251 vector bool short vec_sll (vector bool short, vector unsigned char);
12252 vector pixel vec_sll (vector pixel, vector unsigned int);
12253 vector pixel vec_sll (vector pixel, vector unsigned short);
12254 vector pixel vec_sll (vector pixel, vector unsigned char);
12255 vector signed char vec_sll (vector signed char, vector unsigned int);
12256 vector signed char vec_sll (vector signed char, vector unsigned short);
12257 vector signed char vec_sll (vector signed char, vector unsigned char);
12258 vector unsigned char vec_sll (vector unsigned char,
12259 vector unsigned int);
12260 vector unsigned char vec_sll (vector unsigned char,
12261 vector unsigned short);
12262 vector unsigned char vec_sll (vector unsigned char,
12263 vector unsigned char);
12264 vector bool char vec_sll (vector bool char, vector unsigned int);
12265 vector bool char vec_sll (vector bool char, vector unsigned short);
12266 vector bool char vec_sll (vector bool char, vector unsigned char);
12268 vector float vec_slo (vector float, vector signed char);
12269 vector float vec_slo (vector float, vector unsigned char);
12270 vector signed int vec_slo (vector signed int, vector signed char);
12271 vector signed int vec_slo (vector signed int, vector unsigned char);
12272 vector unsigned int vec_slo (vector unsigned int, vector signed char);
12273 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
12274 vector signed short vec_slo (vector signed short, vector signed char);
12275 vector signed short vec_slo (vector signed short, vector unsigned char);
12276 vector unsigned short vec_slo (vector unsigned short,
12277 vector signed char);
12278 vector unsigned short vec_slo (vector unsigned short,
12279 vector unsigned char);
12280 vector pixel vec_slo (vector pixel, vector signed char);
12281 vector pixel vec_slo (vector pixel, vector unsigned char);
12282 vector signed char vec_slo (vector signed char, vector signed char);
12283 vector signed char vec_slo (vector signed char, vector unsigned char);
12284 vector unsigned char vec_slo (vector unsigned char, vector signed char);
12285 vector unsigned char vec_slo (vector unsigned char,
12286 vector unsigned char);
12288 vector signed char vec_splat (vector signed char, const int);
12289 vector unsigned char vec_splat (vector unsigned char, const int);
12290 vector bool char vec_splat (vector bool char, const int);
12291 vector signed short vec_splat (vector signed short, const int);
12292 vector unsigned short vec_splat (vector unsigned short, const int);
12293 vector bool short vec_splat (vector bool short, const int);
12294 vector pixel vec_splat (vector pixel, const int);
12295 vector float vec_splat (vector float, const int);
12296 vector signed int vec_splat (vector signed int, const int);
12297 vector unsigned int vec_splat (vector unsigned int, const int);
12298 vector bool int vec_splat (vector bool int, const int);
12300 vector float vec_vspltw (vector float, const int);
12301 vector signed int vec_vspltw (vector signed int, const int);
12302 vector unsigned int vec_vspltw (vector unsigned int, const int);
12303 vector bool int vec_vspltw (vector bool int, const int);
12305 vector bool short vec_vsplth (vector bool short, const int);
12306 vector signed short vec_vsplth (vector signed short, const int);
12307 vector unsigned short vec_vsplth (vector unsigned short, const int);
12308 vector pixel vec_vsplth (vector pixel, const int);
12310 vector signed char vec_vspltb (vector signed char, const int);
12311 vector unsigned char vec_vspltb (vector unsigned char, const int);
12312 vector bool char vec_vspltb (vector bool char, const int);
12314 vector signed char vec_splat_s8 (const int);
12316 vector signed short vec_splat_s16 (const int);
12318 vector signed int vec_splat_s32 (const int);
12320 vector unsigned char vec_splat_u8 (const int);
12322 vector unsigned short vec_splat_u16 (const int);
12324 vector unsigned int vec_splat_u32 (const int);
12326 vector signed char vec_sr (vector signed char, vector unsigned char);
12327 vector unsigned char vec_sr (vector unsigned char,
12328 vector unsigned char);
12329 vector signed short vec_sr (vector signed short,
12330 vector unsigned short);
12331 vector unsigned short vec_sr (vector unsigned short,
12332 vector unsigned short);
12333 vector signed int vec_sr (vector signed int, vector unsigned int);
12334 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
12336 vector signed int vec_vsrw (vector signed int, vector unsigned int);
12337 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
12339 vector signed short vec_vsrh (vector signed short,
12340 vector unsigned short);
12341 vector unsigned short vec_vsrh (vector unsigned short,
12342 vector unsigned short);
12344 vector signed char vec_vsrb (vector signed char, vector unsigned char);
12345 vector unsigned char vec_vsrb (vector unsigned char,
12346 vector unsigned char);
12348 vector signed char vec_sra (vector signed char, vector unsigned char);
12349 vector unsigned char vec_sra (vector unsigned char,
12350 vector unsigned char);
12351 vector signed short vec_sra (vector signed short,
12352 vector unsigned short);
12353 vector unsigned short vec_sra (vector unsigned short,
12354 vector unsigned short);
12355 vector signed int vec_sra (vector signed int, vector unsigned int);
12356 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
12358 vector signed int vec_vsraw (vector signed int, vector unsigned int);
12359 vector unsigned int vec_vsraw (vector unsigned int,
12360 vector unsigned int);
12362 vector signed short vec_vsrah (vector signed short,
12363 vector unsigned short);
12364 vector unsigned short vec_vsrah (vector unsigned short,
12365 vector unsigned short);
12367 vector signed char vec_vsrab (vector signed char, vector unsigned char);
12368 vector unsigned char vec_vsrab (vector unsigned char,
12369 vector unsigned char);
12371 vector signed int vec_srl (vector signed int, vector unsigned int);
12372 vector signed int vec_srl (vector signed int, vector unsigned short);
12373 vector signed int vec_srl (vector signed int, vector unsigned char);
12374 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
12375 vector unsigned int vec_srl (vector unsigned int,
12376 vector unsigned short);
12377 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
12378 vector bool int vec_srl (vector bool int, vector unsigned int);
12379 vector bool int vec_srl (vector bool int, vector unsigned short);
12380 vector bool int vec_srl (vector bool int, vector unsigned char);
12381 vector signed short vec_srl (vector signed short, vector unsigned int);
12382 vector signed short vec_srl (vector signed short,
12383 vector unsigned short);
12384 vector signed short vec_srl (vector signed short, vector unsigned char);
12385 vector unsigned short vec_srl (vector unsigned short,
12386 vector unsigned int);
12387 vector unsigned short vec_srl (vector unsigned short,
12388 vector unsigned short);
12389 vector unsigned short vec_srl (vector unsigned short,
12390 vector unsigned char);
12391 vector bool short vec_srl (vector bool short, vector unsigned int);
12392 vector bool short vec_srl (vector bool short, vector unsigned short);
12393 vector bool short vec_srl (vector bool short, vector unsigned char);
12394 vector pixel vec_srl (vector pixel, vector unsigned int);
12395 vector pixel vec_srl (vector pixel, vector unsigned short);
12396 vector pixel vec_srl (vector pixel, vector unsigned char);
12397 vector signed char vec_srl (vector signed char, vector unsigned int);
12398 vector signed char vec_srl (vector signed char, vector unsigned short);
12399 vector signed char vec_srl (vector signed char, vector unsigned char);
12400 vector unsigned char vec_srl (vector unsigned char,
12401 vector unsigned int);
12402 vector unsigned char vec_srl (vector unsigned char,
12403 vector unsigned short);
12404 vector unsigned char vec_srl (vector unsigned char,
12405 vector unsigned char);
12406 vector bool char vec_srl (vector bool char, vector unsigned int);
12407 vector bool char vec_srl (vector bool char, vector unsigned short);
12408 vector bool char vec_srl (vector bool char, vector unsigned char);
12410 vector float vec_sro (vector float, vector signed char);
12411 vector float vec_sro (vector float, vector unsigned char);
12412 vector signed int vec_sro (vector signed int, vector signed char);
12413 vector signed int vec_sro (vector signed int, vector unsigned char);
12414 vector unsigned int vec_sro (vector unsigned int, vector signed char);
12415 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
12416 vector signed short vec_sro (vector signed short, vector signed char);
12417 vector signed short vec_sro (vector signed short, vector unsigned char);
12418 vector unsigned short vec_sro (vector unsigned short,
12419 vector signed char);
12420 vector unsigned short vec_sro (vector unsigned short,
12421 vector unsigned char);
12422 vector pixel vec_sro (vector pixel, vector signed char);
12423 vector pixel vec_sro (vector pixel, vector unsigned char);
12424 vector signed char vec_sro (vector signed char, vector signed char);
12425 vector signed char vec_sro (vector signed char, vector unsigned char);
12426 vector unsigned char vec_sro (vector unsigned char, vector signed char);
12427 vector unsigned char vec_sro (vector unsigned char,
12428 vector unsigned char);
12430 void vec_st (vector float, int, vector float *);
12431 void vec_st (vector float, int, float *);
12432 void vec_st (vector signed int, int, vector signed int *);
12433 void vec_st (vector signed int, int, int *);
12434 void vec_st (vector unsigned int, int, vector unsigned int *);
12435 void vec_st (vector unsigned int, int, unsigned int *);
12436 void vec_st (vector bool int, int, vector bool int *);
12437 void vec_st (vector bool int, int, unsigned int *);
12438 void vec_st (vector bool int, int, int *);
12439 void vec_st (vector signed short, int, vector signed short *);
12440 void vec_st (vector signed short, int, short *);
12441 void vec_st (vector unsigned short, int, vector unsigned short *);
12442 void vec_st (vector unsigned short, int, unsigned short *);
12443 void vec_st (vector bool short, int, vector bool short *);
12444 void vec_st (vector bool short, int, unsigned short *);
12445 void vec_st (vector pixel, int, vector pixel *);
12446 void vec_st (vector pixel, int, unsigned short *);
12447 void vec_st (vector pixel, int, short *);
12448 void vec_st (vector bool short, int, short *);
12449 void vec_st (vector signed char, int, vector signed char *);
12450 void vec_st (vector signed char, int, signed char *);
12451 void vec_st (vector unsigned char, int, vector unsigned char *);
12452 void vec_st (vector unsigned char, int, unsigned char *);
12453 void vec_st (vector bool char, int, vector bool char *);
12454 void vec_st (vector bool char, int, unsigned char *);
12455 void vec_st (vector bool char, int, signed char *);
12457 void vec_ste (vector signed char, int, signed char *);
12458 void vec_ste (vector unsigned char, int, unsigned char *);
12459 void vec_ste (vector bool char, int, signed char *);
12460 void vec_ste (vector bool char, int, unsigned char *);
12461 void vec_ste (vector signed short, int, short *);
12462 void vec_ste (vector unsigned short, int, unsigned short *);
12463 void vec_ste (vector bool short, int, short *);
12464 void vec_ste (vector bool short, int, unsigned short *);
12465 void vec_ste (vector pixel, int, short *);
12466 void vec_ste (vector pixel, int, unsigned short *);
12467 void vec_ste (vector float, int, float *);
12468 void vec_ste (vector signed int, int, int *);
12469 void vec_ste (vector unsigned int, int, unsigned int *);
12470 void vec_ste (vector bool int, int, int *);
12471 void vec_ste (vector bool int, int, unsigned int *);
12473 void vec_stvewx (vector float, int, float *);
12474 void vec_stvewx (vector signed int, int, int *);
12475 void vec_stvewx (vector unsigned int, int, unsigned int *);
12476 void vec_stvewx (vector bool int, int, int *);
12477 void vec_stvewx (vector bool int, int, unsigned int *);
12479 void vec_stvehx (vector signed short, int, short *);
12480 void vec_stvehx (vector unsigned short, int, unsigned short *);
12481 void vec_stvehx (vector bool short, int, short *);
12482 void vec_stvehx (vector bool short, int, unsigned short *);
12483 void vec_stvehx (vector pixel, int, short *);
12484 void vec_stvehx (vector pixel, int, unsigned short *);
12486 void vec_stvebx (vector signed char, int, signed char *);
12487 void vec_stvebx (vector unsigned char, int, unsigned char *);
12488 void vec_stvebx (vector bool char, int, signed char *);
12489 void vec_stvebx (vector bool char, int, unsigned char *);
12491 void vec_stl (vector float, int, vector float *);
12492 void vec_stl (vector float, int, float *);
12493 void vec_stl (vector signed int, int, vector signed int *);
12494 void vec_stl (vector signed int, int, int *);
12495 void vec_stl (vector unsigned int, int, vector unsigned int *);
12496 void vec_stl (vector unsigned int, int, unsigned int *);
12497 void vec_stl (vector bool int, int, vector bool int *);
12498 void vec_stl (vector bool int, int, unsigned int *);
12499 void vec_stl (vector bool int, int, int *);
12500 void vec_stl (vector signed short, int, vector signed short *);
12501 void vec_stl (vector signed short, int, short *);
12502 void vec_stl (vector unsigned short, int, vector unsigned short *);
12503 void vec_stl (vector unsigned short, int, unsigned short *);
12504 void vec_stl (vector bool short, int, vector bool short *);
12505 void vec_stl (vector bool short, int, unsigned short *);
12506 void vec_stl (vector bool short, int, short *);
12507 void vec_stl (vector pixel, int, vector pixel *);
12508 void vec_stl (vector pixel, int, unsigned short *);
12509 void vec_stl (vector pixel, int, short *);
12510 void vec_stl (vector signed char, int, vector signed char *);
12511 void vec_stl (vector signed char, int, signed char *);
12512 void vec_stl (vector unsigned char, int, vector unsigned char *);
12513 void vec_stl (vector unsigned char, int, unsigned char *);
12514 void vec_stl (vector bool char, int, vector bool char *);
12515 void vec_stl (vector bool char, int, unsigned char *);
12516 void vec_stl (vector bool char, int, signed char *);
12518 vector signed char vec_sub (vector bool char, vector signed char);
12519 vector signed char vec_sub (vector signed char, vector bool char);
12520 vector signed char vec_sub (vector signed char, vector signed char);
12521 vector unsigned char vec_sub (vector bool char, vector unsigned char);
12522 vector unsigned char vec_sub (vector unsigned char, vector bool char);
12523 vector unsigned char vec_sub (vector unsigned char,
12524 vector unsigned char);
12525 vector signed short vec_sub (vector bool short, vector signed short);
12526 vector signed short vec_sub (vector signed short, vector bool short);
12527 vector signed short vec_sub (vector signed short, vector signed short);
12528 vector unsigned short vec_sub (vector bool short,
12529 vector unsigned short);
12530 vector unsigned short vec_sub (vector unsigned short,
12531 vector bool short);
12532 vector unsigned short vec_sub (vector unsigned short,
12533 vector unsigned short);
12534 vector signed int vec_sub (vector bool int, vector signed int);
12535 vector signed int vec_sub (vector signed int, vector bool int);
12536 vector signed int vec_sub (vector signed int, vector signed int);
12537 vector unsigned int vec_sub (vector bool int, vector unsigned int);
12538 vector unsigned int vec_sub (vector unsigned int, vector bool int);
12539 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
12540 vector float vec_sub (vector float, vector float);
12542 vector float vec_vsubfp (vector float, vector float);
12544 vector signed int vec_vsubuwm (vector bool int, vector signed int);
12545 vector signed int vec_vsubuwm (vector signed int, vector bool int);
12546 vector signed int vec_vsubuwm (vector signed int, vector signed int);
12547 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
12548 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
12549 vector unsigned int vec_vsubuwm (vector unsigned int,
12550 vector unsigned int);
12552 vector signed short vec_vsubuhm (vector bool short,
12553 vector signed short);
12554 vector signed short vec_vsubuhm (vector signed short,
12555 vector bool short);
12556 vector signed short vec_vsubuhm (vector signed short,
12557 vector signed short);
12558 vector unsigned short vec_vsubuhm (vector bool short,
12559 vector unsigned short);
12560 vector unsigned short vec_vsubuhm (vector unsigned short,
12561 vector bool short);
12562 vector unsigned short vec_vsubuhm (vector unsigned short,
12563 vector unsigned short);
12565 vector signed char vec_vsububm (vector bool char, vector signed char);
12566 vector signed char vec_vsububm (vector signed char, vector bool char);
12567 vector signed char vec_vsububm (vector signed char, vector signed char);
12568 vector unsigned char vec_vsububm (vector bool char,
12569 vector unsigned char);
12570 vector unsigned char vec_vsububm (vector unsigned char,
12572 vector unsigned char vec_vsububm (vector unsigned char,
12573 vector unsigned char);
12575 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
12577 vector unsigned char vec_subs (vector bool char, vector unsigned char);
12578 vector unsigned char vec_subs (vector unsigned char, vector bool char);
12579 vector unsigned char vec_subs (vector unsigned char,
12580 vector unsigned char);
12581 vector signed char vec_subs (vector bool char, vector signed char);
12582 vector signed char vec_subs (vector signed char, vector bool char);
12583 vector signed char vec_subs (vector signed char, vector signed char);
12584 vector unsigned short vec_subs (vector bool short,
12585 vector unsigned short);
12586 vector unsigned short vec_subs (vector unsigned short,
12587 vector bool short);
12588 vector unsigned short vec_subs (vector unsigned short,
12589 vector unsigned short);
12590 vector signed short vec_subs (vector bool short, vector signed short);
12591 vector signed short vec_subs (vector signed short, vector bool short);
12592 vector signed short vec_subs (vector signed short, vector signed short);
12593 vector unsigned int vec_subs (vector bool int, vector unsigned int);
12594 vector unsigned int vec_subs (vector unsigned int, vector bool int);
12595 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
12596 vector signed int vec_subs (vector bool int, vector signed int);
12597 vector signed int vec_subs (vector signed int, vector bool int);
12598 vector signed int vec_subs (vector signed int, vector signed int);
12600 vector signed int vec_vsubsws (vector bool int, vector signed int);
12601 vector signed int vec_vsubsws (vector signed int, vector bool int);
12602 vector signed int vec_vsubsws (vector signed int, vector signed int);
12604 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
12605 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
12606 vector unsigned int vec_vsubuws (vector unsigned int,
12607 vector unsigned int);
12609 vector signed short vec_vsubshs (vector bool short,
12610 vector signed short);
12611 vector signed short vec_vsubshs (vector signed short,
12612 vector bool short);
12613 vector signed short vec_vsubshs (vector signed short,
12614 vector signed short);
12616 vector unsigned short vec_vsubuhs (vector bool short,
12617 vector unsigned short);
12618 vector unsigned short vec_vsubuhs (vector unsigned short,
12619 vector bool short);
12620 vector unsigned short vec_vsubuhs (vector unsigned short,
12621 vector unsigned short);
12623 vector signed char vec_vsubsbs (vector bool char, vector signed char);
12624 vector signed char vec_vsubsbs (vector signed char, vector bool char);
12625 vector signed char vec_vsubsbs (vector signed char, vector signed char);
12627 vector unsigned char vec_vsububs (vector bool char,
12628 vector unsigned char);
12629 vector unsigned char vec_vsububs (vector unsigned char,
12631 vector unsigned char vec_vsububs (vector unsigned char,
12632 vector unsigned char);
12634 vector unsigned int vec_sum4s (vector unsigned char,
12635 vector unsigned int);
12636 vector signed int vec_sum4s (vector signed char, vector signed int);
12637 vector signed int vec_sum4s (vector signed short, vector signed int);
12639 vector signed int vec_vsum4shs (vector signed short, vector signed int);
12641 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
12643 vector unsigned int vec_vsum4ubs (vector unsigned char,
12644 vector unsigned int);
12646 vector signed int vec_sum2s (vector signed int, vector signed int);
12648 vector signed int vec_sums (vector signed int, vector signed int);
12650 vector float vec_trunc (vector float);
12652 vector signed short vec_unpackh (vector signed char);
12653 vector bool short vec_unpackh (vector bool char);
12654 vector signed int vec_unpackh (vector signed short);
12655 vector bool int vec_unpackh (vector bool short);
12656 vector unsigned int vec_unpackh (vector pixel);
12658 vector bool int vec_vupkhsh (vector bool short);
12659 vector signed int vec_vupkhsh (vector signed short);
12661 vector unsigned int vec_vupkhpx (vector pixel);
12663 vector bool short vec_vupkhsb (vector bool char);
12664 vector signed short vec_vupkhsb (vector signed char);
12666 vector signed short vec_unpackl (vector signed char);
12667 vector bool short vec_unpackl (vector bool char);
12668 vector unsigned int vec_unpackl (vector pixel);
12669 vector signed int vec_unpackl (vector signed short);
12670 vector bool int vec_unpackl (vector bool short);
12672 vector unsigned int vec_vupklpx (vector pixel);
12674 vector bool int vec_vupklsh (vector bool short);
12675 vector signed int vec_vupklsh (vector signed short);
12677 vector bool short vec_vupklsb (vector bool char);
12678 vector signed short vec_vupklsb (vector signed char);
12680 vector float vec_xor (vector float, vector float);
12681 vector float vec_xor (vector float, vector bool int);
12682 vector float vec_xor (vector bool int, vector float);
12683 vector bool int vec_xor (vector bool int, vector bool int);
12684 vector signed int vec_xor (vector bool int, vector signed int);
12685 vector signed int vec_xor (vector signed int, vector bool int);
12686 vector signed int vec_xor (vector signed int, vector signed int);
12687 vector unsigned int vec_xor (vector bool int, vector unsigned int);
12688 vector unsigned int vec_xor (vector unsigned int, vector bool int);
12689 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
12690 vector bool short vec_xor (vector bool short, vector bool short);
12691 vector signed short vec_xor (vector bool short, vector signed short);
12692 vector signed short vec_xor (vector signed short, vector bool short);
12693 vector signed short vec_xor (vector signed short, vector signed short);
12694 vector unsigned short vec_xor (vector bool short,
12695 vector unsigned short);
12696 vector unsigned short vec_xor (vector unsigned short,
12697 vector bool short);
12698 vector unsigned short vec_xor (vector unsigned short,
12699 vector unsigned short);
12700 vector signed char vec_xor (vector bool char, vector signed char);
12701 vector bool char vec_xor (vector bool char, vector bool char);
12702 vector signed char vec_xor (vector signed char, vector bool char);
12703 vector signed char vec_xor (vector signed char, vector signed char);
12704 vector unsigned char vec_xor (vector bool char, vector unsigned char);
12705 vector unsigned char vec_xor (vector unsigned char, vector bool char);
12706 vector unsigned char vec_xor (vector unsigned char,
12707 vector unsigned char);
12709 int vec_all_eq (vector signed char, vector bool char);
12710 int vec_all_eq (vector signed char, vector signed char);
12711 int vec_all_eq (vector unsigned char, vector bool char);
12712 int vec_all_eq (vector unsigned char, vector unsigned char);
12713 int vec_all_eq (vector bool char, vector bool char);
12714 int vec_all_eq (vector bool char, vector unsigned char);
12715 int vec_all_eq (vector bool char, vector signed char);
12716 int vec_all_eq (vector signed short, vector bool short);
12717 int vec_all_eq (vector signed short, vector signed short);
12718 int vec_all_eq (vector unsigned short, vector bool short);
12719 int vec_all_eq (vector unsigned short, vector unsigned short);
12720 int vec_all_eq (vector bool short, vector bool short);
12721 int vec_all_eq (vector bool short, vector unsigned short);
12722 int vec_all_eq (vector bool short, vector signed short);
12723 int vec_all_eq (vector pixel, vector pixel);
12724 int vec_all_eq (vector signed int, vector bool int);
12725 int vec_all_eq (vector signed int, vector signed int);
12726 int vec_all_eq (vector unsigned int, vector bool int);
12727 int vec_all_eq (vector unsigned int, vector unsigned int);
12728 int vec_all_eq (vector bool int, vector bool int);
12729 int vec_all_eq (vector bool int, vector unsigned int);
12730 int vec_all_eq (vector bool int, vector signed int);
12731 int vec_all_eq (vector float, vector float);
12733 int vec_all_ge (vector bool char, vector unsigned char);
12734 int vec_all_ge (vector unsigned char, vector bool char);
12735 int vec_all_ge (vector unsigned char, vector unsigned char);
12736 int vec_all_ge (vector bool char, vector signed char);
12737 int vec_all_ge (vector signed char, vector bool char);
12738 int vec_all_ge (vector signed char, vector signed char);
12739 int vec_all_ge (vector bool short, vector unsigned short);
12740 int vec_all_ge (vector unsigned short, vector bool short);
12741 int vec_all_ge (vector unsigned short, vector unsigned short);
12742 int vec_all_ge (vector signed short, vector signed short);
12743 int vec_all_ge (vector bool short, vector signed short);
12744 int vec_all_ge (vector signed short, vector bool short);
12745 int vec_all_ge (vector bool int, vector unsigned int);
12746 int vec_all_ge (vector unsigned int, vector bool int);
12747 int vec_all_ge (vector unsigned int, vector unsigned int);
12748 int vec_all_ge (vector bool int, vector signed int);
12749 int vec_all_ge (vector signed int, vector bool int);
12750 int vec_all_ge (vector signed int, vector signed int);
12751 int vec_all_ge (vector float, vector float);
12753 int vec_all_gt (vector bool char, vector unsigned char);
12754 int vec_all_gt (vector unsigned char, vector bool char);
12755 int vec_all_gt (vector unsigned char, vector unsigned char);
12756 int vec_all_gt (vector bool char, vector signed char);
12757 int vec_all_gt (vector signed char, vector bool char);
12758 int vec_all_gt (vector signed char, vector signed char);
12759 int vec_all_gt (vector bool short, vector unsigned short);
12760 int vec_all_gt (vector unsigned short, vector bool short);
12761 int vec_all_gt (vector unsigned short, vector unsigned short);
12762 int vec_all_gt (vector bool short, vector signed short);
12763 int vec_all_gt (vector signed short, vector bool short);
12764 int vec_all_gt (vector signed short, vector signed short);
12765 int vec_all_gt (vector bool int, vector unsigned int);
12766 int vec_all_gt (vector unsigned int, vector bool int);
12767 int vec_all_gt (vector unsigned int, vector unsigned int);
12768 int vec_all_gt (vector bool int, vector signed int);
12769 int vec_all_gt (vector signed int, vector bool int);
12770 int vec_all_gt (vector signed int, vector signed int);
12771 int vec_all_gt (vector float, vector float);
12773 int vec_all_in (vector float, vector float);
12775 int vec_all_le (vector bool char, vector unsigned char);
12776 int vec_all_le (vector unsigned char, vector bool char);
12777 int vec_all_le (vector unsigned char, vector unsigned char);
12778 int vec_all_le (vector bool char, vector signed char);
12779 int vec_all_le (vector signed char, vector bool char);
12780 int vec_all_le (vector signed char, vector signed char);
12781 int vec_all_le (vector bool short, vector unsigned short);
12782 int vec_all_le (vector unsigned short, vector bool short);
12783 int vec_all_le (vector unsigned short, vector unsigned short);
12784 int vec_all_le (vector bool short, vector signed short);
12785 int vec_all_le (vector signed short, vector bool short);
12786 int vec_all_le (vector signed short, vector signed short);
12787 int vec_all_le (vector bool int, vector unsigned int);
12788 int vec_all_le (vector unsigned int, vector bool int);
12789 int vec_all_le (vector unsigned int, vector unsigned int);
12790 int vec_all_le (vector bool int, vector signed int);
12791 int vec_all_le (vector signed int, vector bool int);
12792 int vec_all_le (vector signed int, vector signed int);
12793 int vec_all_le (vector float, vector float);
12795 int vec_all_lt (vector bool char, vector unsigned char);
12796 int vec_all_lt (vector unsigned char, vector bool char);
12797 int vec_all_lt (vector unsigned char, vector unsigned char);
12798 int vec_all_lt (vector bool char, vector signed char);
12799 int vec_all_lt (vector signed char, vector bool char);
12800 int vec_all_lt (vector signed char, vector signed char);
12801 int vec_all_lt (vector bool short, vector unsigned short);
12802 int vec_all_lt (vector unsigned short, vector bool short);
12803 int vec_all_lt (vector unsigned short, vector unsigned short);
12804 int vec_all_lt (vector bool short, vector signed short);
12805 int vec_all_lt (vector signed short, vector bool short);
12806 int vec_all_lt (vector signed short, vector signed short);
12807 int vec_all_lt (vector bool int, vector unsigned int);
12808 int vec_all_lt (vector unsigned int, vector bool int);
12809 int vec_all_lt (vector unsigned int, vector unsigned int);
12810 int vec_all_lt (vector bool int, vector signed int);
12811 int vec_all_lt (vector signed int, vector bool int);
12812 int vec_all_lt (vector signed int, vector signed int);
12813 int vec_all_lt (vector float, vector float);
12815 int vec_all_nan (vector float);
12817 int vec_all_ne (vector signed char, vector bool char);
12818 int vec_all_ne (vector signed char, vector signed char);
12819 int vec_all_ne (vector unsigned char, vector bool char);
12820 int vec_all_ne (vector unsigned char, vector unsigned char);
12821 int vec_all_ne (vector bool char, vector bool char);
12822 int vec_all_ne (vector bool char, vector unsigned char);
12823 int vec_all_ne (vector bool char, vector signed char);
12824 int vec_all_ne (vector signed short, vector bool short);
12825 int vec_all_ne (vector signed short, vector signed short);
12826 int vec_all_ne (vector unsigned short, vector bool short);
12827 int vec_all_ne (vector unsigned short, vector unsigned short);
12828 int vec_all_ne (vector bool short, vector bool short);
12829 int vec_all_ne (vector bool short, vector unsigned short);
12830 int vec_all_ne (vector bool short, vector signed short);
12831 int vec_all_ne (vector pixel, vector pixel);
12832 int vec_all_ne (vector signed int, vector bool int);
12833 int vec_all_ne (vector signed int, vector signed int);
12834 int vec_all_ne (vector unsigned int, vector bool int);
12835 int vec_all_ne (vector unsigned int, vector unsigned int);
12836 int vec_all_ne (vector bool int, vector bool int);
12837 int vec_all_ne (vector bool int, vector unsigned int);
12838 int vec_all_ne (vector bool int, vector signed int);
12839 int vec_all_ne (vector float, vector float);
12841 int vec_all_nge (vector float, vector float);
12843 int vec_all_ngt (vector float, vector float);
12845 int vec_all_nle (vector float, vector float);
12847 int vec_all_nlt (vector float, vector float);
12849 int vec_all_numeric (vector float);
12851 int vec_any_eq (vector signed char, vector bool char);
12852 int vec_any_eq (vector signed char, vector signed char);
12853 int vec_any_eq (vector unsigned char, vector bool char);
12854 int vec_any_eq (vector unsigned char, vector unsigned char);
12855 int vec_any_eq (vector bool char, vector bool char);
12856 int vec_any_eq (vector bool char, vector unsigned char);
12857 int vec_any_eq (vector bool char, vector signed char);
12858 int vec_any_eq (vector signed short, vector bool short);
12859 int vec_any_eq (vector signed short, vector signed short);
12860 int vec_any_eq (vector unsigned short, vector bool short);
12861 int vec_any_eq (vector unsigned short, vector unsigned short);
12862 int vec_any_eq (vector bool short, vector bool short);
12863 int vec_any_eq (vector bool short, vector unsigned short);
12864 int vec_any_eq (vector bool short, vector signed short);
12865 int vec_any_eq (vector pixel, vector pixel);
12866 int vec_any_eq (vector signed int, vector bool int);
12867 int vec_any_eq (vector signed int, vector signed int);
12868 int vec_any_eq (vector unsigned int, vector bool int);
12869 int vec_any_eq (vector unsigned int, vector unsigned int);
12870 int vec_any_eq (vector bool int, vector bool int);
12871 int vec_any_eq (vector bool int, vector unsigned int);
12872 int vec_any_eq (vector bool int, vector signed int);
12873 int vec_any_eq (vector float, vector float);
12875 int vec_any_ge (vector signed char, vector bool char);
12876 int vec_any_ge (vector unsigned char, vector bool char);
12877 int vec_any_ge (vector unsigned char, vector unsigned char);
12878 int vec_any_ge (vector signed char, vector signed char);
12879 int vec_any_ge (vector bool char, vector unsigned char);
12880 int vec_any_ge (vector bool char, vector signed char);
12881 int vec_any_ge (vector unsigned short, vector bool short);
12882 int vec_any_ge (vector unsigned short, vector unsigned short);
12883 int vec_any_ge (vector signed short, vector signed short);
12884 int vec_any_ge (vector signed short, vector bool short);
12885 int vec_any_ge (vector bool short, vector unsigned short);
12886 int vec_any_ge (vector bool short, vector signed short);
12887 int vec_any_ge (vector signed int, vector bool int);
12888 int vec_any_ge (vector unsigned int, vector bool int);
12889 int vec_any_ge (vector unsigned int, vector unsigned int);
12890 int vec_any_ge (vector signed int, vector signed int);
12891 int vec_any_ge (vector bool int, vector unsigned int);
12892 int vec_any_ge (vector bool int, vector signed int);
12893 int vec_any_ge (vector float, vector float);
12895 int vec_any_gt (vector bool char, vector unsigned char);
12896 int vec_any_gt (vector unsigned char, vector bool char);
12897 int vec_any_gt (vector unsigned char, vector unsigned char);
12898 int vec_any_gt (vector bool char, vector signed char);
12899 int vec_any_gt (vector signed char, vector bool char);
12900 int vec_any_gt (vector signed char, vector signed char);
12901 int vec_any_gt (vector bool short, vector unsigned short);
12902 int vec_any_gt (vector unsigned short, vector bool short);
12903 int vec_any_gt (vector unsigned short, vector unsigned short);
12904 int vec_any_gt (vector bool short, vector signed short);
12905 int vec_any_gt (vector signed short, vector bool short);
12906 int vec_any_gt (vector signed short, vector signed short);
12907 int vec_any_gt (vector bool int, vector unsigned int);
12908 int vec_any_gt (vector unsigned int, vector bool int);
12909 int vec_any_gt (vector unsigned int, vector unsigned int);
12910 int vec_any_gt (vector bool int, vector signed int);
12911 int vec_any_gt (vector signed int, vector bool int);
12912 int vec_any_gt (vector signed int, vector signed int);
12913 int vec_any_gt (vector float, vector float);
12915 int vec_any_le (vector bool char, vector unsigned char);
12916 int vec_any_le (vector unsigned char, vector bool char);
12917 int vec_any_le (vector unsigned char, vector unsigned char);
12918 int vec_any_le (vector bool char, vector signed char);
12919 int vec_any_le (vector signed char, vector bool char);
12920 int vec_any_le (vector signed char, vector signed char);
12921 int vec_any_le (vector bool short, vector unsigned short);
12922 int vec_any_le (vector unsigned short, vector bool short);
12923 int vec_any_le (vector unsigned short, vector unsigned short);
12924 int vec_any_le (vector bool short, vector signed short);
12925 int vec_any_le (vector signed short, vector bool short);
12926 int vec_any_le (vector signed short, vector signed short);
12927 int vec_any_le (vector bool int, vector unsigned int);
12928 int vec_any_le (vector unsigned int, vector bool int);
12929 int vec_any_le (vector unsigned int, vector unsigned int);
12930 int vec_any_le (vector bool int, vector signed int);
12931 int vec_any_le (vector signed int, vector bool int);
12932 int vec_any_le (vector signed int, vector signed int);
12933 int vec_any_le (vector float, vector float);
12935 int vec_any_lt (vector bool char, vector unsigned char);
12936 int vec_any_lt (vector unsigned char, vector bool char);
12937 int vec_any_lt (vector unsigned char, vector unsigned char);
12938 int vec_any_lt (vector bool char, vector signed char);
12939 int vec_any_lt (vector signed char, vector bool char);
12940 int vec_any_lt (vector signed char, vector signed char);
12941 int vec_any_lt (vector bool short, vector unsigned short);
12942 int vec_any_lt (vector unsigned short, vector bool short);
12943 int vec_any_lt (vector unsigned short, vector unsigned short);
12944 int vec_any_lt (vector bool short, vector signed short);
12945 int vec_any_lt (vector signed short, vector bool short);
12946 int vec_any_lt (vector signed short, vector signed short);
12947 int vec_any_lt (vector bool int, vector unsigned int);
12948 int vec_any_lt (vector unsigned int, vector bool int);
12949 int vec_any_lt (vector unsigned int, vector unsigned int);
12950 int vec_any_lt (vector bool int, vector signed int);
12951 int vec_any_lt (vector signed int, vector bool int);
12952 int vec_any_lt (vector signed int, vector signed int);
12953 int vec_any_lt (vector float, vector float);
12955 int vec_any_nan (vector float);
12957 int vec_any_ne (vector signed char, vector bool char);
12958 int vec_any_ne (vector signed char, vector signed char);
12959 int vec_any_ne (vector unsigned char, vector bool char);
12960 int vec_any_ne (vector unsigned char, vector unsigned char);
12961 int vec_any_ne (vector bool char, vector bool char);
12962 int vec_any_ne (vector bool char, vector unsigned char);
12963 int vec_any_ne (vector bool char, vector signed char);
12964 int vec_any_ne (vector signed short, vector bool short);
12965 int vec_any_ne (vector signed short, vector signed short);
12966 int vec_any_ne (vector unsigned short, vector bool short);
12967 int vec_any_ne (vector unsigned short, vector unsigned short);
12968 int vec_any_ne (vector bool short, vector bool short);
12969 int vec_any_ne (vector bool short, vector unsigned short);
12970 int vec_any_ne (vector bool short, vector signed short);
12971 int vec_any_ne (vector pixel, vector pixel);
12972 int vec_any_ne (vector signed int, vector bool int);
12973 int vec_any_ne (vector signed int, vector signed int);
12974 int vec_any_ne (vector unsigned int, vector bool int);
12975 int vec_any_ne (vector unsigned int, vector unsigned int);
12976 int vec_any_ne (vector bool int, vector bool int);
12977 int vec_any_ne (vector bool int, vector unsigned int);
12978 int vec_any_ne (vector bool int, vector signed int);
12979 int vec_any_ne (vector float, vector float);
12981 int vec_any_nge (vector float, vector float);
12983 int vec_any_ngt (vector float, vector float);
12985 int vec_any_nle (vector float, vector float);
12987 int vec_any_nlt (vector float, vector float);
12989 int vec_any_numeric (vector float);
12991 int vec_any_out (vector float, vector float);
12994 If the vector/scalar (VSX) instruction set is available, the following
12995 additional functions are available:
12998 vector double vec_abs (vector double);
12999 vector double vec_add (vector double, vector double);
13000 vector double vec_and (vector double, vector double);
13001 vector double vec_and (vector double, vector bool long);
13002 vector double vec_and (vector bool long, vector double);
13003 vector double vec_andc (vector double, vector double);
13004 vector double vec_andc (vector double, vector bool long);
13005 vector double vec_andc (vector bool long, vector double);
13006 vector double vec_ceil (vector double);
13007 vector bool long vec_cmpeq (vector double, vector double);
13008 vector bool long vec_cmpge (vector double, vector double);
13009 vector bool long vec_cmpgt (vector double, vector double);
13010 vector bool long vec_cmple (vector double, vector double);
13011 vector bool long vec_cmplt (vector double, vector double);
13012 vector float vec_div (vector float, vector float);
13013 vector double vec_div (vector double, vector double);
13014 vector double vec_floor (vector double);
13015 vector double vec_ld (int, const vector double *);
13016 vector double vec_ld (int, const double *);
13017 vector double vec_ldl (int, const vector double *);
13018 vector double vec_ldl (int, const double *);
13019 vector unsigned char vec_lvsl (int, const volatile double *);
13020 vector unsigned char vec_lvsr (int, const volatile double *);
13021 vector double vec_madd (vector double, vector double, vector double);
13022 vector double vec_max (vector double, vector double);
13023 vector double vec_min (vector double, vector double);
13024 vector float vec_msub (vector float, vector float, vector float);
13025 vector double vec_msub (vector double, vector double, vector double);
13026 vector float vec_mul (vector float, vector float);
13027 vector double vec_mul (vector double, vector double);
13028 vector float vec_nearbyint (vector float);
13029 vector double vec_nearbyint (vector double);
13030 vector float vec_nmadd (vector float, vector float, vector float);
13031 vector double vec_nmadd (vector double, vector double, vector double);
13032 vector double vec_nmsub (vector double, vector double, vector double);
13033 vector double vec_nor (vector double, vector double);
13034 vector double vec_or (vector double, vector double);
13035 vector double vec_or (vector double, vector bool long);
13036 vector double vec_or (vector bool long, vector double);
13037 vector double vec_perm (vector double,
13039 vector unsigned char);
13040 vector double vec_rint (vector double);
13041 vector double vec_recip (vector double, vector double);
13042 vector double vec_rsqrt (vector double);
13043 vector double vec_rsqrte (vector double);
13044 vector double vec_sel (vector double, vector double, vector bool long);
13045 vector double vec_sel (vector double, vector double, vector unsigned long);
13046 vector double vec_sub (vector double, vector double);
13047 vector float vec_sqrt (vector float);
13048 vector double vec_sqrt (vector double);
13049 void vec_st (vector double, int, vector double *);
13050 void vec_st (vector double, int, double *);
13051 vector double vec_trunc (vector double);
13052 vector double vec_xor (vector double, vector double);
13053 vector double vec_xor (vector double, vector bool long);
13054 vector double vec_xor (vector bool long, vector double);
13055 int vec_all_eq (vector double, vector double);
13056 int vec_all_ge (vector double, vector double);
13057 int vec_all_gt (vector double, vector double);
13058 int vec_all_le (vector double, vector double);
13059 int vec_all_lt (vector double, vector double);
13060 int vec_all_nan (vector double);
13061 int vec_all_ne (vector double, vector double);
13062 int vec_all_nge (vector double, vector double);
13063 int vec_all_ngt (vector double, vector double);
13064 int vec_all_nle (vector double, vector double);
13065 int vec_all_nlt (vector double, vector double);
13066 int vec_all_numeric (vector double);
13067 int vec_any_eq (vector double, vector double);
13068 int vec_any_ge (vector double, vector double);
13069 int vec_any_gt (vector double, vector double);
13070 int vec_any_le (vector double, vector double);
13071 int vec_any_lt (vector double, vector double);
13072 int vec_any_nan (vector double);
13073 int vec_any_ne (vector double, vector double);
13074 int vec_any_nge (vector double, vector double);
13075 int vec_any_ngt (vector double, vector double);
13076 int vec_any_nle (vector double, vector double);
13077 int vec_any_nlt (vector double, vector double);
13078 int vec_any_numeric (vector double);
13080 vector double vec_vsx_ld (int, const vector double *);
13081 vector double vec_vsx_ld (int, const double *);
13082 vector float vec_vsx_ld (int, const vector float *);
13083 vector float vec_vsx_ld (int, const float *);
13084 vector bool int vec_vsx_ld (int, const vector bool int *);
13085 vector signed int vec_vsx_ld (int, const vector signed int *);
13086 vector signed int vec_vsx_ld (int, const int *);
13087 vector signed int vec_vsx_ld (int, const long *);
13088 vector unsigned int vec_vsx_ld (int, const vector unsigned int *);
13089 vector unsigned int vec_vsx_ld (int, const unsigned int *);
13090 vector unsigned int vec_vsx_ld (int, const unsigned long *);
13091 vector bool short vec_vsx_ld (int, const vector bool short *);
13092 vector pixel vec_vsx_ld (int, const vector pixel *);
13093 vector signed short vec_vsx_ld (int, const vector signed short *);
13094 vector signed short vec_vsx_ld (int, const short *);
13095 vector unsigned short vec_vsx_ld (int, const vector unsigned short *);
13096 vector unsigned short vec_vsx_ld (int, const unsigned short *);
13097 vector bool char vec_vsx_ld (int, const vector bool char *);
13098 vector signed char vec_vsx_ld (int, const vector signed char *);
13099 vector signed char vec_vsx_ld (int, const signed char *);
13100 vector unsigned char vec_vsx_ld (int, const vector unsigned char *);
13101 vector unsigned char vec_vsx_ld (int, const unsigned char *);
13103 void vec_vsx_st (vector double, int, vector double *);
13104 void vec_vsx_st (vector double, int, double *);
13105 void vec_vsx_st (vector float, int, vector float *);
13106 void vec_vsx_st (vector float, int, float *);
13107 void vec_vsx_st (vector signed int, int, vector signed int *);
13108 void vec_vsx_st (vector signed int, int, int *);
13109 void vec_vsx_st (vector unsigned int, int, vector unsigned int *);
13110 void vec_vsx_st (vector unsigned int, int, unsigned int *);
13111 void vec_vsx_st (vector bool int, int, vector bool int *);
13112 void vec_vsx_st (vector bool int, int, unsigned int *);
13113 void vec_vsx_st (vector bool int, int, int *);
13114 void vec_vsx_st (vector signed short, int, vector signed short *);
13115 void vec_vsx_st (vector signed short, int, short *);
13116 void vec_vsx_st (vector unsigned short, int, vector unsigned short *);
13117 void vec_vsx_st (vector unsigned short, int, unsigned short *);
13118 void vec_vsx_st (vector bool short, int, vector bool short *);
13119 void vec_vsx_st (vector bool short, int, unsigned short *);
13120 void vec_vsx_st (vector pixel, int, vector pixel *);
13121 void vec_vsx_st (vector pixel, int, unsigned short *);
13122 void vec_vsx_st (vector pixel, int, short *);
13123 void vec_vsx_st (vector bool short, int, short *);
13124 void vec_vsx_st (vector signed char, int, vector signed char *);
13125 void vec_vsx_st (vector signed char, int, signed char *);
13126 void vec_vsx_st (vector unsigned char, int, vector unsigned char *);
13127 void vec_vsx_st (vector unsigned char, int, unsigned char *);
13128 void vec_vsx_st (vector bool char, int, vector bool char *);
13129 void vec_vsx_st (vector bool char, int, unsigned char *);
13130 void vec_vsx_st (vector bool char, int, signed char *);
13133 Note that the @samp{vec_ld} and @samp{vec_st} builtins will always
13134 generate the Altivec @samp{LVX} and @samp{STVX} instructions even
13135 if the VSX instruction set is available. The @samp{vec_vsx_ld} and
13136 @samp{vec_vsx_st} builtins will always generate the VSX @samp{LXVD2X},
13137 @samp{LXVW4X}, @samp{STXVD2X}, and @samp{STXVW4X} instructions.
13139 GCC provides a few other builtins on Powerpc to access certain instructions:
13141 float __builtin_recipdivf (float, float);
13142 float __builtin_rsqrtf (float);
13143 double __builtin_recipdiv (double, double);
13144 double __builtin_rsqrt (double);
13145 long __builtin_bpermd (long, long);
13146 int __builtin_bswap16 (int);
13149 The @code{vec_rsqrt}, @code{__builtin_rsqrt}, and
13150 @code{__builtin_rsqrtf} functions generate multiple instructions to
13151 implement the reciprocal sqrt functionality using reciprocal sqrt
13152 estimate instructions.
13154 The @code{__builtin_recipdiv}, and @code{__builtin_recipdivf}
13155 functions generate multiple instructions to implement division using
13156 the reciprocal estimate instructions.
13158 @node RX Built-in Functions
13159 @subsection RX Built-in Functions
13160 GCC supports some of the RX instructions which cannot be expressed in
13161 the C programming language via the use of built-in functions. The
13162 following functions are supported:
13164 @deftypefn {Built-in Function} void __builtin_rx_brk (void)
13165 Generates the @code{brk} machine instruction.
13168 @deftypefn {Built-in Function} void __builtin_rx_clrpsw (int)
13169 Generates the @code{clrpsw} machine instruction to clear the specified
13170 bit in the processor status word.
13173 @deftypefn {Built-in Function} void __builtin_rx_int (int)
13174 Generates the @code{int} machine instruction to generate an interrupt
13175 with the specified value.
13178 @deftypefn {Built-in Function} void __builtin_rx_machi (int, int)
13179 Generates the @code{machi} machine instruction to add the result of
13180 multiplying the top 16-bits of the two arguments into the
13184 @deftypefn {Built-in Function} void __builtin_rx_maclo (int, int)
13185 Generates the @code{maclo} machine instruction to add the result of
13186 multiplying the bottom 16-bits of the two arguments into the
13190 @deftypefn {Built-in Function} void __builtin_rx_mulhi (int, int)
13191 Generates the @code{mulhi} machine instruction to place the result of
13192 multiplying the top 16-bits of the two arguments into the
13196 @deftypefn {Built-in Function} void __builtin_rx_mullo (int, int)
13197 Generates the @code{mullo} machine instruction to place the result of
13198 multiplying the bottom 16-bits of the two arguments into the
13202 @deftypefn {Built-in Function} int __builtin_rx_mvfachi (void)
13203 Generates the @code{mvfachi} machine instruction to read the top
13204 32-bits of the accumulator.
13207 @deftypefn {Built-in Function} int __builtin_rx_mvfacmi (void)
13208 Generates the @code{mvfacmi} machine instruction to read the middle
13209 32-bits of the accumulator.
13212 @deftypefn {Built-in Function} int __builtin_rx_mvfc (int)
13213 Generates the @code{mvfc} machine instruction which reads the control
13214 register specified in its argument and returns its value.
13217 @deftypefn {Built-in Function} void __builtin_rx_mvtachi (int)
13218 Generates the @code{mvtachi} machine instruction to set the top
13219 32-bits of the accumulator.
13222 @deftypefn {Built-in Function} void __builtin_rx_mvtaclo (int)
13223 Generates the @code{mvtaclo} machine instruction to set the bottom
13224 32-bits of the accumulator.
13227 @deftypefn {Built-in Function} void __builtin_rx_mvtc (int reg, int val)
13228 Generates the @code{mvtc} machine instruction which sets control
13229 register number @code{reg} to @code{val}.
13232 @deftypefn {Built-in Function} void __builtin_rx_mvtipl (int)
13233 Generates the @code{mvtipl} machine instruction set the interrupt
13237 @deftypefn {Built-in Function} void __builtin_rx_racw (int)
13238 Generates the @code{racw} machine instruction to round the accumulator
13239 according to the specified mode.
13242 @deftypefn {Built-in Function} int __builtin_rx_revw (int)
13243 Generates the @code{revw} machine instruction which swaps the bytes in
13244 the argument so that bits 0--7 now occupy bits 8--15 and vice versa,
13245 and also bits 16--23 occupy bits 24--31 and vice versa.
13248 @deftypefn {Built-in Function} void __builtin_rx_rmpa (void)
13249 Generates the @code{rmpa} machine instruction which initiates a
13250 repeated multiply and accumulate sequence.
13253 @deftypefn {Built-in Function} void __builtin_rx_round (float)
13254 Generates the @code{round} machine instruction which returns the
13255 floating point argument rounded according to the current rounding mode
13256 set in the floating point status word register.
13259 @deftypefn {Built-in Function} int __builtin_rx_sat (int)
13260 Generates the @code{sat} machine instruction which returns the
13261 saturated value of the argument.
13264 @deftypefn {Built-in Function} void __builtin_rx_setpsw (int)
13265 Generates the @code{setpsw} machine instruction to set the specified
13266 bit in the processor status word.
13269 @deftypefn {Built-in Function} void __builtin_rx_wait (void)
13270 Generates the @code{wait} machine instruction.
13273 @node SPARC VIS Built-in Functions
13274 @subsection SPARC VIS Built-in Functions
13276 GCC supports SIMD operations on the SPARC using both the generic vector
13277 extensions (@pxref{Vector Extensions}) as well as built-in functions for
13278 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
13279 switch, the VIS extension is exposed as the following built-in functions:
13282 typedef int v1si __attribute__ ((vector_size (4)));
13283 typedef int v2si __attribute__ ((vector_size (8)));
13284 typedef short v4hi __attribute__ ((vector_size (8)));
13285 typedef short v2hi __attribute__ ((vector_size (4)));
13286 typedef unsigned char v8qi __attribute__ ((vector_size (8)));
13287 typedef unsigned char v4qi __attribute__ ((vector_size (4)));
13289 void __builtin_vis_write_gsr (int64_t);
13290 int64_t __builtin_vis_read_gsr (void);
13292 void * __builtin_vis_alignaddr (void *, long);
13293 void * __builtin_vis_alignaddrl (void *, long);
13294 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
13295 v2si __builtin_vis_faligndatav2si (v2si, v2si);
13296 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
13297 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
13299 v4hi __builtin_vis_fexpand (v4qi);
13301 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
13302 v4hi __builtin_vis_fmul8x16au (v4qi, v2hi);
13303 v4hi __builtin_vis_fmul8x16al (v4qi, v2hi);
13304 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
13305 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
13306 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
13307 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
13309 v4qi __builtin_vis_fpack16 (v4hi);
13310 v8qi __builtin_vis_fpack32 (v2si, v8qi);
13311 v2hi __builtin_vis_fpackfix (v2si);
13312 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
13314 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
13316 long __builtin_vis_edge8 (void *, void *);
13317 long __builtin_vis_edge8l (void *, void *);
13318 long __builtin_vis_edge16 (void *, void *);
13319 long __builtin_vis_edge16l (void *, void *);
13320 long __builtin_vis_edge32 (void *, void *);
13321 long __builtin_vis_edge32l (void *, void *);
13323 long __builtin_vis_fcmple16 (v4hi, v4hi);
13324 long __builtin_vis_fcmple32 (v2si, v2si);
13325 long __builtin_vis_fcmpne16 (v4hi, v4hi);
13326 long __builtin_vis_fcmpne32 (v2si, v2si);
13327 long __builtin_vis_fcmpgt16 (v4hi, v4hi);
13328 long __builtin_vis_fcmpgt32 (v2si, v2si);
13329 long __builtin_vis_fcmpeq16 (v4hi, v4hi);
13330 long __builtin_vis_fcmpeq32 (v2si, v2si);
13332 v4hi __builtin_vis_fpadd16 (v4hi, v4hi);
13333 v2hi __builtin_vis_fpadd16s (v2hi, v2hi);
13334 v2si __builtin_vis_fpadd32 (v2si, v2si);
13335 v1si __builtin_vis_fpadd32s (v1si, v1si);
13336 v4hi __builtin_vis_fpsub16 (v4hi, v4hi);
13337 v2hi __builtin_vis_fpsub16s (v2hi, v2hi);
13338 v2si __builtin_vis_fpsub32 (v2si, v2si);
13339 v1si __builtin_vis_fpsub32s (v1si, v1si);
13341 long __builtin_vis_array8 (long, long);
13342 long __builtin_vis_array16 (long, long);
13343 long __builtin_vis_array32 (long, long);
13346 When you use the @option{-mvis2} switch, the VIS version 2.0 built-in
13347 functions also become available:
13350 long __builtin_vis_bmask (long, long);
13351 int64_t __builtin_vis_bshuffledi (int64_t, int64_t);
13352 v2si __builtin_vis_bshufflev2si (v2si, v2si);
13353 v4hi __builtin_vis_bshufflev2si (v4hi, v4hi);
13354 v8qi __builtin_vis_bshufflev2si (v8qi, v8qi);
13356 long __builtin_vis_edge8n (void *, void *);
13357 long __builtin_vis_edge8ln (void *, void *);
13358 long __builtin_vis_edge16n (void *, void *);
13359 long __builtin_vis_edge16ln (void *, void *);
13360 long __builtin_vis_edge32n (void *, void *);
13361 long __builtin_vis_edge32ln (void *, void *);
13364 When you use the @option{-mvis3} switch, the VIS version 3.0 built-in
13365 functions also become available:
13368 void __builtin_vis_cmask8 (long);
13369 void __builtin_vis_cmask16 (long);
13370 void __builtin_vis_cmask32 (long);
13372 v4hi __builtin_vis_fchksm16 (v4hi, v4hi);
13374 v4hi __builtin_vis_fsll16 (v4hi, v4hi);
13375 v4hi __builtin_vis_fslas16 (v4hi, v4hi);
13376 v4hi __builtin_vis_fsrl16 (v4hi, v4hi);
13377 v4hi __builtin_vis_fsra16 (v4hi, v4hi);
13378 v2si __builtin_vis_fsll16 (v2si, v2si);
13379 v2si __builtin_vis_fslas16 (v2si, v2si);
13380 v2si __builtin_vis_fsrl16 (v2si, v2si);
13381 v2si __builtin_vis_fsra16 (v2si, v2si);
13383 long __builtin_vis_pdistn (v8qi, v8qi);
13385 v4hi __builtin_vis_fmean16 (v4hi, v4hi);
13387 int64_t __builtin_vis_fpadd64 (int64_t, int64_t);
13388 int64_t __builtin_vis_fpsub64 (int64_t, int64_t);
13390 v4hi __builtin_vis_fpadds16 (v4hi, v4hi);
13391 v2hi __builtin_vis_fpadds16s (v2hi, v2hi);
13392 v4hi __builtin_vis_fpsubs16 (v4hi, v4hi);
13393 v2hi __builtin_vis_fpsubs16s (v2hi, v2hi);
13394 v2si __builtin_vis_fpadds32 (v2si, v2si);
13395 v1si __builtin_vis_fpadds32s (v1si, v1si);
13396 v2si __builtin_vis_fpsubs32 (v2si, v2si);
13397 v1si __builtin_vis_fpsubs32s (v1si, v1si);
13399 long __builtin_vis_fucmple8 (v8qi, v8qi);
13400 long __builtin_vis_fucmpne8 (v8qi, v8qi);
13401 long __builtin_vis_fucmpgt8 (v8qi, v8qi);
13402 long __builtin_vis_fucmpeq8 (v8qi, v8qi);
13404 float __builtin_vis_fhadds (float, float);
13405 double __builtin_vis_fhaddd (double, double);
13406 float __builtin_vis_fhsubs (float, float);
13407 double __builtin_vis_fhsubd (double, double);
13408 float __builtin_vis_fnhadds (float, float);
13409 double __builtin_vis_fnhaddd (double, double);
13411 int64_t __builtin_vis_umulxhi (int64_t, int64_t);
13412 int64_t __builtin_vis_xmulx (int64_t, int64_t);
13413 int64_t __builtin_vis_xmulxhi (int64_t, int64_t);
13416 @node SPU Built-in Functions
13417 @subsection SPU Built-in Functions
13419 GCC provides extensions for the SPU processor as described in the
13420 Sony/Toshiba/IBM SPU Language Extensions Specification, which can be
13421 found at @uref{http://cell.scei.co.jp/} or
13422 @uref{http://www.ibm.com/developerworks/power/cell/}. GCC's
13423 implementation differs in several ways.
13428 The optional extension of specifying vector constants in parentheses is
13432 A vector initializer requires no cast if the vector constant is of the
13433 same type as the variable it is initializing.
13436 If @code{signed} or @code{unsigned} is omitted, the signedness of the
13437 vector type is the default signedness of the base type. The default
13438 varies depending on the operating system, so a portable program should
13439 always specify the signedness.
13442 By default, the keyword @code{__vector} is added. The macro
13443 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
13447 GCC allows using a @code{typedef} name as the type specifier for a
13451 For C, overloaded functions are implemented with macros so the following
13455 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
13458 Since @code{spu_add} is a macro, the vector constant in the example
13459 is treated as four separate arguments. Wrap the entire argument in
13460 parentheses for this to work.
13463 The extended version of @code{__builtin_expect} is not supported.
13467 @emph{Note:} Only the interface described in the aforementioned
13468 specification is supported. Internally, GCC uses built-in functions to
13469 implement the required functionality, but these are not supported and
13470 are subject to change without notice.
13472 @node TI C6X Built-in Functions
13473 @subsection TI C6X Built-in Functions
13475 GCC provides intrinsics to access certain instructions of the TI C6X
13476 processors. These intrinsics, listed below, are available after
13477 inclusion of the @code{c6x_intrinsics.h} header file. They map directly
13478 to C6X instructions.
13482 int _sadd (int, int)
13483 int _ssub (int, int)
13484 int _sadd2 (int, int)
13485 int _ssub2 (int, int)
13486 long long _mpy2 (int, int)
13487 long long _smpy2 (int, int)
13488 int _add4 (int, int)
13489 int _sub4 (int, int)
13490 int _saddu4 (int, int)
13492 int _smpy (int, int)
13493 int _smpyh (int, int)
13494 int _smpyhl (int, int)
13495 int _smpylh (int, int)
13497 int _sshl (int, int)
13498 int _subc (int, int)
13500 int _avg2 (int, int)
13501 int _avgu4 (int, int)
13503 int _clrr (int, int)
13504 int _extr (int, int)
13505 int _extru (int, int)
13511 @node Target Format Checks
13512 @section Format Checks Specific to Particular Target Machines
13514 For some target machines, GCC supports additional options to the
13516 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
13519 * Solaris Format Checks::
13520 * Darwin Format Checks::
13523 @node Solaris Format Checks
13524 @subsection Solaris Format Checks
13526 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
13527 check. @code{cmn_err} accepts a subset of the standard @code{printf}
13528 conversions, and the two-argument @code{%b} conversion for displaying
13529 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
13531 @node Darwin Format Checks
13532 @subsection Darwin Format Checks
13534 Darwin targets support the @code{CFString} (or @code{__CFString__}) in the format
13535 attribute context. Declarations made with such attribution will be parsed for correct syntax
13536 and format argument types. However, parsing of the format string itself is currently undefined
13537 and will not be carried out by this version of the compiler.
13539 Additionally, @code{CFStringRefs} (defined by the @code{CoreFoundation} headers) may
13540 also be used as format arguments. Note that the relevant headers are only likely to be
13541 available on Darwin (OSX) installations. On such installations, the XCode and system
13542 documentation provide descriptions of @code{CFString}, @code{CFStringRefs} and
13543 associated functions.
13546 @section Pragmas Accepted by GCC
13548 @cindex @code{#pragma}
13550 GCC supports several types of pragmas, primarily in order to compile
13551 code originally written for other compilers. Note that in general
13552 we do not recommend the use of pragmas; @xref{Function Attributes},
13553 for further explanation.
13559 * RS/6000 and PowerPC Pragmas::
13561 * Solaris Pragmas::
13562 * Symbol-Renaming Pragmas::
13563 * Structure-Packing Pragmas::
13565 * Diagnostic Pragmas::
13566 * Visibility Pragmas::
13567 * Push/Pop Macro Pragmas::
13568 * Function Specific Option Pragmas::
13572 @subsection ARM Pragmas
13574 The ARM target defines pragmas for controlling the default addition of
13575 @code{long_call} and @code{short_call} attributes to functions.
13576 @xref{Function Attributes}, for information about the effects of these
13581 @cindex pragma, long_calls
13582 Set all subsequent functions to have the @code{long_call} attribute.
13584 @item no_long_calls
13585 @cindex pragma, no_long_calls
13586 Set all subsequent functions to have the @code{short_call} attribute.
13588 @item long_calls_off
13589 @cindex pragma, long_calls_off
13590 Do not affect the @code{long_call} or @code{short_call} attributes of
13591 subsequent functions.
13595 @subsection M32C Pragmas
13598 @item GCC memregs @var{number}
13599 @cindex pragma, memregs
13600 Overrides the command-line option @code{-memregs=} for the current
13601 file. Use with care! This pragma must be before any function in the
13602 file, and mixing different memregs values in different objects may
13603 make them incompatible. This pragma is useful when a
13604 performance-critical function uses a memreg for temporary values,
13605 as it may allow you to reduce the number of memregs used.
13607 @item ADDRESS @var{name} @var{address}
13608 @cindex pragma, address
13609 For any declared symbols matching @var{name}, this does three things
13610 to that symbol: it forces the symbol to be located at the given
13611 address (a number), it forces the symbol to be volatile, and it
13612 changes the symbol's scope to be static. This pragma exists for
13613 compatibility with other compilers, but note that the common
13614 @code{1234H} numeric syntax is not supported (use @code{0x1234}
13618 #pragma ADDRESS port3 0x103
13625 @subsection MeP Pragmas
13629 @item custom io_volatile (on|off)
13630 @cindex pragma, custom io_volatile
13631 Overrides the command line option @code{-mio-volatile} for the current
13632 file. Note that for compatibility with future GCC releases, this
13633 option should only be used once before any @code{io} variables in each
13636 @item GCC coprocessor available @var{registers}
13637 @cindex pragma, coprocessor available
13638 Specifies which coprocessor registers are available to the register
13639 allocator. @var{registers} may be a single register, register range
13640 separated by ellipses, or comma-separated list of those. Example:
13643 #pragma GCC coprocessor available $c0...$c10, $c28
13646 @item GCC coprocessor call_saved @var{registers}
13647 @cindex pragma, coprocessor call_saved
13648 Specifies which coprocessor registers are to be saved and restored by
13649 any function using them. @var{registers} may be a single register,
13650 register range separated by ellipses, or comma-separated list of
13654 #pragma GCC coprocessor call_saved $c4...$c6, $c31
13657 @item GCC coprocessor subclass '(A|B|C|D)' = @var{registers}
13658 @cindex pragma, coprocessor subclass
13659 Creates and defines a register class. These register classes can be
13660 used by inline @code{asm} constructs. @var{registers} may be a single
13661 register, register range separated by ellipses, or comma-separated
13662 list of those. Example:
13665 #pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6
13667 asm ("cpfoo %0" : "=B" (x));
13670 @item GCC disinterrupt @var{name} , @var{name} @dots{}
13671 @cindex pragma, disinterrupt
13672 For the named functions, the compiler adds code to disable interrupts
13673 for the duration of those functions. Any functions so named, which
13674 are not encountered in the source, cause a warning that the pragma was
13675 not used. Examples:
13678 #pragma disinterrupt foo
13679 #pragma disinterrupt bar, grill
13680 int foo () @{ @dots{} @}
13683 @item GCC call @var{name} , @var{name} @dots{}
13684 @cindex pragma, call
13685 For the named functions, the compiler always uses a register-indirect
13686 call model when calling the named functions. Examples:
13695 @node RS/6000 and PowerPC Pragmas
13696 @subsection RS/6000 and PowerPC Pragmas
13698 The RS/6000 and PowerPC targets define one pragma for controlling
13699 whether or not the @code{longcall} attribute is added to function
13700 declarations by default. This pragma overrides the @option{-mlongcall}
13701 option, but not the @code{longcall} and @code{shortcall} attributes.
13702 @xref{RS/6000 and PowerPC Options}, for more information about when long
13703 calls are and are not necessary.
13707 @cindex pragma, longcall
13708 Apply the @code{longcall} attribute to all subsequent function
13712 Do not apply the @code{longcall} attribute to subsequent function
13716 @c Describe h8300 pragmas here.
13717 @c Describe sh pragmas here.
13718 @c Describe v850 pragmas here.
13720 @node Darwin Pragmas
13721 @subsection Darwin Pragmas
13723 The following pragmas are available for all architectures running the
13724 Darwin operating system. These are useful for compatibility with other
13728 @item mark @var{tokens}@dots{}
13729 @cindex pragma, mark
13730 This pragma is accepted, but has no effect.
13732 @item options align=@var{alignment}
13733 @cindex pragma, options align
13734 This pragma sets the alignment of fields in structures. The values of
13735 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
13736 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
13737 properly; to restore the previous setting, use @code{reset} for the
13740 @item segment @var{tokens}@dots{}
13741 @cindex pragma, segment
13742 This pragma is accepted, but has no effect.
13744 @item unused (@var{var} [, @var{var}]@dots{})
13745 @cindex pragma, unused
13746 This pragma declares variables to be possibly unused. GCC will not
13747 produce warnings for the listed variables. The effect is similar to
13748 that of the @code{unused} attribute, except that this pragma may appear
13749 anywhere within the variables' scopes.
13752 @node Solaris Pragmas
13753 @subsection Solaris Pragmas
13755 The Solaris target supports @code{#pragma redefine_extname}
13756 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
13757 @code{#pragma} directives for compatibility with the system compiler.
13760 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
13761 @cindex pragma, align
13763 Increase the minimum alignment of each @var{variable} to @var{alignment}.
13764 This is the same as GCC's @code{aligned} attribute @pxref{Variable
13765 Attributes}). Macro expansion occurs on the arguments to this pragma
13766 when compiling C and Objective-C@. It does not currently occur when
13767 compiling C++, but this is a bug which may be fixed in a future
13770 @item fini (@var{function} [, @var{function}]...)
13771 @cindex pragma, fini
13773 This pragma causes each listed @var{function} to be called after
13774 main, or during shared module unloading, by adding a call to the
13775 @code{.fini} section.
13777 @item init (@var{function} [, @var{function}]...)
13778 @cindex pragma, init
13780 This pragma causes each listed @var{function} to be called during
13781 initialization (before @code{main}) or during shared module loading, by
13782 adding a call to the @code{.init} section.
13786 @node Symbol-Renaming Pragmas
13787 @subsection Symbol-Renaming Pragmas
13789 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
13790 supports two @code{#pragma} directives which change the name used in
13791 assembly for a given declaration. @code{#pragma extern_prefix} is only
13792 available on platforms whose system headers need it. To get this effect
13793 on all platforms supported by GCC, use the asm labels extension (@pxref{Asm
13797 @item redefine_extname @var{oldname} @var{newname}
13798 @cindex pragma, redefine_extname
13800 This pragma gives the C function @var{oldname} the assembly symbol
13801 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
13802 will be defined if this pragma is available (currently on all platforms).
13804 @item extern_prefix @var{string}
13805 @cindex pragma, extern_prefix
13807 This pragma causes all subsequent external function and variable
13808 declarations to have @var{string} prepended to their assembly symbols.
13809 This effect may be terminated with another @code{extern_prefix} pragma
13810 whose argument is an empty string. The preprocessor macro
13811 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
13812 available (currently only on Tru64 UNIX)@.
13815 These pragmas and the asm labels extension interact in a complicated
13816 manner. Here are some corner cases you may want to be aware of.
13819 @item Both pragmas silently apply only to declarations with external
13820 linkage. Asm labels do not have this restriction.
13822 @item In C++, both pragmas silently apply only to declarations with
13823 ``C'' linkage. Again, asm labels do not have this restriction.
13825 @item If any of the three ways of changing the assembly name of a
13826 declaration is applied to a declaration whose assembly name has
13827 already been determined (either by a previous use of one of these
13828 features, or because the compiler needed the assembly name in order to
13829 generate code), and the new name is different, a warning issues and
13830 the name does not change.
13832 @item The @var{oldname} used by @code{#pragma redefine_extname} is
13833 always the C-language name.
13835 @item If @code{#pragma extern_prefix} is in effect, and a declaration
13836 occurs with an asm label attached, the prefix is silently ignored for
13839 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
13840 apply to the same declaration, whichever triggered first wins, and a
13841 warning issues if they contradict each other. (We would like to have
13842 @code{#pragma redefine_extname} always win, for consistency with asm
13843 labels, but if @code{#pragma extern_prefix} triggers first we have no
13844 way of knowing that that happened.)
13847 @node Structure-Packing Pragmas
13848 @subsection Structure-Packing Pragmas
13850 For compatibility with Microsoft Windows compilers, GCC supports a
13851 set of @code{#pragma} directives which change the maximum alignment of
13852 members of structures (other than zero-width bitfields), unions, and
13853 classes subsequently defined. The @var{n} value below always is required
13854 to be a small power of two and specifies the new alignment in bytes.
13857 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
13858 @item @code{#pragma pack()} sets the alignment to the one that was in
13859 effect when compilation started (see also command-line option
13860 @option{-fpack-struct[=@var{n}]} @pxref{Code Gen Options}).
13861 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
13862 setting on an internal stack and then optionally sets the new alignment.
13863 @item @code{#pragma pack(pop)} restores the alignment setting to the one
13864 saved at the top of the internal stack (and removes that stack entry).
13865 Note that @code{#pragma pack([@var{n}])} does not influence this internal
13866 stack; thus it is possible to have @code{#pragma pack(push)} followed by
13867 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
13868 @code{#pragma pack(pop)}.
13871 Some targets, e.g.@: i386 and powerpc, support the @code{ms_struct}
13872 @code{#pragma} which lays out a structure as the documented
13873 @code{__attribute__ ((ms_struct))}.
13875 @item @code{#pragma ms_struct on} turns on the layout for structures
13877 @item @code{#pragma ms_struct off} turns off the layout for structures
13879 @item @code{#pragma ms_struct reset} goes back to the default layout.
13883 @subsection Weak Pragmas
13885 For compatibility with SVR4, GCC supports a set of @code{#pragma}
13886 directives for declaring symbols to be weak, and defining weak
13890 @item #pragma weak @var{symbol}
13891 @cindex pragma, weak
13892 This pragma declares @var{symbol} to be weak, as if the declaration
13893 had the attribute of the same name. The pragma may appear before
13894 or after the declaration of @var{symbol}. It is not an error for
13895 @var{symbol} to never be defined at all.
13897 @item #pragma weak @var{symbol1} = @var{symbol2}
13898 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
13899 It is an error if @var{symbol2} is not defined in the current
13903 @node Diagnostic Pragmas
13904 @subsection Diagnostic Pragmas
13906 GCC allows the user to selectively enable or disable certain types of
13907 diagnostics, and change the kind of the diagnostic. For example, a
13908 project's policy might require that all sources compile with
13909 @option{-Werror} but certain files might have exceptions allowing
13910 specific types of warnings. Or, a project might selectively enable
13911 diagnostics and treat them as errors depending on which preprocessor
13912 macros are defined.
13915 @item #pragma GCC diagnostic @var{kind} @var{option}
13916 @cindex pragma, diagnostic
13918 Modifies the disposition of a diagnostic. Note that not all
13919 diagnostics are modifiable; at the moment only warnings (normally
13920 controlled by @samp{-W@dots{}}) can be controlled, and not all of them.
13921 Use @option{-fdiagnostics-show-option} to determine which diagnostics
13922 are controllable and which option controls them.
13924 @var{kind} is @samp{error} to treat this diagnostic as an error,
13925 @samp{warning} to treat it like a warning (even if @option{-Werror} is
13926 in effect), or @samp{ignored} if the diagnostic is to be ignored.
13927 @var{option} is a double quoted string which matches the command-line
13931 #pragma GCC diagnostic warning "-Wformat"
13932 #pragma GCC diagnostic error "-Wformat"
13933 #pragma GCC diagnostic ignored "-Wformat"
13936 Note that these pragmas override any command-line options. GCC keeps
13937 track of the location of each pragma, and issues diagnostics according
13938 to the state as of that point in the source file. Thus, pragmas occurring
13939 after a line do not affect diagnostics caused by that line.
13941 @item #pragma GCC diagnostic push
13942 @itemx #pragma GCC diagnostic pop
13944 Causes GCC to remember the state of the diagnostics as of each
13945 @code{push}, and restore to that point at each @code{pop}. If a
13946 @code{pop} has no matching @code{push}, the command line options are
13950 #pragma GCC diagnostic error "-Wuninitialized"
13951 foo(a); /* error is given for this one */
13952 #pragma GCC diagnostic push
13953 #pragma GCC diagnostic ignored "-Wuninitialized"
13954 foo(b); /* no diagnostic for this one */
13955 #pragma GCC diagnostic pop
13956 foo(c); /* error is given for this one */
13957 #pragma GCC diagnostic pop
13958 foo(d); /* depends on command line options */
13963 GCC also offers a simple mechanism for printing messages during
13967 @item #pragma message @var{string}
13968 @cindex pragma, diagnostic
13970 Prints @var{string} as a compiler message on compilation. The message
13971 is informational only, and is neither a compilation warning nor an error.
13974 #pragma message "Compiling " __FILE__ "..."
13977 @var{string} may be parenthesized, and is printed with location
13978 information. For example,
13981 #define DO_PRAGMA(x) _Pragma (#x)
13982 #define TODO(x) DO_PRAGMA(message ("TODO - " #x))
13984 TODO(Remember to fix this)
13987 prints @samp{/tmp/file.c:4: note: #pragma message:
13988 TODO - Remember to fix this}.
13992 @node Visibility Pragmas
13993 @subsection Visibility Pragmas
13996 @item #pragma GCC visibility push(@var{visibility})
13997 @itemx #pragma GCC visibility pop
13998 @cindex pragma, visibility
14000 This pragma allows the user to set the visibility for multiple
14001 declarations without having to give each a visibility attribute
14002 @xref{Function Attributes}, for more information about visibility and
14003 the attribute syntax.
14005 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
14006 declarations. Class members and template specializations are not
14007 affected; if you want to override the visibility for a particular
14008 member or instantiation, you must use an attribute.
14013 @node Push/Pop Macro Pragmas
14014 @subsection Push/Pop Macro Pragmas
14016 For compatibility with Microsoft Windows compilers, GCC supports
14017 @samp{#pragma push_macro(@var{"macro_name"})}
14018 and @samp{#pragma pop_macro(@var{"macro_name"})}.
14021 @item #pragma push_macro(@var{"macro_name"})
14022 @cindex pragma, push_macro
14023 This pragma saves the value of the macro named as @var{macro_name} to
14024 the top of the stack for this macro.
14026 @item #pragma pop_macro(@var{"macro_name"})
14027 @cindex pragma, pop_macro
14028 This pragma sets the value of the macro named as @var{macro_name} to
14029 the value on top of the stack for this macro. If the stack for
14030 @var{macro_name} is empty, the value of the macro remains unchanged.
14037 #pragma push_macro("X")
14040 #pragma pop_macro("X")
14044 In this example, the definition of X as 1 is saved by @code{#pragma
14045 push_macro} and restored by @code{#pragma pop_macro}.
14047 @node Function Specific Option Pragmas
14048 @subsection Function Specific Option Pragmas
14051 @item #pragma GCC target (@var{"string"}...)
14052 @cindex pragma GCC target
14054 This pragma allows you to set target specific options for functions
14055 defined later in the source file. One or more strings can be
14056 specified. Each function that is defined after this point will be as
14057 if @code{attribute((target("STRING")))} was specified for that
14058 function. The parenthesis around the options is optional.
14059 @xref{Function Attributes}, for more information about the
14060 @code{target} attribute and the attribute syntax.
14062 The @code{#pragma GCC target} attribute is not implemented in GCC versions earlier
14063 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC backends. At
14064 present, it is not implemented for other backends.
14068 @item #pragma GCC optimize (@var{"string"}...)
14069 @cindex pragma GCC optimize
14071 This pragma allows you to set global optimization options for functions
14072 defined later in the source file. One or more strings can be
14073 specified. Each function that is defined after this point will be as
14074 if @code{attribute((optimize("STRING")))} was specified for that
14075 function. The parenthesis around the options is optional.
14076 @xref{Function Attributes}, for more information about the
14077 @code{optimize} attribute and the attribute syntax.
14079 The @samp{#pragma GCC optimize} pragma is not implemented in GCC
14080 versions earlier than 4.4.
14084 @item #pragma GCC push_options
14085 @itemx #pragma GCC pop_options
14086 @cindex pragma GCC push_options
14087 @cindex pragma GCC pop_options
14089 These pragmas maintain a stack of the current target and optimization
14090 options. It is intended for include files where you temporarily want
14091 to switch to using a different @samp{#pragma GCC target} or
14092 @samp{#pragma GCC optimize} and then to pop back to the previous
14095 The @samp{#pragma GCC push_options} and @samp{#pragma GCC pop_options}
14096 pragmas are not implemented in GCC versions earlier than 4.4.
14100 @item #pragma GCC reset_options
14101 @cindex pragma GCC reset_options
14103 This pragma clears the current @code{#pragma GCC target} and
14104 @code{#pragma GCC optimize} to use the default switches as specified
14105 on the command line.
14107 The @samp{#pragma GCC reset_options} pragma is not implemented in GCC
14108 versions earlier than 4.4.
14111 @node Unnamed Fields
14112 @section Unnamed struct/union fields within structs/unions
14113 @cindex @code{struct}
14114 @cindex @code{union}
14116 As permitted by ISO C1X and for compatibility with other compilers,
14117 GCC allows you to define
14118 a structure or union that contains, as fields, structures and unions
14119 without names. For example:
14132 In this example, the user would be able to access members of the unnamed
14133 union with code like @samp{foo.b}. Note that only unnamed structs and
14134 unions are allowed, you may not have, for example, an unnamed
14137 You must never create such structures that cause ambiguous field definitions.
14138 For example, this structure:
14149 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
14150 The compiler gives errors for such constructs.
14152 @opindex fms-extensions
14153 Unless @option{-fms-extensions} is used, the unnamed field must be a
14154 structure or union definition without a tag (for example, @samp{struct
14155 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
14156 also be a definition with a tag such as @samp{struct foo @{ int a;
14157 @};}, a reference to a previously defined structure or union such as
14158 @samp{struct foo;}, or a reference to a @code{typedef} name for a
14159 previously defined structure or union type.
14161 @opindex fplan9-extensions
14162 The option @option{-fplan9-extensions} enables
14163 @option{-fms-extensions} as well as two other extensions. First, a
14164 pointer to a structure is automatically converted to a pointer to an
14165 anonymous field for assignments and function calls. For example:
14168 struct s1 @{ int a; @};
14169 struct s2 @{ struct s1; @};
14170 extern void f1 (struct s1 *);
14171 void f2 (struct s2 *p) @{ f1 (p); @}
14174 In the call to @code{f1} inside @code{f2}, the pointer @code{p} is
14175 converted into a pointer to the anonymous field.
14177 Second, when the type of an anonymous field is a @code{typedef} for a
14178 @code{struct} or @code{union}, code may refer to the field using the
14179 name of the @code{typedef}.
14182 typedef struct @{ int a; @} s1;
14183 struct s2 @{ s1; @};
14184 s1 f1 (struct s2 *p) @{ return p->s1; @}
14187 These usages are only permitted when they are not ambiguous.
14190 @section Thread-Local Storage
14191 @cindex Thread-Local Storage
14192 @cindex @acronym{TLS}
14193 @cindex @code{__thread}
14195 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
14196 are allocated such that there is one instance of the variable per extant
14197 thread. The run-time model GCC uses to implement this originates
14198 in the IA-64 processor-specific ABI, but has since been migrated
14199 to other processors as well. It requires significant support from
14200 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
14201 system libraries (@file{libc.so} and @file{libpthread.so}), so it
14202 is not available everywhere.
14204 At the user level, the extension is visible with a new storage
14205 class keyword: @code{__thread}. For example:
14209 extern __thread struct state s;
14210 static __thread char *p;
14213 The @code{__thread} specifier may be used alone, with the @code{extern}
14214 or @code{static} specifiers, but with no other storage class specifier.
14215 When used with @code{extern} or @code{static}, @code{__thread} must appear
14216 immediately after the other storage class specifier.
14218 The @code{__thread} specifier may be applied to any global, file-scoped
14219 static, function-scoped static, or static data member of a class. It may
14220 not be applied to block-scoped automatic or non-static data member.
14222 When the address-of operator is applied to a thread-local variable, it is
14223 evaluated at run-time and returns the address of the current thread's
14224 instance of that variable. An address so obtained may be used by any
14225 thread. When a thread terminates, any pointers to thread-local variables
14226 in that thread become invalid.
14228 No static initialization may refer to the address of a thread-local variable.
14230 In C++, if an initializer is present for a thread-local variable, it must
14231 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
14234 See @uref{http://www.akkadia.org/drepper/tls.pdf,
14235 ELF Handling For Thread-Local Storage} for a detailed explanation of
14236 the four thread-local storage addressing models, and how the run-time
14237 is expected to function.
14240 * C99 Thread-Local Edits::
14241 * C++98 Thread-Local Edits::
14244 @node C99 Thread-Local Edits
14245 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
14247 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
14248 that document the exact semantics of the language extension.
14252 @cite{5.1.2 Execution environments}
14254 Add new text after paragraph 1
14257 Within either execution environment, a @dfn{thread} is a flow of
14258 control within a program. It is implementation defined whether
14259 or not there may be more than one thread associated with a program.
14260 It is implementation defined how threads beyond the first are
14261 created, the name and type of the function called at thread
14262 startup, and how threads may be terminated. However, objects
14263 with thread storage duration shall be initialized before thread
14268 @cite{6.2.4 Storage durations of objects}
14270 Add new text before paragraph 3
14273 An object whose identifier is declared with the storage-class
14274 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
14275 Its lifetime is the entire execution of the thread, and its
14276 stored value is initialized only once, prior to thread startup.
14280 @cite{6.4.1 Keywords}
14282 Add @code{__thread}.
14285 @cite{6.7.1 Storage-class specifiers}
14287 Add @code{__thread} to the list of storage class specifiers in
14290 Change paragraph 2 to
14293 With the exception of @code{__thread}, at most one storage-class
14294 specifier may be given [@dots{}]. The @code{__thread} specifier may
14295 be used alone, or immediately following @code{extern} or
14299 Add new text after paragraph 6
14302 The declaration of an identifier for a variable that has
14303 block scope that specifies @code{__thread} shall also
14304 specify either @code{extern} or @code{static}.
14306 The @code{__thread} specifier shall be used only with
14311 @node C++98 Thread-Local Edits
14312 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
14314 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
14315 that document the exact semantics of the language extension.
14319 @b{[intro.execution]}
14321 New text after paragraph 4
14324 A @dfn{thread} is a flow of control within the abstract machine.
14325 It is implementation defined whether or not there may be more than
14329 New text after paragraph 7
14332 It is unspecified whether additional action must be taken to
14333 ensure when and whether side effects are visible to other threads.
14339 Add @code{__thread}.
14342 @b{[basic.start.main]}
14344 Add after paragraph 5
14347 The thread that begins execution at the @code{main} function is called
14348 the @dfn{main thread}. It is implementation defined how functions
14349 beginning threads other than the main thread are designated or typed.
14350 A function so designated, as well as the @code{main} function, is called
14351 a @dfn{thread startup function}. It is implementation defined what
14352 happens if a thread startup function returns. It is implementation
14353 defined what happens to other threads when any thread calls @code{exit}.
14357 @b{[basic.start.init]}
14359 Add after paragraph 4
14362 The storage for an object of thread storage duration shall be
14363 statically initialized before the first statement of the thread startup
14364 function. An object of thread storage duration shall not require
14365 dynamic initialization.
14369 @b{[basic.start.term]}
14371 Add after paragraph 3
14374 The type of an object with thread storage duration shall not have a
14375 non-trivial destructor, nor shall it be an array type whose elements
14376 (directly or indirectly) have non-trivial destructors.
14382 Add ``thread storage duration'' to the list in paragraph 1.
14387 Thread, static, and automatic storage durations are associated with
14388 objects introduced by declarations [@dots{}].
14391 Add @code{__thread} to the list of specifiers in paragraph 3.
14394 @b{[basic.stc.thread]}
14396 New section before @b{[basic.stc.static]}
14399 The keyword @code{__thread} applied to a non-local object gives the
14400 object thread storage duration.
14402 A local variable or class data member declared both @code{static}
14403 and @code{__thread} gives the variable or member thread storage
14408 @b{[basic.stc.static]}
14413 All objects which have neither thread storage duration, dynamic
14414 storage duration nor are local [@dots{}].
14420 Add @code{__thread} to the list in paragraph 1.
14425 With the exception of @code{__thread}, at most one
14426 @var{storage-class-specifier} shall appear in a given
14427 @var{decl-specifier-seq}. The @code{__thread} specifier may
14428 be used alone, or immediately following the @code{extern} or
14429 @code{static} specifiers. [@dots{}]
14432 Add after paragraph 5
14435 The @code{__thread} specifier can be applied only to the names of objects
14436 and to anonymous unions.
14442 Add after paragraph 6
14445 Non-@code{static} members shall not be @code{__thread}.
14449 @node Binary constants
14450 @section Binary constants using the @samp{0b} prefix
14451 @cindex Binary constants using the @samp{0b} prefix
14453 Integer constants can be written as binary constants, consisting of a
14454 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
14455 @samp{0B}. This is particularly useful in environments that operate a
14456 lot on the bit-level (like microcontrollers).
14458 The following statements are identical:
14467 The type of these constants follows the same rules as for octal or
14468 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
14471 @node C++ Extensions
14472 @chapter Extensions to the C++ Language
14473 @cindex extensions, C++ language
14474 @cindex C++ language extensions
14476 The GNU compiler provides these extensions to the C++ language (and you
14477 can also use most of the C language extensions in your C++ programs). If you
14478 want to write code that checks whether these features are available, you can
14479 test for the GNU compiler the same way as for C programs: check for a
14480 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
14481 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
14482 Predefined Macros,cpp,The GNU C Preprocessor}).
14485 * C++ Volatiles:: What constitutes an access to a volatile object.
14486 * Restricted Pointers:: C99 restricted pointers and references.
14487 * Vague Linkage:: Where G++ puts inlines, vtables and such.
14488 * C++ Interface:: You can use a single C++ header file for both
14489 declarations and definitions.
14490 * Template Instantiation:: Methods for ensuring that exactly one copy of
14491 each needed template instantiation is emitted.
14492 * Bound member functions:: You can extract a function pointer to the
14493 method denoted by a @samp{->*} or @samp{.*} expression.
14494 * C++ Attributes:: Variable, function, and type attributes for C++ only.
14495 * Namespace Association:: Strong using-directives for namespace association.
14496 * Type Traits:: Compiler support for type traits
14497 * Java Exceptions:: Tweaking exception handling to work with Java.
14498 * Deprecated Features:: Things will disappear from g++.
14499 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
14502 @node C++ Volatiles
14503 @section When is a Volatile C++ Object Accessed?
14504 @cindex accessing volatiles
14505 @cindex volatile read
14506 @cindex volatile write
14507 @cindex volatile access
14509 The C++ standard differs from the C standard in its treatment of
14510 volatile objects. It fails to specify what constitutes a volatile
14511 access, except to say that C++ should behave in a similar manner to C
14512 with respect to volatiles, where possible. However, the different
14513 lvalueness of expressions between C and C++ complicate the behavior.
14514 G++ behaves the same as GCC for volatile access, @xref{C
14515 Extensions,,Volatiles}, for a description of GCC's behavior.
14517 The C and C++ language specifications differ when an object is
14518 accessed in a void context:
14521 volatile int *src = @var{somevalue};
14525 The C++ standard specifies that such expressions do not undergo lvalue
14526 to rvalue conversion, and that the type of the dereferenced object may
14527 be incomplete. The C++ standard does not specify explicitly that it
14528 is lvalue to rvalue conversion which is responsible for causing an
14529 access. There is reason to believe that it is, because otherwise
14530 certain simple expressions become undefined. However, because it
14531 would surprise most programmers, G++ treats dereferencing a pointer to
14532 volatile object of complete type as GCC would do for an equivalent
14533 type in C@. When the object has incomplete type, G++ issues a
14534 warning; if you wish to force an error, you must force a conversion to
14535 rvalue with, for instance, a static cast.
14537 When using a reference to volatile, G++ does not treat equivalent
14538 expressions as accesses to volatiles, but instead issues a warning that
14539 no volatile is accessed. The rationale for this is that otherwise it
14540 becomes difficult to determine where volatile access occur, and not
14541 possible to ignore the return value from functions returning volatile
14542 references. Again, if you wish to force a read, cast the reference to
14545 G++ implements the same behavior as GCC does when assigning to a
14546 volatile object -- there is no reread of the assigned-to object, the
14547 assigned rvalue is reused. Note that in C++ assignment expressions
14548 are lvalues, and if used as an lvalue, the volatile object will be
14549 referred to. For instance, @var{vref} will refer to @var{vobj}, as
14550 expected, in the following example:
14554 volatile int &vref = vobj = @var{something};
14557 @node Restricted Pointers
14558 @section Restricting Pointer Aliasing
14559 @cindex restricted pointers
14560 @cindex restricted references
14561 @cindex restricted this pointer
14563 As with the C front end, G++ understands the C99 feature of restricted pointers,
14564 specified with the @code{__restrict__}, or @code{__restrict} type
14565 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
14566 language flag, @code{restrict} is not a keyword in C++.
14568 In addition to allowing restricted pointers, you can specify restricted
14569 references, which indicate that the reference is not aliased in the local
14573 void fn (int *__restrict__ rptr, int &__restrict__ rref)
14580 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
14581 @var{rref} refers to a (different) unaliased integer.
14583 You may also specify whether a member function's @var{this} pointer is
14584 unaliased by using @code{__restrict__} as a member function qualifier.
14587 void T::fn () __restrict__
14594 Within the body of @code{T::fn}, @var{this} will have the effective
14595 definition @code{T *__restrict__ const this}. Notice that the
14596 interpretation of a @code{__restrict__} member function qualifier is
14597 different to that of @code{const} or @code{volatile} qualifier, in that it
14598 is applied to the pointer rather than the object. This is consistent with
14599 other compilers which implement restricted pointers.
14601 As with all outermost parameter qualifiers, @code{__restrict__} is
14602 ignored in function definition matching. This means you only need to
14603 specify @code{__restrict__} in a function definition, rather than
14604 in a function prototype as well.
14606 @node Vague Linkage
14607 @section Vague Linkage
14608 @cindex vague linkage
14610 There are several constructs in C++ which require space in the object
14611 file but are not clearly tied to a single translation unit. We say that
14612 these constructs have ``vague linkage''. Typically such constructs are
14613 emitted wherever they are needed, though sometimes we can be more
14617 @item Inline Functions
14618 Inline functions are typically defined in a header file which can be
14619 included in many different compilations. Hopefully they can usually be
14620 inlined, but sometimes an out-of-line copy is necessary, if the address
14621 of the function is taken or if inlining fails. In general, we emit an
14622 out-of-line copy in all translation units where one is needed. As an
14623 exception, we only emit inline virtual functions with the vtable, since
14624 it will always require a copy.
14626 Local static variables and string constants used in an inline function
14627 are also considered to have vague linkage, since they must be shared
14628 between all inlined and out-of-line instances of the function.
14632 C++ virtual functions are implemented in most compilers using a lookup
14633 table, known as a vtable. The vtable contains pointers to the virtual
14634 functions provided by a class, and each object of the class contains a
14635 pointer to its vtable (or vtables, in some multiple-inheritance
14636 situations). If the class declares any non-inline, non-pure virtual
14637 functions, the first one is chosen as the ``key method'' for the class,
14638 and the vtable is only emitted in the translation unit where the key
14641 @emph{Note:} If the chosen key method is later defined as inline, the
14642 vtable will still be emitted in every translation unit which defines it.
14643 Make sure that any inline virtuals are declared inline in the class
14644 body, even if they are not defined there.
14646 @item @code{type_info} objects
14647 @cindex @code{type_info}
14649 C++ requires information about types to be written out in order to
14650 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
14651 For polymorphic classes (classes with virtual functions), the @samp{type_info}
14652 object is written out along with the vtable so that @samp{dynamic_cast}
14653 can determine the dynamic type of a class object at runtime. For all
14654 other types, we write out the @samp{type_info} object when it is used: when
14655 applying @samp{typeid} to an expression, throwing an object, or
14656 referring to a type in a catch clause or exception specification.
14658 @item Template Instantiations
14659 Most everything in this section also applies to template instantiations,
14660 but there are other options as well.
14661 @xref{Template Instantiation,,Where's the Template?}.
14665 When used with GNU ld version 2.8 or later on an ELF system such as
14666 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
14667 these constructs will be discarded at link time. This is known as
14670 On targets that don't support COMDAT, but do support weak symbols, GCC
14671 will use them. This way one copy will override all the others, but
14672 the unused copies will still take up space in the executable.
14674 For targets which do not support either COMDAT or weak symbols,
14675 most entities with vague linkage will be emitted as local symbols to
14676 avoid duplicate definition errors from the linker. This will not happen
14677 for local statics in inlines, however, as having multiple copies will
14678 almost certainly break things.
14680 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
14681 another way to control placement of these constructs.
14683 @node C++ Interface
14684 @section #pragma interface and implementation
14686 @cindex interface and implementation headers, C++
14687 @cindex C++ interface and implementation headers
14688 @cindex pragmas, interface and implementation
14690 @code{#pragma interface} and @code{#pragma implementation} provide the
14691 user with a way of explicitly directing the compiler to emit entities
14692 with vague linkage (and debugging information) in a particular
14695 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
14696 most cases, because of COMDAT support and the ``key method'' heuristic
14697 mentioned in @ref{Vague Linkage}. Using them can actually cause your
14698 program to grow due to unnecessary out-of-line copies of inline
14699 functions. Currently (3.4) the only benefit of these
14700 @code{#pragma}s is reduced duplication of debugging information, and
14701 that should be addressed soon on DWARF 2 targets with the use of
14705 @item #pragma interface
14706 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
14707 @kindex #pragma interface
14708 Use this directive in @emph{header files} that define object classes, to save
14709 space in most of the object files that use those classes. Normally,
14710 local copies of certain information (backup copies of inline member
14711 functions, debugging information, and the internal tables that implement
14712 virtual functions) must be kept in each object file that includes class
14713 definitions. You can use this pragma to avoid such duplication. When a
14714 header file containing @samp{#pragma interface} is included in a
14715 compilation, this auxiliary information will not be generated (unless
14716 the main input source file itself uses @samp{#pragma implementation}).
14717 Instead, the object files will contain references to be resolved at link
14720 The second form of this directive is useful for the case where you have
14721 multiple headers with the same name in different directories. If you
14722 use this form, you must specify the same string to @samp{#pragma
14725 @item #pragma implementation
14726 @itemx #pragma implementation "@var{objects}.h"
14727 @kindex #pragma implementation
14728 Use this pragma in a @emph{main input file}, when you want full output from
14729 included header files to be generated (and made globally visible). The
14730 included header file, in turn, should use @samp{#pragma interface}.
14731 Backup copies of inline member functions, debugging information, and the
14732 internal tables used to implement virtual functions are all generated in
14733 implementation files.
14735 @cindex implied @code{#pragma implementation}
14736 @cindex @code{#pragma implementation}, implied
14737 @cindex naming convention, implementation headers
14738 If you use @samp{#pragma implementation} with no argument, it applies to
14739 an include file with the same basename@footnote{A file's @dfn{basename}
14740 was the name stripped of all leading path information and of trailing
14741 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
14742 file. For example, in @file{allclass.cc}, giving just
14743 @samp{#pragma implementation}
14744 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
14746 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
14747 an implementation file whenever you would include it from
14748 @file{allclass.cc} even if you never specified @samp{#pragma
14749 implementation}. This was deemed to be more trouble than it was worth,
14750 however, and disabled.
14752 Use the string argument if you want a single implementation file to
14753 include code from multiple header files. (You must also use
14754 @samp{#include} to include the header file; @samp{#pragma
14755 implementation} only specifies how to use the file---it doesn't actually
14758 There is no way to split up the contents of a single header file into
14759 multiple implementation files.
14762 @cindex inlining and C++ pragmas
14763 @cindex C++ pragmas, effect on inlining
14764 @cindex pragmas in C++, effect on inlining
14765 @samp{#pragma implementation} and @samp{#pragma interface} also have an
14766 effect on function inlining.
14768 If you define a class in a header file marked with @samp{#pragma
14769 interface}, the effect on an inline function defined in that class is
14770 similar to an explicit @code{extern} declaration---the compiler emits
14771 no code at all to define an independent version of the function. Its
14772 definition is used only for inlining with its callers.
14774 @opindex fno-implement-inlines
14775 Conversely, when you include the same header file in a main source file
14776 that declares it as @samp{#pragma implementation}, the compiler emits
14777 code for the function itself; this defines a version of the function
14778 that can be found via pointers (or by callers compiled without
14779 inlining). If all calls to the function can be inlined, you can avoid
14780 emitting the function by compiling with @option{-fno-implement-inlines}.
14781 If any calls were not inlined, you will get linker errors.
14783 @node Template Instantiation
14784 @section Where's the Template?
14785 @cindex template instantiation
14787 C++ templates are the first language feature to require more
14788 intelligence from the environment than one usually finds on a UNIX
14789 system. Somehow the compiler and linker have to make sure that each
14790 template instance occurs exactly once in the executable if it is needed,
14791 and not at all otherwise. There are two basic approaches to this
14792 problem, which are referred to as the Borland model and the Cfront model.
14795 @item Borland model
14796 Borland C++ solved the template instantiation problem by adding the code
14797 equivalent of common blocks to their linker; the compiler emits template
14798 instances in each translation unit that uses them, and the linker
14799 collapses them together. The advantage of this model is that the linker
14800 only has to consider the object files themselves; there is no external
14801 complexity to worry about. This disadvantage is that compilation time
14802 is increased because the template code is being compiled repeatedly.
14803 Code written for this model tends to include definitions of all
14804 templates in the header file, since they must be seen to be
14808 The AT&T C++ translator, Cfront, solved the template instantiation
14809 problem by creating the notion of a template repository, an
14810 automatically maintained place where template instances are stored. A
14811 more modern version of the repository works as follows: As individual
14812 object files are built, the compiler places any template definitions and
14813 instantiations encountered in the repository. At link time, the link
14814 wrapper adds in the objects in the repository and compiles any needed
14815 instances that were not previously emitted. The advantages of this
14816 model are more optimal compilation speed and the ability to use the
14817 system linker; to implement the Borland model a compiler vendor also
14818 needs to replace the linker. The disadvantages are vastly increased
14819 complexity, and thus potential for error; for some code this can be
14820 just as transparent, but in practice it can been very difficult to build
14821 multiple programs in one directory and one program in multiple
14822 directories. Code written for this model tends to separate definitions
14823 of non-inline member templates into a separate file, which should be
14824 compiled separately.
14827 When used with GNU ld version 2.8 or later on an ELF system such as
14828 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
14829 Borland model. On other systems, G++ implements neither automatic
14832 A future version of G++ will support a hybrid model whereby the compiler
14833 will emit any instantiations for which the template definition is
14834 included in the compile, and store template definitions and
14835 instantiation context information into the object file for the rest.
14836 The link wrapper will extract that information as necessary and invoke
14837 the compiler to produce the remaining instantiations. The linker will
14838 then combine duplicate instantiations.
14840 In the mean time, you have the following options for dealing with
14841 template instantiations:
14846 Compile your template-using code with @option{-frepo}. The compiler will
14847 generate files with the extension @samp{.rpo} listing all of the
14848 template instantiations used in the corresponding object files which
14849 could be instantiated there; the link wrapper, @samp{collect2}, will
14850 then update the @samp{.rpo} files to tell the compiler where to place
14851 those instantiations and rebuild any affected object files. The
14852 link-time overhead is negligible after the first pass, as the compiler
14853 will continue to place the instantiations in the same files.
14855 This is your best option for application code written for the Borland
14856 model, as it will just work. Code written for the Cfront model will
14857 need to be modified so that the template definitions are available at
14858 one or more points of instantiation; usually this is as simple as adding
14859 @code{#include <tmethods.cc>} to the end of each template header.
14861 For library code, if you want the library to provide all of the template
14862 instantiations it needs, just try to link all of its object files
14863 together; the link will fail, but cause the instantiations to be
14864 generated as a side effect. Be warned, however, that this may cause
14865 conflicts if multiple libraries try to provide the same instantiations.
14866 For greater control, use explicit instantiation as described in the next
14870 @opindex fno-implicit-templates
14871 Compile your code with @option{-fno-implicit-templates} to disable the
14872 implicit generation of template instances, and explicitly instantiate
14873 all the ones you use. This approach requires more knowledge of exactly
14874 which instances you need than do the others, but it's less
14875 mysterious and allows greater control. You can scatter the explicit
14876 instantiations throughout your program, perhaps putting them in the
14877 translation units where the instances are used or the translation units
14878 that define the templates themselves; you can put all of the explicit
14879 instantiations you need into one big file; or you can create small files
14886 template class Foo<int>;
14887 template ostream& operator <<
14888 (ostream&, const Foo<int>&);
14891 for each of the instances you need, and create a template instantiation
14892 library from those.
14894 If you are using Cfront-model code, you can probably get away with not
14895 using @option{-fno-implicit-templates} when compiling files that don't
14896 @samp{#include} the member template definitions.
14898 If you use one big file to do the instantiations, you may want to
14899 compile it without @option{-fno-implicit-templates} so you get all of the
14900 instances required by your explicit instantiations (but not by any
14901 other files) without having to specify them as well.
14903 G++ has extended the template instantiation syntax given in the ISO
14904 standard to allow forward declaration of explicit instantiations
14905 (with @code{extern}), instantiation of the compiler support data for a
14906 template class (i.e.@: the vtable) without instantiating any of its
14907 members (with @code{inline}), and instantiation of only the static data
14908 members of a template class, without the support data or member
14909 functions (with (@code{static}):
14912 extern template int max (int, int);
14913 inline template class Foo<int>;
14914 static template class Foo<int>;
14918 Do nothing. Pretend G++ does implement automatic instantiation
14919 management. Code written for the Borland model will work fine, but
14920 each translation unit will contain instances of each of the templates it
14921 uses. In a large program, this can lead to an unacceptable amount of code
14925 @node Bound member functions
14926 @section Extracting the function pointer from a bound pointer to member function
14928 @cindex pointer to member function
14929 @cindex bound pointer to member function
14931 In C++, pointer to member functions (PMFs) are implemented using a wide
14932 pointer of sorts to handle all the possible call mechanisms; the PMF
14933 needs to store information about how to adjust the @samp{this} pointer,
14934 and if the function pointed to is virtual, where to find the vtable, and
14935 where in the vtable to look for the member function. If you are using
14936 PMFs in an inner loop, you should really reconsider that decision. If
14937 that is not an option, you can extract the pointer to the function that
14938 would be called for a given object/PMF pair and call it directly inside
14939 the inner loop, to save a bit of time.
14941 Note that you will still be paying the penalty for the call through a
14942 function pointer; on most modern architectures, such a call defeats the
14943 branch prediction features of the CPU@. This is also true of normal
14944 virtual function calls.
14946 The syntax for this extension is
14950 extern int (A::*fp)();
14951 typedef int (*fptr)(A *);
14953 fptr p = (fptr)(a.*fp);
14956 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
14957 no object is needed to obtain the address of the function. They can be
14958 converted to function pointers directly:
14961 fptr p1 = (fptr)(&A::foo);
14964 @opindex Wno-pmf-conversions
14965 You must specify @option{-Wno-pmf-conversions} to use this extension.
14967 @node C++ Attributes
14968 @section C++-Specific Variable, Function, and Type Attributes
14970 Some attributes only make sense for C++ programs.
14973 @item init_priority (@var{priority})
14974 @cindex @code{init_priority} attribute
14977 In Standard C++, objects defined at namespace scope are guaranteed to be
14978 initialized in an order in strict accordance with that of their definitions
14979 @emph{in a given translation unit}. No guarantee is made for initializations
14980 across translation units. However, GNU C++ allows users to control the
14981 order of initialization of objects defined at namespace scope with the
14982 @code{init_priority} attribute by specifying a relative @var{priority},
14983 a constant integral expression currently bounded between 101 and 65535
14984 inclusive. Lower numbers indicate a higher priority.
14986 In the following example, @code{A} would normally be created before
14987 @code{B}, but the @code{init_priority} attribute has reversed that order:
14990 Some_Class A __attribute__ ((init_priority (2000)));
14991 Some_Class B __attribute__ ((init_priority (543)));
14995 Note that the particular values of @var{priority} do not matter; only their
14998 @item java_interface
14999 @cindex @code{java_interface} attribute
15001 This type attribute informs C++ that the class is a Java interface. It may
15002 only be applied to classes declared within an @code{extern "Java"} block.
15003 Calls to methods declared in this interface will be dispatched using GCJ's
15004 interface table mechanism, instead of regular virtual table dispatch.
15008 See also @ref{Namespace Association}.
15010 @node Namespace Association
15011 @section Namespace Association
15013 @strong{Caution:} The semantics of this extension are not fully
15014 defined. Users should refrain from using this extension as its
15015 semantics may change subtly over time. It is possible that this
15016 extension will be removed in future versions of G++.
15018 A using-directive with @code{__attribute ((strong))} is stronger
15019 than a normal using-directive in two ways:
15023 Templates from the used namespace can be specialized and explicitly
15024 instantiated as though they were members of the using namespace.
15027 The using namespace is considered an associated namespace of all
15028 templates in the used namespace for purposes of argument-dependent
15032 The used namespace must be nested within the using namespace so that
15033 normal unqualified lookup works properly.
15035 This is useful for composing a namespace transparently from
15036 implementation namespaces. For example:
15041 template <class T> struct A @{ @};
15043 using namespace debug __attribute ((__strong__));
15044 template <> struct A<int> @{ @}; // @r{ok to specialize}
15046 template <class T> void f (A<T>);
15051 f (std::A<float>()); // @r{lookup finds} std::f
15057 @section Type Traits
15059 The C++ front-end implements syntactic extensions that allow to
15060 determine at compile time various characteristics of a type (or of a
15064 @item __has_nothrow_assign (type)
15065 If @code{type} is const qualified or is a reference type then the trait is
15066 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
15067 is true, else if @code{type} is a cv class or union type with copy assignment
15068 operators that are known not to throw an exception then the trait is true,
15069 else it is false. Requires: @code{type} shall be a complete type,
15070 (possibly cv-qualified) @code{void}, or an array of unknown bound.
15072 @item __has_nothrow_copy (type)
15073 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
15074 @code{type} is a cv class or union type with copy constructors that
15075 are known not to throw an exception then the trait is true, else it is false.
15076 Requires: @code{type} shall be a complete type, (possibly cv-qualified)
15077 @code{void}, or an array of unknown bound.
15079 @item __has_nothrow_constructor (type)
15080 If @code{__has_trivial_constructor (type)} is true then the trait is
15081 true, else if @code{type} is a cv class or union type (or array
15082 thereof) with a default constructor that is known not to throw an
15083 exception then the trait is true, else it is false. Requires:
15084 @code{type} shall be a complete type, (possibly cv-qualified)
15085 @code{void}, or an array of unknown bound.
15087 @item __has_trivial_assign (type)
15088 If @code{type} is const qualified or is a reference type then the trait is
15089 false. Otherwise if @code{__is_pod (type)} is true then the trait is
15090 true, else if @code{type} is a cv class or union type with a trivial
15091 copy assignment ([class.copy]) then the trait is true, else it is
15092 false. Requires: @code{type} shall be a complete type, (possibly
15093 cv-qualified) @code{void}, or an array of unknown bound.
15095 @item __has_trivial_copy (type)
15096 If @code{__is_pod (type)} is true or @code{type} is a reference type
15097 then the trait is true, else if @code{type} is a cv class or union type
15098 with a trivial copy constructor ([class.copy]) then the trait
15099 is true, else it is false. Requires: @code{type} shall be a complete
15100 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15102 @item __has_trivial_constructor (type)
15103 If @code{__is_pod (type)} is true then the trait is true, else if
15104 @code{type} is a cv class or union type (or array thereof) with a
15105 trivial default constructor ([class.ctor]) then the trait is true,
15106 else it is false. Requires: @code{type} shall be a complete
15107 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15109 @item __has_trivial_destructor (type)
15110 If @code{__is_pod (type)} is true or @code{type} is a reference type then
15111 the trait is true, else if @code{type} is a cv class or union type (or
15112 array thereof) with a trivial destructor ([class.dtor]) then the trait
15113 is true, else it is false. Requires: @code{type} shall be a complete
15114 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15116 @item __has_virtual_destructor (type)
15117 If @code{type} is a class type with a virtual destructor
15118 ([class.dtor]) then the trait is true, else it is false. Requires:
15119 @code{type} shall be a complete type, (possibly cv-qualified)
15120 @code{void}, or an array of unknown bound.
15122 @item __is_abstract (type)
15123 If @code{type} is an abstract class ([class.abstract]) then the trait
15124 is true, else it is false. Requires: @code{type} shall be a complete
15125 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15127 @item __is_base_of (base_type, derived_type)
15128 If @code{base_type} is a base class of @code{derived_type}
15129 ([class.derived]) then the trait is true, otherwise it is false.
15130 Top-level cv qualifications of @code{base_type} and
15131 @code{derived_type} are ignored. For the purposes of this trait, a
15132 class type is considered is own base. Requires: if @code{__is_class
15133 (base_type)} and @code{__is_class (derived_type)} are true and
15134 @code{base_type} and @code{derived_type} are not the same type
15135 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
15136 type. Diagnostic is produced if this requirement is not met.
15138 @item __is_class (type)
15139 If @code{type} is a cv class type, and not a union type
15140 ([basic.compound]) the trait is true, else it is false.
15142 @item __is_empty (type)
15143 If @code{__is_class (type)} is false then the trait is false.
15144 Otherwise @code{type} is considered empty if and only if: @code{type}
15145 has no non-static data members, or all non-static data members, if
15146 any, are bit-fields of length 0, and @code{type} has no virtual
15147 members, and @code{type} has no virtual base classes, and @code{type}
15148 has no base classes @code{base_type} for which
15149 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
15150 be a complete type, (possibly cv-qualified) @code{void}, or an array
15153 @item __is_enum (type)
15154 If @code{type} is a cv enumeration type ([basic.compound]) the trait is
15155 true, else it is false.
15157 @item __is_literal_type (type)
15158 If @code{type} is a literal type ([basic.types]) the trait is
15159 true, else it is false. Requires: @code{type} shall be a complete type,
15160 (possibly cv-qualified) @code{void}, or an array of unknown bound.
15162 @item __is_pod (type)
15163 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
15164 else it is false. Requires: @code{type} shall be a complete type,
15165 (possibly cv-qualified) @code{void}, or an array of unknown bound.
15167 @item __is_polymorphic (type)
15168 If @code{type} is a polymorphic class ([class.virtual]) then the trait
15169 is true, else it is false. Requires: @code{type} shall be a complete
15170 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15172 @item __is_standard_layout (type)
15173 If @code{type} is a standard-layout type ([basic.types]) the trait is
15174 true, else it is false. Requires: @code{type} shall be a complete
15175 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15177 @item __is_trivial (type)
15178 If @code{type} is a trivial type ([basic.types]) the trait is
15179 true, else it is false. Requires: @code{type} shall be a complete
15180 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15182 @item __is_union (type)
15183 If @code{type} is a cv union type ([basic.compound]) the trait is
15184 true, else it is false.
15186 @item __underlying_type (type)
15187 The underlying type of @code{type}. Requires: @code{type} shall be
15188 an enumeration type ([dcl.enum]).
15192 @node Java Exceptions
15193 @section Java Exceptions
15195 The Java language uses a slightly different exception handling model
15196 from C++. Normally, GNU C++ will automatically detect when you are
15197 writing C++ code that uses Java exceptions, and handle them
15198 appropriately. However, if C++ code only needs to execute destructors
15199 when Java exceptions are thrown through it, GCC will guess incorrectly.
15200 Sample problematic code is:
15203 struct S @{ ~S(); @};
15204 extern void bar(); // @r{is written in Java, and may throw exceptions}
15213 The usual effect of an incorrect guess is a link failure, complaining of
15214 a missing routine called @samp{__gxx_personality_v0}.
15216 You can inform the compiler that Java exceptions are to be used in a
15217 translation unit, irrespective of what it might think, by writing
15218 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
15219 @samp{#pragma} must appear before any functions that throw or catch
15220 exceptions, or run destructors when exceptions are thrown through them.
15222 You cannot mix Java and C++ exceptions in the same translation unit. It
15223 is believed to be safe to throw a C++ exception from one file through
15224 another file compiled for the Java exception model, or vice versa, but
15225 there may be bugs in this area.
15227 @node Deprecated Features
15228 @section Deprecated Features
15230 In the past, the GNU C++ compiler was extended to experiment with new
15231 features, at a time when the C++ language was still evolving. Now that
15232 the C++ standard is complete, some of those features are superseded by
15233 superior alternatives. Using the old features might cause a warning in
15234 some cases that the feature will be dropped in the future. In other
15235 cases, the feature might be gone already.
15237 While the list below is not exhaustive, it documents some of the options
15238 that are now deprecated:
15241 @item -fexternal-templates
15242 @itemx -falt-external-templates
15243 These are two of the many ways for G++ to implement template
15244 instantiation. @xref{Template Instantiation}. The C++ standard clearly
15245 defines how template definitions have to be organized across
15246 implementation units. G++ has an implicit instantiation mechanism that
15247 should work just fine for standard-conforming code.
15249 @item -fstrict-prototype
15250 @itemx -fno-strict-prototype
15251 Previously it was possible to use an empty prototype parameter list to
15252 indicate an unspecified number of parameters (like C), rather than no
15253 parameters, as C++ demands. This feature has been removed, except where
15254 it is required for backwards compatibility. @xref{Backwards Compatibility}.
15257 G++ allows a virtual function returning @samp{void *} to be overridden
15258 by one returning a different pointer type. This extension to the
15259 covariant return type rules is now deprecated and will be removed from a
15262 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
15263 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
15264 and are now removed from G++. Code using these operators should be
15265 modified to use @code{std::min} and @code{std::max} instead.
15267 The named return value extension has been deprecated, and is now
15270 The use of initializer lists with new expressions has been deprecated,
15271 and is now removed from G++.
15273 Floating and complex non-type template parameters have been deprecated,
15274 and are now removed from G++.
15276 The implicit typename extension has been deprecated and is now
15279 The use of default arguments in function pointers, function typedefs
15280 and other places where they are not permitted by the standard is
15281 deprecated and will be removed from a future version of G++.
15283 G++ allows floating-point literals to appear in integral constant expressions,
15284 e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} }
15285 This extension is deprecated and will be removed from a future version.
15287 G++ allows static data members of const floating-point type to be declared
15288 with an initializer in a class definition. The standard only allows
15289 initializers for static members of const integral types and const
15290 enumeration types so this extension has been deprecated and will be removed
15291 from a future version.
15293 @node Backwards Compatibility
15294 @section Backwards Compatibility
15295 @cindex Backwards Compatibility
15296 @cindex ARM [Annotated C++ Reference Manual]
15298 Now that there is a definitive ISO standard C++, G++ has a specification
15299 to adhere to. The C++ language evolved over time, and features that
15300 used to be acceptable in previous drafts of the standard, such as the ARM
15301 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
15302 compilation of C++ written to such drafts, G++ contains some backwards
15303 compatibilities. @emph{All such backwards compatibility features are
15304 liable to disappear in future versions of G++.} They should be considered
15305 deprecated. @xref{Deprecated Features}.
15309 If a variable is declared at for scope, it used to remain in scope until
15310 the end of the scope which contained the for statement (rather than just
15311 within the for scope). G++ retains this, but issues a warning, if such a
15312 variable is accessed outside the for scope.
15314 @item Implicit C language
15315 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
15316 scope to set the language. On such systems, all header files are
15317 implicitly scoped inside a C language scope. Also, an empty prototype
15318 @code{()} will be treated as an unspecified number of arguments, rather
15319 than no arguments, as C++ demands.