1 @c Copyright (C) 1988, 1989, 1992, 1993, 1994, 1996, 1998, 1999, 2000, 2001,
2 @c 2002, 2003, 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
4 @c This is part of the GCC manual.
5 @c For copying conditions, see the file gcc.texi.
8 @chapter Extensions to the C Language Family
9 @cindex extensions, C language
10 @cindex C language extensions
13 GNU C provides several language features not found in ISO standard C@.
14 (The @option{-pedantic} option directs GCC to print a warning message if
15 any of these features is used.) To test for the availability of these
16 features in conditional compilation, check for a predefined macro
17 @code{__GNUC__}, which is always defined under GCC@.
19 These extensions are available in C and Objective-C@. Most of them are
20 also available in C++. @xref{C++ Extensions,,Extensions to the
21 C++ Language}, for extensions that apply @emph{only} to C++.
23 Some features that are in ISO C99 but not C89 or C++ are also, as
24 extensions, accepted by GCC in C89 mode and in C++.
27 * Statement Exprs:: Putting statements and declarations inside expressions.
28 * Local Labels:: Labels local to a block.
29 * Labels as Values:: Getting pointers to labels, and computed gotos.
30 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
31 * Constructing Calls:: Dispatching a call to another function.
32 * Typeof:: @code{typeof}: referring to the type of an expression.
33 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
34 * Long Long:: Double-word integers---@code{long long int}.
35 * Complex:: Data types for complex numbers.
36 * Floating Types:: Additional Floating Types.
37 * Decimal Float:: Decimal Floating Types.
38 * Hex Floats:: Hexadecimal floating-point constants.
39 * Fixed-Point:: Fixed-Point Types.
40 * Zero Length:: Zero-length arrays.
41 * Variable Length:: Arrays whose length is computed at run time.
42 * Empty Structures:: Structures with no members.
43 * Variadic Macros:: Macros with a variable number of arguments.
44 * Escaped Newlines:: Slightly looser rules for escaped newlines.
45 * Subscripting:: Any array can be subscripted, even if not an lvalue.
46 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
47 * Initializers:: Non-constant initializers.
48 * Compound Literals:: Compound literals give structures, unions
50 * Designated Inits:: Labeling elements of initializers.
51 * Cast to Union:: Casting to union type from any member of the union.
52 * Case Ranges:: `case 1 ... 9' and such.
53 * Mixed Declarations:: Mixing declarations and code.
54 * Function Attributes:: Declaring that functions have no side effects,
55 or that they can never return.
56 * Attribute Syntax:: Formal syntax for attributes.
57 * Function Prototypes:: Prototype declarations and old-style definitions.
58 * C++ Comments:: C++ comments are recognized.
59 * Dollar Signs:: Dollar sign is allowed in identifiers.
60 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
61 * Variable Attributes:: Specifying attributes of variables.
62 * Type Attributes:: Specifying attributes of types.
63 * Alignment:: Inquiring about the alignment of a type or variable.
64 * Inline:: Defining inline functions (as fast as macros).
65 * Extended Asm:: Assembler instructions with C expressions as operands.
66 (With them you can define ``built-in'' functions.)
67 * Constraints:: Constraints for asm operands
68 * Asm Labels:: Specifying the assembler name to use for a C symbol.
69 * Explicit Reg Vars:: Defining variables residing in specified registers.
70 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
71 * Incomplete Enums:: @code{enum foo;}, with details to follow.
72 * Function Names:: Printable strings which are the name of the current
74 * Return Address:: Getting the return or frame address of a function.
75 * Vector Extensions:: Using vector instructions through built-in functions.
76 * Offsetof:: Special syntax for implementing @code{offsetof}.
77 * Atomic Builtins:: Built-in functions for atomic memory access.
78 * Object Size Checking:: Built-in functions for limited buffer overflow
80 * Other Builtins:: Other built-in functions.
81 * Target Builtins:: Built-in functions specific to particular targets.
82 * Target Format Checks:: Format checks specific to particular targets.
83 * Pragmas:: Pragmas accepted by GCC.
84 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
85 * Thread-Local:: Per-thread variables.
86 * Binary constants:: Binary constants using the @samp{0b} prefix.
90 @section Statements and Declarations in Expressions
91 @cindex statements inside expressions
92 @cindex declarations inside expressions
93 @cindex expressions containing statements
94 @cindex macros, statements in expressions
96 @c the above section title wrapped and causes an underfull hbox.. i
97 @c changed it from "within" to "in". --mew 4feb93
98 A compound statement enclosed in parentheses may appear as an expression
99 in GNU C@. This allows you to use loops, switches, and local variables
100 within an expression.
102 Recall that a compound statement is a sequence of statements surrounded
103 by braces; in this construct, parentheses go around the braces. For
107 (@{ int y = foo (); int z;
114 is a valid (though slightly more complex than necessary) expression
115 for the absolute value of @code{foo ()}.
117 The last thing in the compound statement should be an expression
118 followed by a semicolon; the value of this subexpression serves as the
119 value of the entire construct. (If you use some other kind of statement
120 last within the braces, the construct has type @code{void}, and thus
121 effectively no value.)
123 This feature is especially useful in making macro definitions ``safe'' (so
124 that they evaluate each operand exactly once). For example, the
125 ``maximum'' function is commonly defined as a macro in standard C as
129 #define max(a,b) ((a) > (b) ? (a) : (b))
133 @cindex side effects, macro argument
134 But this definition computes either @var{a} or @var{b} twice, with bad
135 results if the operand has side effects. In GNU C, if you know the
136 type of the operands (here taken as @code{int}), you can define
137 the macro safely as follows:
140 #define maxint(a,b) \
141 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
144 Embedded statements are not allowed in constant expressions, such as
145 the value of an enumeration constant, the width of a bit-field, or
146 the initial value of a static variable.
148 If you don't know the type of the operand, you can still do this, but you
149 must use @code{typeof} (@pxref{Typeof}).
151 In G++, the result value of a statement expression undergoes array and
152 function pointer decay, and is returned by value to the enclosing
153 expression. For instance, if @code{A} is a class, then
162 will construct a temporary @code{A} object to hold the result of the
163 statement expression, and that will be used to invoke @code{Foo}.
164 Therefore the @code{this} pointer observed by @code{Foo} will not be the
167 Any temporaries created within a statement within a statement expression
168 will be destroyed at the statement's end. This makes statement
169 expressions inside macros slightly different from function calls. In
170 the latter case temporaries introduced during argument evaluation will
171 be destroyed at the end of the statement that includes the function
172 call. In the statement expression case they will be destroyed during
173 the statement expression. For instance,
176 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
177 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
187 will have different places where temporaries are destroyed. For the
188 @code{macro} case, the temporary @code{X} will be destroyed just after
189 the initialization of @code{b}. In the @code{function} case that
190 temporary will be destroyed when the function returns.
192 These considerations mean that it is probably a bad idea to use
193 statement-expressions of this form in header files that are designed to
194 work with C++. (Note that some versions of the GNU C Library contained
195 header files using statement-expression that lead to precisely this
198 Jumping into a statement expression with @code{goto} or using a
199 @code{switch} statement outside the statement expression with a
200 @code{case} or @code{default} label inside the statement expression is
201 not permitted. Jumping into a statement expression with a computed
202 @code{goto} (@pxref{Labels as Values}) yields undefined behavior.
203 Jumping out of a statement expression is permitted, but if the
204 statement expression is part of a larger expression then it is
205 unspecified which other subexpressions of that expression have been
206 evaluated except where the language definition requires certain
207 subexpressions to be evaluated before or after the statement
208 expression. In any case, as with a function call the evaluation of a
209 statement expression is not interleaved with the evaluation of other
210 parts of the containing expression. For example,
213 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
217 will call @code{foo} and @code{bar1} and will not call @code{baz} but
218 may or may not call @code{bar2}. If @code{bar2} is called, it will be
219 called after @code{foo} and before @code{bar1}
222 @section Locally Declared Labels
224 @cindex macros, local labels
226 GCC allows you to declare @dfn{local labels} in any nested block
227 scope. A local label is just like an ordinary label, but you can
228 only reference it (with a @code{goto} statement, or by taking its
229 address) within the block in which it was declared.
231 A local label declaration looks like this:
234 __label__ @var{label};
241 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
244 Local label declarations must come at the beginning of the block,
245 before any ordinary declarations or statements.
247 The label declaration defines the label @emph{name}, but does not define
248 the label itself. You must do this in the usual way, with
249 @code{@var{label}:}, within the statements of the statement expression.
251 The local label feature is useful for complex macros. If a macro
252 contains nested loops, a @code{goto} can be useful for breaking out of
253 them. However, an ordinary label whose scope is the whole function
254 cannot be used: if the macro can be expanded several times in one
255 function, the label will be multiply defined in that function. A
256 local label avoids this problem. For example:
259 #define SEARCH(value, array, target) \
262 typeof (target) _SEARCH_target = (target); \
263 typeof (*(array)) *_SEARCH_array = (array); \
266 for (i = 0; i < max; i++) \
267 for (j = 0; j < max; j++) \
268 if (_SEARCH_array[i][j] == _SEARCH_target) \
269 @{ (value) = i; goto found; @} \
275 This could also be written using a statement-expression:
278 #define SEARCH(array, target) \
281 typeof (target) _SEARCH_target = (target); \
282 typeof (*(array)) *_SEARCH_array = (array); \
285 for (i = 0; i < max; i++) \
286 for (j = 0; j < max; j++) \
287 if (_SEARCH_array[i][j] == _SEARCH_target) \
288 @{ value = i; goto found; @} \
295 Local label declarations also make the labels they declare visible to
296 nested functions, if there are any. @xref{Nested Functions}, for details.
298 @node Labels as Values
299 @section Labels as Values
300 @cindex labels as values
301 @cindex computed gotos
302 @cindex goto with computed label
303 @cindex address of a label
305 You can get the address of a label defined in the current function
306 (or a containing function) with the unary operator @samp{&&}. The
307 value has type @code{void *}. This value is a constant and can be used
308 wherever a constant of that type is valid. For example:
316 To use these values, you need to be able to jump to one. This is done
317 with the computed goto statement@footnote{The analogous feature in
318 Fortran is called an assigned goto, but that name seems inappropriate in
319 C, where one can do more than simply store label addresses in label
320 variables.}, @code{goto *@var{exp};}. For example,
327 Any expression of type @code{void *} is allowed.
329 One way of using these constants is in initializing a static array that
330 will serve as a jump table:
333 static void *array[] = @{ &&foo, &&bar, &&hack @};
336 Then you can select a label with indexing, like this:
343 Note that this does not check whether the subscript is in bounds---array
344 indexing in C never does that.
346 Such an array of label values serves a purpose much like that of the
347 @code{switch} statement. The @code{switch} statement is cleaner, so
348 use that rather than an array unless the problem does not fit a
349 @code{switch} statement very well.
351 Another use of label values is in an interpreter for threaded code.
352 The labels within the interpreter function can be stored in the
353 threaded code for super-fast dispatching.
355 You may not use this mechanism to jump to code in a different function.
356 If you do that, totally unpredictable things will happen. The best way to
357 avoid this is to store the label address only in automatic variables and
358 never pass it as an argument.
360 An alternate way to write the above example is
363 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
365 goto *(&&foo + array[i]);
369 This is more friendly to code living in shared libraries, as it reduces
370 the number of dynamic relocations that are needed, and by consequence,
371 allows the data to be read-only.
373 The @code{&&foo} expressions for the same label might have different values
374 if the containing function is inlined or cloned. If a program relies on
375 them being always the same, @code{__attribute__((__noinline__))} should
376 be used to prevent inlining. If @code{&&foo} is used
377 in a static variable initializer, inlining is forbidden.
379 @node Nested Functions
380 @section Nested Functions
381 @cindex nested functions
382 @cindex downward funargs
385 A @dfn{nested function} is a function defined inside another function.
386 (Nested functions are not supported for GNU C++.) The nested function's
387 name is local to the block where it is defined. For example, here we
388 define a nested function named @code{square}, and call it twice:
392 foo (double a, double b)
394 double square (double z) @{ return z * z; @}
396 return square (a) + square (b);
401 The nested function can access all the variables of the containing
402 function that are visible at the point of its definition. This is
403 called @dfn{lexical scoping}. For example, here we show a nested
404 function which uses an inherited variable named @code{offset}:
408 bar (int *array, int offset, int size)
410 int access (int *array, int index)
411 @{ return array[index + offset]; @}
414 for (i = 0; i < size; i++)
415 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
420 Nested function definitions are permitted within functions in the places
421 where variable definitions are allowed; that is, in any block, mixed
422 with the other declarations and statements in the block.
424 It is possible to call the nested function from outside the scope of its
425 name by storing its address or passing the address to another function:
428 hack (int *array, int size)
430 void store (int index, int value)
431 @{ array[index] = value; @}
433 intermediate (store, size);
437 Here, the function @code{intermediate} receives the address of
438 @code{store} as an argument. If @code{intermediate} calls @code{store},
439 the arguments given to @code{store} are used to store into @code{array}.
440 But this technique works only so long as the containing function
441 (@code{hack}, in this example) does not exit.
443 If you try to call the nested function through its address after the
444 containing function has exited, all hell will break loose. If you try
445 to call it after a containing scope level has exited, and if it refers
446 to some of the variables that are no longer in scope, you may be lucky,
447 but it's not wise to take the risk. If, however, the nested function
448 does not refer to anything that has gone out of scope, you should be
451 GCC implements taking the address of a nested function using a technique
452 called @dfn{trampolines}. A paper describing them is available as
455 @uref{http://people.debian.org/~aaronl/Usenix88-lexic.pdf}.
457 A nested function can jump to a label inherited from a containing
458 function, provided the label was explicitly declared in the containing
459 function (@pxref{Local Labels}). Such a jump returns instantly to the
460 containing function, exiting the nested function which did the
461 @code{goto} and any intermediate functions as well. Here is an example:
465 bar (int *array, int offset, int size)
468 int access (int *array, int index)
472 return array[index + offset];
476 for (i = 0; i < size; i++)
477 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
481 /* @r{Control comes here from @code{access}
482 if it detects an error.} */
489 A nested function always has no linkage. Declaring one with
490 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
491 before its definition, use @code{auto} (which is otherwise meaningless
492 for function declarations).
495 bar (int *array, int offset, int size)
498 auto int access (int *, int);
500 int access (int *array, int index)
504 return array[index + offset];
510 @node Constructing Calls
511 @section Constructing Function Calls
512 @cindex constructing calls
513 @cindex forwarding calls
515 Using the built-in functions described below, you can record
516 the arguments a function received, and call another function
517 with the same arguments, without knowing the number or types
520 You can also record the return value of that function call,
521 and later return that value, without knowing what data type
522 the function tried to return (as long as your caller expects
525 However, these built-in functions may interact badly with some
526 sophisticated features or other extensions of the language. It
527 is, therefore, not recommended to use them outside very simple
528 functions acting as mere forwarders for their arguments.
530 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
531 This built-in function returns a pointer to data
532 describing how to perform a call with the same arguments as were passed
533 to the current function.
535 The function saves the arg pointer register, structure value address,
536 and all registers that might be used to pass arguments to a function
537 into a block of memory allocated on the stack. Then it returns the
538 address of that block.
541 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
542 This built-in function invokes @var{function}
543 with a copy of the parameters described by @var{arguments}
546 The value of @var{arguments} should be the value returned by
547 @code{__builtin_apply_args}. The argument @var{size} specifies the size
548 of the stack argument data, in bytes.
550 This function returns a pointer to data describing
551 how to return whatever value was returned by @var{function}. The data
552 is saved in a block of memory allocated on the stack.
554 It is not always simple to compute the proper value for @var{size}. The
555 value is used by @code{__builtin_apply} to compute the amount of data
556 that should be pushed on the stack and copied from the incoming argument
560 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
561 This built-in function returns the value described by @var{result} from
562 the containing function. You should specify, for @var{result}, a value
563 returned by @code{__builtin_apply}.
566 @deftypefn {Built-in Function} __builtin_va_arg_pack ()
567 This built-in function represents all anonymous arguments of an inline
568 function. It can be used only in inline functions which will be always
569 inlined, never compiled as a separate function, such as those using
570 @code{__attribute__ ((__always_inline__))} or
571 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
572 It must be only passed as last argument to some other function
573 with variable arguments. This is useful for writing small wrapper
574 inlines for variable argument functions, when using preprocessor
575 macros is undesirable. For example:
577 extern int myprintf (FILE *f, const char *format, ...);
578 extern inline __attribute__ ((__gnu_inline__)) int
579 myprintf (FILE *f, const char *format, ...)
581 int r = fprintf (f, "myprintf: ");
584 int s = fprintf (f, format, __builtin_va_arg_pack ());
592 @deftypefn {Built-in Function} __builtin_va_arg_pack_len ()
593 This built-in function returns the number of anonymous arguments of
594 an inline function. It can be used only in inline functions which
595 will be always inlined, never compiled as a separate function, such
596 as those using @code{__attribute__ ((__always_inline__))} or
597 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
598 For example following will do link or runtime checking of open
599 arguments for optimized code:
602 extern inline __attribute__((__gnu_inline__)) int
603 myopen (const char *path, int oflag, ...)
605 if (__builtin_va_arg_pack_len () > 1)
606 warn_open_too_many_arguments ();
608 if (__builtin_constant_p (oflag))
610 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1)
612 warn_open_missing_mode ();
613 return __open_2 (path, oflag);
615 return open (path, oflag, __builtin_va_arg_pack ());
618 if (__builtin_va_arg_pack_len () < 1)
619 return __open_2 (path, oflag);
621 return open (path, oflag, __builtin_va_arg_pack ());
628 @section Referring to a Type with @code{typeof}
631 @cindex macros, types of arguments
633 Another way to refer to the type of an expression is with @code{typeof}.
634 The syntax of using of this keyword looks like @code{sizeof}, but the
635 construct acts semantically like a type name defined with @code{typedef}.
637 There are two ways of writing the argument to @code{typeof}: with an
638 expression or with a type. Here is an example with an expression:
645 This assumes that @code{x} is an array of pointers to functions;
646 the type described is that of the values of the functions.
648 Here is an example with a typename as the argument:
655 Here the type described is that of pointers to @code{int}.
657 If you are writing a header file that must work when included in ISO C
658 programs, write @code{__typeof__} instead of @code{typeof}.
659 @xref{Alternate Keywords}.
661 A @code{typeof}-construct can be used anywhere a typedef name could be
662 used. For example, you can use it in a declaration, in a cast, or inside
663 of @code{sizeof} or @code{typeof}.
665 @code{typeof} is often useful in conjunction with the
666 statements-within-expressions feature. Here is how the two together can
667 be used to define a safe ``maximum'' macro that operates on any
668 arithmetic type and evaluates each of its arguments exactly once:
672 (@{ typeof (a) _a = (a); \
673 typeof (b) _b = (b); \
674 _a > _b ? _a : _b; @})
677 @cindex underscores in variables in macros
678 @cindex @samp{_} in variables in macros
679 @cindex local variables in macros
680 @cindex variables, local, in macros
681 @cindex macros, local variables in
683 The reason for using names that start with underscores for the local
684 variables is to avoid conflicts with variable names that occur within the
685 expressions that are substituted for @code{a} and @code{b}. Eventually we
686 hope to design a new form of declaration syntax that allows you to declare
687 variables whose scopes start only after their initializers; this will be a
688 more reliable way to prevent such conflicts.
691 Some more examples of the use of @code{typeof}:
695 This declares @code{y} with the type of what @code{x} points to.
702 This declares @code{y} as an array of such values.
709 This declares @code{y} as an array of pointers to characters:
712 typeof (typeof (char *)[4]) y;
716 It is equivalent to the following traditional C declaration:
722 To see the meaning of the declaration using @code{typeof}, and why it
723 might be a useful way to write, rewrite it with these macros:
726 #define pointer(T) typeof(T *)
727 #define array(T, N) typeof(T [N])
731 Now the declaration can be rewritten this way:
734 array (pointer (char), 4) y;
738 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
739 pointers to @code{char}.
742 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
743 a more limited extension which permitted one to write
746 typedef @var{T} = @var{expr};
750 with the effect of declaring @var{T} to have the type of the expression
751 @var{expr}. This extension does not work with GCC 3 (versions between
752 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
753 relies on it should be rewritten to use @code{typeof}:
756 typedef typeof(@var{expr}) @var{T};
760 This will work with all versions of GCC@.
763 @section Conditionals with Omitted Operands
764 @cindex conditional expressions, extensions
765 @cindex omitted middle-operands
766 @cindex middle-operands, omitted
767 @cindex extensions, @code{?:}
768 @cindex @code{?:} extensions
770 The middle operand in a conditional expression may be omitted. Then
771 if the first operand is nonzero, its value is the value of the conditional
774 Therefore, the expression
781 has the value of @code{x} if that is nonzero; otherwise, the value of
784 This example is perfectly equivalent to
790 @cindex side effect in ?:
791 @cindex ?: side effect
793 In this simple case, the ability to omit the middle operand is not
794 especially useful. When it becomes useful is when the first operand does,
795 or may (if it is a macro argument), contain a side effect. Then repeating
796 the operand in the middle would perform the side effect twice. Omitting
797 the middle operand uses the value already computed without the undesirable
798 effects of recomputing it.
801 @section Double-Word Integers
802 @cindex @code{long long} data types
803 @cindex double-word arithmetic
804 @cindex multiprecision arithmetic
805 @cindex @code{LL} integer suffix
806 @cindex @code{ULL} integer suffix
808 ISO C99 supports data types for integers that are at least 64 bits wide,
809 and as an extension GCC supports them in C89 mode and in C++.
810 Simply write @code{long long int} for a signed integer, or
811 @code{unsigned long long int} for an unsigned integer. To make an
812 integer constant of type @code{long long int}, add the suffix @samp{LL}
813 to the integer. To make an integer constant of type @code{unsigned long
814 long int}, add the suffix @samp{ULL} to the integer.
816 You can use these types in arithmetic like any other integer types.
817 Addition, subtraction, and bitwise boolean operations on these types
818 are open-coded on all types of machines. Multiplication is open-coded
819 if the machine supports fullword-to-doubleword a widening multiply
820 instruction. Division and shifts are open-coded only on machines that
821 provide special support. The operations that are not open-coded use
822 special library routines that come with GCC@.
824 There may be pitfalls when you use @code{long long} types for function
825 arguments, unless you declare function prototypes. If a function
826 expects type @code{int} for its argument, and you pass a value of type
827 @code{long long int}, confusion will result because the caller and the
828 subroutine will disagree about the number of bytes for the argument.
829 Likewise, if the function expects @code{long long int} and you pass
830 @code{int}. The best way to avoid such problems is to use prototypes.
833 @section Complex Numbers
834 @cindex complex numbers
835 @cindex @code{_Complex} keyword
836 @cindex @code{__complex__} keyword
838 ISO C99 supports complex floating data types, and as an extension GCC
839 supports them in C89 mode and in C++, and supports complex integer data
840 types which are not part of ISO C99. You can declare complex types
841 using the keyword @code{_Complex}. As an extension, the older GNU
842 keyword @code{__complex__} is also supported.
844 For example, @samp{_Complex double x;} declares @code{x} as a
845 variable whose real part and imaginary part are both of type
846 @code{double}. @samp{_Complex short int y;} declares @code{y} to
847 have real and imaginary parts of type @code{short int}; this is not
848 likely to be useful, but it shows that the set of complex types is
851 To write a constant with a complex data type, use the suffix @samp{i} or
852 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
853 has type @code{_Complex float} and @code{3i} has type
854 @code{_Complex int}. Such a constant always has a pure imaginary
855 value, but you can form any complex value you like by adding one to a
856 real constant. This is a GNU extension; if you have an ISO C99
857 conforming C library (such as GNU libc), and want to construct complex
858 constants of floating type, you should include @code{<complex.h>} and
859 use the macros @code{I} or @code{_Complex_I} instead.
861 @cindex @code{__real__} keyword
862 @cindex @code{__imag__} keyword
863 To extract the real part of a complex-valued expression @var{exp}, write
864 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
865 extract the imaginary part. This is a GNU extension; for values of
866 floating type, you should use the ISO C99 functions @code{crealf},
867 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
868 @code{cimagl}, declared in @code{<complex.h>} and also provided as
869 built-in functions by GCC@.
871 @cindex complex conjugation
872 The operator @samp{~} performs complex conjugation when used on a value
873 with a complex type. This is a GNU extension; for values of
874 floating type, you should use the ISO C99 functions @code{conjf},
875 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
876 provided as built-in functions by GCC@.
878 GCC can allocate complex automatic variables in a noncontiguous
879 fashion; it's even possible for the real part to be in a register while
880 the imaginary part is on the stack (or vice-versa). Only the DWARF2
881 debug info format can represent this, so use of DWARF2 is recommended.
882 If you are using the stabs debug info format, GCC describes a noncontiguous
883 complex variable as if it were two separate variables of noncomplex type.
884 If the variable's actual name is @code{foo}, the two fictitious
885 variables are named @code{foo$real} and @code{foo$imag}. You can
886 examine and set these two fictitious variables with your debugger.
889 @section Additional Floating Types
890 @cindex additional floating types
891 @cindex @code{__float80} data type
892 @cindex @code{__float128} data type
893 @cindex @code{w} floating point suffix
894 @cindex @code{q} floating point suffix
895 @cindex @code{W} floating point suffix
896 @cindex @code{Q} floating point suffix
898 As an extension, the GNU C compiler supports additional floating
899 types, @code{__float80} and @code{__float128} to support 80bit
900 (@code{XFmode}) and 128 bit (@code{TFmode}) floating types.
901 Support for additional types includes the arithmetic operators:
902 add, subtract, multiply, divide; unary arithmetic operators;
903 relational operators; equality operators; and conversions to and from
904 integer and other floating types. Use a suffix @samp{w} or @samp{W}
905 in a literal constant of type @code{__float80} and @samp{q} or @samp{Q}
906 for @code{_float128}. You can declare complex types using the
907 corresponding internal complex type, @code{XCmode} for @code{__float80}
908 type and @code{TCmode} for @code{__float128} type:
911 typedef _Complex float __attribute__((mode(TC))) _Complex128;
912 typedef _Complex float __attribute__((mode(XC))) _Complex80;
915 Not all targets support additional floating point types. @code{__float80}
916 is supported on i386, x86_64 and ia64 targets and target @code{__float128}
917 is supported on x86_64 and ia64 targets.
920 @section Decimal Floating Types
921 @cindex decimal floating types
922 @cindex @code{_Decimal32} data type
923 @cindex @code{_Decimal64} data type
924 @cindex @code{_Decimal128} data type
925 @cindex @code{df} integer suffix
926 @cindex @code{dd} integer suffix
927 @cindex @code{dl} integer suffix
928 @cindex @code{DF} integer suffix
929 @cindex @code{DD} integer suffix
930 @cindex @code{DL} integer suffix
932 As an extension, the GNU C compiler supports decimal floating types as
933 defined in the N1176 draft of ISO/IEC WDTR24732. Support for decimal
934 floating types in GCC will evolve as the draft technical report changes.
935 Calling conventions for any target might also change. Not all targets
936 support decimal floating types.
938 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
939 @code{_Decimal128}. They use a radix of ten, unlike the floating types
940 @code{float}, @code{double}, and @code{long double} whose radix is not
941 specified by the C standard but is usually two.
943 Support for decimal floating types includes the arithmetic operators
944 add, subtract, multiply, divide; unary arithmetic operators;
945 relational operators; equality operators; and conversions to and from
946 integer and other floating types. Use a suffix @samp{df} or
947 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
948 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
951 GCC support of decimal float as specified by the draft technical report
956 Translation time data type (TTDT) is not supported.
959 When the value of a decimal floating type cannot be represented in the
960 integer type to which it is being converted, the result is undefined
961 rather than the result value specified by the draft technical report.
964 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
965 are supported by the DWARF2 debug information format.
971 ISO C99 supports floating-point numbers written not only in the usual
972 decimal notation, such as @code{1.55e1}, but also numbers such as
973 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
974 supports this in C89 mode (except in some cases when strictly
975 conforming) and in C++. In that format the
976 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
977 mandatory. The exponent is a decimal number that indicates the power of
978 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
985 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
986 is the same as @code{1.55e1}.
988 Unlike for floating-point numbers in the decimal notation the exponent
989 is always required in the hexadecimal notation. Otherwise the compiler
990 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
991 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
992 extension for floating-point constants of type @code{float}.
995 @section Fixed-Point Types
996 @cindex fixed-point types
997 @cindex @code{_Fract} data type
998 @cindex @code{_Accum} data type
999 @cindex @code{_Sat} data type
1000 @cindex @code{hr} fixed-suffix
1001 @cindex @code{r} fixed-suffix
1002 @cindex @code{lr} fixed-suffix
1003 @cindex @code{llr} fixed-suffix
1004 @cindex @code{uhr} fixed-suffix
1005 @cindex @code{ur} fixed-suffix
1006 @cindex @code{ulr} fixed-suffix
1007 @cindex @code{ullr} fixed-suffix
1008 @cindex @code{hk} fixed-suffix
1009 @cindex @code{k} fixed-suffix
1010 @cindex @code{lk} fixed-suffix
1011 @cindex @code{llk} fixed-suffix
1012 @cindex @code{uhk} fixed-suffix
1013 @cindex @code{uk} fixed-suffix
1014 @cindex @code{ulk} fixed-suffix
1015 @cindex @code{ullk} fixed-suffix
1016 @cindex @code{HR} fixed-suffix
1017 @cindex @code{R} fixed-suffix
1018 @cindex @code{LR} fixed-suffix
1019 @cindex @code{LLR} fixed-suffix
1020 @cindex @code{UHR} fixed-suffix
1021 @cindex @code{UR} fixed-suffix
1022 @cindex @code{ULR} fixed-suffix
1023 @cindex @code{ULLR} fixed-suffix
1024 @cindex @code{HK} fixed-suffix
1025 @cindex @code{K} fixed-suffix
1026 @cindex @code{LK} fixed-suffix
1027 @cindex @code{LLK} fixed-suffix
1028 @cindex @code{UHK} fixed-suffix
1029 @cindex @code{UK} fixed-suffix
1030 @cindex @code{ULK} fixed-suffix
1031 @cindex @code{ULLK} fixed-suffix
1033 As an extension, the GNU C compiler supports fixed-point types as
1034 defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point
1035 types in GCC will evolve as the draft technical report changes.
1036 Calling conventions for any target might also change. Not all targets
1037 support fixed-point types.
1039 The fixed-point types are
1040 @code{short _Fract},
1043 @code{long long _Fract},
1044 @code{unsigned short _Fract},
1045 @code{unsigned _Fract},
1046 @code{unsigned long _Fract},
1047 @code{unsigned long long _Fract},
1048 @code{_Sat short _Fract},
1050 @code{_Sat long _Fract},
1051 @code{_Sat long long _Fract},
1052 @code{_Sat unsigned short _Fract},
1053 @code{_Sat unsigned _Fract},
1054 @code{_Sat unsigned long _Fract},
1055 @code{_Sat unsigned long long _Fract},
1056 @code{short _Accum},
1059 @code{long long _Accum},
1060 @code{unsigned short _Accum},
1061 @code{unsigned _Accum},
1062 @code{unsigned long _Accum},
1063 @code{unsigned long long _Accum},
1064 @code{_Sat short _Accum},
1066 @code{_Sat long _Accum},
1067 @code{_Sat long long _Accum},
1068 @code{_Sat unsigned short _Accum},
1069 @code{_Sat unsigned _Accum},
1070 @code{_Sat unsigned long _Accum},
1071 @code{_Sat unsigned long long _Accum}.
1072 Fixed-point data values contain fractional and optional integral parts.
1073 The format of fixed-point data varies and depends on the target machine.
1075 Support for fixed-point types includes prefix and postfix increment
1076 and decrement operators (@code{++}, @code{--}); unary arithmetic operators
1077 (@code{+}, @code{-}, @code{!}); binary arithmetic operators (@code{+},
1078 @code{-}, @code{*}, @code{/}); binary shift operators (@code{<<}, @code{>>});
1079 relational operators (@code{<}, @code{<=}, @code{>=}, @code{>});
1080 equality operators (@code{==}, @code{!=}); assignment operators
1081 (@code{+=}, @code{-=}, @code{*=}, @code{/=}, @code{<<=}, @code{>>=});
1082 and conversions to and from integer, floating-point, or fixed-point types.
1084 Use a suffix @samp{hr} or @samp{HR} in a literal constant of type
1085 @code{short _Fract} and @code{_Sat short _Fract},
1086 @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract},
1087 @samp{lr} or @samp{LR} for @code{long _Fract} and @code{_Sat long _Fract},
1088 @samp{llr} or @samp{LLR} for @code{long long _Fract} and
1089 @code{_Sat long long _Fract},
1090 @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and
1091 @code{_Sat unsigned short _Fract},
1092 @samp{ur} or @samp{UR} for @code{unsigned _Fract} and
1093 @code{_Sat unsigned _Fract},
1094 @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and
1095 @code{_Sat unsigned long _Fract},
1096 @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract}
1097 and @code{_Sat unsigned long long _Fract},
1098 @samp{hk} or @samp{HK} for @code{short _Accum} and @code{_Sat short _Accum},
1099 @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum},
1100 @samp{lk} or @samp{LK} for @code{long _Accum} and @code{_Sat long _Accum},
1101 @samp{llk} or @samp{LLK} for @code{long long _Accum} and
1102 @code{_Sat long long _Accum},
1103 @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and
1104 @code{_Sat unsigned short _Accum},
1105 @samp{uk} or @samp{UK} for @code{unsigned _Accum} and
1106 @code{_Sat unsigned _Accum},
1107 @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and
1108 @code{_Sat unsigned long _Accum},
1109 and @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum}
1110 and @code{_Sat unsigned long long _Accum}.
1112 GCC support of fixed-point types as specified by the draft technical report
1117 Pragmas to control overflow and rounding behaviors are not implemented.
1120 Fixed-point types are supported by the DWARF2 debug information format.
1123 @section Arrays of Length Zero
1124 @cindex arrays of length zero
1125 @cindex zero-length arrays
1126 @cindex length-zero arrays
1127 @cindex flexible array members
1129 Zero-length arrays are allowed in GNU C@. They are very useful as the
1130 last element of a structure which is really a header for a variable-length
1139 struct line *thisline = (struct line *)
1140 malloc (sizeof (struct line) + this_length);
1141 thisline->length = this_length;
1144 In ISO C90, you would have to give @code{contents} a length of 1, which
1145 means either you waste space or complicate the argument to @code{malloc}.
1147 In ISO C99, you would use a @dfn{flexible array member}, which is
1148 slightly different in syntax and semantics:
1152 Flexible array members are written as @code{contents[]} without
1156 Flexible array members have incomplete type, and so the @code{sizeof}
1157 operator may not be applied. As a quirk of the original implementation
1158 of zero-length arrays, @code{sizeof} evaluates to zero.
1161 Flexible array members may only appear as the last member of a
1162 @code{struct} that is otherwise non-empty.
1165 A structure containing a flexible array member, or a union containing
1166 such a structure (possibly recursively), may not be a member of a
1167 structure or an element of an array. (However, these uses are
1168 permitted by GCC as extensions.)
1171 GCC versions before 3.0 allowed zero-length arrays to be statically
1172 initialized, as if they were flexible arrays. In addition to those
1173 cases that were useful, it also allowed initializations in situations
1174 that would corrupt later data. Non-empty initialization of zero-length
1175 arrays is now treated like any case where there are more initializer
1176 elements than the array holds, in that a suitable warning about "excess
1177 elements in array" is given, and the excess elements (all of them, in
1178 this case) are ignored.
1180 Instead GCC allows static initialization of flexible array members.
1181 This is equivalent to defining a new structure containing the original
1182 structure followed by an array of sufficient size to contain the data.
1183 I.e.@: in the following, @code{f1} is constructed as if it were declared
1189 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1192 struct f1 f1; int data[3];
1193 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1197 The convenience of this extension is that @code{f1} has the desired
1198 type, eliminating the need to consistently refer to @code{f2.f1}.
1200 This has symmetry with normal static arrays, in that an array of
1201 unknown size is also written with @code{[]}.
1203 Of course, this extension only makes sense if the extra data comes at
1204 the end of a top-level object, as otherwise we would be overwriting
1205 data at subsequent offsets. To avoid undue complication and confusion
1206 with initialization of deeply nested arrays, we simply disallow any
1207 non-empty initialization except when the structure is the top-level
1208 object. For example:
1211 struct foo @{ int x; int y[]; @};
1212 struct bar @{ struct foo z; @};
1214 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1215 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1216 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1217 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1220 @node Empty Structures
1221 @section Structures With No Members
1222 @cindex empty structures
1223 @cindex zero-size structures
1225 GCC permits a C structure to have no members:
1232 The structure will have size zero. In C++, empty structures are part
1233 of the language. G++ treats empty structures as if they had a single
1234 member of type @code{char}.
1236 @node Variable Length
1237 @section Arrays of Variable Length
1238 @cindex variable-length arrays
1239 @cindex arrays of variable length
1242 Variable-length automatic arrays are allowed in ISO C99, and as an
1243 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1244 implementation of variable-length arrays does not yet conform in detail
1245 to the ISO C99 standard.) These arrays are
1246 declared like any other automatic arrays, but with a length that is not
1247 a constant expression. The storage is allocated at the point of
1248 declaration and deallocated when the brace-level is exited. For
1253 concat_fopen (char *s1, char *s2, char *mode)
1255 char str[strlen (s1) + strlen (s2) + 1];
1258 return fopen (str, mode);
1262 @cindex scope of a variable length array
1263 @cindex variable-length array scope
1264 @cindex deallocating variable length arrays
1265 Jumping or breaking out of the scope of the array name deallocates the
1266 storage. Jumping into the scope is not allowed; you get an error
1269 @cindex @code{alloca} vs variable-length arrays
1270 You can use the function @code{alloca} to get an effect much like
1271 variable-length arrays. The function @code{alloca} is available in
1272 many other C implementations (but not in all). On the other hand,
1273 variable-length arrays are more elegant.
1275 There are other differences between these two methods. Space allocated
1276 with @code{alloca} exists until the containing @emph{function} returns.
1277 The space for a variable-length array is deallocated as soon as the array
1278 name's scope ends. (If you use both variable-length arrays and
1279 @code{alloca} in the same function, deallocation of a variable-length array
1280 will also deallocate anything more recently allocated with @code{alloca}.)
1282 You can also use variable-length arrays as arguments to functions:
1286 tester (int len, char data[len][len])
1292 The length of an array is computed once when the storage is allocated
1293 and is remembered for the scope of the array in case you access it with
1296 If you want to pass the array first and the length afterward, you can
1297 use a forward declaration in the parameter list---another GNU extension.
1301 tester (int len; char data[len][len], int len)
1307 @cindex parameter forward declaration
1308 The @samp{int len} before the semicolon is a @dfn{parameter forward
1309 declaration}, and it serves the purpose of making the name @code{len}
1310 known when the declaration of @code{data} is parsed.
1312 You can write any number of such parameter forward declarations in the
1313 parameter list. They can be separated by commas or semicolons, but the
1314 last one must end with a semicolon, which is followed by the ``real''
1315 parameter declarations. Each forward declaration must match a ``real''
1316 declaration in parameter name and data type. ISO C99 does not support
1317 parameter forward declarations.
1319 @node Variadic Macros
1320 @section Macros with a Variable Number of Arguments.
1321 @cindex variable number of arguments
1322 @cindex macro with variable arguments
1323 @cindex rest argument (in macro)
1324 @cindex variadic macros
1326 In the ISO C standard of 1999, a macro can be declared to accept a
1327 variable number of arguments much as a function can. The syntax for
1328 defining the macro is similar to that of a function. Here is an
1332 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1335 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1336 such a macro, it represents the zero or more tokens until the closing
1337 parenthesis that ends the invocation, including any commas. This set of
1338 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1339 wherever it appears. See the CPP manual for more information.
1341 GCC has long supported variadic macros, and used a different syntax that
1342 allowed you to give a name to the variable arguments just like any other
1343 argument. Here is an example:
1346 #define debug(format, args...) fprintf (stderr, format, args)
1349 This is in all ways equivalent to the ISO C example above, but arguably
1350 more readable and descriptive.
1352 GNU CPP has two further variadic macro extensions, and permits them to
1353 be used with either of the above forms of macro definition.
1355 In standard C, you are not allowed to leave the variable argument out
1356 entirely; but you are allowed to pass an empty argument. For example,
1357 this invocation is invalid in ISO C, because there is no comma after
1364 GNU CPP permits you to completely omit the variable arguments in this
1365 way. In the above examples, the compiler would complain, though since
1366 the expansion of the macro still has the extra comma after the format
1369 To help solve this problem, CPP behaves specially for variable arguments
1370 used with the token paste operator, @samp{##}. If instead you write
1373 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1376 and if the variable arguments are omitted or empty, the @samp{##}
1377 operator causes the preprocessor to remove the comma before it. If you
1378 do provide some variable arguments in your macro invocation, GNU CPP
1379 does not complain about the paste operation and instead places the
1380 variable arguments after the comma. Just like any other pasted macro
1381 argument, these arguments are not macro expanded.
1383 @node Escaped Newlines
1384 @section Slightly Looser Rules for Escaped Newlines
1385 @cindex escaped newlines
1386 @cindex newlines (escaped)
1388 Recently, the preprocessor has relaxed its treatment of escaped
1389 newlines. Previously, the newline had to immediately follow a
1390 backslash. The current implementation allows whitespace in the form
1391 of spaces, horizontal and vertical tabs, and form feeds between the
1392 backslash and the subsequent newline. The preprocessor issues a
1393 warning, but treats it as a valid escaped newline and combines the two
1394 lines to form a single logical line. This works within comments and
1395 tokens, as well as between tokens. Comments are @emph{not} treated as
1396 whitespace for the purposes of this relaxation, since they have not
1397 yet been replaced with spaces.
1400 @section Non-Lvalue Arrays May Have Subscripts
1401 @cindex subscripting
1402 @cindex arrays, non-lvalue
1404 @cindex subscripting and function values
1405 In ISO C99, arrays that are not lvalues still decay to pointers, and
1406 may be subscripted, although they may not be modified or used after
1407 the next sequence point and the unary @samp{&} operator may not be
1408 applied to them. As an extension, GCC allows such arrays to be
1409 subscripted in C89 mode, though otherwise they do not decay to
1410 pointers outside C99 mode. For example,
1411 this is valid in GNU C though not valid in C89:
1415 struct foo @{int a[4];@};
1421 return f().a[index];
1427 @section Arithmetic on @code{void}- and Function-Pointers
1428 @cindex void pointers, arithmetic
1429 @cindex void, size of pointer to
1430 @cindex function pointers, arithmetic
1431 @cindex function, size of pointer to
1433 In GNU C, addition and subtraction operations are supported on pointers to
1434 @code{void} and on pointers to functions. This is done by treating the
1435 size of a @code{void} or of a function as 1.
1437 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1438 and on function types, and returns 1.
1440 @opindex Wpointer-arith
1441 The option @option{-Wpointer-arith} requests a warning if these extensions
1445 @section Non-Constant Initializers
1446 @cindex initializers, non-constant
1447 @cindex non-constant initializers
1449 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1450 automatic variable are not required to be constant expressions in GNU C@.
1451 Here is an example of an initializer with run-time varying elements:
1454 foo (float f, float g)
1456 float beat_freqs[2] = @{ f-g, f+g @};
1461 @node Compound Literals
1462 @section Compound Literals
1463 @cindex constructor expressions
1464 @cindex initializations in expressions
1465 @cindex structures, constructor expression
1466 @cindex expressions, constructor
1467 @cindex compound literals
1468 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1470 ISO C99 supports compound literals. A compound literal looks like
1471 a cast containing an initializer. Its value is an object of the
1472 type specified in the cast, containing the elements specified in
1473 the initializer; it is an lvalue. As an extension, GCC supports
1474 compound literals in C89 mode and in C++.
1476 Usually, the specified type is a structure. Assume that
1477 @code{struct foo} and @code{structure} are declared as shown:
1480 struct foo @{int a; char b[2];@} structure;
1484 Here is an example of constructing a @code{struct foo} with a compound literal:
1487 structure = ((struct foo) @{x + y, 'a', 0@});
1491 This is equivalent to writing the following:
1495 struct foo temp = @{x + y, 'a', 0@};
1500 You can also construct an array. If all the elements of the compound literal
1501 are (made up of) simple constant expressions, suitable for use in
1502 initializers of objects of static storage duration, then the compound
1503 literal can be coerced to a pointer to its first element and used in
1504 such an initializer, as shown here:
1507 char **foo = (char *[]) @{ "x", "y", "z" @};
1510 Compound literals for scalar types and union types are is
1511 also allowed, but then the compound literal is equivalent
1514 As a GNU extension, GCC allows initialization of objects with static storage
1515 duration by compound literals (which is not possible in ISO C99, because
1516 the initializer is not a constant).
1517 It is handled as if the object was initialized only with the bracket
1518 enclosed list if the types of the compound literal and the object match.
1519 The initializer list of the compound literal must be constant.
1520 If the object being initialized has array type of unknown size, the size is
1521 determined by compound literal size.
1524 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1525 static int y[] = (int []) @{1, 2, 3@};
1526 static int z[] = (int [3]) @{1@};
1530 The above lines are equivalent to the following:
1532 static struct foo x = @{1, 'a', 'b'@};
1533 static int y[] = @{1, 2, 3@};
1534 static int z[] = @{1, 0, 0@};
1537 @node Designated Inits
1538 @section Designated Initializers
1539 @cindex initializers with labeled elements
1540 @cindex labeled elements in initializers
1541 @cindex case labels in initializers
1542 @cindex designated initializers
1544 Standard C89 requires the elements of an initializer to appear in a fixed
1545 order, the same as the order of the elements in the array or structure
1548 In ISO C99 you can give the elements in any order, specifying the array
1549 indices or structure field names they apply to, and GNU C allows this as
1550 an extension in C89 mode as well. This extension is not
1551 implemented in GNU C++.
1553 To specify an array index, write
1554 @samp{[@var{index}] =} before the element value. For example,
1557 int a[6] = @{ [4] = 29, [2] = 15 @};
1564 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1568 The index values must be constant expressions, even if the array being
1569 initialized is automatic.
1571 An alternative syntax for this which has been obsolete since GCC 2.5 but
1572 GCC still accepts is to write @samp{[@var{index}]} before the element
1573 value, with no @samp{=}.
1575 To initialize a range of elements to the same value, write
1576 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1577 extension. For example,
1580 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1584 If the value in it has side-effects, the side-effects will happen only once,
1585 not for each initialized field by the range initializer.
1588 Note that the length of the array is the highest value specified
1591 In a structure initializer, specify the name of a field to initialize
1592 with @samp{.@var{fieldname} =} before the element value. For example,
1593 given the following structure,
1596 struct point @{ int x, y; @};
1600 the following initialization
1603 struct point p = @{ .y = yvalue, .x = xvalue @};
1610 struct point p = @{ xvalue, yvalue @};
1613 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1614 @samp{@var{fieldname}:}, as shown here:
1617 struct point p = @{ y: yvalue, x: xvalue @};
1621 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1622 @dfn{designator}. You can also use a designator (or the obsolete colon
1623 syntax) when initializing a union, to specify which element of the union
1624 should be used. For example,
1627 union foo @{ int i; double d; @};
1629 union foo f = @{ .d = 4 @};
1633 will convert 4 to a @code{double} to store it in the union using
1634 the second element. By contrast, casting 4 to type @code{union foo}
1635 would store it into the union as the integer @code{i}, since it is
1636 an integer. (@xref{Cast to Union}.)
1638 You can combine this technique of naming elements with ordinary C
1639 initialization of successive elements. Each initializer element that
1640 does not have a designator applies to the next consecutive element of the
1641 array or structure. For example,
1644 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1651 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1654 Labeling the elements of an array initializer is especially useful
1655 when the indices are characters or belong to an @code{enum} type.
1660 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1661 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1664 @cindex designator lists
1665 You can also write a series of @samp{.@var{fieldname}} and
1666 @samp{[@var{index}]} designators before an @samp{=} to specify a
1667 nested subobject to initialize; the list is taken relative to the
1668 subobject corresponding to the closest surrounding brace pair. For
1669 example, with the @samp{struct point} declaration above:
1672 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1676 If the same field is initialized multiple times, it will have value from
1677 the last initialization. If any such overridden initialization has
1678 side-effect, it is unspecified whether the side-effect happens or not.
1679 Currently, GCC will discard them and issue a warning.
1682 @section Case Ranges
1684 @cindex ranges in case statements
1686 You can specify a range of consecutive values in a single @code{case} label,
1690 case @var{low} ... @var{high}:
1694 This has the same effect as the proper number of individual @code{case}
1695 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1697 This feature is especially useful for ranges of ASCII character codes:
1703 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1704 it may be parsed wrong when you use it with integer values. For example,
1719 @section Cast to a Union Type
1720 @cindex cast to a union
1721 @cindex union, casting to a
1723 A cast to union type is similar to other casts, except that the type
1724 specified is a union type. You can specify the type either with
1725 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1726 a constructor though, not a cast, and hence does not yield an lvalue like
1727 normal casts. (@xref{Compound Literals}.)
1729 The types that may be cast to the union type are those of the members
1730 of the union. Thus, given the following union and variables:
1733 union foo @{ int i; double d; @};
1739 both @code{x} and @code{y} can be cast to type @code{union foo}.
1741 Using the cast as the right-hand side of an assignment to a variable of
1742 union type is equivalent to storing in a member of the union:
1747 u = (union foo) x @equiv{} u.i = x
1748 u = (union foo) y @equiv{} u.d = y
1751 You can also use the union cast as a function argument:
1754 void hack (union foo);
1756 hack ((union foo) x);
1759 @node Mixed Declarations
1760 @section Mixed Declarations and Code
1761 @cindex mixed declarations and code
1762 @cindex declarations, mixed with code
1763 @cindex code, mixed with declarations
1765 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1766 within compound statements. As an extension, GCC also allows this in
1767 C89 mode. For example, you could do:
1776 Each identifier is visible from where it is declared until the end of
1777 the enclosing block.
1779 @node Function Attributes
1780 @section Declaring Attributes of Functions
1781 @cindex function attributes
1782 @cindex declaring attributes of functions
1783 @cindex functions that never return
1784 @cindex functions that return more than once
1785 @cindex functions that have no side effects
1786 @cindex functions in arbitrary sections
1787 @cindex functions that behave like malloc
1788 @cindex @code{volatile} applied to function
1789 @cindex @code{const} applied to function
1790 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1791 @cindex functions with non-null pointer arguments
1792 @cindex functions that are passed arguments in registers on the 386
1793 @cindex functions that pop the argument stack on the 386
1794 @cindex functions that do not pop the argument stack on the 386
1796 In GNU C, you declare certain things about functions called in your program
1797 which help the compiler optimize function calls and check your code more
1800 The keyword @code{__attribute__} allows you to specify special
1801 attributes when making a declaration. This keyword is followed by an
1802 attribute specification inside double parentheses. The following
1803 attributes are currently defined for functions on all targets:
1804 @code{aligned}, @code{alloc_size}, @code{noreturn},
1805 @code{returns_twice}, @code{noinline}, @code{always_inline},
1806 @code{flatten}, @code{pure}, @code{const}, @code{nothrow},
1807 @code{sentinel}, @code{format}, @code{format_arg},
1808 @code{no_instrument_function}, @code{section}, @code{constructor},
1809 @code{destructor}, @code{used}, @code{unused}, @code{deprecated},
1810 @code{weak}, @code{malloc}, @code{alias}, @code{warn_unused_result},
1811 @code{nonnull}, @code{gnu_inline}, @code{externally_visible},
1812 @code{hot}, @code{cold}, @code{artificial}, @code{error}
1814 Several other attributes are defined for functions on particular
1815 target systems. Other attributes, including @code{section} are
1816 supported for variables declarations (@pxref{Variable Attributes}) and
1817 for types (@pxref{Type Attributes}).
1819 You may also specify attributes with @samp{__} preceding and following
1820 each keyword. This allows you to use them in header files without
1821 being concerned about a possible macro of the same name. For example,
1822 you may use @code{__noreturn__} instead of @code{noreturn}.
1824 @xref{Attribute Syntax}, for details of the exact syntax for using
1828 @c Keep this table alphabetized by attribute name. Treat _ as space.
1830 @item alias ("@var{target}")
1831 @cindex @code{alias} attribute
1832 The @code{alias} attribute causes the declaration to be emitted as an
1833 alias for another symbol, which must be specified. For instance,
1836 void __f () @{ /* @r{Do something.} */; @}
1837 void f () __attribute__ ((weak, alias ("__f")));
1840 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
1841 mangled name for the target must be used. It is an error if @samp{__f}
1842 is not defined in the same translation unit.
1844 Not all target machines support this attribute.
1846 @item aligned (@var{alignment})
1847 @cindex @code{aligned} attribute
1848 This attribute specifies a minimum alignment for the function,
1851 You cannot use this attribute to decrease the alignment of a function,
1852 only to increase it. However, when you explicitly specify a function
1853 alignment this will override the effect of the
1854 @option{-falign-functions} (@pxref{Optimize Options}) option for this
1857 Note that the effectiveness of @code{aligned} attributes may be
1858 limited by inherent limitations in your linker. On many systems, the
1859 linker is only able to arrange for functions to be aligned up to a
1860 certain maximum alignment. (For some linkers, the maximum supported
1861 alignment may be very very small.) See your linker documentation for
1862 further information.
1864 The @code{aligned} attribute can also be used for variables and fields
1865 (@pxref{Variable Attributes}.)
1868 @cindex @code{alloc_size} attribute
1869 The @code{alloc_size} attribute is used to tell the compiler that the
1870 function return value points to memory, where the size is given by
1871 one or two of the functions parameters. GCC uses this
1872 information to improve the correctness of @code{__builtin_object_size}.
1874 The function parameter(s) denoting the allocated size are specified by
1875 one or two integer arguments supplied to the attribute. The allocated size
1876 is either the value of the single function argument specified or the product
1877 of the two function arguments specified. Argument numbering starts at
1883 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
1884 void my_realloc(void*, size_t) __attribute__((alloc_size(2)))
1887 declares that my_calloc will return memory of the size given by
1888 the product of parameter 1 and 2 and that my_realloc will return memory
1889 of the size given by parameter 2.
1892 @cindex @code{always_inline} function attribute
1893 Generally, functions are not inlined unless optimization is specified.
1894 For functions declared inline, this attribute inlines the function even
1895 if no optimization level was specified.
1898 @cindex @code{gnu_inline} function attribute
1899 This attribute should be used with a function which is also declared
1900 with the @code{inline} keyword. It directs GCC to treat the function
1901 as if it were defined in gnu89 mode even when compiling in C99 or
1904 If the function is declared @code{extern}, then this definition of the
1905 function is used only for inlining. In no case is the function
1906 compiled as a standalone function, not even if you take its address
1907 explicitly. Such an address becomes an external reference, as if you
1908 had only declared the function, and had not defined it. This has
1909 almost the effect of a macro. The way to use this is to put a
1910 function definition in a header file with this attribute, and put
1911 another copy of the function, without @code{extern}, in a library
1912 file. The definition in the header file will cause most calls to the
1913 function to be inlined. If any uses of the function remain, they will
1914 refer to the single copy in the library. Note that the two
1915 definitions of the functions need not be precisely the same, although
1916 if they do not have the same effect your program may behave oddly.
1918 In C, if the function is neither @code{extern} nor @code{static}, then
1919 the function is compiled as a standalone function, as well as being
1920 inlined where possible.
1922 This is how GCC traditionally handled functions declared
1923 @code{inline}. Since ISO C99 specifies a different semantics for
1924 @code{inline}, this function attribute is provided as a transition
1925 measure and as a useful feature in its own right. This attribute is
1926 available in GCC 4.1.3 and later. It is available if either of the
1927 preprocessor macros @code{__GNUC_GNU_INLINE__} or
1928 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
1929 Function is As Fast As a Macro}.
1931 In C++, this attribute does not depend on @code{extern} in any way,
1932 but it still requires the @code{inline} keyword to enable its special
1935 @cindex @code{artificial} function attribute
1937 This attribute is useful for small inline wrappers which if possible
1938 should appear during debugging as a unit, depending on the debug
1939 info format it will either mean marking the function as artificial
1940 or using the caller location for all instructions within the inlined
1943 @cindex @code{flatten} function attribute
1945 Generally, inlining into a function is limited. For a function marked with
1946 this attribute, every call inside this function will be inlined, if possible.
1947 Whether the function itself is considered for inlining depends on its size and
1948 the current inlining parameters. The @code{flatten} attribute only works
1949 reliably in unit-at-a-time mode.
1951 @item error ("@var{message}")
1952 @cindex @code{error} function attribute
1953 If this attribute is used on a function declaration and a call to such a function
1954 is not eliminated through dead code elimination or other optimizations, an error
1955 which will include @var{message} will be diagnosed. This is useful
1956 for compile time checking, especially together with @code{__builtin_constant_p}
1957 and inline functions where checking the inline function arguments is not
1958 possible through @code{extern char [(condition) ? 1 : -1];} tricks.
1959 While it is possible to leave the function undefined and thus invoke
1960 a link failure, when using this attribute the problem will be diagnosed
1961 earlier and with exact location of the call even in presence of inline
1962 functions or when not emitting debugging information.
1964 @item warning ("@var{message}")
1965 @cindex @code{warning} function attribute
1966 If this attribute is used on a function declaration and a call to such a function
1967 is not eliminated through dead code elimination or other optimizations, a warning
1968 which will include @var{message} will be diagnosed. This is useful
1969 for compile time checking, especially together with @code{__builtin_constant_p}
1970 and inline functions. While it is possible to define the function with
1971 a message in @code{.gnu.warning*} section, when using this attribute the problem
1972 will be diagnosed earlier and with exact location of the call even in presence
1973 of inline functions or when not emitting debugging information.
1976 @cindex functions that do pop the argument stack on the 386
1978 On the Intel 386, the @code{cdecl} attribute causes the compiler to
1979 assume that the calling function will pop off the stack space used to
1980 pass arguments. This is
1981 useful to override the effects of the @option{-mrtd} switch.
1984 @cindex @code{const} function attribute
1985 Many functions do not examine any values except their arguments, and
1986 have no effects except the return value. Basically this is just slightly
1987 more strict class than the @code{pure} attribute below, since function is not
1988 allowed to read global memory.
1990 @cindex pointer arguments
1991 Note that a function that has pointer arguments and examines the data
1992 pointed to must @emph{not} be declared @code{const}. Likewise, a
1993 function that calls a non-@code{const} function usually must not be
1994 @code{const}. It does not make sense for a @code{const} function to
1997 The attribute @code{const} is not implemented in GCC versions earlier
1998 than 2.5. An alternative way to declare that a function has no side
1999 effects, which works in the current version and in some older versions,
2003 typedef int intfn ();
2005 extern const intfn square;
2008 This approach does not work in GNU C++ from 2.6.0 on, since the language
2009 specifies that the @samp{const} must be attached to the return value.
2013 @itemx constructor (@var{priority})
2014 @itemx destructor (@var{priority})
2015 @cindex @code{constructor} function attribute
2016 @cindex @code{destructor} function attribute
2017 The @code{constructor} attribute causes the function to be called
2018 automatically before execution enters @code{main ()}. Similarly, the
2019 @code{destructor} attribute causes the function to be called
2020 automatically after @code{main ()} has completed or @code{exit ()} has
2021 been called. Functions with these attributes are useful for
2022 initializing data that will be used implicitly during the execution of
2025 You may provide an optional integer priority to control the order in
2026 which constructor and destructor functions are run. A constructor
2027 with a smaller priority number runs before a constructor with a larger
2028 priority number; the opposite relationship holds for destructors. So,
2029 if you have a constructor that allocates a resource and a destructor
2030 that deallocates the same resource, both functions typically have the
2031 same priority. The priorities for constructor and destructor
2032 functions are the same as those specified for namespace-scope C++
2033 objects (@pxref{C++ Attributes}).
2035 These attributes are not currently implemented for Objective-C@.
2038 @cindex @code{deprecated} attribute.
2039 The @code{deprecated} attribute results in a warning if the function
2040 is used anywhere in the source file. This is useful when identifying
2041 functions that are expected to be removed in a future version of a
2042 program. The warning also includes the location of the declaration
2043 of the deprecated function, to enable users to easily find further
2044 information about why the function is deprecated, or what they should
2045 do instead. Note that the warnings only occurs for uses:
2048 int old_fn () __attribute__ ((deprecated));
2050 int (*fn_ptr)() = old_fn;
2053 results in a warning on line 3 but not line 2.
2055 The @code{deprecated} attribute can also be used for variables and
2056 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2059 @cindex @code{__declspec(dllexport)}
2060 On Microsoft Windows targets and Symbian OS targets the
2061 @code{dllexport} attribute causes the compiler to provide a global
2062 pointer to a pointer in a DLL, so that it can be referenced with the
2063 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
2064 name is formed by combining @code{_imp__} and the function or variable
2067 You can use @code{__declspec(dllexport)} as a synonym for
2068 @code{__attribute__ ((dllexport))} for compatibility with other
2071 On systems that support the @code{visibility} attribute, this
2072 attribute also implies ``default'' visibility. It is an error to
2073 explicitly specify any other visibility.
2075 Currently, the @code{dllexport} attribute is ignored for inlined
2076 functions, unless the @option{-fkeep-inline-functions} flag has been
2077 used. The attribute is also ignored for undefined symbols.
2079 When applied to C++ classes, the attribute marks defined non-inlined
2080 member functions and static data members as exports. Static consts
2081 initialized in-class are not marked unless they are also defined
2084 For Microsoft Windows targets there are alternative methods for
2085 including the symbol in the DLL's export table such as using a
2086 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
2087 the @option{--export-all} linker flag.
2090 @cindex @code{__declspec(dllimport)}
2091 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
2092 attribute causes the compiler to reference a function or variable via
2093 a global pointer to a pointer that is set up by the DLL exporting the
2094 symbol. The attribute implies @code{extern}. On Microsoft Windows
2095 targets, the pointer name is formed by combining @code{_imp__} and the
2096 function or variable name.
2098 You can use @code{__declspec(dllimport)} as a synonym for
2099 @code{__attribute__ ((dllimport))} for compatibility with other
2102 On systems that support the @code{visibility} attribute, this
2103 attribute also implies ``default'' visibility. It is an error to
2104 explicitly specify any other visibility.
2106 Currently, the attribute is ignored for inlined functions. If the
2107 attribute is applied to a symbol @emph{definition}, an error is reported.
2108 If a symbol previously declared @code{dllimport} is later defined, the
2109 attribute is ignored in subsequent references, and a warning is emitted.
2110 The attribute is also overridden by a subsequent declaration as
2113 When applied to C++ classes, the attribute marks non-inlined
2114 member functions and static data members as imports. However, the
2115 attribute is ignored for virtual methods to allow creation of vtables
2118 On the SH Symbian OS target the @code{dllimport} attribute also has
2119 another affect---it can cause the vtable and run-time type information
2120 for a class to be exported. This happens when the class has a
2121 dllimport'ed constructor or a non-inline, non-pure virtual function
2122 and, for either of those two conditions, the class also has a inline
2123 constructor or destructor and has a key function that is defined in
2124 the current translation unit.
2126 For Microsoft Windows based targets the use of the @code{dllimport}
2127 attribute on functions is not necessary, but provides a small
2128 performance benefit by eliminating a thunk in the DLL@. The use of the
2129 @code{dllimport} attribute on imported variables was required on older
2130 versions of the GNU linker, but can now be avoided by passing the
2131 @option{--enable-auto-import} switch to the GNU linker. As with
2132 functions, using the attribute for a variable eliminates a thunk in
2135 One drawback to using this attribute is that a pointer to a
2136 @emph{variable} marked as @code{dllimport} cannot be used as a constant
2137 address. However, a pointer to a @emph{function} with the
2138 @code{dllimport} attribute can be used as a constant initializer; in
2139 this case, the address of a stub function in the import lib is
2140 referenced. On Microsoft Windows targets, the attribute can be disabled
2141 for functions by setting the @option{-mnop-fun-dllimport} flag.
2144 @cindex eight bit data on the H8/300, H8/300H, and H8S
2145 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2146 variable should be placed into the eight bit data section.
2147 The compiler will generate more efficient code for certain operations
2148 on data in the eight bit data area. Note the eight bit data area is limited to
2151 You must use GAS and GLD from GNU binutils version 2.7 or later for
2152 this attribute to work correctly.
2154 @item exception_handler
2155 @cindex exception handler functions on the Blackfin processor
2156 Use this attribute on the Blackfin to indicate that the specified function
2157 is an exception handler. The compiler will generate function entry and
2158 exit sequences suitable for use in an exception handler when this
2159 attribute is present.
2162 @cindex functions which handle memory bank switching
2163 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
2164 use a calling convention that takes care of switching memory banks when
2165 entering and leaving a function. This calling convention is also the
2166 default when using the @option{-mlong-calls} option.
2168 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
2169 to call and return from a function.
2171 On 68HC11 the compiler will generate a sequence of instructions
2172 to invoke a board-specific routine to switch the memory bank and call the
2173 real function. The board-specific routine simulates a @code{call}.
2174 At the end of a function, it will jump to a board-specific routine
2175 instead of using @code{rts}. The board-specific return routine simulates
2179 @cindex functions that pop the argument stack on the 386
2180 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2181 pass the first argument (if of integral type) in the register ECX and
2182 the second argument (if of integral type) in the register EDX@. Subsequent
2183 and other typed arguments are passed on the stack. The called function will
2184 pop the arguments off the stack. If the number of arguments is variable all
2185 arguments are pushed on the stack.
2187 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2188 @cindex @code{format} function attribute
2190 The @code{format} attribute specifies that a function takes @code{printf},
2191 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2192 should be type-checked against a format string. For example, the
2197 my_printf (void *my_object, const char *my_format, ...)
2198 __attribute__ ((format (printf, 2, 3)));
2202 causes the compiler to check the arguments in calls to @code{my_printf}
2203 for consistency with the @code{printf} style format string argument
2206 The parameter @var{archetype} determines how the format string is
2207 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime},
2208 @code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or
2209 @code{strfmon}. (You can also use @code{__printf__},
2210 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On
2211 MinGW targets, @code{ms_printf}, @code{ms_scanf}, and
2212 @code{ms_strftime} are also present.
2213 @var{archtype} values such as @code{printf} refer to the formats accepted
2214 by the system's C run-time library, while @code{gnu_} values always refer
2215 to the formats accepted by the GNU C Library. On Microsoft Windows
2216 targets, @code{ms_} values refer to the formats accepted by the
2217 @file{msvcrt.dll} library.
2218 The parameter @var{string-index}
2219 specifies which argument is the format string argument (starting
2220 from 1), while @var{first-to-check} is the number of the first
2221 argument to check against the format string. For functions
2222 where the arguments are not available to be checked (such as
2223 @code{vprintf}), specify the third parameter as zero. In this case the
2224 compiler only checks the format string for consistency. For
2225 @code{strftime} formats, the third parameter is required to be zero.
2226 Since non-static C++ methods have an implicit @code{this} argument, the
2227 arguments of such methods should be counted from two, not one, when
2228 giving values for @var{string-index} and @var{first-to-check}.
2230 In the example above, the format string (@code{my_format}) is the second
2231 argument of the function @code{my_print}, and the arguments to check
2232 start with the third argument, so the correct parameters for the format
2233 attribute are 2 and 3.
2235 @opindex ffreestanding
2236 @opindex fno-builtin
2237 The @code{format} attribute allows you to identify your own functions
2238 which take format strings as arguments, so that GCC can check the
2239 calls to these functions for errors. The compiler always (unless
2240 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2241 for the standard library functions @code{printf}, @code{fprintf},
2242 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2243 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2244 warnings are requested (using @option{-Wformat}), so there is no need to
2245 modify the header file @file{stdio.h}. In C99 mode, the functions
2246 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2247 @code{vsscanf} are also checked. Except in strictly conforming C
2248 standard modes, the X/Open function @code{strfmon} is also checked as
2249 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2250 @xref{C Dialect Options,,Options Controlling C Dialect}.
2252 The target may provide additional types of format checks.
2253 @xref{Target Format Checks,,Format Checks Specific to Particular
2256 @item format_arg (@var{string-index})
2257 @cindex @code{format_arg} function attribute
2258 @opindex Wformat-nonliteral
2259 The @code{format_arg} attribute specifies that a function takes a format
2260 string for a @code{printf}, @code{scanf}, @code{strftime} or
2261 @code{strfmon} style function and modifies it (for example, to translate
2262 it into another language), so the result can be passed to a
2263 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2264 function (with the remaining arguments to the format function the same
2265 as they would have been for the unmodified string). For example, the
2270 my_dgettext (char *my_domain, const char *my_format)
2271 __attribute__ ((format_arg (2)));
2275 causes the compiler to check the arguments in calls to a @code{printf},
2276 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2277 format string argument is a call to the @code{my_dgettext} function, for
2278 consistency with the format string argument @code{my_format}. If the
2279 @code{format_arg} attribute had not been specified, all the compiler
2280 could tell in such calls to format functions would be that the format
2281 string argument is not constant; this would generate a warning when
2282 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2283 without the attribute.
2285 The parameter @var{string-index} specifies which argument is the format
2286 string argument (starting from one). Since non-static C++ methods have
2287 an implicit @code{this} argument, the arguments of such methods should
2288 be counted from two.
2290 The @code{format-arg} attribute allows you to identify your own
2291 functions which modify format strings, so that GCC can check the
2292 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2293 type function whose operands are a call to one of your own function.
2294 The compiler always treats @code{gettext}, @code{dgettext}, and
2295 @code{dcgettext} in this manner except when strict ISO C support is
2296 requested by @option{-ansi} or an appropriate @option{-std} option, or
2297 @option{-ffreestanding} or @option{-fno-builtin}
2298 is used. @xref{C Dialect Options,,Options
2299 Controlling C Dialect}.
2301 @item function_vector
2302 @cindex calling functions through the function vector on H8/300, M16C, M32C and SH2A processors
2303 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2304 function should be called through the function vector. Calling a
2305 function through the function vector will reduce code size, however;
2306 the function vector has a limited size (maximum 128 entries on the H8/300
2307 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
2309 In SH2A target, this attribute declares a function to be called using the
2310 TBR relative addressing mode. The argument to this attribute is the entry
2311 number of the same function in a vector table containing all the TBR
2312 relative addressable functions. For the successful jump, register TBR
2313 should contain the start address of this TBR relative vector table.
2314 In the startup routine of the user application, user needs to care of this
2315 TBR register initialization. The TBR relative vector table can have at
2316 max 256 function entries. The jumps to these functions will be generated
2317 using a SH2A specific, non delayed branch instruction JSR/N @@(disp8,TBR).
2318 You must use GAS and GLD from GNU binutils version 2.7 or later for
2319 this attribute to work correctly.
2321 Please refer the example of M16C target, to see the use of this
2322 attribute while declaring a function,
2324 In an application, for a function being called once, this attribute will
2325 save at least 8 bytes of code; and if other successive calls are being
2326 made to the same function, it will save 2 bytes of code per each of these
2329 On M16C/M32C targets, the @code{function_vector} attribute declares a
2330 special page subroutine call function. Use of this attribute reduces
2331 the code size by 2 bytes for each call generated to the
2332 subroutine. The argument to the attribute is the vector number entry
2333 from the special page vector table which contains the 16 low-order
2334 bits of the subroutine's entry address. Each vector table has special
2335 page number (18 to 255) which are used in @code{jsrs} instruction.
2336 Jump addresses of the routines are generated by adding 0x0F0000 (in
2337 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the 2
2338 byte addresses set in the vector table. Therefore you need to ensure
2339 that all the special page vector routines should get mapped within the
2340 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
2343 In the following example 2 bytes will be saved for each call to
2344 function @code{foo}.
2347 void foo (void) __attribute__((function_vector(0x18)));
2358 If functions are defined in one file and are called in another file,
2359 then be sure to write this declaration in both files.
2361 This attribute is ignored for R8C target.
2364 @cindex interrupt handler functions
2365 Use this attribute on the ARM, AVR, CRX, M32C, M32R/D, m68k, MS1,
2366 and Xstormy16 ports to indicate that the specified function is an
2367 interrupt handler. The compiler will generate function entry and exit
2368 sequences suitable for use in an interrupt handler when this attribute
2371 Note, interrupt handlers for the Blackfin, H8/300, H8/300H, H8S, and
2372 SH processors can be specified via the @code{interrupt_handler} attribute.
2374 Note, on the AVR, interrupts will be enabled inside the function.
2376 Note, for the ARM, you can specify the kind of interrupt to be handled by
2377 adding an optional parameter to the interrupt attribute like this:
2380 void f () __attribute__ ((interrupt ("IRQ")));
2383 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2385 On ARMv7-M the interrupt type is ignored, and the attribute means the function
2386 may be called with a word aligned stack pointer.
2388 @item interrupt_handler
2389 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
2390 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
2391 indicate that the specified function is an interrupt handler. The compiler
2392 will generate function entry and exit sequences suitable for use in an
2393 interrupt handler when this attribute is present.
2395 @item interrupt_thread
2396 @cindex interrupt thread functions on fido
2397 Use this attribute on fido, a subarchitecture of the m68k, to indicate
2398 that the specified function is an interrupt handler that is designed
2399 to run as a thread. The compiler omits generate prologue/epilogue
2400 sequences and replaces the return instruction with a @code{sleep}
2401 instruction. This attribute is available only on fido.
2404 @cindex User stack pointer in interrupts on the Blackfin
2405 When used together with @code{interrupt_handler}, @code{exception_handler}
2406 or @code{nmi_handler}, code will be generated to load the stack pointer
2407 from the USP register in the function prologue.
2410 @cindex @code{l1_text} function attribute
2411 This attribute specifies a function to be placed into L1 Instruction
2412 SRAM@. The function will be put into a specific section named @code{.l1.text}.
2413 With @option{-mfdpic}, function calls with a such function as the callee
2414 or caller will use inlined PLT.
2416 @item long_call/short_call
2417 @cindex indirect calls on ARM
2418 This attribute specifies how a particular function is called on
2419 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2420 command line switch and @code{#pragma long_calls} settings. The
2421 @code{long_call} attribute indicates that the function might be far
2422 away from the call site and require a different (more expensive)
2423 calling sequence. The @code{short_call} attribute always places
2424 the offset to the function from the call site into the @samp{BL}
2425 instruction directly.
2427 @item longcall/shortcall
2428 @cindex functions called via pointer on the RS/6000 and PowerPC
2429 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute
2430 indicates that the function might be far away from the call site and
2431 require a different (more expensive) calling sequence. The
2432 @code{shortcall} attribute indicates that the function is always close
2433 enough for the shorter calling sequence to be used. These attributes
2434 override both the @option{-mlongcall} switch and, on the RS/6000 and
2435 PowerPC, the @code{#pragma longcall} setting.
2437 @xref{RS/6000 and PowerPC Options}, for more information on whether long
2438 calls are necessary.
2440 @item long_call/near/far
2441 @cindex indirect calls on MIPS
2442 These attributes specify how a particular function is called on MIPS@.
2443 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
2444 command-line switch. The @code{long_call} and @code{far} attributes are
2445 synonyms, and cause the compiler to always call
2446 the function by first loading its address into a register, and then using
2447 the contents of that register. The @code{near} attribute has the opposite
2448 effect; it specifies that non-PIC calls should be made using the more
2449 efficient @code{jal} instruction.
2452 @cindex @code{malloc} attribute
2453 The @code{malloc} attribute is used to tell the compiler that a function
2454 may be treated as if any non-@code{NULL} pointer it returns cannot
2455 alias any other pointer valid when the function returns.
2456 This will often improve optimization.
2457 Standard functions with this property include @code{malloc} and
2458 @code{calloc}. @code{realloc}-like functions have this property as
2459 long as the old pointer is never referred to (including comparing it
2460 to the new pointer) after the function returns a non-@code{NULL}
2463 @item mips16/nomips16
2464 @cindex @code{mips16} attribute
2465 @cindex @code{nomips16} attribute
2467 On MIPS targets, you can use the @code{mips16} and @code{nomips16}
2468 function attributes to locally select or turn off MIPS16 code generation.
2469 A function with the @code{mips16} attribute is emitted as MIPS16 code,
2470 while MIPS16 code generation is disabled for functions with the
2471 @code{nomips16} attribute. These attributes override the
2472 @option{-mips16} and @option{-mno-mips16} options on the command line
2473 (@pxref{MIPS Options}).
2475 When compiling files containing mixed MIPS16 and non-MIPS16 code, the
2476 preprocessor symbol @code{__mips16} reflects the setting on the command line,
2477 not that within individual functions. Mixed MIPS16 and non-MIPS16 code
2478 may interact badly with some GCC extensions such as @code{__builtin_apply}
2479 (@pxref{Constructing Calls}).
2481 @item model (@var{model-name})
2482 @cindex function addressability on the M32R/D
2483 @cindex variable addressability on the IA-64
2485 On the M32R/D, use this attribute to set the addressability of an
2486 object, and of the code generated for a function. The identifier
2487 @var{model-name} is one of @code{small}, @code{medium}, or
2488 @code{large}, representing each of the code models.
2490 Small model objects live in the lower 16MB of memory (so that their
2491 addresses can be loaded with the @code{ld24} instruction), and are
2492 callable with the @code{bl} instruction.
2494 Medium model objects may live anywhere in the 32-bit address space (the
2495 compiler will generate @code{seth/add3} instructions to load their addresses),
2496 and are callable with the @code{bl} instruction.
2498 Large model objects may live anywhere in the 32-bit address space (the
2499 compiler will generate @code{seth/add3} instructions to load their addresses),
2500 and may not be reachable with the @code{bl} instruction (the compiler will
2501 generate the much slower @code{seth/add3/jl} instruction sequence).
2503 On IA-64, use this attribute to set the addressability of an object.
2504 At present, the only supported identifier for @var{model-name} is
2505 @code{small}, indicating addressability via ``small'' (22-bit)
2506 addresses (so that their addresses can be loaded with the @code{addl}
2507 instruction). Caveat: such addressing is by definition not position
2508 independent and hence this attribute must not be used for objects
2509 defined by shared libraries.
2511 @item ms_abi/sysv_abi
2512 @cindex @code[ms_abi} attribute
2513 @cindex @code{sysv_abi} attribute
2515 On 64-bit x86_65-*-* targets, you can use an ABI attribute to indicate
2516 which calling convention should be used for a function. The @code{ms_abi}
2517 attribute tells the compiler to use the Microsoft ABI, while the
2518 @code{sysv_abi} attribute tells the compiler to use the ABI used on
2519 GNU/Linux and other systems. The default is to use the Microsoft ABI
2520 when targeting Windows. On all other systems, the default is the AMD ABI.
2522 Note, This feature is currently sorried out for Windows targets trying to
2525 @cindex function without a prologue/epilogue code
2526 Use this attribute on the ARM, AVR, IP2K and SPU ports to indicate that
2527 the specified function does not need prologue/epilogue sequences generated by
2528 the compiler. It is up to the programmer to provide these sequences. The
2529 only statements that can be safely included in naked functions are
2530 @code{asm} statements that do not have operands. All other statements,
2531 including declarations of local variables, @code{if} statements, and so
2532 forth, should be avoided. Naked functions should be used to implement the
2533 body of an assembly function, while allowing the compiler to construct
2534 the requisite function declaration for the assembler.
2537 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2538 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2539 use the normal calling convention based on @code{jsr} and @code{rts}.
2540 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2544 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
2545 Use this attribute together with @code{interrupt_handler},
2546 @code{exception_handler} or @code{nmi_handler} to indicate that the function
2547 entry code should enable nested interrupts or exceptions.
2550 @cindex NMI handler functions on the Blackfin processor
2551 Use this attribute on the Blackfin to indicate that the specified function
2552 is an NMI handler. The compiler will generate function entry and
2553 exit sequences suitable for use in an NMI handler when this
2554 attribute is present.
2556 @item no_instrument_function
2557 @cindex @code{no_instrument_function} function attribute
2558 @opindex finstrument-functions
2559 If @option{-finstrument-functions} is given, profiling function calls will
2560 be generated at entry and exit of most user-compiled functions.
2561 Functions with this attribute will not be so instrumented.
2564 @cindex @code{noinline} function attribute
2565 This function attribute prevents a function from being considered for
2567 @c Don't enumerate the optimizations by name here; we try to be
2568 @c future-compatible with this mechanism.
2569 If the function does not have side-effects, there are optimizations
2570 other than inlining that causes function calls to be optimized away,
2571 although the function call is live. To keep such calls from being
2576 (@pxref{Extended Asm}) in the called function, to serve as a special
2579 @item nonnull (@var{arg-index}, @dots{})
2580 @cindex @code{nonnull} function attribute
2581 The @code{nonnull} attribute specifies that some function parameters should
2582 be non-null pointers. For instance, the declaration:
2586 my_memcpy (void *dest, const void *src, size_t len)
2587 __attribute__((nonnull (1, 2)));
2591 causes the compiler to check that, in calls to @code{my_memcpy},
2592 arguments @var{dest} and @var{src} are non-null. If the compiler
2593 determines that a null pointer is passed in an argument slot marked
2594 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2595 is issued. The compiler may also choose to make optimizations based
2596 on the knowledge that certain function arguments will not be null.
2598 If no argument index list is given to the @code{nonnull} attribute,
2599 all pointer arguments are marked as non-null. To illustrate, the
2600 following declaration is equivalent to the previous example:
2604 my_memcpy (void *dest, const void *src, size_t len)
2605 __attribute__((nonnull));
2609 @cindex @code{noreturn} function attribute
2610 A few standard library functions, such as @code{abort} and @code{exit},
2611 cannot return. GCC knows this automatically. Some programs define
2612 their own functions that never return. You can declare them
2613 @code{noreturn} to tell the compiler this fact. For example,
2617 void fatal () __attribute__ ((noreturn));
2620 fatal (/* @r{@dots{}} */)
2622 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2628 The @code{noreturn} keyword tells the compiler to assume that
2629 @code{fatal} cannot return. It can then optimize without regard to what
2630 would happen if @code{fatal} ever did return. This makes slightly
2631 better code. More importantly, it helps avoid spurious warnings of
2632 uninitialized variables.
2634 The @code{noreturn} keyword does not affect the exceptional path when that
2635 applies: a @code{noreturn}-marked function may still return to the caller
2636 by throwing an exception or calling @code{longjmp}.
2638 Do not assume that registers saved by the calling function are
2639 restored before calling the @code{noreturn} function.
2641 It does not make sense for a @code{noreturn} function to have a return
2642 type other than @code{void}.
2644 The attribute @code{noreturn} is not implemented in GCC versions
2645 earlier than 2.5. An alternative way to declare that a function does
2646 not return, which works in the current version and in some older
2647 versions, is as follows:
2650 typedef void voidfn ();
2652 volatile voidfn fatal;
2655 This approach does not work in GNU C++.
2658 @cindex @code{nothrow} function attribute
2659 The @code{nothrow} attribute is used to inform the compiler that a
2660 function cannot throw an exception. For example, most functions in
2661 the standard C library can be guaranteed not to throw an exception
2662 with the notable exceptions of @code{qsort} and @code{bsearch} that
2663 take function pointer arguments. The @code{nothrow} attribute is not
2664 implemented in GCC versions earlier than 3.3.
2667 @cindex @code{pure} function attribute
2668 Many functions have no effects except the return value and their
2669 return value depends only on the parameters and/or global variables.
2670 Such a function can be subject
2671 to common subexpression elimination and loop optimization just as an
2672 arithmetic operator would be. These functions should be declared
2673 with the attribute @code{pure}. For example,
2676 int square (int) __attribute__ ((pure));
2680 says that the hypothetical function @code{square} is safe to call
2681 fewer times than the program says.
2683 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2684 Interesting non-pure functions are functions with infinite loops or those
2685 depending on volatile memory or other system resource, that may change between
2686 two consecutive calls (such as @code{feof} in a multithreading environment).
2688 The attribute @code{pure} is not implemented in GCC versions earlier
2692 @cindex @code{hot} function attribute
2693 The @code{hot} attribute is used to inform the compiler that a function is a
2694 hot spot of the compiled program. The function is optimized more aggressively
2695 and on many target it is placed into special subsection of the text section so
2696 all hot functions appears close together improving locality.
2698 When profile feedback is available, via @option{-fprofile-use}, hot functions
2699 are automatically detected and this attribute is ignored.
2701 The @code{hot} attribute is not implemented in GCC versions earlier than 4.3.
2704 @cindex @code{cold} function attribute
2705 The @code{cold} attribute is used to inform the compiler that a function is
2706 unlikely executed. The function is optimized for size rather than speed and on
2707 many targets it is placed into special subsection of the text section so all
2708 cold functions appears close together improving code locality of non-cold parts
2709 of program. The paths leading to call of cold functions within code are marked
2710 as unlikely by the branch prediction mechanism. It is thus useful to mark
2711 functions used to handle unlikely conditions, such as @code{perror}, as cold to
2712 improve optimization of hot functions that do call marked functions in rare
2715 When profile feedback is available, via @option{-fprofile-use}, hot functions
2716 are automatically detected and this attribute is ignored.
2718 The @code{hot} attribute is not implemented in GCC versions earlier than 4.3.
2720 @item regparm (@var{number})
2721 @cindex @code{regparm} attribute
2722 @cindex functions that are passed arguments in registers on the 386
2723 On the Intel 386, the @code{regparm} attribute causes the compiler to
2724 pass arguments number one to @var{number} if they are of integral type
2725 in registers EAX, EDX, and ECX instead of on the stack. Functions that
2726 take a variable number of arguments will continue to be passed all of their
2727 arguments on the stack.
2729 Beware that on some ELF systems this attribute is unsuitable for
2730 global functions in shared libraries with lazy binding (which is the
2731 default). Lazy binding will send the first call via resolving code in
2732 the loader, which might assume EAX, EDX and ECX can be clobbered, as
2733 per the standard calling conventions. Solaris 8 is affected by this.
2734 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
2735 safe since the loaders there save all registers. (Lazy binding can be
2736 disabled with the linker or the loader if desired, to avoid the
2740 @cindex @code{sseregparm} attribute
2741 On the Intel 386 with SSE support, the @code{sseregparm} attribute
2742 causes the compiler to pass up to 3 floating point arguments in
2743 SSE registers instead of on the stack. Functions that take a
2744 variable number of arguments will continue to pass all of their
2745 floating point arguments on the stack.
2747 @item force_align_arg_pointer
2748 @cindex @code{force_align_arg_pointer} attribute
2749 On the Intel x86, the @code{force_align_arg_pointer} attribute may be
2750 applied to individual function definitions, generating an alternate
2751 prologue and epilogue that realigns the runtime stack. This supports
2752 mixing legacy codes that run with a 4-byte aligned stack with modern
2753 codes that keep a 16-byte stack for SSE compatibility. The alternate
2754 prologue and epilogue are slower and bigger than the regular ones, and
2755 the alternate prologue requires a scratch register; this lowers the
2756 number of registers available if used in conjunction with the
2757 @code{regparm} attribute. The @code{force_align_arg_pointer}
2758 attribute is incompatible with nested functions; this is considered a
2762 @cindex @code{resbank} attribute
2763 On the SH2A target, this attribute enables the high-speed register
2764 saving and restoration using a register bank for @code{interrupt_handler}
2765 routines. Saving to the bank is performed automatcially after the CPU
2766 accepts an interrupt that uses a register bank.
2768 The nineteen 32-bit registers comprising general register R0 to R14,
2769 control register GBR, and system registers MACH, MACL, and PR and the
2770 vector table address offset are saved into a register bank. Register
2771 banks are stacked in first-in last-out (FILO) sequence. Restoration
2772 from the bank is executed by issuing a RESBANK instruction.
2775 @cindex @code{returns_twice} attribute
2776 The @code{returns_twice} attribute tells the compiler that a function may
2777 return more than one time. The compiler will ensure that all registers
2778 are dead before calling such a function and will emit a warning about
2779 the variables that may be clobbered after the second return from the
2780 function. Examples of such functions are @code{setjmp} and @code{vfork}.
2781 The @code{longjmp}-like counterpart of such function, if any, might need
2782 to be marked with the @code{noreturn} attribute.
2785 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
2786 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
2787 all registers except the stack pointer should be saved in the prologue
2788 regardless of whether they are used or not.
2790 @item section ("@var{section-name}")
2791 @cindex @code{section} function attribute
2792 Normally, the compiler places the code it generates in the @code{text} section.
2793 Sometimes, however, you need additional sections, or you need certain
2794 particular functions to appear in special sections. The @code{section}
2795 attribute specifies that a function lives in a particular section.
2796 For example, the declaration:
2799 extern void foobar (void) __attribute__ ((section ("bar")));
2803 puts the function @code{foobar} in the @code{bar} section.
2805 Some file formats do not support arbitrary sections so the @code{section}
2806 attribute is not available on all platforms.
2807 If you need to map the entire contents of a module to a particular
2808 section, consider using the facilities of the linker instead.
2811 @cindex @code{sentinel} function attribute
2812 This function attribute ensures that a parameter in a function call is
2813 an explicit @code{NULL}. The attribute is only valid on variadic
2814 functions. By default, the sentinel is located at position zero, the
2815 last parameter of the function call. If an optional integer position
2816 argument P is supplied to the attribute, the sentinel must be located at
2817 position P counting backwards from the end of the argument list.
2820 __attribute__ ((sentinel))
2822 __attribute__ ((sentinel(0)))
2825 The attribute is automatically set with a position of 0 for the built-in
2826 functions @code{execl} and @code{execlp}. The built-in function
2827 @code{execle} has the attribute set with a position of 1.
2829 A valid @code{NULL} in this context is defined as zero with any pointer
2830 type. If your system defines the @code{NULL} macro with an integer type
2831 then you need to add an explicit cast. GCC replaces @code{stddef.h}
2832 with a copy that redefines NULL appropriately.
2834 The warnings for missing or incorrect sentinels are enabled with
2838 See long_call/short_call.
2841 See longcall/shortcall.
2844 @cindex signal handler functions on the AVR processors
2845 Use this attribute on the AVR to indicate that the specified
2846 function is a signal handler. The compiler will generate function
2847 entry and exit sequences suitable for use in a signal handler when this
2848 attribute is present. Interrupts will be disabled inside the function.
2851 Use this attribute on the SH to indicate an @code{interrupt_handler}
2852 function should switch to an alternate stack. It expects a string
2853 argument that names a global variable holding the address of the
2858 void f () __attribute__ ((interrupt_handler,
2859 sp_switch ("alt_stack")));
2863 @cindex functions that pop the argument stack on the 386
2864 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2865 assume that the called function will pop off the stack space used to
2866 pass arguments, unless it takes a variable number of arguments.
2869 @cindex tiny data section on the H8/300H and H8S
2870 Use this attribute on the H8/300H and H8S to indicate that the specified
2871 variable should be placed into the tiny data section.
2872 The compiler will generate more efficient code for loads and stores
2873 on data in the tiny data section. Note the tiny data area is limited to
2874 slightly under 32kbytes of data.
2877 Use this attribute on the SH for an @code{interrupt_handler} to return using
2878 @code{trapa} instead of @code{rte}. This attribute expects an integer
2879 argument specifying the trap number to be used.
2882 @cindex @code{unused} attribute.
2883 This attribute, attached to a function, means that the function is meant
2884 to be possibly unused. GCC will not produce a warning for this
2888 @cindex @code{used} attribute.
2889 This attribute, attached to a function, means that code must be emitted
2890 for the function even if it appears that the function is not referenced.
2891 This is useful, for example, when the function is referenced only in
2895 @cindex @code{version_id} attribute on IA64 HP-UX
2896 This attribute, attached to a global variable or function, renames a
2897 symbol to contain a version string, thus allowing for function level
2898 versioning. HP-UX system header files may use version level functioning
2899 for some system calls.
2902 extern int foo () __attribute__((version_id ("20040821")));
2905 Calls to @var{foo} will be mapped to calls to @var{foo@{20040821@}}.
2907 @item visibility ("@var{visibility_type}")
2908 @cindex @code{visibility} attribute
2909 This attribute affects the linkage of the declaration to which it is attached.
2910 There are four supported @var{visibility_type} values: default,
2911 hidden, protected or internal visibility.
2914 void __attribute__ ((visibility ("protected")))
2915 f () @{ /* @r{Do something.} */; @}
2916 int i __attribute__ ((visibility ("hidden")));
2919 The possible values of @var{visibility_type} correspond to the
2920 visibility settings in the ELF gABI.
2923 @c keep this list of visibilities in alphabetical order.
2926 Default visibility is the normal case for the object file format.
2927 This value is available for the visibility attribute to override other
2928 options that may change the assumed visibility of entities.
2930 On ELF, default visibility means that the declaration is visible to other
2931 modules and, in shared libraries, means that the declared entity may be
2934 On Darwin, default visibility means that the declaration is visible to
2937 Default visibility corresponds to ``external linkage'' in the language.
2940 Hidden visibility indicates that the entity declared will have a new
2941 form of linkage, which we'll call ``hidden linkage''. Two
2942 declarations of an object with hidden linkage refer to the same object
2943 if they are in the same shared object.
2946 Internal visibility is like hidden visibility, but with additional
2947 processor specific semantics. Unless otherwise specified by the
2948 psABI, GCC defines internal visibility to mean that a function is
2949 @emph{never} called from another module. Compare this with hidden
2950 functions which, while they cannot be referenced directly by other
2951 modules, can be referenced indirectly via function pointers. By
2952 indicating that a function cannot be called from outside the module,
2953 GCC may for instance omit the load of a PIC register since it is known
2954 that the calling function loaded the correct value.
2957 Protected visibility is like default visibility except that it
2958 indicates that references within the defining module will bind to the
2959 definition in that module. That is, the declared entity cannot be
2960 overridden by another module.
2964 All visibilities are supported on many, but not all, ELF targets
2965 (supported when the assembler supports the @samp{.visibility}
2966 pseudo-op). Default visibility is supported everywhere. Hidden
2967 visibility is supported on Darwin targets.
2969 The visibility attribute should be applied only to declarations which
2970 would otherwise have external linkage. The attribute should be applied
2971 consistently, so that the same entity should not be declared with
2972 different settings of the attribute.
2974 In C++, the visibility attribute applies to types as well as functions
2975 and objects, because in C++ types have linkage. A class must not have
2976 greater visibility than its non-static data member types and bases,
2977 and class members default to the visibility of their class. Also, a
2978 declaration without explicit visibility is limited to the visibility
2981 In C++, you can mark member functions and static member variables of a
2982 class with the visibility attribute. This is useful if if you know a
2983 particular method or static member variable should only be used from
2984 one shared object; then you can mark it hidden while the rest of the
2985 class has default visibility. Care must be taken to avoid breaking
2986 the One Definition Rule; for example, it is usually not useful to mark
2987 an inline method as hidden without marking the whole class as hidden.
2989 A C++ namespace declaration can also have the visibility attribute.
2990 This attribute applies only to the particular namespace body, not to
2991 other definitions of the same namespace; it is equivalent to using
2992 @samp{#pragma GCC visibility} before and after the namespace
2993 definition (@pxref{Visibility Pragmas}).
2995 In C++, if a template argument has limited visibility, this
2996 restriction is implicitly propagated to the template instantiation.
2997 Otherwise, template instantiations and specializations default to the
2998 visibility of their template.
3000 If both the template and enclosing class have explicit visibility, the
3001 visibility from the template is used.
3003 @item warn_unused_result
3004 @cindex @code{warn_unused_result} attribute
3005 The @code{warn_unused_result} attribute causes a warning to be emitted
3006 if a caller of the function with this attribute does not use its
3007 return value. This is useful for functions where not checking
3008 the result is either a security problem or always a bug, such as
3012 int fn () __attribute__ ((warn_unused_result));
3015 if (fn () < 0) return -1;
3021 results in warning on line 5.
3024 @cindex @code{weak} attribute
3025 The @code{weak} attribute causes the declaration to be emitted as a weak
3026 symbol rather than a global. This is primarily useful in defining
3027 library functions which can be overridden in user code, though it can
3028 also be used with non-function declarations. Weak symbols are supported
3029 for ELF targets, and also for a.out targets when using the GNU assembler
3033 @itemx weakref ("@var{target}")
3034 @cindex @code{weakref} attribute
3035 The @code{weakref} attribute marks a declaration as a weak reference.
3036 Without arguments, it should be accompanied by an @code{alias} attribute
3037 naming the target symbol. Optionally, the @var{target} may be given as
3038 an argument to @code{weakref} itself. In either case, @code{weakref}
3039 implicitly marks the declaration as @code{weak}. Without a
3040 @var{target}, given as an argument to @code{weakref} or to @code{alias},
3041 @code{weakref} is equivalent to @code{weak}.
3044 static int x() __attribute__ ((weakref ("y")));
3045 /* is equivalent to... */
3046 static int x() __attribute__ ((weak, weakref, alias ("y")));
3048 static int x() __attribute__ ((weakref));
3049 static int x() __attribute__ ((alias ("y")));
3052 A weak reference is an alias that does not by itself require a
3053 definition to be given for the target symbol. If the target symbol is
3054 only referenced through weak references, then the becomes a @code{weak}
3055 undefined symbol. If it is directly referenced, however, then such
3056 strong references prevail, and a definition will be required for the
3057 symbol, not necessarily in the same translation unit.
3059 The effect is equivalent to moving all references to the alias to a
3060 separate translation unit, renaming the alias to the aliased symbol,
3061 declaring it as weak, compiling the two separate translation units and
3062 performing a reloadable link on them.
3064 At present, a declaration to which @code{weakref} is attached can
3065 only be @code{static}.
3067 @item externally_visible
3068 @cindex @code{externally_visible} attribute.
3069 This attribute, attached to a global variable or function nullify
3070 effect of @option{-fwhole-program} command line option, so the object
3071 remain visible outside the current compilation unit
3075 You can specify multiple attributes in a declaration by separating them
3076 by commas within the double parentheses or by immediately following an
3077 attribute declaration with another attribute declaration.
3079 @cindex @code{#pragma}, reason for not using
3080 @cindex pragma, reason for not using
3081 Some people object to the @code{__attribute__} feature, suggesting that
3082 ISO C's @code{#pragma} should be used instead. At the time
3083 @code{__attribute__} was designed, there were two reasons for not doing
3088 It is impossible to generate @code{#pragma} commands from a macro.
3091 There is no telling what the same @code{#pragma} might mean in another
3095 These two reasons applied to almost any application that might have been
3096 proposed for @code{#pragma}. It was basically a mistake to use
3097 @code{#pragma} for @emph{anything}.
3099 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
3100 to be generated from macros. In addition, a @code{#pragma GCC}
3101 namespace is now in use for GCC-specific pragmas. However, it has been
3102 found convenient to use @code{__attribute__} to achieve a natural
3103 attachment of attributes to their corresponding declarations, whereas
3104 @code{#pragma GCC} is of use for constructs that do not naturally form
3105 part of the grammar. @xref{Other Directives,,Miscellaneous
3106 Preprocessing Directives, cpp, The GNU C Preprocessor}.
3108 @node Attribute Syntax
3109 @section Attribute Syntax
3110 @cindex attribute syntax
3112 This section describes the syntax with which @code{__attribute__} may be
3113 used, and the constructs to which attribute specifiers bind, for the C
3114 language. Some details may vary for C++ and Objective-C@. Because of
3115 infelicities in the grammar for attributes, some forms described here
3116 may not be successfully parsed in all cases.
3118 There are some problems with the semantics of attributes in C++. For
3119 example, there are no manglings for attributes, although they may affect
3120 code generation, so problems may arise when attributed types are used in
3121 conjunction with templates or overloading. Similarly, @code{typeid}
3122 does not distinguish between types with different attributes. Support
3123 for attributes in C++ may be restricted in future to attributes on
3124 declarations only, but not on nested declarators.
3126 @xref{Function Attributes}, for details of the semantics of attributes
3127 applying to functions. @xref{Variable Attributes}, for details of the
3128 semantics of attributes applying to variables. @xref{Type Attributes},
3129 for details of the semantics of attributes applying to structure, union
3130 and enumerated types.
3132 An @dfn{attribute specifier} is of the form
3133 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
3134 is a possibly empty comma-separated sequence of @dfn{attributes}, where
3135 each attribute is one of the following:
3139 Empty. Empty attributes are ignored.
3142 A word (which may be an identifier such as @code{unused}, or a reserved
3143 word such as @code{const}).
3146 A word, followed by, in parentheses, parameters for the attribute.
3147 These parameters take one of the following forms:
3151 An identifier. For example, @code{mode} attributes use this form.
3154 An identifier followed by a comma and a non-empty comma-separated list
3155 of expressions. For example, @code{format} attributes use this form.
3158 A possibly empty comma-separated list of expressions. For example,
3159 @code{format_arg} attributes use this form with the list being a single
3160 integer constant expression, and @code{alias} attributes use this form
3161 with the list being a single string constant.
3165 An @dfn{attribute specifier list} is a sequence of one or more attribute
3166 specifiers, not separated by any other tokens.
3168 In GNU C, an attribute specifier list may appear after the colon following a
3169 label, other than a @code{case} or @code{default} label. The only
3170 attribute it makes sense to use after a label is @code{unused}. This
3171 feature is intended for code generated by programs which contains labels
3172 that may be unused but which is compiled with @option{-Wall}. It would
3173 not normally be appropriate to use in it human-written code, though it
3174 could be useful in cases where the code that jumps to the label is
3175 contained within an @code{#ifdef} conditional. GNU C++ does not permit
3176 such placement of attribute lists, as it is permissible for a
3177 declaration, which could begin with an attribute list, to be labelled in
3178 C++. Declarations cannot be labelled in C90 or C99, so the ambiguity
3179 does not arise there.
3181 An attribute specifier list may appear as part of a @code{struct},
3182 @code{union} or @code{enum} specifier. It may go either immediately
3183 after the @code{struct}, @code{union} or @code{enum} keyword, or after
3184 the closing brace. The former syntax is preferred.
3185 Where attribute specifiers follow the closing brace, they are considered
3186 to relate to the structure, union or enumerated type defined, not to any
3187 enclosing declaration the type specifier appears in, and the type
3188 defined is not complete until after the attribute specifiers.
3189 @c Otherwise, there would be the following problems: a shift/reduce
3190 @c conflict between attributes binding the struct/union/enum and
3191 @c binding to the list of specifiers/qualifiers; and "aligned"
3192 @c attributes could use sizeof for the structure, but the size could be
3193 @c changed later by "packed" attributes.
3195 Otherwise, an attribute specifier appears as part of a declaration,
3196 counting declarations of unnamed parameters and type names, and relates
3197 to that declaration (which may be nested in another declaration, for
3198 example in the case of a parameter declaration), or to a particular declarator
3199 within a declaration. Where an
3200 attribute specifier is applied to a parameter declared as a function or
3201 an array, it should apply to the function or array rather than the
3202 pointer to which the parameter is implicitly converted, but this is not
3203 yet correctly implemented.
3205 Any list of specifiers and qualifiers at the start of a declaration may
3206 contain attribute specifiers, whether or not such a list may in that
3207 context contain storage class specifiers. (Some attributes, however,
3208 are essentially in the nature of storage class specifiers, and only make
3209 sense where storage class specifiers may be used; for example,
3210 @code{section}.) There is one necessary limitation to this syntax: the
3211 first old-style parameter declaration in a function definition cannot
3212 begin with an attribute specifier, because such an attribute applies to
3213 the function instead by syntax described below (which, however, is not
3214 yet implemented in this case). In some other cases, attribute
3215 specifiers are permitted by this grammar but not yet supported by the
3216 compiler. All attribute specifiers in this place relate to the
3217 declaration as a whole. In the obsolescent usage where a type of
3218 @code{int} is implied by the absence of type specifiers, such a list of
3219 specifiers and qualifiers may be an attribute specifier list with no
3220 other specifiers or qualifiers.
3222 At present, the first parameter in a function prototype must have some
3223 type specifier which is not an attribute specifier; this resolves an
3224 ambiguity in the interpretation of @code{void f(int
3225 (__attribute__((foo)) x))}, but is subject to change. At present, if
3226 the parentheses of a function declarator contain only attributes then
3227 those attributes are ignored, rather than yielding an error or warning
3228 or implying a single parameter of type int, but this is subject to
3231 An attribute specifier list may appear immediately before a declarator
3232 (other than the first) in a comma-separated list of declarators in a
3233 declaration of more than one identifier using a single list of
3234 specifiers and qualifiers. Such attribute specifiers apply
3235 only to the identifier before whose declarator they appear. For
3239 __attribute__((noreturn)) void d0 (void),
3240 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
3245 the @code{noreturn} attribute applies to all the functions
3246 declared; the @code{format} attribute only applies to @code{d1}.
3248 An attribute specifier list may appear immediately before the comma,
3249 @code{=} or semicolon terminating the declaration of an identifier other
3250 than a function definition. Such attribute specifiers apply
3251 to the declared object or function. Where an
3252 assembler name for an object or function is specified (@pxref{Asm
3253 Labels}), the attribute must follow the @code{asm}
3256 An attribute specifier list may, in future, be permitted to appear after
3257 the declarator in a function definition (before any old-style parameter
3258 declarations or the function body).
3260 Attribute specifiers may be mixed with type qualifiers appearing inside
3261 the @code{[]} of a parameter array declarator, in the C99 construct by
3262 which such qualifiers are applied to the pointer to which the array is
3263 implicitly converted. Such attribute specifiers apply to the pointer,
3264 not to the array, but at present this is not implemented and they are
3267 An attribute specifier list may appear at the start of a nested
3268 declarator. At present, there are some limitations in this usage: the
3269 attributes correctly apply to the declarator, but for most individual
3270 attributes the semantics this implies are not implemented.
3271 When attribute specifiers follow the @code{*} of a pointer
3272 declarator, they may be mixed with any type qualifiers present.
3273 The following describes the formal semantics of this syntax. It will make the
3274 most sense if you are familiar with the formal specification of
3275 declarators in the ISO C standard.
3277 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
3278 D1}, where @code{T} contains declaration specifiers that specify a type
3279 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
3280 contains an identifier @var{ident}. The type specified for @var{ident}
3281 for derived declarators whose type does not include an attribute
3282 specifier is as in the ISO C standard.
3284 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
3285 and the declaration @code{T D} specifies the type
3286 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
3287 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
3288 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
3290 If @code{D1} has the form @code{*
3291 @var{type-qualifier-and-attribute-specifier-list} D}, and the
3292 declaration @code{T D} specifies the type
3293 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
3294 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
3295 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
3301 void (__attribute__((noreturn)) ****f) (void);
3305 specifies the type ``pointer to pointer to pointer to pointer to
3306 non-returning function returning @code{void}''. As another example,
3309 char *__attribute__((aligned(8))) *f;
3313 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
3314 Note again that this does not work with most attributes; for example,
3315 the usage of @samp{aligned} and @samp{noreturn} attributes given above
3316 is not yet supported.
3318 For compatibility with existing code written for compiler versions that
3319 did not implement attributes on nested declarators, some laxity is
3320 allowed in the placing of attributes. If an attribute that only applies
3321 to types is applied to a declaration, it will be treated as applying to
3322 the type of that declaration. If an attribute that only applies to
3323 declarations is applied to the type of a declaration, it will be treated
3324 as applying to that declaration; and, for compatibility with code
3325 placing the attributes immediately before the identifier declared, such
3326 an attribute applied to a function return type will be treated as
3327 applying to the function type, and such an attribute applied to an array
3328 element type will be treated as applying to the array type. If an
3329 attribute that only applies to function types is applied to a
3330 pointer-to-function type, it will be treated as applying to the pointer
3331 target type; if such an attribute is applied to a function return type
3332 that is not a pointer-to-function type, it will be treated as applying
3333 to the function type.
3335 @node Function Prototypes
3336 @section Prototypes and Old-Style Function Definitions
3337 @cindex function prototype declarations
3338 @cindex old-style function definitions
3339 @cindex promotion of formal parameters
3341 GNU C extends ISO C to allow a function prototype to override a later
3342 old-style non-prototype definition. Consider the following example:
3345 /* @r{Use prototypes unless the compiler is old-fashioned.} */
3352 /* @r{Prototype function declaration.} */
3353 int isroot P((uid_t));
3355 /* @r{Old-style function definition.} */
3357 isroot (x) /* @r{??? lossage here ???} */
3364 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
3365 not allow this example, because subword arguments in old-style
3366 non-prototype definitions are promoted. Therefore in this example the
3367 function definition's argument is really an @code{int}, which does not
3368 match the prototype argument type of @code{short}.
3370 This restriction of ISO C makes it hard to write code that is portable
3371 to traditional C compilers, because the programmer does not know
3372 whether the @code{uid_t} type is @code{short}, @code{int}, or
3373 @code{long}. Therefore, in cases like these GNU C allows a prototype
3374 to override a later old-style definition. More precisely, in GNU C, a
3375 function prototype argument type overrides the argument type specified
3376 by a later old-style definition if the former type is the same as the
3377 latter type before promotion. Thus in GNU C the above example is
3378 equivalent to the following:
3391 GNU C++ does not support old-style function definitions, so this
3392 extension is irrelevant.
3395 @section C++ Style Comments
3397 @cindex C++ comments
3398 @cindex comments, C++ style
3400 In GNU C, you may use C++ style comments, which start with @samp{//} and
3401 continue until the end of the line. Many other C implementations allow
3402 such comments, and they are included in the 1999 C standard. However,
3403 C++ style comments are not recognized if you specify an @option{-std}
3404 option specifying a version of ISO C before C99, or @option{-ansi}
3405 (equivalent to @option{-std=c89}).
3408 @section Dollar Signs in Identifier Names
3410 @cindex dollar signs in identifier names
3411 @cindex identifier names, dollar signs in
3413 In GNU C, you may normally use dollar signs in identifier names.
3414 This is because many traditional C implementations allow such identifiers.
3415 However, dollar signs in identifiers are not supported on a few target
3416 machines, typically because the target assembler does not allow them.
3418 @node Character Escapes
3419 @section The Character @key{ESC} in Constants
3421 You can use the sequence @samp{\e} in a string or character constant to
3422 stand for the ASCII character @key{ESC}.
3425 @section Inquiring on Alignment of Types or Variables
3427 @cindex type alignment
3428 @cindex variable alignment
3430 The keyword @code{__alignof__} allows you to inquire about how an object
3431 is aligned, or the minimum alignment usually required by a type. Its
3432 syntax is just like @code{sizeof}.
3434 For example, if the target machine requires a @code{double} value to be
3435 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
3436 This is true on many RISC machines. On more traditional machine
3437 designs, @code{__alignof__ (double)} is 4 or even 2.
3439 Some machines never actually require alignment; they allow reference to any
3440 data type even at an odd address. For these machines, @code{__alignof__}
3441 reports the smallest alignment that GCC will give the data type, usually as
3442 mandated by the target ABI.
3444 If the operand of @code{__alignof__} is an lvalue rather than a type,
3445 its value is the required alignment for its type, taking into account
3446 any minimum alignment specified with GCC's @code{__attribute__}
3447 extension (@pxref{Variable Attributes}). For example, after this
3451 struct foo @{ int x; char y; @} foo1;
3455 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
3456 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
3458 It is an error to ask for the alignment of an incomplete type.
3460 @node Variable Attributes
3461 @section Specifying Attributes of Variables
3462 @cindex attribute of variables
3463 @cindex variable attributes
3465 The keyword @code{__attribute__} allows you to specify special
3466 attributes of variables or structure fields. This keyword is followed
3467 by an attribute specification inside double parentheses. Some
3468 attributes are currently defined generically for variables.
3469 Other attributes are defined for variables on particular target
3470 systems. Other attributes are available for functions
3471 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
3472 Other front ends might define more attributes
3473 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
3475 You may also specify attributes with @samp{__} preceding and following
3476 each keyword. This allows you to use them in header files without
3477 being concerned about a possible macro of the same name. For example,
3478 you may use @code{__aligned__} instead of @code{aligned}.
3480 @xref{Attribute Syntax}, for details of the exact syntax for using
3484 @cindex @code{aligned} attribute
3485 @item aligned (@var{alignment})
3486 This attribute specifies a minimum alignment for the variable or
3487 structure field, measured in bytes. For example, the declaration:
3490 int x __attribute__ ((aligned (16))) = 0;
3494 causes the compiler to allocate the global variable @code{x} on a
3495 16-byte boundary. On a 68040, this could be used in conjunction with
3496 an @code{asm} expression to access the @code{move16} instruction which
3497 requires 16-byte aligned operands.
3499 You can also specify the alignment of structure fields. For example, to
3500 create a double-word aligned @code{int} pair, you could write:
3503 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
3507 This is an alternative to creating a union with a @code{double} member
3508 that forces the union to be double-word aligned.
3510 As in the preceding examples, you can explicitly specify the alignment
3511 (in bytes) that you wish the compiler to use for a given variable or
3512 structure field. Alternatively, you can leave out the alignment factor
3513 and just ask the compiler to align a variable or field to the maximum
3514 useful alignment for the target machine you are compiling for. For
3515 example, you could write:
3518 short array[3] __attribute__ ((aligned));
3521 Whenever you leave out the alignment factor in an @code{aligned} attribute
3522 specification, the compiler automatically sets the alignment for the declared
3523 variable or field to the largest alignment which is ever used for any data
3524 type on the target machine you are compiling for. Doing this can often make
3525 copy operations more efficient, because the compiler can use whatever
3526 instructions copy the biggest chunks of memory when performing copies to
3527 or from the variables or fields that you have aligned this way.
3529 When used on a struct, or struct member, the @code{aligned} attribute can
3530 only increase the alignment; in order to decrease it, the @code{packed}
3531 attribute must be specified as well. When used as part of a typedef, the
3532 @code{aligned} attribute can both increase and decrease alignment, and
3533 specifying the @code{packed} attribute will generate a warning.
3535 Note that the effectiveness of @code{aligned} attributes may be limited
3536 by inherent limitations in your linker. On many systems, the linker is
3537 only able to arrange for variables to be aligned up to a certain maximum
3538 alignment. (For some linkers, the maximum supported alignment may
3539 be very very small.) If your linker is only able to align variables
3540 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3541 in an @code{__attribute__} will still only provide you with 8 byte
3542 alignment. See your linker documentation for further information.
3544 The @code{aligned} attribute can also be used for functions
3545 (@pxref{Function Attributes}.)
3547 @item cleanup (@var{cleanup_function})
3548 @cindex @code{cleanup} attribute
3549 The @code{cleanup} attribute runs a function when the variable goes
3550 out of scope. This attribute can only be applied to auto function
3551 scope variables; it may not be applied to parameters or variables
3552 with static storage duration. The function must take one parameter,
3553 a pointer to a type compatible with the variable. The return value
3554 of the function (if any) is ignored.
3556 If @option{-fexceptions} is enabled, then @var{cleanup_function}
3557 will be run during the stack unwinding that happens during the
3558 processing of the exception. Note that the @code{cleanup} attribute
3559 does not allow the exception to be caught, only to perform an action.
3560 It is undefined what happens if @var{cleanup_function} does not
3565 @cindex @code{common} attribute
3566 @cindex @code{nocommon} attribute
3569 The @code{common} attribute requests GCC to place a variable in
3570 ``common'' storage. The @code{nocommon} attribute requests the
3571 opposite---to allocate space for it directly.
3573 These attributes override the default chosen by the
3574 @option{-fno-common} and @option{-fcommon} flags respectively.
3577 @cindex @code{deprecated} attribute
3578 The @code{deprecated} attribute results in a warning if the variable
3579 is used anywhere in the source file. This is useful when identifying
3580 variables that are expected to be removed in a future version of a
3581 program. The warning also includes the location of the declaration
3582 of the deprecated variable, to enable users to easily find further
3583 information about why the variable is deprecated, or what they should
3584 do instead. Note that the warning only occurs for uses:
3587 extern int old_var __attribute__ ((deprecated));
3589 int new_fn () @{ return old_var; @}
3592 results in a warning on line 3 but not line 2.
3594 The @code{deprecated} attribute can also be used for functions and
3595 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
3597 @item mode (@var{mode})
3598 @cindex @code{mode} attribute
3599 This attribute specifies the data type for the declaration---whichever
3600 type corresponds to the mode @var{mode}. This in effect lets you
3601 request an integer or floating point type according to its width.
3603 You may also specify a mode of @samp{byte} or @samp{__byte__} to
3604 indicate the mode corresponding to a one-byte integer, @samp{word} or
3605 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
3606 or @samp{__pointer__} for the mode used to represent pointers.
3609 @cindex @code{packed} attribute
3610 The @code{packed} attribute specifies that a variable or structure field
3611 should have the smallest possible alignment---one byte for a variable,
3612 and one bit for a field, unless you specify a larger value with the
3613 @code{aligned} attribute.
3615 Here is a structure in which the field @code{x} is packed, so that it
3616 immediately follows @code{a}:
3622 int x[2] __attribute__ ((packed));
3626 @item section ("@var{section-name}")
3627 @cindex @code{section} variable attribute
3628 Normally, the compiler places the objects it generates in sections like
3629 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
3630 or you need certain particular variables to appear in special sections,
3631 for example to map to special hardware. The @code{section}
3632 attribute specifies that a variable (or function) lives in a particular
3633 section. For example, this small program uses several specific section names:
3636 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
3637 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
3638 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
3639 int init_data __attribute__ ((section ("INITDATA"))) = 0;
3643 /* @r{Initialize stack pointer} */
3644 init_sp (stack + sizeof (stack));
3646 /* @r{Initialize initialized data} */
3647 memcpy (&init_data, &data, &edata - &data);
3649 /* @r{Turn on the serial ports} */
3656 Use the @code{section} attribute with an @emph{initialized} definition
3657 of a @emph{global} variable, as shown in the example. GCC issues
3658 a warning and otherwise ignores the @code{section} attribute in
3659 uninitialized variable declarations.
3661 You may only use the @code{section} attribute with a fully initialized
3662 global definition because of the way linkers work. The linker requires
3663 each object be defined once, with the exception that uninitialized
3664 variables tentatively go in the @code{common} (or @code{bss}) section
3665 and can be multiply ``defined''. You can force a variable to be
3666 initialized with the @option{-fno-common} flag or the @code{nocommon}
3669 Some file formats do not support arbitrary sections so the @code{section}
3670 attribute is not available on all platforms.
3671 If you need to map the entire contents of a module to a particular
3672 section, consider using the facilities of the linker instead.
3675 @cindex @code{shared} variable attribute
3676 On Microsoft Windows, in addition to putting variable definitions in a named
3677 section, the section can also be shared among all running copies of an
3678 executable or DLL@. For example, this small program defines shared data
3679 by putting it in a named section @code{shared} and marking the section
3683 int foo __attribute__((section ("shared"), shared)) = 0;
3688 /* @r{Read and write foo. All running
3689 copies see the same value.} */
3695 You may only use the @code{shared} attribute along with @code{section}
3696 attribute with a fully initialized global definition because of the way
3697 linkers work. See @code{section} attribute for more information.
3699 The @code{shared} attribute is only available on Microsoft Windows@.
3701 @item tls_model ("@var{tls_model}")
3702 @cindex @code{tls_model} attribute
3703 The @code{tls_model} attribute sets thread-local storage model
3704 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
3705 overriding @option{-ftls-model=} command line switch on a per-variable
3707 The @var{tls_model} argument should be one of @code{global-dynamic},
3708 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
3710 Not all targets support this attribute.
3713 This attribute, attached to a variable, means that the variable is meant
3714 to be possibly unused. GCC will not produce a warning for this
3718 This attribute, attached to a variable, means that the variable must be
3719 emitted even if it appears that the variable is not referenced.
3721 @item vector_size (@var{bytes})
3722 This attribute specifies the vector size for the variable, measured in
3723 bytes. For example, the declaration:
3726 int foo __attribute__ ((vector_size (16)));
3730 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
3731 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
3732 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
3734 This attribute is only applicable to integral and float scalars,
3735 although arrays, pointers, and function return values are allowed in
3736 conjunction with this construct.
3738 Aggregates with this attribute are invalid, even if they are of the same
3739 size as a corresponding scalar. For example, the declaration:
3742 struct S @{ int a; @};
3743 struct S __attribute__ ((vector_size (16))) foo;
3747 is invalid even if the size of the structure is the same as the size of
3751 The @code{selectany} attribute causes an initialized global variable to
3752 have link-once semantics. When multiple definitions of the variable are
3753 encountered by the linker, the first is selected and the remainder are
3754 discarded. Following usage by the Microsoft compiler, the linker is told
3755 @emph{not} to warn about size or content differences of the multiple
3758 Although the primary usage of this attribute is for POD types, the
3759 attribute can also be applied to global C++ objects that are initialized
3760 by a constructor. In this case, the static initialization and destruction
3761 code for the object is emitted in each translation defining the object,
3762 but the calls to the constructor and destructor are protected by a
3763 link-once guard variable.
3765 The @code{selectany} attribute is only available on Microsoft Windows
3766 targets. You can use @code{__declspec (selectany)} as a synonym for
3767 @code{__attribute__ ((selectany))} for compatibility with other
3771 The @code{weak} attribute is described in @xref{Function Attributes}.
3774 The @code{dllimport} attribute is described in @xref{Function Attributes}.
3777 The @code{dllexport} attribute is described in @xref{Function Attributes}.
3781 @subsection Blackfin Variable Attributes
3783 Three attributes are currently defined for the Blackfin.
3789 @cindex @code{l1_data} variable attribute
3790 @cindex @code{l1_data_A} variable attribute
3791 @cindex @code{l1_data_B} variable attribute
3792 Use these attributes on the Blackfin to place the variable into L1 Data SRAM.
3793 Variables with @code{l1_data} attribute will be put into the specific section
3794 named @code{.l1.data}. Those with @code{l1_data_A} attribute will be put into
3795 the specific section named @code{.l1.data.A}. Those with @code{l1_data_B}
3796 attribute will be put into the specific section named @code{.l1.data.B}.
3799 @subsection M32R/D Variable Attributes
3801 One attribute is currently defined for the M32R/D@.
3804 @item model (@var{model-name})
3805 @cindex variable addressability on the M32R/D
3806 Use this attribute on the M32R/D to set the addressability of an object.
3807 The identifier @var{model-name} is one of @code{small}, @code{medium},
3808 or @code{large}, representing each of the code models.
3810 Small model objects live in the lower 16MB of memory (so that their
3811 addresses can be loaded with the @code{ld24} instruction).
3813 Medium and large model objects may live anywhere in the 32-bit address space
3814 (the compiler will generate @code{seth/add3} instructions to load their
3818 @anchor{i386 Variable Attributes}
3819 @subsection i386 Variable Attributes
3821 Two attributes are currently defined for i386 configurations:
3822 @code{ms_struct} and @code{gcc_struct}
3827 @cindex @code{ms_struct} attribute
3828 @cindex @code{gcc_struct} attribute
3830 If @code{packed} is used on a structure, or if bit-fields are used
3831 it may be that the Microsoft ABI packs them differently
3832 than GCC would normally pack them. Particularly when moving packed
3833 data between functions compiled with GCC and the native Microsoft compiler
3834 (either via function call or as data in a file), it may be necessary to access
3837 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
3838 compilers to match the native Microsoft compiler.
3840 The Microsoft structure layout algorithm is fairly simple with the exception
3841 of the bitfield packing:
3843 The padding and alignment of members of structures and whether a bit field
3844 can straddle a storage-unit boundary
3847 @item Structure members are stored sequentially in the order in which they are
3848 declared: the first member has the lowest memory address and the last member
3851 @item Every data object has an alignment-requirement. The alignment-requirement
3852 for all data except structures, unions, and arrays is either the size of the
3853 object or the current packing size (specified with either the aligned attribute
3854 or the pack pragma), whichever is less. For structures, unions, and arrays,
3855 the alignment-requirement is the largest alignment-requirement of its members.
3856 Every object is allocated an offset so that:
3858 offset % alignment-requirement == 0
3860 @item Adjacent bit fields are packed into the same 1-, 2-, or 4-byte allocation
3861 unit if the integral types are the same size and if the next bit field fits
3862 into the current allocation unit without crossing the boundary imposed by the
3863 common alignment requirements of the bit fields.
3866 Handling of zero-length bitfields:
3868 MSVC interprets zero-length bitfields in the following ways:
3871 @item If a zero-length bitfield is inserted between two bitfields that would
3872 normally be coalesced, the bitfields will not be coalesced.
3879 unsigned long bf_1 : 12;
3881 unsigned long bf_2 : 12;
3885 The size of @code{t1} would be 8 bytes with the zero-length bitfield. If the
3886 zero-length bitfield were removed, @code{t1}'s size would be 4 bytes.
3888 @item If a zero-length bitfield is inserted after a bitfield, @code{foo}, and the
3889 alignment of the zero-length bitfield is greater than the member that follows it,
3890 @code{bar}, @code{bar} will be aligned as the type of the zero-length bitfield.
3910 For @code{t2}, @code{bar} will be placed at offset 2, rather than offset 1.
3911 Accordingly, the size of @code{t2} will be 4. For @code{t3}, the zero-length
3912 bitfield will not affect the alignment of @code{bar} or, as a result, the size
3915 Taking this into account, it is important to note the following:
3918 @item If a zero-length bitfield follows a normal bitfield, the type of the
3919 zero-length bitfield may affect the alignment of the structure as whole. For
3920 example, @code{t2} has a size of 4 bytes, since the zero-length bitfield follows a
3921 normal bitfield, and is of type short.
3923 @item Even if a zero-length bitfield is not followed by a normal bitfield, it may
3924 still affect the alignment of the structure:
3934 Here, @code{t4} will take up 4 bytes.
3937 @item Zero-length bitfields following non-bitfield members are ignored:
3948 Here, @code{t5} will take up 2 bytes.
3952 @subsection PowerPC Variable Attributes
3954 Three attributes currently are defined for PowerPC configurations:
3955 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
3957 For full documentation of the struct attributes please see the
3958 documentation in the @xref{i386 Variable Attributes}, section.
3960 For documentation of @code{altivec} attribute please see the
3961 documentation in the @xref{PowerPC Type Attributes}, section.
3963 @subsection SPU Variable Attributes
3965 The SPU supports the @code{spu_vector} attribute for variables. For
3966 documentation of this attribute please see the documentation in the
3967 @xref{SPU Type Attributes}, section.
3969 @subsection Xstormy16 Variable Attributes
3971 One attribute is currently defined for xstormy16 configurations:
3976 @cindex @code{below100} attribute
3978 If a variable has the @code{below100} attribute (@code{BELOW100} is
3979 allowed also), GCC will place the variable in the first 0x100 bytes of
3980 memory and use special opcodes to access it. Such variables will be
3981 placed in either the @code{.bss_below100} section or the
3982 @code{.data_below100} section.
3986 @subsection AVR Variable Attributes
3990 @cindex @code{progmem} variable attribute
3991 The @code{progmem} attribute is used on the AVR to place data in the Program
3992 Memory address space. The AVR is a Harvard Architecture processor and data
3993 normally resides in the Data Memory address space.
3996 @node Type Attributes
3997 @section Specifying Attributes of Types
3998 @cindex attribute of types
3999 @cindex type attributes
4001 The keyword @code{__attribute__} allows you to specify special
4002 attributes of @code{struct} and @code{union} types when you define
4003 such types. This keyword is followed by an attribute specification
4004 inside double parentheses. Seven attributes are currently defined for
4005 types: @code{aligned}, @code{packed}, @code{transparent_union},
4006 @code{unused}, @code{deprecated}, @code{visibility}, and
4007 @code{may_alias}. Other attributes are defined for functions
4008 (@pxref{Function Attributes}) and for variables (@pxref{Variable
4011 You may also specify any one of these attributes with @samp{__}
4012 preceding and following its keyword. This allows you to use these
4013 attributes in header files without being concerned about a possible
4014 macro of the same name. For example, you may use @code{__aligned__}
4015 instead of @code{aligned}.
4017 You may specify type attributes in an enum, struct or union type
4018 declaration or definition, or for other types in a @code{typedef}
4021 For an enum, struct or union type, you may specify attributes either
4022 between the enum, struct or union tag and the name of the type, or
4023 just past the closing curly brace of the @emph{definition}. The
4024 former syntax is preferred.
4026 @xref{Attribute Syntax}, for details of the exact syntax for using
4030 @cindex @code{aligned} attribute
4031 @item aligned (@var{alignment})
4032 This attribute specifies a minimum alignment (in bytes) for variables
4033 of the specified type. For example, the declarations:
4036 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
4037 typedef int more_aligned_int __attribute__ ((aligned (8)));
4041 force the compiler to insure (as far as it can) that each variable whose
4042 type is @code{struct S} or @code{more_aligned_int} will be allocated and
4043 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
4044 variables of type @code{struct S} aligned to 8-byte boundaries allows
4045 the compiler to use the @code{ldd} and @code{std} (doubleword load and
4046 store) instructions when copying one variable of type @code{struct S} to
4047 another, thus improving run-time efficiency.
4049 Note that the alignment of any given @code{struct} or @code{union} type
4050 is required by the ISO C standard to be at least a perfect multiple of
4051 the lowest common multiple of the alignments of all of the members of
4052 the @code{struct} or @code{union} in question. This means that you @emph{can}
4053 effectively adjust the alignment of a @code{struct} or @code{union}
4054 type by attaching an @code{aligned} attribute to any one of the members
4055 of such a type, but the notation illustrated in the example above is a
4056 more obvious, intuitive, and readable way to request the compiler to
4057 adjust the alignment of an entire @code{struct} or @code{union} type.
4059 As in the preceding example, you can explicitly specify the alignment
4060 (in bytes) that you wish the compiler to use for a given @code{struct}
4061 or @code{union} type. Alternatively, you can leave out the alignment factor
4062 and just ask the compiler to align a type to the maximum
4063 useful alignment for the target machine you are compiling for. For
4064 example, you could write:
4067 struct S @{ short f[3]; @} __attribute__ ((aligned));
4070 Whenever you leave out the alignment factor in an @code{aligned}
4071 attribute specification, the compiler automatically sets the alignment
4072 for the type to the largest alignment which is ever used for any data
4073 type on the target machine you are compiling for. Doing this can often
4074 make copy operations more efficient, because the compiler can use
4075 whatever instructions copy the biggest chunks of memory when performing
4076 copies to or from the variables which have types that you have aligned
4079 In the example above, if the size of each @code{short} is 2 bytes, then
4080 the size of the entire @code{struct S} type is 6 bytes. The smallest
4081 power of two which is greater than or equal to that is 8, so the
4082 compiler sets the alignment for the entire @code{struct S} type to 8
4085 Note that although you can ask the compiler to select a time-efficient
4086 alignment for a given type and then declare only individual stand-alone
4087 objects of that type, the compiler's ability to select a time-efficient
4088 alignment is primarily useful only when you plan to create arrays of
4089 variables having the relevant (efficiently aligned) type. If you
4090 declare or use arrays of variables of an efficiently-aligned type, then
4091 it is likely that your program will also be doing pointer arithmetic (or
4092 subscripting, which amounts to the same thing) on pointers to the
4093 relevant type, and the code that the compiler generates for these
4094 pointer arithmetic operations will often be more efficient for
4095 efficiently-aligned types than for other types.
4097 The @code{aligned} attribute can only increase the alignment; but you
4098 can decrease it by specifying @code{packed} as well. See below.
4100 Note that the effectiveness of @code{aligned} attributes may be limited
4101 by inherent limitations in your linker. On many systems, the linker is
4102 only able to arrange for variables to be aligned up to a certain maximum
4103 alignment. (For some linkers, the maximum supported alignment may
4104 be very very small.) If your linker is only able to align variables
4105 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
4106 in an @code{__attribute__} will still only provide you with 8 byte
4107 alignment. See your linker documentation for further information.
4110 This attribute, attached to @code{struct} or @code{union} type
4111 definition, specifies that each member (other than zero-width bitfields)
4112 of the structure or union is placed to minimize the memory required. When
4113 attached to an @code{enum} definition, it indicates that the smallest
4114 integral type should be used.
4116 @opindex fshort-enums
4117 Specifying this attribute for @code{struct} and @code{union} types is
4118 equivalent to specifying the @code{packed} attribute on each of the
4119 structure or union members. Specifying the @option{-fshort-enums}
4120 flag on the line is equivalent to specifying the @code{packed}
4121 attribute on all @code{enum} definitions.
4123 In the following example @code{struct my_packed_struct}'s members are
4124 packed closely together, but the internal layout of its @code{s} member
4125 is not packed---to do that, @code{struct my_unpacked_struct} would need to
4129 struct my_unpacked_struct
4135 struct __attribute__ ((__packed__)) my_packed_struct
4139 struct my_unpacked_struct s;
4143 You may only specify this attribute on the definition of a @code{enum},
4144 @code{struct} or @code{union}, not on a @code{typedef} which does not
4145 also define the enumerated type, structure or union.
4147 @item transparent_union
4148 This attribute, attached to a @code{union} type definition, indicates
4149 that any function parameter having that union type causes calls to that
4150 function to be treated in a special way.
4152 First, the argument corresponding to a transparent union type can be of
4153 any type in the union; no cast is required. Also, if the union contains
4154 a pointer type, the corresponding argument can be a null pointer
4155 constant or a void pointer expression; and if the union contains a void
4156 pointer type, the corresponding argument can be any pointer expression.
4157 If the union member type is a pointer, qualifiers like @code{const} on
4158 the referenced type must be respected, just as with normal pointer
4161 Second, the argument is passed to the function using the calling
4162 conventions of the first member of the transparent union, not the calling
4163 conventions of the union itself. All members of the union must have the
4164 same machine representation; this is necessary for this argument passing
4167 Transparent unions are designed for library functions that have multiple
4168 interfaces for compatibility reasons. For example, suppose the
4169 @code{wait} function must accept either a value of type @code{int *} to
4170 comply with Posix, or a value of type @code{union wait *} to comply with
4171 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
4172 @code{wait} would accept both kinds of arguments, but it would also
4173 accept any other pointer type and this would make argument type checking
4174 less useful. Instead, @code{<sys/wait.h>} might define the interface
4178 typedef union __attribute__ ((__transparent_union__))
4182 @} wait_status_ptr_t;
4184 pid_t wait (wait_status_ptr_t);
4187 This interface allows either @code{int *} or @code{union wait *}
4188 arguments to be passed, using the @code{int *} calling convention.
4189 The program can call @code{wait} with arguments of either type:
4192 int w1 () @{ int w; return wait (&w); @}
4193 int w2 () @{ union wait w; return wait (&w); @}
4196 With this interface, @code{wait}'s implementation might look like this:
4199 pid_t wait (wait_status_ptr_t p)
4201 return waitpid (-1, p.__ip, 0);
4206 When attached to a type (including a @code{union} or a @code{struct}),
4207 this attribute means that variables of that type are meant to appear
4208 possibly unused. GCC will not produce a warning for any variables of
4209 that type, even if the variable appears to do nothing. This is often
4210 the case with lock or thread classes, which are usually defined and then
4211 not referenced, but contain constructors and destructors that have
4212 nontrivial bookkeeping functions.
4215 The @code{deprecated} attribute results in a warning if the type
4216 is used anywhere in the source file. This is useful when identifying
4217 types that are expected to be removed in a future version of a program.
4218 If possible, the warning also includes the location of the declaration
4219 of the deprecated type, to enable users to easily find further
4220 information about why the type is deprecated, or what they should do
4221 instead. Note that the warnings only occur for uses and then only
4222 if the type is being applied to an identifier that itself is not being
4223 declared as deprecated.
4226 typedef int T1 __attribute__ ((deprecated));
4230 typedef T1 T3 __attribute__ ((deprecated));
4231 T3 z __attribute__ ((deprecated));
4234 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
4235 warning is issued for line 4 because T2 is not explicitly
4236 deprecated. Line 5 has no warning because T3 is explicitly
4237 deprecated. Similarly for line 6.
4239 The @code{deprecated} attribute can also be used for functions and
4240 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
4243 Accesses to objects with types with this attribute are not subjected to
4244 type-based alias analysis, but are instead assumed to be able to alias
4245 any other type of objects, just like the @code{char} type. See
4246 @option{-fstrict-aliasing} for more information on aliasing issues.
4251 typedef short __attribute__((__may_alias__)) short_a;
4257 short_a *b = (short_a *) &a;
4261 if (a == 0x12345678)
4268 If you replaced @code{short_a} with @code{short} in the variable
4269 declaration, the above program would abort when compiled with
4270 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
4271 above in recent GCC versions.
4274 In C++, attribute visibility (@pxref{Function Attributes}) can also be
4275 applied to class, struct, union and enum types. Unlike other type
4276 attributes, the attribute must appear between the initial keyword and
4277 the name of the type; it cannot appear after the body of the type.
4279 Note that the type visibility is applied to vague linkage entities
4280 associated with the class (vtable, typeinfo node, etc.). In
4281 particular, if a class is thrown as an exception in one shared object
4282 and caught in another, the class must have default visibility.
4283 Otherwise the two shared objects will be unable to use the same
4284 typeinfo node and exception handling will break.
4286 @subsection ARM Type Attributes
4288 On those ARM targets that support @code{dllimport} (such as Symbian
4289 OS), you can use the @code{notshared} attribute to indicate that the
4290 virtual table and other similar data for a class should not be
4291 exported from a DLL@. For example:
4294 class __declspec(notshared) C @{
4296 __declspec(dllimport) C();
4300 __declspec(dllexport)
4304 In this code, @code{C::C} is exported from the current DLL, but the
4305 virtual table for @code{C} is not exported. (You can use
4306 @code{__attribute__} instead of @code{__declspec} if you prefer, but
4307 most Symbian OS code uses @code{__declspec}.)
4309 @anchor{i386 Type Attributes}
4310 @subsection i386 Type Attributes
4312 Two attributes are currently defined for i386 configurations:
4313 @code{ms_struct} and @code{gcc_struct}
4317 @cindex @code{ms_struct}
4318 @cindex @code{gcc_struct}
4320 If @code{packed} is used on a structure, or if bit-fields are used
4321 it may be that the Microsoft ABI packs them differently
4322 than GCC would normally pack them. Particularly when moving packed
4323 data between functions compiled with GCC and the native Microsoft compiler
4324 (either via function call or as data in a file), it may be necessary to access
4327 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
4328 compilers to match the native Microsoft compiler.
4331 To specify multiple attributes, separate them by commas within the
4332 double parentheses: for example, @samp{__attribute__ ((aligned (16),
4335 @anchor{PowerPC Type Attributes}
4336 @subsection PowerPC Type Attributes
4338 Three attributes currently are defined for PowerPC configurations:
4339 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
4341 For full documentation of the struct attributes please see the
4342 documentation in the @xref{i386 Type Attributes}, section.
4344 The @code{altivec} attribute allows one to declare AltiVec vector data
4345 types supported by the AltiVec Programming Interface Manual. The
4346 attribute requires an argument to specify one of three vector types:
4347 @code{vector__}, @code{pixel__} (always followed by unsigned short),
4348 and @code{bool__} (always followed by unsigned).
4351 __attribute__((altivec(vector__)))
4352 __attribute__((altivec(pixel__))) unsigned short
4353 __attribute__((altivec(bool__))) unsigned
4356 These attributes mainly are intended to support the @code{__vector},
4357 @code{__pixel}, and @code{__bool} AltiVec keywords.
4359 @anchor{SPU Type Attributes}
4360 @subsection SPU Type Attributes
4362 The SPU supports the @code{spu_vector} attribute for types. This attribute
4363 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
4364 Language Extensions Specification. It is intended to support the
4365 @code{__vector} keyword.
4369 @section An Inline Function is As Fast As a Macro
4370 @cindex inline functions
4371 @cindex integrating function code
4373 @cindex macros, inline alternative
4375 By declaring a function inline, you can direct GCC to make
4376 calls to that function faster. One way GCC can achieve this is to
4377 integrate that function's code into the code for its callers. This
4378 makes execution faster by eliminating the function-call overhead; in
4379 addition, if any of the actual argument values are constant, their
4380 known values may permit simplifications at compile time so that not
4381 all of the inline function's code needs to be included. The effect on
4382 code size is less predictable; object code may be larger or smaller
4383 with function inlining, depending on the particular case. You can
4384 also direct GCC to try to integrate all ``simple enough'' functions
4385 into their callers with the option @option{-finline-functions}.
4387 GCC implements three different semantics of declaring a function
4388 inline. One is available with @option{-std=gnu89} or
4389 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
4390 on all inline declarations, another when @option{-std=c99} or
4391 @option{-std=gnu99} (without @option{-fgnu89-inline}), and the third
4392 is used when compiling C++.
4394 To declare a function inline, use the @code{inline} keyword in its
4395 declaration, like this:
4405 If you are writing a header file to be included in ISO C89 programs, write
4406 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
4408 The three types of inlining behave similarly in two important cases:
4409 when the @code{inline} keyword is used on a @code{static} function,
4410 like the example above, and when a function is first declared without
4411 using the @code{inline} keyword and then is defined with
4412 @code{inline}, like this:
4415 extern int inc (int *a);
4423 In both of these common cases, the program behaves the same as if you
4424 had not used the @code{inline} keyword, except for its speed.
4426 @cindex inline functions, omission of
4427 @opindex fkeep-inline-functions
4428 When a function is both inline and @code{static}, if all calls to the
4429 function are integrated into the caller, and the function's address is
4430 never used, then the function's own assembler code is never referenced.
4431 In this case, GCC does not actually output assembler code for the
4432 function, unless you specify the option @option{-fkeep-inline-functions}.
4433 Some calls cannot be integrated for various reasons (in particular,
4434 calls that precede the function's definition cannot be integrated, and
4435 neither can recursive calls within the definition). If there is a
4436 nonintegrated call, then the function is compiled to assembler code as
4437 usual. The function must also be compiled as usual if the program
4438 refers to its address, because that can't be inlined.
4441 Note that certain usages in a function definition can make it unsuitable
4442 for inline substitution. Among these usages are: use of varargs, use of
4443 alloca, use of variable sized data types (@pxref{Variable Length}),
4444 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
4445 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
4446 will warn when a function marked @code{inline} could not be substituted,
4447 and will give the reason for the failure.
4449 @cindex automatic @code{inline} for C++ member fns
4450 @cindex @code{inline} automatic for C++ member fns
4451 @cindex member fns, automatically @code{inline}
4452 @cindex C++ member fns, automatically @code{inline}
4453 @opindex fno-default-inline
4454 As required by ISO C++, GCC considers member functions defined within
4455 the body of a class to be marked inline even if they are
4456 not explicitly declared with the @code{inline} keyword. You can
4457 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
4458 Options,,Options Controlling C++ Dialect}.
4460 GCC does not inline any functions when not optimizing unless you specify
4461 the @samp{always_inline} attribute for the function, like this:
4464 /* @r{Prototype.} */
4465 inline void foo (const char) __attribute__((always_inline));
4468 The remainder of this section is specific to GNU C89 inlining.
4470 @cindex non-static inline function
4471 When an inline function is not @code{static}, then the compiler must assume
4472 that there may be calls from other source files; since a global symbol can
4473 be defined only once in any program, the function must not be defined in
4474 the other source files, so the calls therein cannot be integrated.
4475 Therefore, a non-@code{static} inline function is always compiled on its
4476 own in the usual fashion.
4478 If you specify both @code{inline} and @code{extern} in the function
4479 definition, then the definition is used only for inlining. In no case
4480 is the function compiled on its own, not even if you refer to its
4481 address explicitly. Such an address becomes an external reference, as
4482 if you had only declared the function, and had not defined it.
4484 This combination of @code{inline} and @code{extern} has almost the
4485 effect of a macro. The way to use it is to put a function definition in
4486 a header file with these keywords, and put another copy of the
4487 definition (lacking @code{inline} and @code{extern}) in a library file.
4488 The definition in the header file will cause most calls to the function
4489 to be inlined. If any uses of the function remain, they will refer to
4490 the single copy in the library.
4493 @section Assembler Instructions with C Expression Operands
4494 @cindex extended @code{asm}
4495 @cindex @code{asm} expressions
4496 @cindex assembler instructions
4499 In an assembler instruction using @code{asm}, you can specify the
4500 operands of the instruction using C expressions. This means you need not
4501 guess which registers or memory locations will contain the data you want
4504 You must specify an assembler instruction template much like what
4505 appears in a machine description, plus an operand constraint string for
4508 For example, here is how to use the 68881's @code{fsinx} instruction:
4511 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
4515 Here @code{angle} is the C expression for the input operand while
4516 @code{result} is that of the output operand. Each has @samp{"f"} as its
4517 operand constraint, saying that a floating point register is required.
4518 The @samp{=} in @samp{=f} indicates that the operand is an output; all
4519 output operands' constraints must use @samp{=}. The constraints use the
4520 same language used in the machine description (@pxref{Constraints}).
4522 Each operand is described by an operand-constraint string followed by
4523 the C expression in parentheses. A colon separates the assembler
4524 template from the first output operand and another separates the last
4525 output operand from the first input, if any. Commas separate the
4526 operands within each group. The total number of operands is currently
4527 limited to 30; this limitation may be lifted in some future version of
4530 If there are no output operands but there are input operands, you must
4531 place two consecutive colons surrounding the place where the output
4534 As of GCC version 3.1, it is also possible to specify input and output
4535 operands using symbolic names which can be referenced within the
4536 assembler code. These names are specified inside square brackets
4537 preceding the constraint string, and can be referenced inside the
4538 assembler code using @code{%[@var{name}]} instead of a percentage sign
4539 followed by the operand number. Using named operands the above example
4543 asm ("fsinx %[angle],%[output]"
4544 : [output] "=f" (result)
4545 : [angle] "f" (angle));
4549 Note that the symbolic operand names have no relation whatsoever to
4550 other C identifiers. You may use any name you like, even those of
4551 existing C symbols, but you must ensure that no two operands within the same
4552 assembler construct use the same symbolic name.
4554 Output operand expressions must be lvalues; the compiler can check this.
4555 The input operands need not be lvalues. The compiler cannot check
4556 whether the operands have data types that are reasonable for the
4557 instruction being executed. It does not parse the assembler instruction
4558 template and does not know what it means or even whether it is valid
4559 assembler input. The extended @code{asm} feature is most often used for
4560 machine instructions the compiler itself does not know exist. If
4561 the output expression cannot be directly addressed (for example, it is a
4562 bit-field), your constraint must allow a register. In that case, GCC
4563 will use the register as the output of the @code{asm}, and then store
4564 that register into the output.
4566 The ordinary output operands must be write-only; GCC will assume that
4567 the values in these operands before the instruction are dead and need
4568 not be generated. Extended asm supports input-output or read-write
4569 operands. Use the constraint character @samp{+} to indicate such an
4570 operand and list it with the output operands. You should only use
4571 read-write operands when the constraints for the operand (or the
4572 operand in which only some of the bits are to be changed) allow a
4575 You may, as an alternative, logically split its function into two
4576 separate operands, one input operand and one write-only output
4577 operand. The connection between them is expressed by constraints
4578 which say they need to be in the same location when the instruction
4579 executes. You can use the same C expression for both operands, or
4580 different expressions. For example, here we write the (fictitious)
4581 @samp{combine} instruction with @code{bar} as its read-only source
4582 operand and @code{foo} as its read-write destination:
4585 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
4589 The constraint @samp{"0"} for operand 1 says that it must occupy the
4590 same location as operand 0. A number in constraint is allowed only in
4591 an input operand and it must refer to an output operand.
4593 Only a number in the constraint can guarantee that one operand will be in
4594 the same place as another. The mere fact that @code{foo} is the value
4595 of both operands is not enough to guarantee that they will be in the
4596 same place in the generated assembler code. The following would not
4600 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
4603 Various optimizations or reloading could cause operands 0 and 1 to be in
4604 different registers; GCC knows no reason not to do so. For example, the
4605 compiler might find a copy of the value of @code{foo} in one register and
4606 use it for operand 1, but generate the output operand 0 in a different
4607 register (copying it afterward to @code{foo}'s own address). Of course,
4608 since the register for operand 1 is not even mentioned in the assembler
4609 code, the result will not work, but GCC can't tell that.
4611 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
4612 the operand number for a matching constraint. For example:
4615 asm ("cmoveq %1,%2,%[result]"
4616 : [result] "=r"(result)
4617 : "r" (test), "r"(new), "[result]"(old));
4620 Sometimes you need to make an @code{asm} operand be a specific register,
4621 but there's no matching constraint letter for that register @emph{by
4622 itself}. To force the operand into that register, use a local variable
4623 for the operand and specify the register in the variable declaration.
4624 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
4625 register constraint letter that matches the register:
4628 register int *p1 asm ("r0") = @dots{};
4629 register int *p2 asm ("r1") = @dots{};
4630 register int *result asm ("r0");
4631 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
4634 @anchor{Example of asm with clobbered asm reg}
4635 In the above example, beware that a register that is call-clobbered by
4636 the target ABI will be overwritten by any function call in the
4637 assignment, including library calls for arithmetic operators.
4638 Assuming it is a call-clobbered register, this may happen to @code{r0}
4639 above by the assignment to @code{p2}. If you have to use such a
4640 register, use temporary variables for expressions between the register
4645 register int *p1 asm ("r0") = @dots{};
4646 register int *p2 asm ("r1") = t1;
4647 register int *result asm ("r0");
4648 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
4651 Some instructions clobber specific hard registers. To describe this,
4652 write a third colon after the input operands, followed by the names of
4653 the clobbered hard registers (given as strings). Here is a realistic
4654 example for the VAX:
4657 asm volatile ("movc3 %0,%1,%2"
4658 : /* @r{no outputs} */
4659 : "g" (from), "g" (to), "g" (count)
4660 : "r0", "r1", "r2", "r3", "r4", "r5");
4663 You may not write a clobber description in a way that overlaps with an
4664 input or output operand. For example, you may not have an operand
4665 describing a register class with one member if you mention that register
4666 in the clobber list. Variables declared to live in specific registers
4667 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
4668 have no part mentioned in the clobber description.
4669 There is no way for you to specify that an input
4670 operand is modified without also specifying it as an output
4671 operand. Note that if all the output operands you specify are for this
4672 purpose (and hence unused), you will then also need to specify
4673 @code{volatile} for the @code{asm} construct, as described below, to
4674 prevent GCC from deleting the @code{asm} statement as unused.
4676 If you refer to a particular hardware register from the assembler code,
4677 you will probably have to list the register after the third colon to
4678 tell the compiler the register's value is modified. In some assemblers,
4679 the register names begin with @samp{%}; to produce one @samp{%} in the
4680 assembler code, you must write @samp{%%} in the input.
4682 If your assembler instruction can alter the condition code register, add
4683 @samp{cc} to the list of clobbered registers. GCC on some machines
4684 represents the condition codes as a specific hardware register;
4685 @samp{cc} serves to name this register. On other machines, the
4686 condition code is handled differently, and specifying @samp{cc} has no
4687 effect. But it is valid no matter what the machine.
4689 If your assembler instructions access memory in an unpredictable
4690 fashion, add @samp{memory} to the list of clobbered registers. This
4691 will cause GCC to not keep memory values cached in registers across the
4692 assembler instruction and not optimize stores or loads to that memory.
4693 You will also want to add the @code{volatile} keyword if the memory
4694 affected is not listed in the inputs or outputs of the @code{asm}, as
4695 the @samp{memory} clobber does not count as a side-effect of the
4696 @code{asm}. If you know how large the accessed memory is, you can add
4697 it as input or output but if this is not known, you should add
4698 @samp{memory}. As an example, if you access ten bytes of a string, you
4699 can use a memory input like:
4702 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
4705 Note that in the following example the memory input is necessary,
4706 otherwise GCC might optimize the store to @code{x} away:
4713 asm ("magic stuff accessing an 'int' pointed to by '%1'"
4714 "=&d" (r) : "a" (y), "m" (*y));
4719 You can put multiple assembler instructions together in a single
4720 @code{asm} template, separated by the characters normally used in assembly
4721 code for the system. A combination that works in most places is a newline
4722 to break the line, plus a tab character to move to the instruction field
4723 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
4724 assembler allows semicolons as a line-breaking character. Note that some
4725 assembler dialects use semicolons to start a comment.
4726 The input operands are guaranteed not to use any of the clobbered
4727 registers, and neither will the output operands' addresses, so you can
4728 read and write the clobbered registers as many times as you like. Here
4729 is an example of multiple instructions in a template; it assumes the
4730 subroutine @code{_foo} accepts arguments in registers 9 and 10:
4733 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
4735 : "g" (from), "g" (to)
4739 Unless an output operand has the @samp{&} constraint modifier, GCC
4740 may allocate it in the same register as an unrelated input operand, on
4741 the assumption the inputs are consumed before the outputs are produced.
4742 This assumption may be false if the assembler code actually consists of
4743 more than one instruction. In such a case, use @samp{&} for each output
4744 operand that may not overlap an input. @xref{Modifiers}.
4746 If you want to test the condition code produced by an assembler
4747 instruction, you must include a branch and a label in the @code{asm}
4748 construct, as follows:
4751 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
4757 This assumes your assembler supports local labels, as the GNU assembler
4758 and most Unix assemblers do.
4760 Speaking of labels, jumps from one @code{asm} to another are not
4761 supported. The compiler's optimizers do not know about these jumps, and
4762 therefore they cannot take account of them when deciding how to
4765 @cindex macros containing @code{asm}
4766 Usually the most convenient way to use these @code{asm} instructions is to
4767 encapsulate them in macros that look like functions. For example,
4771 (@{ double __value, __arg = (x); \
4772 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
4777 Here the variable @code{__arg} is used to make sure that the instruction
4778 operates on a proper @code{double} value, and to accept only those
4779 arguments @code{x} which can convert automatically to a @code{double}.
4781 Another way to make sure the instruction operates on the correct data
4782 type is to use a cast in the @code{asm}. This is different from using a
4783 variable @code{__arg} in that it converts more different types. For
4784 example, if the desired type were @code{int}, casting the argument to
4785 @code{int} would accept a pointer with no complaint, while assigning the
4786 argument to an @code{int} variable named @code{__arg} would warn about
4787 using a pointer unless the caller explicitly casts it.
4789 If an @code{asm} has output operands, GCC assumes for optimization
4790 purposes the instruction has no side effects except to change the output
4791 operands. This does not mean instructions with a side effect cannot be
4792 used, but you must be careful, because the compiler may eliminate them
4793 if the output operands aren't used, or move them out of loops, or
4794 replace two with one if they constitute a common subexpression. Also,
4795 if your instruction does have a side effect on a variable that otherwise
4796 appears not to change, the old value of the variable may be reused later
4797 if it happens to be found in a register.
4799 You can prevent an @code{asm} instruction from being deleted
4800 by writing the keyword @code{volatile} after
4801 the @code{asm}. For example:
4804 #define get_and_set_priority(new) \
4806 asm volatile ("get_and_set_priority %0, %1" \
4807 : "=g" (__old) : "g" (new)); \
4812 The @code{volatile} keyword indicates that the instruction has
4813 important side-effects. GCC will not delete a volatile @code{asm} if
4814 it is reachable. (The instruction can still be deleted if GCC can
4815 prove that control-flow will never reach the location of the
4816 instruction.) Note that even a volatile @code{asm} instruction
4817 can be moved relative to other code, including across jump
4818 instructions. For example, on many targets there is a system
4819 register which can be set to control the rounding mode of
4820 floating point operations. You might try
4821 setting it with a volatile @code{asm}, like this PowerPC example:
4824 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
4829 This will not work reliably, as the compiler may move the addition back
4830 before the volatile @code{asm}. To make it work you need to add an
4831 artificial dependency to the @code{asm} referencing a variable in the code
4832 you don't want moved, for example:
4835 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
4839 Similarly, you can't expect a
4840 sequence of volatile @code{asm} instructions to remain perfectly
4841 consecutive. If you want consecutive output, use a single @code{asm}.
4842 Also, GCC will perform some optimizations across a volatile @code{asm}
4843 instruction; GCC does not ``forget everything'' when it encounters
4844 a volatile @code{asm} instruction the way some other compilers do.
4846 An @code{asm} instruction without any output operands will be treated
4847 identically to a volatile @code{asm} instruction.
4849 It is a natural idea to look for a way to give access to the condition
4850 code left by the assembler instruction. However, when we attempted to
4851 implement this, we found no way to make it work reliably. The problem
4852 is that output operands might need reloading, which would result in
4853 additional following ``store'' instructions. On most machines, these
4854 instructions would alter the condition code before there was time to
4855 test it. This problem doesn't arise for ordinary ``test'' and
4856 ``compare'' instructions because they don't have any output operands.
4858 For reasons similar to those described above, it is not possible to give
4859 an assembler instruction access to the condition code left by previous
4862 If you are writing a header file that should be includable in ISO C
4863 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
4866 @subsection Size of an @code{asm}
4868 Some targets require that GCC track the size of each instruction used in
4869 order to generate correct code. Because the final length of an
4870 @code{asm} is only known by the assembler, GCC must make an estimate as
4871 to how big it will be. The estimate is formed by counting the number of
4872 statements in the pattern of the @code{asm} and multiplying that by the
4873 length of the longest instruction on that processor. Statements in the
4874 @code{asm} are identified by newline characters and whatever statement
4875 separator characters are supported by the assembler; on most processors
4876 this is the `@code{;}' character.
4878 Normally, GCC's estimate is perfectly adequate to ensure that correct
4879 code is generated, but it is possible to confuse the compiler if you use
4880 pseudo instructions or assembler macros that expand into multiple real
4881 instructions or if you use assembler directives that expand to more
4882 space in the object file than would be needed for a single instruction.
4883 If this happens then the assembler will produce a diagnostic saying that
4884 a label is unreachable.
4886 @subsection i386 floating point asm operands
4888 There are several rules on the usage of stack-like regs in
4889 asm_operands insns. These rules apply only to the operands that are
4894 Given a set of input regs that die in an asm_operands, it is
4895 necessary to know which are implicitly popped by the asm, and
4896 which must be explicitly popped by gcc.
4898 An input reg that is implicitly popped by the asm must be
4899 explicitly clobbered, unless it is constrained to match an
4903 For any input reg that is implicitly popped by an asm, it is
4904 necessary to know how to adjust the stack to compensate for the pop.
4905 If any non-popped input is closer to the top of the reg-stack than
4906 the implicitly popped reg, it would not be possible to know what the
4907 stack looked like---it's not clear how the rest of the stack ``slides
4910 All implicitly popped input regs must be closer to the top of
4911 the reg-stack than any input that is not implicitly popped.
4913 It is possible that if an input dies in an insn, reload might
4914 use the input reg for an output reload. Consider this example:
4917 asm ("foo" : "=t" (a) : "f" (b));
4920 This asm says that input B is not popped by the asm, and that
4921 the asm pushes a result onto the reg-stack, i.e., the stack is one
4922 deeper after the asm than it was before. But, it is possible that
4923 reload will think that it can use the same reg for both the input and
4924 the output, if input B dies in this insn.
4926 If any input operand uses the @code{f} constraint, all output reg
4927 constraints must use the @code{&} earlyclobber.
4929 The asm above would be written as
4932 asm ("foo" : "=&t" (a) : "f" (b));
4936 Some operands need to be in particular places on the stack. All
4937 output operands fall in this category---there is no other way to
4938 know which regs the outputs appear in unless the user indicates
4939 this in the constraints.
4941 Output operands must specifically indicate which reg an output
4942 appears in after an asm. @code{=f} is not allowed: the operand
4943 constraints must select a class with a single reg.
4946 Output operands may not be ``inserted'' between existing stack regs.
4947 Since no 387 opcode uses a read/write operand, all output operands
4948 are dead before the asm_operands, and are pushed by the asm_operands.
4949 It makes no sense to push anywhere but the top of the reg-stack.
4951 Output operands must start at the top of the reg-stack: output
4952 operands may not ``skip'' a reg.
4955 Some asm statements may need extra stack space for internal
4956 calculations. This can be guaranteed by clobbering stack registers
4957 unrelated to the inputs and outputs.
4961 Here are a couple of reasonable asms to want to write. This asm
4962 takes one input, which is internally popped, and produces two outputs.
4965 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
4968 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
4969 and replaces them with one output. The user must code the @code{st(1)}
4970 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
4973 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
4979 @section Controlling Names Used in Assembler Code
4980 @cindex assembler names for identifiers
4981 @cindex names used in assembler code
4982 @cindex identifiers, names in assembler code
4984 You can specify the name to be used in the assembler code for a C
4985 function or variable by writing the @code{asm} (or @code{__asm__})
4986 keyword after the declarator as follows:
4989 int foo asm ("myfoo") = 2;
4993 This specifies that the name to be used for the variable @code{foo} in
4994 the assembler code should be @samp{myfoo} rather than the usual
4997 On systems where an underscore is normally prepended to the name of a C
4998 function or variable, this feature allows you to define names for the
4999 linker that do not start with an underscore.
5001 It does not make sense to use this feature with a non-static local
5002 variable since such variables do not have assembler names. If you are
5003 trying to put the variable in a particular register, see @ref{Explicit
5004 Reg Vars}. GCC presently accepts such code with a warning, but will
5005 probably be changed to issue an error, rather than a warning, in the
5008 You cannot use @code{asm} in this way in a function @emph{definition}; but
5009 you can get the same effect by writing a declaration for the function
5010 before its definition and putting @code{asm} there, like this:
5013 extern func () asm ("FUNC");
5020 It is up to you to make sure that the assembler names you choose do not
5021 conflict with any other assembler symbols. Also, you must not use a
5022 register name; that would produce completely invalid assembler code. GCC
5023 does not as yet have the ability to store static variables in registers.
5024 Perhaps that will be added.
5026 @node Explicit Reg Vars
5027 @section Variables in Specified Registers
5028 @cindex explicit register variables
5029 @cindex variables in specified registers
5030 @cindex specified registers
5031 @cindex registers, global allocation
5033 GNU C allows you to put a few global variables into specified hardware
5034 registers. You can also specify the register in which an ordinary
5035 register variable should be allocated.
5039 Global register variables reserve registers throughout the program.
5040 This may be useful in programs such as programming language
5041 interpreters which have a couple of global variables that are accessed
5045 Local register variables in specific registers do not reserve the
5046 registers, except at the point where they are used as input or output
5047 operands in an @code{asm} statement and the @code{asm} statement itself is
5048 not deleted. The compiler's data flow analysis is capable of determining
5049 where the specified registers contain live values, and where they are
5050 available for other uses. Stores into local register variables may be deleted
5051 when they appear to be dead according to dataflow analysis. References
5052 to local register variables may be deleted or moved or simplified.
5054 These local variables are sometimes convenient for use with the extended
5055 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
5056 output of the assembler instruction directly into a particular register.
5057 (This will work provided the register you specify fits the constraints
5058 specified for that operand in the @code{asm}.)
5066 @node Global Reg Vars
5067 @subsection Defining Global Register Variables
5068 @cindex global register variables
5069 @cindex registers, global variables in
5071 You can define a global register variable in GNU C like this:
5074 register int *foo asm ("a5");
5078 Here @code{a5} is the name of the register which should be used. Choose a
5079 register which is normally saved and restored by function calls on your
5080 machine, so that library routines will not clobber it.
5082 Naturally the register name is cpu-dependent, so you would need to
5083 conditionalize your program according to cpu type. The register
5084 @code{a5} would be a good choice on a 68000 for a variable of pointer
5085 type. On machines with register windows, be sure to choose a ``global''
5086 register that is not affected magically by the function call mechanism.
5088 In addition, operating systems on one type of cpu may differ in how they
5089 name the registers; then you would need additional conditionals. For
5090 example, some 68000 operating systems call this register @code{%a5}.
5092 Eventually there may be a way of asking the compiler to choose a register
5093 automatically, but first we need to figure out how it should choose and
5094 how to enable you to guide the choice. No solution is evident.
5096 Defining a global register variable in a certain register reserves that
5097 register entirely for this use, at least within the current compilation.
5098 The register will not be allocated for any other purpose in the functions
5099 in the current compilation. The register will not be saved and restored by
5100 these functions. Stores into this register are never deleted even if they
5101 would appear to be dead, but references may be deleted or moved or
5104 It is not safe to access the global register variables from signal
5105 handlers, or from more than one thread of control, because the system
5106 library routines may temporarily use the register for other things (unless
5107 you recompile them specially for the task at hand).
5109 @cindex @code{qsort}, and global register variables
5110 It is not safe for one function that uses a global register variable to
5111 call another such function @code{foo} by way of a third function
5112 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
5113 different source file in which the variable wasn't declared). This is
5114 because @code{lose} might save the register and put some other value there.
5115 For example, you can't expect a global register variable to be available in
5116 the comparison-function that you pass to @code{qsort}, since @code{qsort}
5117 might have put something else in that register. (If you are prepared to
5118 recompile @code{qsort} with the same global register variable, you can
5119 solve this problem.)
5121 If you want to recompile @code{qsort} or other source files which do not
5122 actually use your global register variable, so that they will not use that
5123 register for any other purpose, then it suffices to specify the compiler
5124 option @option{-ffixed-@var{reg}}. You need not actually add a global
5125 register declaration to their source code.
5127 A function which can alter the value of a global register variable cannot
5128 safely be called from a function compiled without this variable, because it
5129 could clobber the value the caller expects to find there on return.
5130 Therefore, the function which is the entry point into the part of the
5131 program that uses the global register variable must explicitly save and
5132 restore the value which belongs to its caller.
5134 @cindex register variable after @code{longjmp}
5135 @cindex global register after @code{longjmp}
5136 @cindex value after @code{longjmp}
5139 On most machines, @code{longjmp} will restore to each global register
5140 variable the value it had at the time of the @code{setjmp}. On some
5141 machines, however, @code{longjmp} will not change the value of global
5142 register variables. To be portable, the function that called @code{setjmp}
5143 should make other arrangements to save the values of the global register
5144 variables, and to restore them in a @code{longjmp}. This way, the same
5145 thing will happen regardless of what @code{longjmp} does.
5147 All global register variable declarations must precede all function
5148 definitions. If such a declaration could appear after function
5149 definitions, the declaration would be too late to prevent the register from
5150 being used for other purposes in the preceding functions.
5152 Global register variables may not have initial values, because an
5153 executable file has no means to supply initial contents for a register.
5155 On the SPARC, there are reports that g3 @dots{} g7 are suitable
5156 registers, but certain library functions, such as @code{getwd}, as well
5157 as the subroutines for division and remainder, modify g3 and g4. g1 and
5158 g2 are local temporaries.
5160 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
5161 Of course, it will not do to use more than a few of those.
5163 @node Local Reg Vars
5164 @subsection Specifying Registers for Local Variables
5165 @cindex local variables, specifying registers
5166 @cindex specifying registers for local variables
5167 @cindex registers for local variables
5169 You can define a local register variable with a specified register
5173 register int *foo asm ("a5");
5177 Here @code{a5} is the name of the register which should be used. Note
5178 that this is the same syntax used for defining global register
5179 variables, but for a local variable it would appear within a function.
5181 Naturally the register name is cpu-dependent, but this is not a
5182 problem, since specific registers are most often useful with explicit
5183 assembler instructions (@pxref{Extended Asm}). Both of these things
5184 generally require that you conditionalize your program according to
5187 In addition, operating systems on one type of cpu may differ in how they
5188 name the registers; then you would need additional conditionals. For
5189 example, some 68000 operating systems call this register @code{%a5}.
5191 Defining such a register variable does not reserve the register; it
5192 remains available for other uses in places where flow control determines
5193 the variable's value is not live.
5195 This option does not guarantee that GCC will generate code that has
5196 this variable in the register you specify at all times. You may not
5197 code an explicit reference to this register in the @emph{assembler
5198 instruction template} part of an @code{asm} statement and assume it will
5199 always refer to this variable. However, using the variable as an
5200 @code{asm} @emph{operand} guarantees that the specified register is used
5203 Stores into local register variables may be deleted when they appear to be dead
5204 according to dataflow analysis. References to local register variables may
5205 be deleted or moved or simplified.
5207 As for global register variables, it's recommended that you choose a
5208 register which is normally saved and restored by function calls on
5209 your machine, so that library routines will not clobber it. A common
5210 pitfall is to initialize multiple call-clobbered registers with
5211 arbitrary expressions, where a function call or library call for an
5212 arithmetic operator will overwrite a register value from a previous
5213 assignment, for example @code{r0} below:
5215 register int *p1 asm ("r0") = @dots{};
5216 register int *p2 asm ("r1") = @dots{};
5218 In those cases, a solution is to use a temporary variable for
5219 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
5221 @node Alternate Keywords
5222 @section Alternate Keywords
5223 @cindex alternate keywords
5224 @cindex keywords, alternate
5226 @option{-ansi} and the various @option{-std} options disable certain
5227 keywords. This causes trouble when you want to use GNU C extensions, or
5228 a general-purpose header file that should be usable by all programs,
5229 including ISO C programs. The keywords @code{asm}, @code{typeof} and
5230 @code{inline} are not available in programs compiled with
5231 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
5232 program compiled with @option{-std=c99}). The ISO C99 keyword
5233 @code{restrict} is only available when @option{-std=gnu99} (which will
5234 eventually be the default) or @option{-std=c99} (or the equivalent
5235 @option{-std=iso9899:1999}) is used.
5237 The way to solve these problems is to put @samp{__} at the beginning and
5238 end of each problematical keyword. For example, use @code{__asm__}
5239 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
5241 Other C compilers won't accept these alternative keywords; if you want to
5242 compile with another compiler, you can define the alternate keywords as
5243 macros to replace them with the customary keywords. It looks like this:
5251 @findex __extension__
5253 @option{-pedantic} and other options cause warnings for many GNU C extensions.
5255 prevent such warnings within one expression by writing
5256 @code{__extension__} before the expression. @code{__extension__} has no
5257 effect aside from this.
5259 @node Incomplete Enums
5260 @section Incomplete @code{enum} Types
5262 You can define an @code{enum} tag without specifying its possible values.
5263 This results in an incomplete type, much like what you get if you write
5264 @code{struct foo} without describing the elements. A later declaration
5265 which does specify the possible values completes the type.
5267 You can't allocate variables or storage using the type while it is
5268 incomplete. However, you can work with pointers to that type.
5270 This extension may not be very useful, but it makes the handling of
5271 @code{enum} more consistent with the way @code{struct} and @code{union}
5274 This extension is not supported by GNU C++.
5276 @node Function Names
5277 @section Function Names as Strings
5278 @cindex @code{__func__} identifier
5279 @cindex @code{__FUNCTION__} identifier
5280 @cindex @code{__PRETTY_FUNCTION__} identifier
5282 GCC provides three magic variables which hold the name of the current
5283 function, as a string. The first of these is @code{__func__}, which
5284 is part of the C99 standard:
5287 The identifier @code{__func__} is implicitly declared by the translator
5288 as if, immediately following the opening brace of each function
5289 definition, the declaration
5292 static const char __func__[] = "function-name";
5295 appeared, where function-name is the name of the lexically-enclosing
5296 function. This name is the unadorned name of the function.
5299 @code{__FUNCTION__} is another name for @code{__func__}. Older
5300 versions of GCC recognize only this name. However, it is not
5301 standardized. For maximum portability, we recommend you use
5302 @code{__func__}, but provide a fallback definition with the
5306 #if __STDC_VERSION__ < 199901L
5308 # define __func__ __FUNCTION__
5310 # define __func__ "<unknown>"
5315 In C, @code{__PRETTY_FUNCTION__} is yet another name for
5316 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
5317 the type signature of the function as well as its bare name. For
5318 example, this program:
5322 extern int printf (char *, ...);
5329 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
5330 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
5348 __PRETTY_FUNCTION__ = void a::sub(int)
5351 These identifiers are not preprocessor macros. In GCC 3.3 and
5352 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
5353 were treated as string literals; they could be used to initialize
5354 @code{char} arrays, and they could be concatenated with other string
5355 literals. GCC 3.4 and later treat them as variables, like
5356 @code{__func__}. In C++, @code{__FUNCTION__} and
5357 @code{__PRETTY_FUNCTION__} have always been variables.
5359 @node Return Address
5360 @section Getting the Return or Frame Address of a Function
5362 These functions may be used to get information about the callers of a
5365 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
5366 This function returns the return address of the current function, or of
5367 one of its callers. The @var{level} argument is number of frames to
5368 scan up the call stack. A value of @code{0} yields the return address
5369 of the current function, a value of @code{1} yields the return address
5370 of the caller of the current function, and so forth. When inlining
5371 the expected behavior is that the function will return the address of
5372 the function that will be returned to. To work around this behavior use
5373 the @code{noinline} function attribute.
5375 The @var{level} argument must be a constant integer.
5377 On some machines it may be impossible to determine the return address of
5378 any function other than the current one; in such cases, or when the top
5379 of the stack has been reached, this function will return @code{0} or a
5380 random value. In addition, @code{__builtin_frame_address} may be used
5381 to determine if the top of the stack has been reached.
5383 This function should only be used with a nonzero argument for debugging
5387 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
5388 This function is similar to @code{__builtin_return_address}, but it
5389 returns the address of the function frame rather than the return address
5390 of the function. Calling @code{__builtin_frame_address} with a value of
5391 @code{0} yields the frame address of the current function, a value of
5392 @code{1} yields the frame address of the caller of the current function,
5395 The frame is the area on the stack which holds local variables and saved
5396 registers. The frame address is normally the address of the first word
5397 pushed on to the stack by the function. However, the exact definition
5398 depends upon the processor and the calling convention. If the processor
5399 has a dedicated frame pointer register, and the function has a frame,
5400 then @code{__builtin_frame_address} will return the value of the frame
5403 On some machines it may be impossible to determine the frame address of
5404 any function other than the current one; in such cases, or when the top
5405 of the stack has been reached, this function will return @code{0} if
5406 the first frame pointer is properly initialized by the startup code.
5408 This function should only be used with a nonzero argument for debugging
5412 @node Vector Extensions
5413 @section Using vector instructions through built-in functions
5415 On some targets, the instruction set contains SIMD vector instructions that
5416 operate on multiple values contained in one large register at the same time.
5417 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
5420 The first step in using these extensions is to provide the necessary data
5421 types. This should be done using an appropriate @code{typedef}:
5424 typedef int v4si __attribute__ ((vector_size (16)));
5427 The @code{int} type specifies the base type, while the attribute specifies
5428 the vector size for the variable, measured in bytes. For example, the
5429 declaration above causes the compiler to set the mode for the @code{v4si}
5430 type to be 16 bytes wide and divided into @code{int} sized units. For
5431 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
5432 corresponding mode of @code{foo} will be @acronym{V4SI}.
5434 The @code{vector_size} attribute is only applicable to integral and
5435 float scalars, although arrays, pointers, and function return values
5436 are allowed in conjunction with this construct.
5438 All the basic integer types can be used as base types, both as signed
5439 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
5440 @code{long long}. In addition, @code{float} and @code{double} can be
5441 used to build floating-point vector types.
5443 Specifying a combination that is not valid for the current architecture
5444 will cause GCC to synthesize the instructions using a narrower mode.
5445 For example, if you specify a variable of type @code{V4SI} and your
5446 architecture does not allow for this specific SIMD type, GCC will
5447 produce code that uses 4 @code{SIs}.
5449 The types defined in this manner can be used with a subset of normal C
5450 operations. Currently, GCC will allow using the following operators
5451 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~}@.
5453 The operations behave like C++ @code{valarrays}. Addition is defined as
5454 the addition of the corresponding elements of the operands. For
5455 example, in the code below, each of the 4 elements in @var{a} will be
5456 added to the corresponding 4 elements in @var{b} and the resulting
5457 vector will be stored in @var{c}.
5460 typedef int v4si __attribute__ ((vector_size (16)));
5467 Subtraction, multiplication, division, and the logical operations
5468 operate in a similar manner. Likewise, the result of using the unary
5469 minus or complement operators on a vector type is a vector whose
5470 elements are the negative or complemented values of the corresponding
5471 elements in the operand.
5473 You can declare variables and use them in function calls and returns, as
5474 well as in assignments and some casts. You can specify a vector type as
5475 a return type for a function. Vector types can also be used as function
5476 arguments. It is possible to cast from one vector type to another,
5477 provided they are of the same size (in fact, you can also cast vectors
5478 to and from other datatypes of the same size).
5480 You cannot operate between vectors of different lengths or different
5481 signedness without a cast.
5483 A port that supports hardware vector operations, usually provides a set
5484 of built-in functions that can be used to operate on vectors. For
5485 example, a function to add two vectors and multiply the result by a
5486 third could look like this:
5489 v4si f (v4si a, v4si b, v4si c)
5491 v4si tmp = __builtin_addv4si (a, b);
5492 return __builtin_mulv4si (tmp, c);
5499 @findex __builtin_offsetof
5501 GCC implements for both C and C++ a syntactic extension to implement
5502 the @code{offsetof} macro.
5506 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
5508 offsetof_member_designator:
5510 | offsetof_member_designator "." @code{identifier}
5511 | offsetof_member_designator "[" @code{expr} "]"
5514 This extension is sufficient such that
5517 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
5520 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
5521 may be dependent. In either case, @var{member} may consist of a single
5522 identifier, or a sequence of member accesses and array references.
5524 @node Atomic Builtins
5525 @section Built-in functions for atomic memory access
5527 The following builtins are intended to be compatible with those described
5528 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
5529 section 7.4. As such, they depart from the normal GCC practice of using
5530 the ``__builtin_'' prefix, and further that they are overloaded such that
5531 they work on multiple types.
5533 The definition given in the Intel documentation allows only for the use of
5534 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
5535 counterparts. GCC will allow any integral scalar or pointer type that is
5536 1, 2, 4 or 8 bytes in length.
5538 Not all operations are supported by all target processors. If a particular
5539 operation cannot be implemented on the target processor, a warning will be
5540 generated and a call an external function will be generated. The external
5541 function will carry the same name as the builtin, with an additional suffix
5542 @samp{_@var{n}} where @var{n} is the size of the data type.
5544 @c ??? Should we have a mechanism to suppress this warning? This is almost
5545 @c useful for implementing the operation under the control of an external
5548 In most cases, these builtins are considered a @dfn{full barrier}. That is,
5549 no memory operand will be moved across the operation, either forward or
5550 backward. Further, instructions will be issued as necessary to prevent the
5551 processor from speculating loads across the operation and from queuing stores
5552 after the operation.
5554 All of the routines are are described in the Intel documentation to take
5555 ``an optional list of variables protected by the memory barrier''. It's
5556 not clear what is meant by that; it could mean that @emph{only} the
5557 following variables are protected, or it could mean that these variables
5558 should in addition be protected. At present GCC ignores this list and
5559 protects all variables which are globally accessible. If in the future
5560 we make some use of this list, an empty list will continue to mean all
5561 globally accessible variables.
5564 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
5565 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
5566 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
5567 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
5568 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
5569 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
5570 @findex __sync_fetch_and_add
5571 @findex __sync_fetch_and_sub
5572 @findex __sync_fetch_and_or
5573 @findex __sync_fetch_and_and
5574 @findex __sync_fetch_and_xor
5575 @findex __sync_fetch_and_nand
5576 These builtins perform the operation suggested by the name, and
5577 returns the value that had previously been in memory. That is,
5580 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
5581 @{ tmp = *ptr; *ptr = ~tmp & value; return tmp; @} // nand
5584 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
5585 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
5586 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
5587 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
5588 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
5589 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
5590 @findex __sync_add_and_fetch
5591 @findex __sync_sub_and_fetch
5592 @findex __sync_or_and_fetch
5593 @findex __sync_and_and_fetch
5594 @findex __sync_xor_and_fetch
5595 @findex __sync_nand_and_fetch
5596 These builtins perform the operation suggested by the name, and
5597 return the new value. That is,
5600 @{ *ptr @var{op}= value; return *ptr; @}
5601 @{ *ptr = ~*ptr & value; return *ptr; @} // nand
5604 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
5605 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
5606 @findex __sync_bool_compare_and_swap
5607 @findex __sync_val_compare_and_swap
5608 These builtins perform an atomic compare and swap. That is, if the current
5609 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
5612 The ``bool'' version returns true if the comparison is successful and
5613 @var{newval} was written. The ``val'' version returns the contents
5614 of @code{*@var{ptr}} before the operation.
5616 @item __sync_synchronize (...)
5617 @findex __sync_synchronize
5618 This builtin issues a full memory barrier.
5620 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
5621 @findex __sync_lock_test_and_set
5622 This builtin, as described by Intel, is not a traditional test-and-set
5623 operation, but rather an atomic exchange operation. It writes @var{value}
5624 into @code{*@var{ptr}}, and returns the previous contents of
5627 Many targets have only minimal support for such locks, and do not support
5628 a full exchange operation. In this case, a target may support reduced
5629 functionality here by which the @emph{only} valid value to store is the
5630 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
5631 is implementation defined.
5633 This builtin is not a full barrier, but rather an @dfn{acquire barrier}.
5634 This means that references after the builtin cannot move to (or be
5635 speculated to) before the builtin, but previous memory stores may not
5636 be globally visible yet, and previous memory loads may not yet be
5639 @item void __sync_lock_release (@var{type} *ptr, ...)
5640 @findex __sync_lock_release
5641 This builtin releases the lock acquired by @code{__sync_lock_test_and_set}.
5642 Normally this means writing the constant 0 to @code{*@var{ptr}}.
5644 This builtin is not a full barrier, but rather a @dfn{release barrier}.
5645 This means that all previous memory stores are globally visible, and all
5646 previous memory loads have been satisfied, but following memory reads
5647 are not prevented from being speculated to before the barrier.
5650 @node Object Size Checking
5651 @section Object Size Checking Builtins
5652 @findex __builtin_object_size
5653 @findex __builtin___memcpy_chk
5654 @findex __builtin___mempcpy_chk
5655 @findex __builtin___memmove_chk
5656 @findex __builtin___memset_chk
5657 @findex __builtin___strcpy_chk
5658 @findex __builtin___stpcpy_chk
5659 @findex __builtin___strncpy_chk
5660 @findex __builtin___strcat_chk
5661 @findex __builtin___strncat_chk
5662 @findex __builtin___sprintf_chk
5663 @findex __builtin___snprintf_chk
5664 @findex __builtin___vsprintf_chk
5665 @findex __builtin___vsnprintf_chk
5666 @findex __builtin___printf_chk
5667 @findex __builtin___vprintf_chk
5668 @findex __builtin___fprintf_chk
5669 @findex __builtin___vfprintf_chk
5671 GCC implements a limited buffer overflow protection mechanism
5672 that can prevent some buffer overflow attacks.
5674 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
5675 is a built-in construct that returns a constant number of bytes from
5676 @var{ptr} to the end of the object @var{ptr} pointer points to
5677 (if known at compile time). @code{__builtin_object_size} never evaluates
5678 its arguments for side-effects. If there are any side-effects in them, it
5679 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
5680 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
5681 point to and all of them are known at compile time, the returned number
5682 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
5683 0 and minimum if nonzero. If it is not possible to determine which objects
5684 @var{ptr} points to at compile time, @code{__builtin_object_size} should
5685 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
5686 for @var{type} 2 or 3.
5688 @var{type} is an integer constant from 0 to 3. If the least significant
5689 bit is clear, objects are whole variables, if it is set, a closest
5690 surrounding subobject is considered the object a pointer points to.
5691 The second bit determines if maximum or minimum of remaining bytes
5695 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
5696 char *p = &var.buf1[1], *q = &var.b;
5698 /* Here the object p points to is var. */
5699 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
5700 /* The subobject p points to is var.buf1. */
5701 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
5702 /* The object q points to is var. */
5703 assert (__builtin_object_size (q, 0)
5704 == (char *) (&var + 1) - (char *) &var.b);
5705 /* The subobject q points to is var.b. */
5706 assert (__builtin_object_size (q, 1) == sizeof (var.b));
5710 There are built-in functions added for many common string operation
5711 functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk}
5712 built-in is provided. This built-in has an additional last argument,
5713 which is the number of bytes remaining in object the @var{dest}
5714 argument points to or @code{(size_t) -1} if the size is not known.
5716 The built-in functions are optimized into the normal string functions
5717 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
5718 it is known at compile time that the destination object will not
5719 be overflown. If the compiler can determine at compile time the
5720 object will be always overflown, it issues a warning.
5722 The intended use can be e.g.
5726 #define bos0(dest) __builtin_object_size (dest, 0)
5727 #define memcpy(dest, src, n) \
5728 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
5732 /* It is unknown what object p points to, so this is optimized
5733 into plain memcpy - no checking is possible. */
5734 memcpy (p, "abcde", n);
5735 /* Destination is known and length too. It is known at compile
5736 time there will be no overflow. */
5737 memcpy (&buf[5], "abcde", 5);
5738 /* Destination is known, but the length is not known at compile time.
5739 This will result in __memcpy_chk call that can check for overflow
5741 memcpy (&buf[5], "abcde", n);
5742 /* Destination is known and it is known at compile time there will
5743 be overflow. There will be a warning and __memcpy_chk call that
5744 will abort the program at runtime. */
5745 memcpy (&buf[6], "abcde", 5);
5748 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
5749 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
5750 @code{strcat} and @code{strncat}.
5752 There are also checking built-in functions for formatted output functions.
5754 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
5755 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
5756 const char *fmt, ...);
5757 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
5759 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
5760 const char *fmt, va_list ap);
5763 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
5764 etc.@: functions and can contain implementation specific flags on what
5765 additional security measures the checking function might take, such as
5766 handling @code{%n} differently.
5768 The @var{os} argument is the object size @var{s} points to, like in the
5769 other built-in functions. There is a small difference in the behavior
5770 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
5771 optimized into the non-checking functions only if @var{flag} is 0, otherwise
5772 the checking function is called with @var{os} argument set to
5775 In addition to this, there are checking built-in functions
5776 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
5777 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
5778 These have just one additional argument, @var{flag}, right before
5779 format string @var{fmt}. If the compiler is able to optimize them to
5780 @code{fputc} etc.@: functions, it will, otherwise the checking function
5781 should be called and the @var{flag} argument passed to it.
5783 @node Other Builtins
5784 @section Other built-in functions provided by GCC
5785 @cindex built-in functions
5786 @findex __builtin_fpclassify
5787 @findex __builtin_isfinite
5788 @findex __builtin_isnormal
5789 @findex __builtin_isgreater
5790 @findex __builtin_isgreaterequal
5791 @findex __builtin_isinf_sign
5792 @findex __builtin_isless
5793 @findex __builtin_islessequal
5794 @findex __builtin_islessgreater
5795 @findex __builtin_isunordered
5796 @findex __builtin_powi
5797 @findex __builtin_powif
5798 @findex __builtin_powil
5956 @findex fprintf_unlocked
5958 @findex fputs_unlocked
6075 @findex printf_unlocked
6107 @findex significandf
6108 @findex significandl
6179 GCC provides a large number of built-in functions other than the ones
6180 mentioned above. Some of these are for internal use in the processing
6181 of exceptions or variable-length argument lists and will not be
6182 documented here because they may change from time to time; we do not
6183 recommend general use of these functions.
6185 The remaining functions are provided for optimization purposes.
6187 @opindex fno-builtin
6188 GCC includes built-in versions of many of the functions in the standard
6189 C library. The versions prefixed with @code{__builtin_} will always be
6190 treated as having the same meaning as the C library function even if you
6191 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
6192 Many of these functions are only optimized in certain cases; if they are
6193 not optimized in a particular case, a call to the library function will
6198 Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or
6199 @option{-std=c99}), the functions
6200 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
6201 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
6202 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
6203 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
6204 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
6205 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
6206 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
6207 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
6208 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
6209 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
6210 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
6211 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
6212 @code{signbitd64}, @code{signbitd128}, @code{significandf},
6213 @code{significandl}, @code{significand}, @code{sincosf},
6214 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
6215 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
6216 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
6217 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
6219 may be handled as built-in functions.
6220 All these functions have corresponding versions
6221 prefixed with @code{__builtin_}, which may be used even in strict C89
6224 The ISO C99 functions
6225 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
6226 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
6227 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
6228 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
6229 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
6230 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
6231 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
6232 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
6233 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
6234 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
6235 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
6236 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
6237 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
6238 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
6239 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
6240 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
6241 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
6242 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
6243 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
6244 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
6245 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
6246 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
6247 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
6248 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
6249 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
6250 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
6251 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
6252 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
6253 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
6254 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
6255 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
6256 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
6257 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
6258 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
6259 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
6260 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
6261 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
6262 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
6263 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
6264 are handled as built-in functions
6265 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
6267 There are also built-in versions of the ISO C99 functions
6268 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
6269 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
6270 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
6271 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
6272 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
6273 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
6274 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
6275 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
6276 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
6277 that are recognized in any mode since ISO C90 reserves these names for
6278 the purpose to which ISO C99 puts them. All these functions have
6279 corresponding versions prefixed with @code{__builtin_}.
6281 The ISO C94 functions
6282 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
6283 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
6284 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
6286 are handled as built-in functions
6287 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
6289 The ISO C90 functions
6290 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
6291 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
6292 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
6293 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
6294 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
6295 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
6296 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
6297 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
6298 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
6299 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
6300 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
6301 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
6302 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
6303 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
6304 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
6305 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
6306 are all recognized as built-in functions unless
6307 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
6308 is specified for an individual function). All of these functions have
6309 corresponding versions prefixed with @code{__builtin_}.
6311 GCC provides built-in versions of the ISO C99 floating point comparison
6312 macros that avoid raising exceptions for unordered operands. They have
6313 the same names as the standard macros ( @code{isgreater},
6314 @code{isgreaterequal}, @code{isless}, @code{islessequal},
6315 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
6316 prefixed. We intend for a library implementor to be able to simply
6317 @code{#define} each standard macro to its built-in equivalent.
6318 In the same fashion, GCC provides @code{fpclassify}, @code{isfinite},
6319 @code{isinf_sign} and @code{isnormal} built-ins used with
6320 @code{__builtin_} prefixed. The @code{isinf} and @code{isnan}
6321 builtins appear both with and without the @code{__builtin_} prefix.
6323 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
6325 You can use the built-in function @code{__builtin_types_compatible_p} to
6326 determine whether two types are the same.
6328 This built-in function returns 1 if the unqualified versions of the
6329 types @var{type1} and @var{type2} (which are types, not expressions) are
6330 compatible, 0 otherwise. The result of this built-in function can be
6331 used in integer constant expressions.
6333 This built-in function ignores top level qualifiers (e.g., @code{const},
6334 @code{volatile}). For example, @code{int} is equivalent to @code{const
6337 The type @code{int[]} and @code{int[5]} are compatible. On the other
6338 hand, @code{int} and @code{char *} are not compatible, even if the size
6339 of their types, on the particular architecture are the same. Also, the
6340 amount of pointer indirection is taken into account when determining
6341 similarity. Consequently, @code{short *} is not similar to
6342 @code{short **}. Furthermore, two types that are typedefed are
6343 considered compatible if their underlying types are compatible.
6345 An @code{enum} type is not considered to be compatible with another
6346 @code{enum} type even if both are compatible with the same integer
6347 type; this is what the C standard specifies.
6348 For example, @code{enum @{foo, bar@}} is not similar to
6349 @code{enum @{hot, dog@}}.
6351 You would typically use this function in code whose execution varies
6352 depending on the arguments' types. For example:
6357 typeof (x) tmp = (x); \
6358 if (__builtin_types_compatible_p (typeof (x), long double)) \
6359 tmp = foo_long_double (tmp); \
6360 else if (__builtin_types_compatible_p (typeof (x), double)) \
6361 tmp = foo_double (tmp); \
6362 else if (__builtin_types_compatible_p (typeof (x), float)) \
6363 tmp = foo_float (tmp); \
6370 @emph{Note:} This construct is only available for C@.
6374 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
6376 You can use the built-in function @code{__builtin_choose_expr} to
6377 evaluate code depending on the value of a constant expression. This
6378 built-in function returns @var{exp1} if @var{const_exp}, which is a
6379 constant expression that must be able to be determined at compile time,
6380 is nonzero. Otherwise it returns 0.
6382 This built-in function is analogous to the @samp{? :} operator in C,
6383 except that the expression returned has its type unaltered by promotion
6384 rules. Also, the built-in function does not evaluate the expression
6385 that was not chosen. For example, if @var{const_exp} evaluates to true,
6386 @var{exp2} is not evaluated even if it has side-effects.
6388 This built-in function can return an lvalue if the chosen argument is an
6391 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
6392 type. Similarly, if @var{exp2} is returned, its return type is the same
6399 __builtin_choose_expr ( \
6400 __builtin_types_compatible_p (typeof (x), double), \
6402 __builtin_choose_expr ( \
6403 __builtin_types_compatible_p (typeof (x), float), \
6405 /* @r{The void expression results in a compile-time error} \
6406 @r{when assigning the result to something.} */ \
6410 @emph{Note:} This construct is only available for C@. Furthermore, the
6411 unused expression (@var{exp1} or @var{exp2} depending on the value of
6412 @var{const_exp}) may still generate syntax errors. This may change in
6417 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
6418 You can use the built-in function @code{__builtin_constant_p} to
6419 determine if a value is known to be constant at compile-time and hence
6420 that GCC can perform constant-folding on expressions involving that
6421 value. The argument of the function is the value to test. The function
6422 returns the integer 1 if the argument is known to be a compile-time
6423 constant and 0 if it is not known to be a compile-time constant. A
6424 return of 0 does not indicate that the value is @emph{not} a constant,
6425 but merely that GCC cannot prove it is a constant with the specified
6426 value of the @option{-O} option.
6428 You would typically use this function in an embedded application where
6429 memory was a critical resource. If you have some complex calculation,
6430 you may want it to be folded if it involves constants, but need to call
6431 a function if it does not. For example:
6434 #define Scale_Value(X) \
6435 (__builtin_constant_p (X) \
6436 ? ((X) * SCALE + OFFSET) : Scale (X))
6439 You may use this built-in function in either a macro or an inline
6440 function. However, if you use it in an inlined function and pass an
6441 argument of the function as the argument to the built-in, GCC will
6442 never return 1 when you call the inline function with a string constant
6443 or compound literal (@pxref{Compound Literals}) and will not return 1
6444 when you pass a constant numeric value to the inline function unless you
6445 specify the @option{-O} option.
6447 You may also use @code{__builtin_constant_p} in initializers for static
6448 data. For instance, you can write
6451 static const int table[] = @{
6452 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
6458 This is an acceptable initializer even if @var{EXPRESSION} is not a
6459 constant expression. GCC must be more conservative about evaluating the
6460 built-in in this case, because it has no opportunity to perform
6463 Previous versions of GCC did not accept this built-in in data
6464 initializers. The earliest version where it is completely safe is
6468 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
6469 @opindex fprofile-arcs
6470 You may use @code{__builtin_expect} to provide the compiler with
6471 branch prediction information. In general, you should prefer to
6472 use actual profile feedback for this (@option{-fprofile-arcs}), as
6473 programmers are notoriously bad at predicting how their programs
6474 actually perform. However, there are applications in which this
6475 data is hard to collect.
6477 The return value is the value of @var{exp}, which should be an integral
6478 expression. The semantics of the built-in are that it is expected that
6479 @var{exp} == @var{c}. For example:
6482 if (__builtin_expect (x, 0))
6487 would indicate that we do not expect to call @code{foo}, since
6488 we expect @code{x} to be zero. Since you are limited to integral
6489 expressions for @var{exp}, you should use constructions such as
6492 if (__builtin_expect (ptr != NULL, 1))
6497 when testing pointer or floating-point values.
6500 @deftypefn {Built-in Function} void __builtin_trap (void)
6501 This function causes the program to exit abnormally. GCC implements
6502 this function by using a target-dependent mechanism (such as
6503 intentionally executing an illegal instruction) or by calling
6504 @code{abort}. The mechanism used may vary from release to release so
6505 you should not rely on any particular implementation.
6508 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
6509 This function is used to flush the processor's instruction cache for
6510 the region of memory between @var{begin} inclusive and @var{end}
6511 exclusive. Some targets require that the instruction cache be
6512 flushed, after modifying memory containing code, in order to obtain
6513 deterministic behavior.
6515 If the target does not require instruction cache flushes,
6516 @code{__builtin___clear_cache} has no effect. Otherwise either
6517 instructions are emitted in-line to clear the instruction cache or a
6518 call to the @code{__clear_cache} function in libgcc is made.
6521 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
6522 This function is used to minimize cache-miss latency by moving data into
6523 a cache before it is accessed.
6524 You can insert calls to @code{__builtin_prefetch} into code for which
6525 you know addresses of data in memory that is likely to be accessed soon.
6526 If the target supports them, data prefetch instructions will be generated.
6527 If the prefetch is done early enough before the access then the data will
6528 be in the cache by the time it is accessed.
6530 The value of @var{addr} is the address of the memory to prefetch.
6531 There are two optional arguments, @var{rw} and @var{locality}.
6532 The value of @var{rw} is a compile-time constant one or zero; one
6533 means that the prefetch is preparing for a write to the memory address
6534 and zero, the default, means that the prefetch is preparing for a read.
6535 The value @var{locality} must be a compile-time constant integer between
6536 zero and three. A value of zero means that the data has no temporal
6537 locality, so it need not be left in the cache after the access. A value
6538 of three means that the data has a high degree of temporal locality and
6539 should be left in all levels of cache possible. Values of one and two
6540 mean, respectively, a low or moderate degree of temporal locality. The
6544 for (i = 0; i < n; i++)
6547 __builtin_prefetch (&a[i+j], 1, 1);
6548 __builtin_prefetch (&b[i+j], 0, 1);
6553 Data prefetch does not generate faults if @var{addr} is invalid, but
6554 the address expression itself must be valid. For example, a prefetch
6555 of @code{p->next} will not fault if @code{p->next} is not a valid
6556 address, but evaluation will fault if @code{p} is not a valid address.
6558 If the target does not support data prefetch, the address expression
6559 is evaluated if it includes side effects but no other code is generated
6560 and GCC does not issue a warning.
6563 @deftypefn {Built-in Function} double __builtin_huge_val (void)
6564 Returns a positive infinity, if supported by the floating-point format,
6565 else @code{DBL_MAX}. This function is suitable for implementing the
6566 ISO C macro @code{HUGE_VAL}.
6569 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
6570 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
6573 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
6574 Similar to @code{__builtin_huge_val}, except the return
6575 type is @code{long double}.
6578 @deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...)
6579 This built-in implements the C99 fpclassify functionality. The first
6580 five int arguments should be the target library's notion of the
6581 possible FP classes and are used for return values. They must be
6582 constant values and they must appear in this order: @code{FP_NAN},
6583 @code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and
6584 @code{FP_ZERO}. The ellipsis is for exactly one floating point value
6585 to classify. GCC treats the last argument as type-generic, which
6586 means it does not do default promotion from float to double.
6589 @deftypefn {Built-in Function} double __builtin_inf (void)
6590 Similar to @code{__builtin_huge_val}, except a warning is generated
6591 if the target floating-point format does not support infinities.
6594 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
6595 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
6598 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
6599 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
6602 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
6603 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
6606 @deftypefn {Built-in Function} float __builtin_inff (void)
6607 Similar to @code{__builtin_inf}, except the return type is @code{float}.
6608 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
6611 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
6612 Similar to @code{__builtin_inf}, except the return
6613 type is @code{long double}.
6616 @deftypefn {Built-in Function} int __builtin_isinf_sign (...)
6617 Similar to @code{isinf}, except the return value will be negative for
6618 an argument of @code{-Inf}. Note while the parameter list is an
6619 ellipsis, this function only accepts exactly one floating point
6620 argument. GCC treats this parameter as type-generic, which means it
6621 does not do default promotion from float to double.
6624 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
6625 This is an implementation of the ISO C99 function @code{nan}.
6627 Since ISO C99 defines this function in terms of @code{strtod}, which we
6628 do not implement, a description of the parsing is in order. The string
6629 is parsed as by @code{strtol}; that is, the base is recognized by
6630 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
6631 in the significand such that the least significant bit of the number
6632 is at the least significant bit of the significand. The number is
6633 truncated to fit the significand field provided. The significand is
6634 forced to be a quiet NaN@.
6636 This function, if given a string literal all of which would have been
6637 consumed by strtol, is evaluated early enough that it is considered a
6638 compile-time constant.
6641 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
6642 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
6645 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
6646 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
6649 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
6650 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
6653 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
6654 Similar to @code{__builtin_nan}, except the return type is @code{float}.
6657 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
6658 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
6661 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
6662 Similar to @code{__builtin_nan}, except the significand is forced
6663 to be a signaling NaN@. The @code{nans} function is proposed by
6664 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
6667 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
6668 Similar to @code{__builtin_nans}, except the return type is @code{float}.
6671 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
6672 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
6675 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
6676 Returns one plus the index of the least significant 1-bit of @var{x}, or
6677 if @var{x} is zero, returns zero.
6680 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
6681 Returns the number of leading 0-bits in @var{x}, starting at the most
6682 significant bit position. If @var{x} is 0, the result is undefined.
6685 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
6686 Returns the number of trailing 0-bits in @var{x}, starting at the least
6687 significant bit position. If @var{x} is 0, the result is undefined.
6690 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
6691 Returns the number of 1-bits in @var{x}.
6694 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
6695 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
6699 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
6700 Similar to @code{__builtin_ffs}, except the argument type is
6701 @code{unsigned long}.
6704 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
6705 Similar to @code{__builtin_clz}, except the argument type is
6706 @code{unsigned long}.
6709 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
6710 Similar to @code{__builtin_ctz}, except the argument type is
6711 @code{unsigned long}.
6714 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
6715 Similar to @code{__builtin_popcount}, except the argument type is
6716 @code{unsigned long}.
6719 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
6720 Similar to @code{__builtin_parity}, except the argument type is
6721 @code{unsigned long}.
6724 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
6725 Similar to @code{__builtin_ffs}, except the argument type is
6726 @code{unsigned long long}.
6729 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
6730 Similar to @code{__builtin_clz}, except the argument type is
6731 @code{unsigned long long}.
6734 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
6735 Similar to @code{__builtin_ctz}, except the argument type is
6736 @code{unsigned long long}.
6739 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
6740 Similar to @code{__builtin_popcount}, except the argument type is
6741 @code{unsigned long long}.
6744 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
6745 Similar to @code{__builtin_parity}, except the argument type is
6746 @code{unsigned long long}.
6749 @deftypefn {Built-in Function} double __builtin_powi (double, int)
6750 Returns the first argument raised to the power of the second. Unlike the
6751 @code{pow} function no guarantees about precision and rounding are made.
6754 @deftypefn {Built-in Function} float __builtin_powif (float, int)
6755 Similar to @code{__builtin_powi}, except the argument and return types
6759 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
6760 Similar to @code{__builtin_powi}, except the argument and return types
6761 are @code{long double}.
6764 @deftypefn {Built-in Function} int32_t __builtin_bswap32 (int32_t x)
6765 Returns @var{x} with the order of the bytes reversed; for example,
6766 @code{0xaabbccdd} becomes @code{0xddccbbaa}. Byte here always means
6770 @deftypefn {Built-in Function} int64_t __builtin_bswap64 (int64_t x)
6771 Similar to @code{__builtin_bswap32}, except the argument and return types
6775 @node Target Builtins
6776 @section Built-in Functions Specific to Particular Target Machines
6778 On some target machines, GCC supports many built-in functions specific
6779 to those machines. Generally these generate calls to specific machine
6780 instructions, but allow the compiler to schedule those calls.
6783 * Alpha Built-in Functions::
6784 * ARM iWMMXt Built-in Functions::
6785 * ARM NEON Intrinsics::
6786 * Blackfin Built-in Functions::
6787 * FR-V Built-in Functions::
6788 * X86 Built-in Functions::
6789 * MIPS DSP Built-in Functions::
6790 * MIPS Paired-Single Support::
6791 * PowerPC AltiVec Built-in Functions::
6792 * SPARC VIS Built-in Functions::
6793 * SPU Built-in Functions::
6796 @node Alpha Built-in Functions
6797 @subsection Alpha Built-in Functions
6799 These built-in functions are available for the Alpha family of
6800 processors, depending on the command-line switches used.
6802 The following built-in functions are always available. They
6803 all generate the machine instruction that is part of the name.
6806 long __builtin_alpha_implver (void)
6807 long __builtin_alpha_rpcc (void)
6808 long __builtin_alpha_amask (long)
6809 long __builtin_alpha_cmpbge (long, long)
6810 long __builtin_alpha_extbl (long, long)
6811 long __builtin_alpha_extwl (long, long)
6812 long __builtin_alpha_extll (long, long)
6813 long __builtin_alpha_extql (long, long)
6814 long __builtin_alpha_extwh (long, long)
6815 long __builtin_alpha_extlh (long, long)
6816 long __builtin_alpha_extqh (long, long)
6817 long __builtin_alpha_insbl (long, long)
6818 long __builtin_alpha_inswl (long, long)
6819 long __builtin_alpha_insll (long, long)
6820 long __builtin_alpha_insql (long, long)
6821 long __builtin_alpha_inswh (long, long)
6822 long __builtin_alpha_inslh (long, long)
6823 long __builtin_alpha_insqh (long, long)
6824 long __builtin_alpha_mskbl (long, long)
6825 long __builtin_alpha_mskwl (long, long)
6826 long __builtin_alpha_mskll (long, long)
6827 long __builtin_alpha_mskql (long, long)
6828 long __builtin_alpha_mskwh (long, long)
6829 long __builtin_alpha_msklh (long, long)
6830 long __builtin_alpha_mskqh (long, long)
6831 long __builtin_alpha_umulh (long, long)
6832 long __builtin_alpha_zap (long, long)
6833 long __builtin_alpha_zapnot (long, long)
6836 The following built-in functions are always with @option{-mmax}
6837 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
6838 later. They all generate the machine instruction that is part
6842 long __builtin_alpha_pklb (long)
6843 long __builtin_alpha_pkwb (long)
6844 long __builtin_alpha_unpkbl (long)
6845 long __builtin_alpha_unpkbw (long)
6846 long __builtin_alpha_minub8 (long, long)
6847 long __builtin_alpha_minsb8 (long, long)
6848 long __builtin_alpha_minuw4 (long, long)
6849 long __builtin_alpha_minsw4 (long, long)
6850 long __builtin_alpha_maxub8 (long, long)
6851 long __builtin_alpha_maxsb8 (long, long)
6852 long __builtin_alpha_maxuw4 (long, long)
6853 long __builtin_alpha_maxsw4 (long, long)
6854 long __builtin_alpha_perr (long, long)
6857 The following built-in functions are always with @option{-mcix}
6858 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
6859 later. They all generate the machine instruction that is part
6863 long __builtin_alpha_cttz (long)
6864 long __builtin_alpha_ctlz (long)
6865 long __builtin_alpha_ctpop (long)
6868 The following builtins are available on systems that use the OSF/1
6869 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
6870 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
6871 @code{rdval} and @code{wrval}.
6874 void *__builtin_thread_pointer (void)
6875 void __builtin_set_thread_pointer (void *)
6878 @node ARM iWMMXt Built-in Functions
6879 @subsection ARM iWMMXt Built-in Functions
6881 These built-in functions are available for the ARM family of
6882 processors when the @option{-mcpu=iwmmxt} switch is used:
6885 typedef int v2si __attribute__ ((vector_size (8)));
6886 typedef short v4hi __attribute__ ((vector_size (8)));
6887 typedef char v8qi __attribute__ ((vector_size (8)));
6889 int __builtin_arm_getwcx (int)
6890 void __builtin_arm_setwcx (int, int)
6891 int __builtin_arm_textrmsb (v8qi, int)
6892 int __builtin_arm_textrmsh (v4hi, int)
6893 int __builtin_arm_textrmsw (v2si, int)
6894 int __builtin_arm_textrmub (v8qi, int)
6895 int __builtin_arm_textrmuh (v4hi, int)
6896 int __builtin_arm_textrmuw (v2si, int)
6897 v8qi __builtin_arm_tinsrb (v8qi, int)
6898 v4hi __builtin_arm_tinsrh (v4hi, int)
6899 v2si __builtin_arm_tinsrw (v2si, int)
6900 long long __builtin_arm_tmia (long long, int, int)
6901 long long __builtin_arm_tmiabb (long long, int, int)
6902 long long __builtin_arm_tmiabt (long long, int, int)
6903 long long __builtin_arm_tmiaph (long long, int, int)
6904 long long __builtin_arm_tmiatb (long long, int, int)
6905 long long __builtin_arm_tmiatt (long long, int, int)
6906 int __builtin_arm_tmovmskb (v8qi)
6907 int __builtin_arm_tmovmskh (v4hi)
6908 int __builtin_arm_tmovmskw (v2si)
6909 long long __builtin_arm_waccb (v8qi)
6910 long long __builtin_arm_wacch (v4hi)
6911 long long __builtin_arm_waccw (v2si)
6912 v8qi __builtin_arm_waddb (v8qi, v8qi)
6913 v8qi __builtin_arm_waddbss (v8qi, v8qi)
6914 v8qi __builtin_arm_waddbus (v8qi, v8qi)
6915 v4hi __builtin_arm_waddh (v4hi, v4hi)
6916 v4hi __builtin_arm_waddhss (v4hi, v4hi)
6917 v4hi __builtin_arm_waddhus (v4hi, v4hi)
6918 v2si __builtin_arm_waddw (v2si, v2si)
6919 v2si __builtin_arm_waddwss (v2si, v2si)
6920 v2si __builtin_arm_waddwus (v2si, v2si)
6921 v8qi __builtin_arm_walign (v8qi, v8qi, int)
6922 long long __builtin_arm_wand(long long, long long)
6923 long long __builtin_arm_wandn (long long, long long)
6924 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
6925 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
6926 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
6927 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
6928 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
6929 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
6930 v2si __builtin_arm_wcmpeqw (v2si, v2si)
6931 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
6932 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
6933 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
6934 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
6935 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
6936 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
6937 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
6938 long long __builtin_arm_wmacsz (v4hi, v4hi)
6939 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
6940 long long __builtin_arm_wmacuz (v4hi, v4hi)
6941 v4hi __builtin_arm_wmadds (v4hi, v4hi)
6942 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
6943 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
6944 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
6945 v2si __builtin_arm_wmaxsw (v2si, v2si)
6946 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
6947 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
6948 v2si __builtin_arm_wmaxuw (v2si, v2si)
6949 v8qi __builtin_arm_wminsb (v8qi, v8qi)
6950 v4hi __builtin_arm_wminsh (v4hi, v4hi)
6951 v2si __builtin_arm_wminsw (v2si, v2si)
6952 v8qi __builtin_arm_wminub (v8qi, v8qi)
6953 v4hi __builtin_arm_wminuh (v4hi, v4hi)
6954 v2si __builtin_arm_wminuw (v2si, v2si)
6955 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
6956 v4hi __builtin_arm_wmulul (v4hi, v4hi)
6957 v4hi __builtin_arm_wmulum (v4hi, v4hi)
6958 long long __builtin_arm_wor (long long, long long)
6959 v2si __builtin_arm_wpackdss (long long, long long)
6960 v2si __builtin_arm_wpackdus (long long, long long)
6961 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
6962 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
6963 v4hi __builtin_arm_wpackwss (v2si, v2si)
6964 v4hi __builtin_arm_wpackwus (v2si, v2si)
6965 long long __builtin_arm_wrord (long long, long long)
6966 long long __builtin_arm_wrordi (long long, int)
6967 v4hi __builtin_arm_wrorh (v4hi, long long)
6968 v4hi __builtin_arm_wrorhi (v4hi, int)
6969 v2si __builtin_arm_wrorw (v2si, long long)
6970 v2si __builtin_arm_wrorwi (v2si, int)
6971 v2si __builtin_arm_wsadb (v8qi, v8qi)
6972 v2si __builtin_arm_wsadbz (v8qi, v8qi)
6973 v2si __builtin_arm_wsadh (v4hi, v4hi)
6974 v2si __builtin_arm_wsadhz (v4hi, v4hi)
6975 v4hi __builtin_arm_wshufh (v4hi, int)
6976 long long __builtin_arm_wslld (long long, long long)
6977 long long __builtin_arm_wslldi (long long, int)
6978 v4hi __builtin_arm_wsllh (v4hi, long long)
6979 v4hi __builtin_arm_wsllhi (v4hi, int)
6980 v2si __builtin_arm_wsllw (v2si, long long)
6981 v2si __builtin_arm_wsllwi (v2si, int)
6982 long long __builtin_arm_wsrad (long long, long long)
6983 long long __builtin_arm_wsradi (long long, int)
6984 v4hi __builtin_arm_wsrah (v4hi, long long)
6985 v4hi __builtin_arm_wsrahi (v4hi, int)
6986 v2si __builtin_arm_wsraw (v2si, long long)
6987 v2si __builtin_arm_wsrawi (v2si, int)
6988 long long __builtin_arm_wsrld (long long, long long)
6989 long long __builtin_arm_wsrldi (long long, int)
6990 v4hi __builtin_arm_wsrlh (v4hi, long long)
6991 v4hi __builtin_arm_wsrlhi (v4hi, int)
6992 v2si __builtin_arm_wsrlw (v2si, long long)
6993 v2si __builtin_arm_wsrlwi (v2si, int)
6994 v8qi __builtin_arm_wsubb (v8qi, v8qi)
6995 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
6996 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
6997 v4hi __builtin_arm_wsubh (v4hi, v4hi)
6998 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
6999 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
7000 v2si __builtin_arm_wsubw (v2si, v2si)
7001 v2si __builtin_arm_wsubwss (v2si, v2si)
7002 v2si __builtin_arm_wsubwus (v2si, v2si)
7003 v4hi __builtin_arm_wunpckehsb (v8qi)
7004 v2si __builtin_arm_wunpckehsh (v4hi)
7005 long long __builtin_arm_wunpckehsw (v2si)
7006 v4hi __builtin_arm_wunpckehub (v8qi)
7007 v2si __builtin_arm_wunpckehuh (v4hi)
7008 long long __builtin_arm_wunpckehuw (v2si)
7009 v4hi __builtin_arm_wunpckelsb (v8qi)
7010 v2si __builtin_arm_wunpckelsh (v4hi)
7011 long long __builtin_arm_wunpckelsw (v2si)
7012 v4hi __builtin_arm_wunpckelub (v8qi)
7013 v2si __builtin_arm_wunpckeluh (v4hi)
7014 long long __builtin_arm_wunpckeluw (v2si)
7015 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
7016 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
7017 v2si __builtin_arm_wunpckihw (v2si, v2si)
7018 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
7019 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
7020 v2si __builtin_arm_wunpckilw (v2si, v2si)
7021 long long __builtin_arm_wxor (long long, long long)
7022 long long __builtin_arm_wzero ()
7025 @node ARM NEON Intrinsics
7026 @subsection ARM NEON Intrinsics
7028 These built-in intrinsics for the ARM Advanced SIMD extension are available
7029 when the @option{-mfpu=neon} switch is used:
7031 @include arm-neon-intrinsics.texi
7033 @node Blackfin Built-in Functions
7034 @subsection Blackfin Built-in Functions
7036 Currently, there are two Blackfin-specific built-in functions. These are
7037 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
7038 using inline assembly; by using these built-in functions the compiler can
7039 automatically add workarounds for hardware errata involving these
7040 instructions. These functions are named as follows:
7043 void __builtin_bfin_csync (void)
7044 void __builtin_bfin_ssync (void)
7047 @node FR-V Built-in Functions
7048 @subsection FR-V Built-in Functions
7050 GCC provides many FR-V-specific built-in functions. In general,
7051 these functions are intended to be compatible with those described
7052 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
7053 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
7054 @code{__MBTOHE}, the gcc forms of which pass 128-bit values by
7055 pointer rather than by value.
7057 Most of the functions are named after specific FR-V instructions.
7058 Such functions are said to be ``directly mapped'' and are summarized
7059 here in tabular form.
7063 * Directly-mapped Integer Functions::
7064 * Directly-mapped Media Functions::
7065 * Raw read/write Functions::
7066 * Other Built-in Functions::
7069 @node Argument Types
7070 @subsubsection Argument Types
7072 The arguments to the built-in functions can be divided into three groups:
7073 register numbers, compile-time constants and run-time values. In order
7074 to make this classification clear at a glance, the arguments and return
7075 values are given the following pseudo types:
7077 @multitable @columnfractions .20 .30 .15 .35
7078 @item Pseudo type @tab Real C type @tab Constant? @tab Description
7079 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
7080 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
7081 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
7082 @item @code{uw2} @tab @code{unsigned long long} @tab No
7083 @tab an unsigned doubleword
7084 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
7085 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
7086 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
7087 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
7090 These pseudo types are not defined by GCC, they are simply a notational
7091 convenience used in this manual.
7093 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
7094 and @code{sw2} are evaluated at run time. They correspond to
7095 register operands in the underlying FR-V instructions.
7097 @code{const} arguments represent immediate operands in the underlying
7098 FR-V instructions. They must be compile-time constants.
7100 @code{acc} arguments are evaluated at compile time and specify the number
7101 of an accumulator register. For example, an @code{acc} argument of 2
7102 will select the ACC2 register.
7104 @code{iacc} arguments are similar to @code{acc} arguments but specify the
7105 number of an IACC register. See @pxref{Other Built-in Functions}
7108 @node Directly-mapped Integer Functions
7109 @subsubsection Directly-mapped Integer Functions
7111 The functions listed below map directly to FR-V I-type instructions.
7113 @multitable @columnfractions .45 .32 .23
7114 @item Function prototype @tab Example usage @tab Assembly output
7115 @item @code{sw1 __ADDSS (sw1, sw1)}
7116 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
7117 @tab @code{ADDSS @var{a},@var{b},@var{c}}
7118 @item @code{sw1 __SCAN (sw1, sw1)}
7119 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
7120 @tab @code{SCAN @var{a},@var{b},@var{c}}
7121 @item @code{sw1 __SCUTSS (sw1)}
7122 @tab @code{@var{b} = __SCUTSS (@var{a})}
7123 @tab @code{SCUTSS @var{a},@var{b}}
7124 @item @code{sw1 __SLASS (sw1, sw1)}
7125 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
7126 @tab @code{SLASS @var{a},@var{b},@var{c}}
7127 @item @code{void __SMASS (sw1, sw1)}
7128 @tab @code{__SMASS (@var{a}, @var{b})}
7129 @tab @code{SMASS @var{a},@var{b}}
7130 @item @code{void __SMSSS (sw1, sw1)}
7131 @tab @code{__SMSSS (@var{a}, @var{b})}
7132 @tab @code{SMSSS @var{a},@var{b}}
7133 @item @code{void __SMU (sw1, sw1)}
7134 @tab @code{__SMU (@var{a}, @var{b})}
7135 @tab @code{SMU @var{a},@var{b}}
7136 @item @code{sw2 __SMUL (sw1, sw1)}
7137 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
7138 @tab @code{SMUL @var{a},@var{b},@var{c}}
7139 @item @code{sw1 __SUBSS (sw1, sw1)}
7140 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
7141 @tab @code{SUBSS @var{a},@var{b},@var{c}}
7142 @item @code{uw2 __UMUL (uw1, uw1)}
7143 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
7144 @tab @code{UMUL @var{a},@var{b},@var{c}}
7147 @node Directly-mapped Media Functions
7148 @subsubsection Directly-mapped Media Functions
7150 The functions listed below map directly to FR-V M-type instructions.
7152 @multitable @columnfractions .45 .32 .23
7153 @item Function prototype @tab Example usage @tab Assembly output
7154 @item @code{uw1 __MABSHS (sw1)}
7155 @tab @code{@var{b} = __MABSHS (@var{a})}
7156 @tab @code{MABSHS @var{a},@var{b}}
7157 @item @code{void __MADDACCS (acc, acc)}
7158 @tab @code{__MADDACCS (@var{b}, @var{a})}
7159 @tab @code{MADDACCS @var{a},@var{b}}
7160 @item @code{sw1 __MADDHSS (sw1, sw1)}
7161 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
7162 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
7163 @item @code{uw1 __MADDHUS (uw1, uw1)}
7164 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
7165 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
7166 @item @code{uw1 __MAND (uw1, uw1)}
7167 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
7168 @tab @code{MAND @var{a},@var{b},@var{c}}
7169 @item @code{void __MASACCS (acc, acc)}
7170 @tab @code{__MASACCS (@var{b}, @var{a})}
7171 @tab @code{MASACCS @var{a},@var{b}}
7172 @item @code{uw1 __MAVEH (uw1, uw1)}
7173 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
7174 @tab @code{MAVEH @var{a},@var{b},@var{c}}
7175 @item @code{uw2 __MBTOH (uw1)}
7176 @tab @code{@var{b} = __MBTOH (@var{a})}
7177 @tab @code{MBTOH @var{a},@var{b}}
7178 @item @code{void __MBTOHE (uw1 *, uw1)}
7179 @tab @code{__MBTOHE (&@var{b}, @var{a})}
7180 @tab @code{MBTOHE @var{a},@var{b}}
7181 @item @code{void __MCLRACC (acc)}
7182 @tab @code{__MCLRACC (@var{a})}
7183 @tab @code{MCLRACC @var{a}}
7184 @item @code{void __MCLRACCA (void)}
7185 @tab @code{__MCLRACCA ()}
7186 @tab @code{MCLRACCA}
7187 @item @code{uw1 __Mcop1 (uw1, uw1)}
7188 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
7189 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
7190 @item @code{uw1 __Mcop2 (uw1, uw1)}
7191 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
7192 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
7193 @item @code{uw1 __MCPLHI (uw2, const)}
7194 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
7195 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
7196 @item @code{uw1 __MCPLI (uw2, const)}
7197 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
7198 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
7199 @item @code{void __MCPXIS (acc, sw1, sw1)}
7200 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
7201 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
7202 @item @code{void __MCPXIU (acc, uw1, uw1)}
7203 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
7204 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
7205 @item @code{void __MCPXRS (acc, sw1, sw1)}
7206 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
7207 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
7208 @item @code{void __MCPXRU (acc, uw1, uw1)}
7209 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
7210 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
7211 @item @code{uw1 __MCUT (acc, uw1)}
7212 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
7213 @tab @code{MCUT @var{a},@var{b},@var{c}}
7214 @item @code{uw1 __MCUTSS (acc, sw1)}
7215 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
7216 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
7217 @item @code{void __MDADDACCS (acc, acc)}
7218 @tab @code{__MDADDACCS (@var{b}, @var{a})}
7219 @tab @code{MDADDACCS @var{a},@var{b}}
7220 @item @code{void __MDASACCS (acc, acc)}
7221 @tab @code{__MDASACCS (@var{b}, @var{a})}
7222 @tab @code{MDASACCS @var{a},@var{b}}
7223 @item @code{uw2 __MDCUTSSI (acc, const)}
7224 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
7225 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
7226 @item @code{uw2 __MDPACKH (uw2, uw2)}
7227 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
7228 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
7229 @item @code{uw2 __MDROTLI (uw2, const)}
7230 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
7231 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
7232 @item @code{void __MDSUBACCS (acc, acc)}
7233 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
7234 @tab @code{MDSUBACCS @var{a},@var{b}}
7235 @item @code{void __MDUNPACKH (uw1 *, uw2)}
7236 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
7237 @tab @code{MDUNPACKH @var{a},@var{b}}
7238 @item @code{uw2 __MEXPDHD (uw1, const)}
7239 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
7240 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
7241 @item @code{uw1 __MEXPDHW (uw1, const)}
7242 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
7243 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
7244 @item @code{uw1 __MHDSETH (uw1, const)}
7245 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
7246 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
7247 @item @code{sw1 __MHDSETS (const)}
7248 @tab @code{@var{b} = __MHDSETS (@var{a})}
7249 @tab @code{MHDSETS #@var{a},@var{b}}
7250 @item @code{uw1 __MHSETHIH (uw1, const)}
7251 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
7252 @tab @code{MHSETHIH #@var{a},@var{b}}
7253 @item @code{sw1 __MHSETHIS (sw1, const)}
7254 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
7255 @tab @code{MHSETHIS #@var{a},@var{b}}
7256 @item @code{uw1 __MHSETLOH (uw1, const)}
7257 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
7258 @tab @code{MHSETLOH #@var{a},@var{b}}
7259 @item @code{sw1 __MHSETLOS (sw1, const)}
7260 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
7261 @tab @code{MHSETLOS #@var{a},@var{b}}
7262 @item @code{uw1 __MHTOB (uw2)}
7263 @tab @code{@var{b} = __MHTOB (@var{a})}
7264 @tab @code{MHTOB @var{a},@var{b}}
7265 @item @code{void __MMACHS (acc, sw1, sw1)}
7266 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
7267 @tab @code{MMACHS @var{a},@var{b},@var{c}}
7268 @item @code{void __MMACHU (acc, uw1, uw1)}
7269 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
7270 @tab @code{MMACHU @var{a},@var{b},@var{c}}
7271 @item @code{void __MMRDHS (acc, sw1, sw1)}
7272 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
7273 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
7274 @item @code{void __MMRDHU (acc, uw1, uw1)}
7275 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
7276 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
7277 @item @code{void __MMULHS (acc, sw1, sw1)}
7278 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
7279 @tab @code{MMULHS @var{a},@var{b},@var{c}}
7280 @item @code{void __MMULHU (acc, uw1, uw1)}
7281 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
7282 @tab @code{MMULHU @var{a},@var{b},@var{c}}
7283 @item @code{void __MMULXHS (acc, sw1, sw1)}
7284 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
7285 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
7286 @item @code{void __MMULXHU (acc, uw1, uw1)}
7287 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
7288 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
7289 @item @code{uw1 __MNOT (uw1)}
7290 @tab @code{@var{b} = __MNOT (@var{a})}
7291 @tab @code{MNOT @var{a},@var{b}}
7292 @item @code{uw1 __MOR (uw1, uw1)}
7293 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
7294 @tab @code{MOR @var{a},@var{b},@var{c}}
7295 @item @code{uw1 __MPACKH (uh, uh)}
7296 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
7297 @tab @code{MPACKH @var{a},@var{b},@var{c}}
7298 @item @code{sw2 __MQADDHSS (sw2, sw2)}
7299 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
7300 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
7301 @item @code{uw2 __MQADDHUS (uw2, uw2)}
7302 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
7303 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
7304 @item @code{void __MQCPXIS (acc, sw2, sw2)}
7305 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
7306 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
7307 @item @code{void __MQCPXIU (acc, uw2, uw2)}
7308 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
7309 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
7310 @item @code{void __MQCPXRS (acc, sw2, sw2)}
7311 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
7312 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
7313 @item @code{void __MQCPXRU (acc, uw2, uw2)}
7314 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
7315 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
7316 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
7317 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
7318 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
7319 @item @code{sw2 __MQLMTHS (sw2, sw2)}
7320 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
7321 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
7322 @item @code{void __MQMACHS (acc, sw2, sw2)}
7323 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
7324 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
7325 @item @code{void __MQMACHU (acc, uw2, uw2)}
7326 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
7327 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
7328 @item @code{void __MQMACXHS (acc, sw2, sw2)}
7329 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
7330 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
7331 @item @code{void __MQMULHS (acc, sw2, sw2)}
7332 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
7333 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
7334 @item @code{void __MQMULHU (acc, uw2, uw2)}
7335 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
7336 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
7337 @item @code{void __MQMULXHS (acc, sw2, sw2)}
7338 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
7339 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
7340 @item @code{void __MQMULXHU (acc, uw2, uw2)}
7341 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
7342 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
7343 @item @code{sw2 __MQSATHS (sw2, sw2)}
7344 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
7345 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
7346 @item @code{uw2 __MQSLLHI (uw2, int)}
7347 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
7348 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
7349 @item @code{sw2 __MQSRAHI (sw2, int)}
7350 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
7351 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
7352 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
7353 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
7354 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
7355 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
7356 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
7357 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
7358 @item @code{void __MQXMACHS (acc, sw2, sw2)}
7359 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
7360 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
7361 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
7362 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
7363 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
7364 @item @code{uw1 __MRDACC (acc)}
7365 @tab @code{@var{b} = __MRDACC (@var{a})}
7366 @tab @code{MRDACC @var{a},@var{b}}
7367 @item @code{uw1 __MRDACCG (acc)}
7368 @tab @code{@var{b} = __MRDACCG (@var{a})}
7369 @tab @code{MRDACCG @var{a},@var{b}}
7370 @item @code{uw1 __MROTLI (uw1, const)}
7371 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
7372 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
7373 @item @code{uw1 __MROTRI (uw1, const)}
7374 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
7375 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
7376 @item @code{sw1 __MSATHS (sw1, sw1)}
7377 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
7378 @tab @code{MSATHS @var{a},@var{b},@var{c}}
7379 @item @code{uw1 __MSATHU (uw1, uw1)}
7380 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
7381 @tab @code{MSATHU @var{a},@var{b},@var{c}}
7382 @item @code{uw1 __MSLLHI (uw1, const)}
7383 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
7384 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
7385 @item @code{sw1 __MSRAHI (sw1, const)}
7386 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
7387 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
7388 @item @code{uw1 __MSRLHI (uw1, const)}
7389 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
7390 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
7391 @item @code{void __MSUBACCS (acc, acc)}
7392 @tab @code{__MSUBACCS (@var{b}, @var{a})}
7393 @tab @code{MSUBACCS @var{a},@var{b}}
7394 @item @code{sw1 __MSUBHSS (sw1, sw1)}
7395 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
7396 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
7397 @item @code{uw1 __MSUBHUS (uw1, uw1)}
7398 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
7399 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
7400 @item @code{void __MTRAP (void)}
7401 @tab @code{__MTRAP ()}
7403 @item @code{uw2 __MUNPACKH (uw1)}
7404 @tab @code{@var{b} = __MUNPACKH (@var{a})}
7405 @tab @code{MUNPACKH @var{a},@var{b}}
7406 @item @code{uw1 __MWCUT (uw2, uw1)}
7407 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
7408 @tab @code{MWCUT @var{a},@var{b},@var{c}}
7409 @item @code{void __MWTACC (acc, uw1)}
7410 @tab @code{__MWTACC (@var{b}, @var{a})}
7411 @tab @code{MWTACC @var{a},@var{b}}
7412 @item @code{void __MWTACCG (acc, uw1)}
7413 @tab @code{__MWTACCG (@var{b}, @var{a})}
7414 @tab @code{MWTACCG @var{a},@var{b}}
7415 @item @code{uw1 __MXOR (uw1, uw1)}
7416 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
7417 @tab @code{MXOR @var{a},@var{b},@var{c}}
7420 @node Raw read/write Functions
7421 @subsubsection Raw read/write Functions
7423 This sections describes built-in functions related to read and write
7424 instructions to access memory. These functions generate
7425 @code{membar} instructions to flush the I/O load and stores where
7426 appropriate, as described in Fujitsu's manual described above.
7430 @item unsigned char __builtin_read8 (void *@var{data})
7431 @item unsigned short __builtin_read16 (void *@var{data})
7432 @item unsigned long __builtin_read32 (void *@var{data})
7433 @item unsigned long long __builtin_read64 (void *@var{data})
7435 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
7436 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
7437 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
7438 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
7441 @node Other Built-in Functions
7442 @subsubsection Other Built-in Functions
7444 This section describes built-in functions that are not named after
7445 a specific FR-V instruction.
7448 @item sw2 __IACCreadll (iacc @var{reg})
7449 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
7450 for future expansion and must be 0.
7452 @item sw1 __IACCreadl (iacc @var{reg})
7453 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
7454 Other values of @var{reg} are rejected as invalid.
7456 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
7457 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
7458 is reserved for future expansion and must be 0.
7460 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
7461 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
7462 is 1. Other values of @var{reg} are rejected as invalid.
7464 @item void __data_prefetch0 (const void *@var{x})
7465 Use the @code{dcpl} instruction to load the contents of address @var{x}
7466 into the data cache.
7468 @item void __data_prefetch (const void *@var{x})
7469 Use the @code{nldub} instruction to load the contents of address @var{x}
7470 into the data cache. The instruction will be issued in slot I1@.
7473 @node X86 Built-in Functions
7474 @subsection X86 Built-in Functions
7476 These built-in functions are available for the i386 and x86-64 family
7477 of computers, depending on the command-line switches used.
7479 Note that, if you specify command-line switches such as @option{-msse},
7480 the compiler could use the extended instruction sets even if the built-ins
7481 are not used explicitly in the program. For this reason, applications
7482 which perform runtime CPU detection must compile separate files for each
7483 supported architecture, using the appropriate flags. In particular,
7484 the file containing the CPU detection code should be compiled without
7487 The following machine modes are available for use with MMX built-in functions
7488 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
7489 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
7490 vector of eight 8-bit integers. Some of the built-in functions operate on
7491 MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode.
7493 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
7494 of two 32-bit floating point values.
7496 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
7497 floating point values. Some instructions use a vector of four 32-bit
7498 integers, these use @code{V4SI}. Finally, some instructions operate on an
7499 entire vector register, interpreting it as a 128-bit integer, these use mode
7502 In 64-bit mode, the x86-64 family of processors uses additional built-in
7503 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
7504 floating point and @code{TC} 128-bit complex floating point values.
7506 The following floating point built-in functions are available in 64-bit
7507 mode. All of them implement the function that is part of the name.
7510 __float128 __builtin_fabsq (__float128)
7511 __float128 __builtin_copysignq (__float128, __float128)
7514 The following floating point built-in functions are made available in the
7518 @item __float128 __builtin_infq (void)
7519 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
7522 The following built-in functions are made available by @option{-mmmx}.
7523 All of them generate the machine instruction that is part of the name.
7526 v8qi __builtin_ia32_paddb (v8qi, v8qi)
7527 v4hi __builtin_ia32_paddw (v4hi, v4hi)
7528 v2si __builtin_ia32_paddd (v2si, v2si)
7529 v8qi __builtin_ia32_psubb (v8qi, v8qi)
7530 v4hi __builtin_ia32_psubw (v4hi, v4hi)
7531 v2si __builtin_ia32_psubd (v2si, v2si)
7532 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
7533 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
7534 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
7535 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
7536 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
7537 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
7538 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
7539 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
7540 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
7541 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
7542 di __builtin_ia32_pand (di, di)
7543 di __builtin_ia32_pandn (di,di)
7544 di __builtin_ia32_por (di, di)
7545 di __builtin_ia32_pxor (di, di)
7546 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
7547 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
7548 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
7549 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
7550 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
7551 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
7552 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
7553 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
7554 v2si __builtin_ia32_punpckhdq (v2si, v2si)
7555 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
7556 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
7557 v2si __builtin_ia32_punpckldq (v2si, v2si)
7558 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
7559 v4hi __builtin_ia32_packssdw (v2si, v2si)
7560 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
7562 v4hi __builtin_ia32_psllw (v4hi, v4hi)
7563 v2si __builtin_ia32_pslld (v2si, v2si)
7564 v1di __builtin_ia32_psllq (v1di, v1di)
7565 v4hi __builtin_ia32_psrlw (v4hi, v4hi)
7566 v2si __builtin_ia32_psrld (v2si, v2si)
7567 v1di __builtin_ia32_psrlq (v1di, v1di)
7568 v4hi __builtin_ia32_psraw (v4hi, v4hi)
7569 v2si __builtin_ia32_psrad (v2si, v2si)
7570 v4hi __builtin_ia32_psllwi (v4hi, int)
7571 v2si __builtin_ia32_pslldi (v2si, int)
7572 v1di __builtin_ia32_psllqi (v1di, int)
7573 v4hi __builtin_ia32_psrlwi (v4hi, int)
7574 v2si __builtin_ia32_psrldi (v2si, int)
7575 v1di __builtin_ia32_psrlqi (v1di, int)
7576 v4hi __builtin_ia32_psrawi (v4hi, int)
7577 v2si __builtin_ia32_psradi (v2si, int)
7581 The following built-in functions are made available either with
7582 @option{-msse}, or with a combination of @option{-m3dnow} and
7583 @option{-march=athlon}. All of them generate the machine
7584 instruction that is part of the name.
7587 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
7588 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
7589 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
7590 v1di __builtin_ia32_psadbw (v8qi, v8qi)
7591 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
7592 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
7593 v8qi __builtin_ia32_pminub (v8qi, v8qi)
7594 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
7595 int __builtin_ia32_pextrw (v4hi, int)
7596 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
7597 int __builtin_ia32_pmovmskb (v8qi)
7598 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
7599 void __builtin_ia32_movntq (di *, di)
7600 void __builtin_ia32_sfence (void)
7603 The following built-in functions are available when @option{-msse} is used.
7604 All of them generate the machine instruction that is part of the name.
7607 int __builtin_ia32_comieq (v4sf, v4sf)
7608 int __builtin_ia32_comineq (v4sf, v4sf)
7609 int __builtin_ia32_comilt (v4sf, v4sf)
7610 int __builtin_ia32_comile (v4sf, v4sf)
7611 int __builtin_ia32_comigt (v4sf, v4sf)
7612 int __builtin_ia32_comige (v4sf, v4sf)
7613 int __builtin_ia32_ucomieq (v4sf, v4sf)
7614 int __builtin_ia32_ucomineq (v4sf, v4sf)
7615 int __builtin_ia32_ucomilt (v4sf, v4sf)
7616 int __builtin_ia32_ucomile (v4sf, v4sf)
7617 int __builtin_ia32_ucomigt (v4sf, v4sf)
7618 int __builtin_ia32_ucomige (v4sf, v4sf)
7619 v4sf __builtin_ia32_addps (v4sf, v4sf)
7620 v4sf __builtin_ia32_subps (v4sf, v4sf)
7621 v4sf __builtin_ia32_mulps (v4sf, v4sf)
7622 v4sf __builtin_ia32_divps (v4sf, v4sf)
7623 v4sf __builtin_ia32_addss (v4sf, v4sf)
7624 v4sf __builtin_ia32_subss (v4sf, v4sf)
7625 v4sf __builtin_ia32_mulss (v4sf, v4sf)
7626 v4sf __builtin_ia32_divss (v4sf, v4sf)
7627 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
7628 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
7629 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
7630 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
7631 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
7632 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
7633 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
7634 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
7635 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
7636 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
7637 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
7638 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
7639 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
7640 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
7641 v4si __builtin_ia32_cmpless (v4sf, v4sf)
7642 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
7643 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
7644 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
7645 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
7646 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
7647 v4sf __builtin_ia32_maxps (v4sf, v4sf)
7648 v4sf __builtin_ia32_maxss (v4sf, v4sf)
7649 v4sf __builtin_ia32_minps (v4sf, v4sf)
7650 v4sf __builtin_ia32_minss (v4sf, v4sf)
7651 v4sf __builtin_ia32_andps (v4sf, v4sf)
7652 v4sf __builtin_ia32_andnps (v4sf, v4sf)
7653 v4sf __builtin_ia32_orps (v4sf, v4sf)
7654 v4sf __builtin_ia32_xorps (v4sf, v4sf)
7655 v4sf __builtin_ia32_movss (v4sf, v4sf)
7656 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
7657 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
7658 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
7659 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
7660 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
7661 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
7662 v2si __builtin_ia32_cvtps2pi (v4sf)
7663 int __builtin_ia32_cvtss2si (v4sf)
7664 v2si __builtin_ia32_cvttps2pi (v4sf)
7665 int __builtin_ia32_cvttss2si (v4sf)
7666 v4sf __builtin_ia32_rcpps (v4sf)
7667 v4sf __builtin_ia32_rsqrtps (v4sf)
7668 v4sf __builtin_ia32_sqrtps (v4sf)
7669 v4sf __builtin_ia32_rcpss (v4sf)
7670 v4sf __builtin_ia32_rsqrtss (v4sf)
7671 v4sf __builtin_ia32_sqrtss (v4sf)
7672 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
7673 void __builtin_ia32_movntps (float *, v4sf)
7674 int __builtin_ia32_movmskps (v4sf)
7677 The following built-in functions are available when @option{-msse} is used.
7680 @item v4sf __builtin_ia32_loadaps (float *)
7681 Generates the @code{movaps} machine instruction as a load from memory.
7682 @item void __builtin_ia32_storeaps (float *, v4sf)
7683 Generates the @code{movaps} machine instruction as a store to memory.
7684 @item v4sf __builtin_ia32_loadups (float *)
7685 Generates the @code{movups} machine instruction as a load from memory.
7686 @item void __builtin_ia32_storeups (float *, v4sf)
7687 Generates the @code{movups} machine instruction as a store to memory.
7688 @item v4sf __builtin_ia32_loadsss (float *)
7689 Generates the @code{movss} machine instruction as a load from memory.
7690 @item void __builtin_ia32_storess (float *, v4sf)
7691 Generates the @code{movss} machine instruction as a store to memory.
7692 @item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)
7693 Generates the @code{movhps} machine instruction as a load from memory.
7694 @item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)
7695 Generates the @code{movlps} machine instruction as a load from memory
7696 @item void __builtin_ia32_storehps (v2sf *, v4sf)
7697 Generates the @code{movhps} machine instruction as a store to memory.
7698 @item void __builtin_ia32_storelps (v2sf *, v4sf)
7699 Generates the @code{movlps} machine instruction as a store to memory.
7702 The following built-in functions are available when @option{-msse2} is used.
7703 All of them generate the machine instruction that is part of the name.
7706 int __builtin_ia32_comisdeq (v2df, v2df)
7707 int __builtin_ia32_comisdlt (v2df, v2df)
7708 int __builtin_ia32_comisdle (v2df, v2df)
7709 int __builtin_ia32_comisdgt (v2df, v2df)
7710 int __builtin_ia32_comisdge (v2df, v2df)
7711 int __builtin_ia32_comisdneq (v2df, v2df)
7712 int __builtin_ia32_ucomisdeq (v2df, v2df)
7713 int __builtin_ia32_ucomisdlt (v2df, v2df)
7714 int __builtin_ia32_ucomisdle (v2df, v2df)
7715 int __builtin_ia32_ucomisdgt (v2df, v2df)
7716 int __builtin_ia32_ucomisdge (v2df, v2df)
7717 int __builtin_ia32_ucomisdneq (v2df, v2df)
7718 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
7719 v2df __builtin_ia32_cmpltpd (v2df, v2df)
7720 v2df __builtin_ia32_cmplepd (v2df, v2df)
7721 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
7722 v2df __builtin_ia32_cmpgepd (v2df, v2df)
7723 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
7724 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
7725 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
7726 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
7727 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
7728 v2df __builtin_ia32_cmpngepd (v2df, v2df)
7729 v2df __builtin_ia32_cmpordpd (v2df, v2df)
7730 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
7731 v2df __builtin_ia32_cmpltsd (v2df, v2df)
7732 v2df __builtin_ia32_cmplesd (v2df, v2df)
7733 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
7734 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
7735 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
7736 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
7737 v2df __builtin_ia32_cmpordsd (v2df, v2df)
7738 v2di __builtin_ia32_paddq (v2di, v2di)
7739 v2di __builtin_ia32_psubq (v2di, v2di)
7740 v2df __builtin_ia32_addpd (v2df, v2df)
7741 v2df __builtin_ia32_subpd (v2df, v2df)
7742 v2df __builtin_ia32_mulpd (v2df, v2df)
7743 v2df __builtin_ia32_divpd (v2df, v2df)
7744 v2df __builtin_ia32_addsd (v2df, v2df)
7745 v2df __builtin_ia32_subsd (v2df, v2df)
7746 v2df __builtin_ia32_mulsd (v2df, v2df)
7747 v2df __builtin_ia32_divsd (v2df, v2df)
7748 v2df __builtin_ia32_minpd (v2df, v2df)
7749 v2df __builtin_ia32_maxpd (v2df, v2df)
7750 v2df __builtin_ia32_minsd (v2df, v2df)
7751 v2df __builtin_ia32_maxsd (v2df, v2df)
7752 v2df __builtin_ia32_andpd (v2df, v2df)
7753 v2df __builtin_ia32_andnpd (v2df, v2df)
7754 v2df __builtin_ia32_orpd (v2df, v2df)
7755 v2df __builtin_ia32_xorpd (v2df, v2df)
7756 v2df __builtin_ia32_movsd (v2df, v2df)
7757 v2df __builtin_ia32_unpckhpd (v2df, v2df)
7758 v2df __builtin_ia32_unpcklpd (v2df, v2df)
7759 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
7760 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
7761 v4si __builtin_ia32_paddd128 (v4si, v4si)
7762 v2di __builtin_ia32_paddq128 (v2di, v2di)
7763 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
7764 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
7765 v4si __builtin_ia32_psubd128 (v4si, v4si)
7766 v2di __builtin_ia32_psubq128 (v2di, v2di)
7767 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
7768 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
7769 v2di __builtin_ia32_pand128 (v2di, v2di)
7770 v2di __builtin_ia32_pandn128 (v2di, v2di)
7771 v2di __builtin_ia32_por128 (v2di, v2di)
7772 v2di __builtin_ia32_pxor128 (v2di, v2di)
7773 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
7774 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
7775 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
7776 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
7777 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
7778 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
7779 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
7780 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
7781 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
7782 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
7783 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
7784 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
7785 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
7786 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
7787 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
7788 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
7789 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
7790 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
7791 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
7792 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
7793 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi)
7794 v8hi __builtin_ia32_packssdw128 (v4si, v4si)
7795 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi)
7796 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
7797 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
7798 v2df __builtin_ia32_loadupd (double *)
7799 void __builtin_ia32_storeupd (double *, v2df)
7800 v2df __builtin_ia32_loadhpd (v2df, double const *)
7801 v2df __builtin_ia32_loadlpd (v2df, double const *)
7802 int __builtin_ia32_movmskpd (v2df)
7803 int __builtin_ia32_pmovmskb128 (v16qi)
7804 void __builtin_ia32_movnti (int *, int)
7805 void __builtin_ia32_movntpd (double *, v2df)
7806 void __builtin_ia32_movntdq (v2df *, v2df)
7807 v4si __builtin_ia32_pshufd (v4si, int)
7808 v8hi __builtin_ia32_pshuflw (v8hi, int)
7809 v8hi __builtin_ia32_pshufhw (v8hi, int)
7810 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
7811 v2df __builtin_ia32_sqrtpd (v2df)
7812 v2df __builtin_ia32_sqrtsd (v2df)
7813 v2df __builtin_ia32_shufpd (v2df, v2df, int)
7814 v2df __builtin_ia32_cvtdq2pd (v4si)
7815 v4sf __builtin_ia32_cvtdq2ps (v4si)
7816 v4si __builtin_ia32_cvtpd2dq (v2df)
7817 v2si __builtin_ia32_cvtpd2pi (v2df)
7818 v4sf __builtin_ia32_cvtpd2ps (v2df)
7819 v4si __builtin_ia32_cvttpd2dq (v2df)
7820 v2si __builtin_ia32_cvttpd2pi (v2df)
7821 v2df __builtin_ia32_cvtpi2pd (v2si)
7822 int __builtin_ia32_cvtsd2si (v2df)
7823 int __builtin_ia32_cvttsd2si (v2df)
7824 long long __builtin_ia32_cvtsd2si64 (v2df)
7825 long long __builtin_ia32_cvttsd2si64 (v2df)
7826 v4si __builtin_ia32_cvtps2dq (v4sf)
7827 v2df __builtin_ia32_cvtps2pd (v4sf)
7828 v4si __builtin_ia32_cvttps2dq (v4sf)
7829 v2df __builtin_ia32_cvtsi2sd (v2df, int)
7830 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
7831 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
7832 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
7833 void __builtin_ia32_clflush (const void *)
7834 void __builtin_ia32_lfence (void)
7835 void __builtin_ia32_mfence (void)
7836 v16qi __builtin_ia32_loaddqu (const char *)
7837 void __builtin_ia32_storedqu (char *, v16qi)
7838 v1di __builtin_ia32_pmuludq (v2si, v2si)
7839 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
7840 v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
7841 v4si __builtin_ia32_pslld128 (v4si, v4si)
7842 v2di __builtin_ia32_psllq128 (v2di, v2di)
7843 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
7844 v4si __builtin_ia32_psrld128 (v4si, v4si)
7845 v2di __builtin_ia32_psrlq128 (v2di, v2di)
7846 v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
7847 v4si __builtin_ia32_psrad128 (v4si, v4si)
7848 v2di __builtin_ia32_pslldqi128 (v2di, int)
7849 v8hi __builtin_ia32_psllwi128 (v8hi, int)
7850 v4si __builtin_ia32_pslldi128 (v4si, int)
7851 v2di __builtin_ia32_psllqi128 (v2di, int)
7852 v2di __builtin_ia32_psrldqi128 (v2di, int)
7853 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
7854 v4si __builtin_ia32_psrldi128 (v4si, int)
7855 v2di __builtin_ia32_psrlqi128 (v2di, int)
7856 v8hi __builtin_ia32_psrawi128 (v8hi, int)
7857 v4si __builtin_ia32_psradi128 (v4si, int)
7858 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
7861 The following built-in functions are available when @option{-msse3} is used.
7862 All of them generate the machine instruction that is part of the name.
7865 v2df __builtin_ia32_addsubpd (v2df, v2df)
7866 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
7867 v2df __builtin_ia32_haddpd (v2df, v2df)
7868 v4sf __builtin_ia32_haddps (v4sf, v4sf)
7869 v2df __builtin_ia32_hsubpd (v2df, v2df)
7870 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
7871 v16qi __builtin_ia32_lddqu (char const *)
7872 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
7873 v2df __builtin_ia32_movddup (v2df)
7874 v4sf __builtin_ia32_movshdup (v4sf)
7875 v4sf __builtin_ia32_movsldup (v4sf)
7876 void __builtin_ia32_mwait (unsigned int, unsigned int)
7879 The following built-in functions are available when @option{-msse3} is used.
7882 @item v2df __builtin_ia32_loadddup (double const *)
7883 Generates the @code{movddup} machine instruction as a load from memory.
7886 The following built-in functions are available when @option{-mssse3} is used.
7887 All of them generate the machine instruction that is part of the name
7891 v2si __builtin_ia32_phaddd (v2si, v2si)
7892 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
7893 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
7894 v2si __builtin_ia32_phsubd (v2si, v2si)
7895 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
7896 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
7897 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi)
7898 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
7899 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
7900 v8qi __builtin_ia32_psignb (v8qi, v8qi)
7901 v2si __builtin_ia32_psignd (v2si, v2si)
7902 v4hi __builtin_ia32_psignw (v4hi, v4hi)
7903 v1di __builtin_ia32_palignr (v1di, v1di, int)
7904 v8qi __builtin_ia32_pabsb (v8qi)
7905 v2si __builtin_ia32_pabsd (v2si)
7906 v4hi __builtin_ia32_pabsw (v4hi)
7909 The following built-in functions are available when @option{-mssse3} is used.
7910 All of them generate the machine instruction that is part of the name
7914 v4si __builtin_ia32_phaddd128 (v4si, v4si)
7915 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
7916 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
7917 v4si __builtin_ia32_phsubd128 (v4si, v4si)
7918 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
7919 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
7920 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
7921 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
7922 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
7923 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
7924 v4si __builtin_ia32_psignd128 (v4si, v4si)
7925 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
7926 v2di __builtin_ia32_palignr128 (v2di, v2di, int)
7927 v16qi __builtin_ia32_pabsb128 (v16qi)
7928 v4si __builtin_ia32_pabsd128 (v4si)
7929 v8hi __builtin_ia32_pabsw128 (v8hi)
7932 The following built-in functions are available when @option{-msse4.1} is
7933 used. All of them generate the machine instruction that is part of the
7937 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
7938 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
7939 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
7940 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
7941 v2df __builtin_ia32_dppd (v2df, v2df, const int)
7942 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
7943 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
7944 v2di __builtin_ia32_movntdqa (v2di *);
7945 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
7946 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
7947 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
7948 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
7949 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
7950 v8hi __builtin_ia32_phminposuw128 (v8hi)
7951 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
7952 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
7953 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
7954 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
7955 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
7956 v4si __builtin_ia32_pminsd128 (v4si, v4si)
7957 v4si __builtin_ia32_pminud128 (v4si, v4si)
7958 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
7959 v4si __builtin_ia32_pmovsxbd128 (v16qi)
7960 v2di __builtin_ia32_pmovsxbq128 (v16qi)
7961 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
7962 v2di __builtin_ia32_pmovsxdq128 (v4si)
7963 v4si __builtin_ia32_pmovsxwd128 (v8hi)
7964 v2di __builtin_ia32_pmovsxwq128 (v8hi)
7965 v4si __builtin_ia32_pmovzxbd128 (v16qi)
7966 v2di __builtin_ia32_pmovzxbq128 (v16qi)
7967 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
7968 v2di __builtin_ia32_pmovzxdq128 (v4si)
7969 v4si __builtin_ia32_pmovzxwd128 (v8hi)
7970 v2di __builtin_ia32_pmovzxwq128 (v8hi)
7971 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
7972 v4si __builtin_ia32_pmulld128 (v4si, v4si)
7973 int __builtin_ia32_ptestc128 (v2di, v2di)
7974 int __builtin_ia32_ptestnzc128 (v2di, v2di)
7975 int __builtin_ia32_ptestz128 (v2di, v2di)
7976 v2df __builtin_ia32_roundpd (v2df, const int)
7977 v4sf __builtin_ia32_roundps (v4sf, const int)
7978 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
7979 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
7982 The following built-in functions are available when @option{-msse4.1} is
7986 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
7987 Generates the @code{insertps} machine instruction.
7988 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
7989 Generates the @code{pextrb} machine instruction.
7990 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
7991 Generates the @code{pinsrb} machine instruction.
7992 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
7993 Generates the @code{pinsrd} machine instruction.
7994 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
7995 Generates the @code{pinsrq} machine instruction in 64bit mode.
7998 The following built-in functions are changed to generate new SSE4.1
7999 instructions when @option{-msse4.1} is used.
8002 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
8003 Generates the @code{extractps} machine instruction.
8004 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
8005 Generates the @code{pextrd} machine instruction.
8006 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
8007 Generates the @code{pextrq} machine instruction in 64bit mode.
8010 The following built-in functions are available when @option{-msse4.2} is
8011 used. All of them generate the machine instruction that is part of the
8015 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
8016 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
8017 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
8018 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
8019 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
8020 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
8021 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
8022 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
8023 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
8024 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
8025 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
8026 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
8027 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
8028 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
8029 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
8032 The following built-in functions are available when @option{-msse4.2} is
8036 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
8037 Generates the @code{crc32b} machine instruction.
8038 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
8039 Generates the @code{crc32w} machine instruction.
8040 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
8041 Generates the @code{crc32l} machine instruction.
8042 @item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)
8045 The following built-in functions are changed to generate new SSE4.2
8046 instructions when @option{-msse4.2} is used.
8049 @item int __builtin_popcount (unsigned int)
8050 Generates the @code{popcntl} machine instruction.
8051 @item int __builtin_popcountl (unsigned long)
8052 Generates the @code{popcntl} or @code{popcntq} machine instruction,
8053 depending on the size of @code{unsigned long}.
8054 @item int __builtin_popcountll (unsigned long long)
8055 Generates the @code{popcntq} machine instruction.
8058 The following built-in functions are available when @option{-maes} is
8059 used. All of them generate the machine instruction that is part of the
8063 v2di __builtin_ia32_aesenc128 (v2di, v2di)
8064 v2di __builtin_ia32_aesenclast128 (v2di, v2di)
8065 v2di __builtin_ia32_aesdec128 (v2di, v2di)
8066 v2di __builtin_ia32_aesdeclast128 (v2di, v2di)
8067 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int)
8068 v2di __builtin_ia32_aesimc128 (v2di)
8071 The following built-in function is available when @option{-mpclmul} is
8075 @item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)
8076 Generates the @code{pclmulqdq} machine instruction.
8079 The following built-in functions are available when @option{-msse4a} is used.
8080 All of them generate the machine instruction that is part of the name.
8083 void __builtin_ia32_movntsd (double *, v2df)
8084 void __builtin_ia32_movntss (float *, v4sf)
8085 v2di __builtin_ia32_extrq (v2di, v16qi)
8086 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
8087 v2di __builtin_ia32_insertq (v2di, v2di)
8088 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
8091 The following built-in functions are available when @option{-msse5} is used.
8092 All of them generate the machine instruction that is part of the name
8096 v2df __builtin_ia32_comeqpd (v2df, v2df)
8097 v2df __builtin_ia32_comeqps (v2df, v2df)
8098 v4sf __builtin_ia32_comeqsd (v4sf, v4sf)
8099 v4sf __builtin_ia32_comeqss (v4sf, v4sf)
8100 v2df __builtin_ia32_comfalsepd (v2df, v2df)
8101 v2df __builtin_ia32_comfalseps (v2df, v2df)
8102 v4sf __builtin_ia32_comfalsesd (v4sf, v4sf)
8103 v4sf __builtin_ia32_comfalsess (v4sf, v4sf)
8104 v2df __builtin_ia32_comgepd (v2df, v2df)
8105 v2df __builtin_ia32_comgeps (v2df, v2df)
8106 v4sf __builtin_ia32_comgesd (v4sf, v4sf)
8107 v4sf __builtin_ia32_comgess (v4sf, v4sf)
8108 v2df __builtin_ia32_comgtpd (v2df, v2df)
8109 v2df __builtin_ia32_comgtps (v2df, v2df)
8110 v4sf __builtin_ia32_comgtsd (v4sf, v4sf)
8111 v4sf __builtin_ia32_comgtss (v4sf, v4sf)
8112 v2df __builtin_ia32_comlepd (v2df, v2df)
8113 v2df __builtin_ia32_comleps (v2df, v2df)
8114 v4sf __builtin_ia32_comlesd (v4sf, v4sf)
8115 v4sf __builtin_ia32_comless (v4sf, v4sf)
8116 v2df __builtin_ia32_comltpd (v2df, v2df)
8117 v2df __builtin_ia32_comltps (v2df, v2df)
8118 v4sf __builtin_ia32_comltsd (v4sf, v4sf)
8119 v4sf __builtin_ia32_comltss (v4sf, v4sf)
8120 v2df __builtin_ia32_comnepd (v2df, v2df)
8121 v2df __builtin_ia32_comneps (v2df, v2df)
8122 v4sf __builtin_ia32_comnesd (v4sf, v4sf)
8123 v4sf __builtin_ia32_comness (v4sf, v4sf)
8124 v2df __builtin_ia32_comordpd (v2df, v2df)
8125 v2df __builtin_ia32_comordps (v2df, v2df)
8126 v4sf __builtin_ia32_comordsd (v4sf, v4sf)
8127 v4sf __builtin_ia32_comordss (v4sf, v4sf)
8128 v2df __builtin_ia32_comtruepd (v2df, v2df)
8129 v2df __builtin_ia32_comtrueps (v2df, v2df)
8130 v4sf __builtin_ia32_comtruesd (v4sf, v4sf)
8131 v4sf __builtin_ia32_comtruess (v4sf, v4sf)
8132 v2df __builtin_ia32_comueqpd (v2df, v2df)
8133 v2df __builtin_ia32_comueqps (v2df, v2df)
8134 v4sf __builtin_ia32_comueqsd (v4sf, v4sf)
8135 v4sf __builtin_ia32_comueqss (v4sf, v4sf)
8136 v2df __builtin_ia32_comugepd (v2df, v2df)
8137 v2df __builtin_ia32_comugeps (v2df, v2df)
8138 v4sf __builtin_ia32_comugesd (v4sf, v4sf)
8139 v4sf __builtin_ia32_comugess (v4sf, v4sf)
8140 v2df __builtin_ia32_comugtpd (v2df, v2df)
8141 v2df __builtin_ia32_comugtps (v2df, v2df)
8142 v4sf __builtin_ia32_comugtsd (v4sf, v4sf)
8143 v4sf __builtin_ia32_comugtss (v4sf, v4sf)
8144 v2df __builtin_ia32_comulepd (v2df, v2df)
8145 v2df __builtin_ia32_comuleps (v2df, v2df)
8146 v4sf __builtin_ia32_comulesd (v4sf, v4sf)
8147 v4sf __builtin_ia32_comuless (v4sf, v4sf)
8148 v2df __builtin_ia32_comultpd (v2df, v2df)
8149 v2df __builtin_ia32_comultps (v2df, v2df)
8150 v4sf __builtin_ia32_comultsd (v4sf, v4sf)
8151 v4sf __builtin_ia32_comultss (v4sf, v4sf)
8152 v2df __builtin_ia32_comunepd (v2df, v2df)
8153 v2df __builtin_ia32_comuneps (v2df, v2df)
8154 v4sf __builtin_ia32_comunesd (v4sf, v4sf)
8155 v4sf __builtin_ia32_comuness (v4sf, v4sf)
8156 v2df __builtin_ia32_comunordpd (v2df, v2df)
8157 v2df __builtin_ia32_comunordps (v2df, v2df)
8158 v4sf __builtin_ia32_comunordsd (v4sf, v4sf)
8159 v4sf __builtin_ia32_comunordss (v4sf, v4sf)
8160 v2df __builtin_ia32_fmaddpd (v2df, v2df, v2df)
8161 v4sf __builtin_ia32_fmaddps (v4sf, v4sf, v4sf)
8162 v2df __builtin_ia32_fmaddsd (v2df, v2df, v2df)
8163 v4sf __builtin_ia32_fmaddss (v4sf, v4sf, v4sf)
8164 v2df __builtin_ia32_fmsubpd (v2df, v2df, v2df)
8165 v4sf __builtin_ia32_fmsubps (v4sf, v4sf, v4sf)
8166 v2df __builtin_ia32_fmsubsd (v2df, v2df, v2df)
8167 v4sf __builtin_ia32_fmsubss (v4sf, v4sf, v4sf)
8168 v2df __builtin_ia32_fnmaddpd (v2df, v2df, v2df)
8169 v4sf __builtin_ia32_fnmaddps (v4sf, v4sf, v4sf)
8170 v2df __builtin_ia32_fnmaddsd (v2df, v2df, v2df)
8171 v4sf __builtin_ia32_fnmaddss (v4sf, v4sf, v4sf)
8172 v2df __builtin_ia32_fnmsubpd (v2df, v2df, v2df)
8173 v4sf __builtin_ia32_fnmsubps (v4sf, v4sf, v4sf)
8174 v2df __builtin_ia32_fnmsubsd (v2df, v2df, v2df)
8175 v4sf __builtin_ia32_fnmsubss (v4sf, v4sf, v4sf)
8176 v2df __builtin_ia32_frczpd (v2df)
8177 v4sf __builtin_ia32_frczps (v4sf)
8178 v2df __builtin_ia32_frczsd (v2df, v2df)
8179 v4sf __builtin_ia32_frczss (v4sf, v4sf)
8180 v2di __builtin_ia32_pcmov (v2di, v2di, v2di)
8181 v2di __builtin_ia32_pcmov_v2di (v2di, v2di, v2di)
8182 v4si __builtin_ia32_pcmov_v4si (v4si, v4si, v4si)
8183 v8hi __builtin_ia32_pcmov_v8hi (v8hi, v8hi, v8hi)
8184 v16qi __builtin_ia32_pcmov_v16qi (v16qi, v16qi, v16qi)
8185 v2df __builtin_ia32_pcmov_v2df (v2df, v2df, v2df)
8186 v4sf __builtin_ia32_pcmov_v4sf (v4sf, v4sf, v4sf)
8187 v16qi __builtin_ia32_pcomeqb (v16qi, v16qi)
8188 v8hi __builtin_ia32_pcomeqw (v8hi, v8hi)
8189 v4si __builtin_ia32_pcomeqd (v4si, v4si)
8190 v2di __builtin_ia32_pcomeqq (v2di, v2di)
8191 v16qi __builtin_ia32_pcomequb (v16qi, v16qi)
8192 v4si __builtin_ia32_pcomequd (v4si, v4si)
8193 v2di __builtin_ia32_pcomequq (v2di, v2di)
8194 v8hi __builtin_ia32_pcomequw (v8hi, v8hi)
8195 v8hi __builtin_ia32_pcomeqw (v8hi, v8hi)
8196 v16qi __builtin_ia32_pcomfalseb (v16qi, v16qi)
8197 v4si __builtin_ia32_pcomfalsed (v4si, v4si)
8198 v2di __builtin_ia32_pcomfalseq (v2di, v2di)
8199 v16qi __builtin_ia32_pcomfalseub (v16qi, v16qi)
8200 v4si __builtin_ia32_pcomfalseud (v4si, v4si)
8201 v2di __builtin_ia32_pcomfalseuq (v2di, v2di)
8202 v8hi __builtin_ia32_pcomfalseuw (v8hi, v8hi)
8203 v8hi __builtin_ia32_pcomfalsew (v8hi, v8hi)
8204 v16qi __builtin_ia32_pcomgeb (v16qi, v16qi)
8205 v4si __builtin_ia32_pcomged (v4si, v4si)
8206 v2di __builtin_ia32_pcomgeq (v2di, v2di)
8207 v16qi __builtin_ia32_pcomgeub (v16qi, v16qi)
8208 v4si __builtin_ia32_pcomgeud (v4si, v4si)
8209 v2di __builtin_ia32_pcomgeuq (v2di, v2di)
8210 v8hi __builtin_ia32_pcomgeuw (v8hi, v8hi)
8211 v8hi __builtin_ia32_pcomgew (v8hi, v8hi)
8212 v16qi __builtin_ia32_pcomgtb (v16qi, v16qi)
8213 v4si __builtin_ia32_pcomgtd (v4si, v4si)
8214 v2di __builtin_ia32_pcomgtq (v2di, v2di)
8215 v16qi __builtin_ia32_pcomgtub (v16qi, v16qi)
8216 v4si __builtin_ia32_pcomgtud (v4si, v4si)
8217 v2di __builtin_ia32_pcomgtuq (v2di, v2di)
8218 v8hi __builtin_ia32_pcomgtuw (v8hi, v8hi)
8219 v8hi __builtin_ia32_pcomgtw (v8hi, v8hi)
8220 v16qi __builtin_ia32_pcomleb (v16qi, v16qi)
8221 v4si __builtin_ia32_pcomled (v4si, v4si)
8222 v2di __builtin_ia32_pcomleq (v2di, v2di)
8223 v16qi __builtin_ia32_pcomleub (v16qi, v16qi)
8224 v4si __builtin_ia32_pcomleud (v4si, v4si)
8225 v2di __builtin_ia32_pcomleuq (v2di, v2di)
8226 v8hi __builtin_ia32_pcomleuw (v8hi, v8hi)
8227 v8hi __builtin_ia32_pcomlew (v8hi, v8hi)
8228 v16qi __builtin_ia32_pcomltb (v16qi, v16qi)
8229 v4si __builtin_ia32_pcomltd (v4si, v4si)
8230 v2di __builtin_ia32_pcomltq (v2di, v2di)
8231 v16qi __builtin_ia32_pcomltub (v16qi, v16qi)
8232 v4si __builtin_ia32_pcomltud (v4si, v4si)
8233 v2di __builtin_ia32_pcomltuq (v2di, v2di)
8234 v8hi __builtin_ia32_pcomltuw (v8hi, v8hi)
8235 v8hi __builtin_ia32_pcomltw (v8hi, v8hi)
8236 v16qi __builtin_ia32_pcomneb (v16qi, v16qi)
8237 v4si __builtin_ia32_pcomned (v4si, v4si)
8238 v2di __builtin_ia32_pcomneq (v2di, v2di)
8239 v16qi __builtin_ia32_pcomneub (v16qi, v16qi)
8240 v4si __builtin_ia32_pcomneud (v4si, v4si)
8241 v2di __builtin_ia32_pcomneuq (v2di, v2di)
8242 v8hi __builtin_ia32_pcomneuw (v8hi, v8hi)
8243 v8hi __builtin_ia32_pcomnew (v8hi, v8hi)
8244 v16qi __builtin_ia32_pcomtrueb (v16qi, v16qi)
8245 v4si __builtin_ia32_pcomtrued (v4si, v4si)
8246 v2di __builtin_ia32_pcomtrueq (v2di, v2di)
8247 v16qi __builtin_ia32_pcomtrueub (v16qi, v16qi)
8248 v4si __builtin_ia32_pcomtrueud (v4si, v4si)
8249 v2di __builtin_ia32_pcomtrueuq (v2di, v2di)
8250 v8hi __builtin_ia32_pcomtrueuw (v8hi, v8hi)
8251 v8hi __builtin_ia32_pcomtruew (v8hi, v8hi)
8252 v4df __builtin_ia32_permpd (v2df, v2df, v16qi)
8253 v4sf __builtin_ia32_permps (v4sf, v4sf, v16qi)
8254 v4si __builtin_ia32_phaddbd (v16qi)
8255 v2di __builtin_ia32_phaddbq (v16qi)
8256 v8hi __builtin_ia32_phaddbw (v16qi)
8257 v2di __builtin_ia32_phadddq (v4si)
8258 v4si __builtin_ia32_phaddubd (v16qi)
8259 v2di __builtin_ia32_phaddubq (v16qi)
8260 v8hi __builtin_ia32_phaddubw (v16qi)
8261 v2di __builtin_ia32_phaddudq (v4si)
8262 v4si __builtin_ia32_phadduwd (v8hi)
8263 v2di __builtin_ia32_phadduwq (v8hi)
8264 v4si __builtin_ia32_phaddwd (v8hi)
8265 v2di __builtin_ia32_phaddwq (v8hi)
8266 v8hi __builtin_ia32_phsubbw (v16qi)
8267 v2di __builtin_ia32_phsubdq (v4si)
8268 v4si __builtin_ia32_phsubwd (v8hi)
8269 v4si __builtin_ia32_pmacsdd (v4si, v4si, v4si)
8270 v2di __builtin_ia32_pmacsdqh (v4si, v4si, v2di)
8271 v2di __builtin_ia32_pmacsdql (v4si, v4si, v2di)
8272 v4si __builtin_ia32_pmacssdd (v4si, v4si, v4si)
8273 v2di __builtin_ia32_pmacssdqh (v4si, v4si, v2di)
8274 v2di __builtin_ia32_pmacssdql (v4si, v4si, v2di)
8275 v4si __builtin_ia32_pmacsswd (v8hi, v8hi, v4si)
8276 v8hi __builtin_ia32_pmacssww (v8hi, v8hi, v8hi)
8277 v4si __builtin_ia32_pmacswd (v8hi, v8hi, v4si)
8278 v8hi __builtin_ia32_pmacsww (v8hi, v8hi, v8hi)
8279 v4si __builtin_ia32_pmadcsswd (v8hi, v8hi, v4si)
8280 v4si __builtin_ia32_pmadcswd (v8hi, v8hi, v4si)
8281 v16qi __builtin_ia32_pperm (v16qi, v16qi, v16qi)
8282 v16qi __builtin_ia32_protb (v16qi, v16qi)
8283 v4si __builtin_ia32_protd (v4si, v4si)
8284 v2di __builtin_ia32_protq (v2di, v2di)
8285 v8hi __builtin_ia32_protw (v8hi, v8hi)
8286 v16qi __builtin_ia32_pshab (v16qi, v16qi)
8287 v4si __builtin_ia32_pshad (v4si, v4si)
8288 v2di __builtin_ia32_pshaq (v2di, v2di)
8289 v8hi __builtin_ia32_pshaw (v8hi, v8hi)
8290 v16qi __builtin_ia32_pshlb (v16qi, v16qi)
8291 v4si __builtin_ia32_pshld (v4si, v4si)
8292 v2di __builtin_ia32_pshlq (v2di, v2di)
8293 v8hi __builtin_ia32_pshlw (v8hi, v8hi)
8296 The following builtin-in functions are available when @option{-msse5}
8297 is used. The second argument must be an integer constant and generate
8298 the machine instruction that is part of the name with the @samp{_imm}
8302 v16qi __builtin_ia32_protb_imm (v16qi, int)
8303 v4si __builtin_ia32_protd_imm (v4si, int)
8304 v2di __builtin_ia32_protq_imm (v2di, int)
8305 v8hi __builtin_ia32_protw_imm (v8hi, int)
8308 The following built-in functions are available when @option{-m3dnow} is used.
8309 All of them generate the machine instruction that is part of the name.
8312 void __builtin_ia32_femms (void)
8313 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
8314 v2si __builtin_ia32_pf2id (v2sf)
8315 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
8316 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
8317 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
8318 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
8319 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
8320 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
8321 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
8322 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
8323 v2sf __builtin_ia32_pfrcp (v2sf)
8324 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
8325 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
8326 v2sf __builtin_ia32_pfrsqrt (v2sf)
8327 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
8328 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
8329 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
8330 v2sf __builtin_ia32_pi2fd (v2si)
8331 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
8334 The following built-in functions are available when both @option{-m3dnow}
8335 and @option{-march=athlon} are used. All of them generate the machine
8336 instruction that is part of the name.
8339 v2si __builtin_ia32_pf2iw (v2sf)
8340 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
8341 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
8342 v2sf __builtin_ia32_pi2fw (v2si)
8343 v2sf __builtin_ia32_pswapdsf (v2sf)
8344 v2si __builtin_ia32_pswapdsi (v2si)
8347 @node MIPS DSP Built-in Functions
8348 @subsection MIPS DSP Built-in Functions
8350 The MIPS DSP Application-Specific Extension (ASE) includes new
8351 instructions that are designed to improve the performance of DSP and
8352 media applications. It provides instructions that operate on packed
8353 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
8355 GCC supports MIPS DSP operations using both the generic
8356 vector extensions (@pxref{Vector Extensions}) and a collection of
8357 MIPS-specific built-in functions. Both kinds of support are
8358 enabled by the @option{-mdsp} command-line option.
8360 Revision 2 of the ASE was introduced in the second half of 2006.
8361 This revision adds extra instructions to the original ASE, but is
8362 otherwise backwards-compatible with it. You can select revision 2
8363 using the command-line option @option{-mdspr2}; this option implies
8366 At present, GCC only provides support for operations on 32-bit
8367 vectors. The vector type associated with 8-bit integer data is
8368 usually called @code{v4i8}, the vector type associated with Q7
8369 is usually called @code{v4q7}, the vector type associated with 16-bit
8370 integer data is usually called @code{v2i16}, and the vector type
8371 associated with Q15 is usually called @code{v2q15}. They can be
8372 defined in C as follows:
8375 typedef signed char v4i8 __attribute__ ((vector_size(4)));
8376 typedef signed char v4q7 __attribute__ ((vector_size(4)));
8377 typedef short v2i16 __attribute__ ((vector_size(4)));
8378 typedef short v2q15 __attribute__ ((vector_size(4)));
8381 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
8382 initialized in the same way as aggregates. For example:
8385 v4i8 a = @{1, 2, 3, 4@};
8387 b = (v4i8) @{5, 6, 7, 8@};
8389 v2q15 c = @{0x0fcb, 0x3a75@};
8391 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
8394 @emph{Note:} The CPU's endianness determines the order in which values
8395 are packed. On little-endian targets, the first value is the least
8396 significant and the last value is the most significant. The opposite
8397 order applies to big-endian targets. For example, the code above will
8398 set the lowest byte of @code{a} to @code{1} on little-endian targets
8399 and @code{4} on big-endian targets.
8401 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
8402 representation. As shown in this example, the integer representation
8403 of a Q7 value can be obtained by multiplying the fractional value by
8404 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
8405 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
8408 The table below lists the @code{v4i8} and @code{v2q15} operations for which
8409 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
8410 and @code{c} and @code{d} are @code{v2q15} values.
8412 @multitable @columnfractions .50 .50
8413 @item C code @tab MIPS instruction
8414 @item @code{a + b} @tab @code{addu.qb}
8415 @item @code{c + d} @tab @code{addq.ph}
8416 @item @code{a - b} @tab @code{subu.qb}
8417 @item @code{c - d} @tab @code{subq.ph}
8420 The table below lists the @code{v2i16} operation for which
8421 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
8422 @code{v2i16} values.
8424 @multitable @columnfractions .50 .50
8425 @item C code @tab MIPS instruction
8426 @item @code{e * f} @tab @code{mul.ph}
8429 It is easier to describe the DSP built-in functions if we first define
8430 the following types:
8435 typedef unsigned int ui32;
8436 typedef long long a64;
8439 @code{q31} and @code{i32} are actually the same as @code{int}, but we
8440 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
8441 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
8442 @code{long long}, but we use @code{a64} to indicate values that will
8443 be placed in one of the four DSP accumulators (@code{$ac0},
8444 @code{$ac1}, @code{$ac2} or @code{$ac3}).
8446 Also, some built-in functions prefer or require immediate numbers as
8447 parameters, because the corresponding DSP instructions accept both immediate
8448 numbers and register operands, or accept immediate numbers only. The
8449 immediate parameters are listed as follows.
8458 imm_n32_31: -32 to 31.
8459 imm_n512_511: -512 to 511.
8462 The following built-in functions map directly to a particular MIPS DSP
8463 instruction. Please refer to the architecture specification
8464 for details on what each instruction does.
8467 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
8468 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
8469 q31 __builtin_mips_addq_s_w (q31, q31)
8470 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
8471 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
8472 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
8473 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
8474 q31 __builtin_mips_subq_s_w (q31, q31)
8475 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
8476 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
8477 i32 __builtin_mips_addsc (i32, i32)
8478 i32 __builtin_mips_addwc (i32, i32)
8479 i32 __builtin_mips_modsub (i32, i32)
8480 i32 __builtin_mips_raddu_w_qb (v4i8)
8481 v2q15 __builtin_mips_absq_s_ph (v2q15)
8482 q31 __builtin_mips_absq_s_w (q31)
8483 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
8484 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
8485 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
8486 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
8487 q31 __builtin_mips_preceq_w_phl (v2q15)
8488 q31 __builtin_mips_preceq_w_phr (v2q15)
8489 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
8490 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
8491 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
8492 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
8493 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
8494 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
8495 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
8496 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
8497 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
8498 v4i8 __builtin_mips_shll_qb (v4i8, i32)
8499 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
8500 v2q15 __builtin_mips_shll_ph (v2q15, i32)
8501 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
8502 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
8503 q31 __builtin_mips_shll_s_w (q31, imm0_31)
8504 q31 __builtin_mips_shll_s_w (q31, i32)
8505 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
8506 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
8507 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
8508 v2q15 __builtin_mips_shra_ph (v2q15, i32)
8509 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
8510 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
8511 q31 __builtin_mips_shra_r_w (q31, imm0_31)
8512 q31 __builtin_mips_shra_r_w (q31, i32)
8513 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
8514 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
8515 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
8516 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
8517 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
8518 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
8519 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
8520 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
8521 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
8522 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
8523 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
8524 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
8525 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
8526 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
8527 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
8528 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
8529 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
8530 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
8531 i32 __builtin_mips_bitrev (i32)
8532 i32 __builtin_mips_insv (i32, i32)
8533 v4i8 __builtin_mips_repl_qb (imm0_255)
8534 v4i8 __builtin_mips_repl_qb (i32)
8535 v2q15 __builtin_mips_repl_ph (imm_n512_511)
8536 v2q15 __builtin_mips_repl_ph (i32)
8537 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
8538 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
8539 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
8540 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
8541 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
8542 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
8543 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
8544 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
8545 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
8546 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
8547 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
8548 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
8549 i32 __builtin_mips_extr_w (a64, imm0_31)
8550 i32 __builtin_mips_extr_w (a64, i32)
8551 i32 __builtin_mips_extr_r_w (a64, imm0_31)
8552 i32 __builtin_mips_extr_s_h (a64, i32)
8553 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
8554 i32 __builtin_mips_extr_rs_w (a64, i32)
8555 i32 __builtin_mips_extr_s_h (a64, imm0_31)
8556 i32 __builtin_mips_extr_r_w (a64, i32)
8557 i32 __builtin_mips_extp (a64, imm0_31)
8558 i32 __builtin_mips_extp (a64, i32)
8559 i32 __builtin_mips_extpdp (a64, imm0_31)
8560 i32 __builtin_mips_extpdp (a64, i32)
8561 a64 __builtin_mips_shilo (a64, imm_n32_31)
8562 a64 __builtin_mips_shilo (a64, i32)
8563 a64 __builtin_mips_mthlip (a64, i32)
8564 void __builtin_mips_wrdsp (i32, imm0_63)
8565 i32 __builtin_mips_rddsp (imm0_63)
8566 i32 __builtin_mips_lbux (void *, i32)
8567 i32 __builtin_mips_lhx (void *, i32)
8568 i32 __builtin_mips_lwx (void *, i32)
8569 i32 __builtin_mips_bposge32 (void)
8572 The following built-in functions map directly to a particular MIPS DSP REV 2
8573 instruction. Please refer to the architecture specification
8574 for details on what each instruction does.
8577 v4q7 __builtin_mips_absq_s_qb (v4q7);
8578 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
8579 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
8580 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
8581 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
8582 i32 __builtin_mips_append (i32, i32, imm0_31);
8583 i32 __builtin_mips_balign (i32, i32, imm0_3);
8584 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
8585 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
8586 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
8587 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
8588 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
8589 a64 __builtin_mips_madd (a64, i32, i32);
8590 a64 __builtin_mips_maddu (a64, ui32, ui32);
8591 a64 __builtin_mips_msub (a64, i32, i32);
8592 a64 __builtin_mips_msubu (a64, ui32, ui32);
8593 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
8594 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
8595 q31 __builtin_mips_mulq_rs_w (q31, q31);
8596 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
8597 q31 __builtin_mips_mulq_s_w (q31, q31);
8598 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
8599 a64 __builtin_mips_mult (i32, i32);
8600 a64 __builtin_mips_multu (ui32, ui32);
8601 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
8602 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
8603 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
8604 i32 __builtin_mips_prepend (i32, i32, imm0_31);
8605 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
8606 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
8607 v4i8 __builtin_mips_shra_qb (v4i8, i32);
8608 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
8609 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
8610 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
8611 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
8612 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
8613 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
8614 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
8615 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
8616 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
8617 q31 __builtin_mips_addqh_w (q31, q31);
8618 q31 __builtin_mips_addqh_r_w (q31, q31);
8619 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
8620 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
8621 q31 __builtin_mips_subqh_w (q31, q31);
8622 q31 __builtin_mips_subqh_r_w (q31, q31);
8623 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
8624 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
8625 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
8626 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
8627 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
8628 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
8632 @node MIPS Paired-Single Support
8633 @subsection MIPS Paired-Single Support
8635 The MIPS64 architecture includes a number of instructions that
8636 operate on pairs of single-precision floating-point values.
8637 Each pair is packed into a 64-bit floating-point register,
8638 with one element being designated the ``upper half'' and
8639 the other being designated the ``lower half''.
8641 GCC supports paired-single operations using both the generic
8642 vector extensions (@pxref{Vector Extensions}) and a collection of
8643 MIPS-specific built-in functions. Both kinds of support are
8644 enabled by the @option{-mpaired-single} command-line option.
8646 The vector type associated with paired-single values is usually
8647 called @code{v2sf}. It can be defined in C as follows:
8650 typedef float v2sf __attribute__ ((vector_size (8)));
8653 @code{v2sf} values are initialized in the same way as aggregates.
8657 v2sf a = @{1.5, 9.1@};
8660 b = (v2sf) @{e, f@};
8663 @emph{Note:} The CPU's endianness determines which value is stored in
8664 the upper half of a register and which value is stored in the lower half.
8665 On little-endian targets, the first value is the lower one and the second
8666 value is the upper one. The opposite order applies to big-endian targets.
8667 For example, the code above will set the lower half of @code{a} to
8668 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
8671 * Paired-Single Arithmetic::
8672 * Paired-Single Built-in Functions::
8673 * MIPS-3D Built-in Functions::
8676 @node Paired-Single Arithmetic
8677 @subsubsection Paired-Single Arithmetic
8679 The table below lists the @code{v2sf} operations for which hardware
8680 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
8681 values and @code{x} is an integral value.
8683 @multitable @columnfractions .50 .50
8684 @item C code @tab MIPS instruction
8685 @item @code{a + b} @tab @code{add.ps}
8686 @item @code{a - b} @tab @code{sub.ps}
8687 @item @code{-a} @tab @code{neg.ps}
8688 @item @code{a * b} @tab @code{mul.ps}
8689 @item @code{a * b + c} @tab @code{madd.ps}
8690 @item @code{a * b - c} @tab @code{msub.ps}
8691 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
8692 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
8693 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
8696 Note that the multiply-accumulate instructions can be disabled
8697 using the command-line option @code{-mno-fused-madd}.
8699 @node Paired-Single Built-in Functions
8700 @subsubsection Paired-Single Built-in Functions
8702 The following paired-single functions map directly to a particular
8703 MIPS instruction. Please refer to the architecture specification
8704 for details on what each instruction does.
8707 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
8708 Pair lower lower (@code{pll.ps}).
8710 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
8711 Pair upper lower (@code{pul.ps}).
8713 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
8714 Pair lower upper (@code{plu.ps}).
8716 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
8717 Pair upper upper (@code{puu.ps}).
8719 @item v2sf __builtin_mips_cvt_ps_s (float, float)
8720 Convert pair to paired single (@code{cvt.ps.s}).
8722 @item float __builtin_mips_cvt_s_pl (v2sf)
8723 Convert pair lower to single (@code{cvt.s.pl}).
8725 @item float __builtin_mips_cvt_s_pu (v2sf)
8726 Convert pair upper to single (@code{cvt.s.pu}).
8728 @item v2sf __builtin_mips_abs_ps (v2sf)
8729 Absolute value (@code{abs.ps}).
8731 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
8732 Align variable (@code{alnv.ps}).
8734 @emph{Note:} The value of the third parameter must be 0 or 4
8735 modulo 8, otherwise the result will be unpredictable. Please read the
8736 instruction description for details.
8739 The following multi-instruction functions are also available.
8740 In each case, @var{cond} can be any of the 16 floating-point conditions:
8741 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
8742 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
8743 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
8746 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8747 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8748 Conditional move based on floating point comparison (@code{c.@var{cond}.ps},
8749 @code{movt.ps}/@code{movf.ps}).
8751 The @code{movt} functions return the value @var{x} computed by:
8754 c.@var{cond}.ps @var{cc},@var{a},@var{b}
8755 mov.ps @var{x},@var{c}
8756 movt.ps @var{x},@var{d},@var{cc}
8759 The @code{movf} functions are similar but use @code{movf.ps} instead
8762 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8763 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8764 Comparison of two paired-single values (@code{c.@var{cond}.ps},
8765 @code{bc1t}/@code{bc1f}).
8767 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
8768 and return either the upper or lower half of the result. For example:
8772 if (__builtin_mips_upper_c_eq_ps (a, b))
8773 upper_halves_are_equal ();
8775 upper_halves_are_unequal ();
8777 if (__builtin_mips_lower_c_eq_ps (a, b))
8778 lower_halves_are_equal ();
8780 lower_halves_are_unequal ();
8784 @node MIPS-3D Built-in Functions
8785 @subsubsection MIPS-3D Built-in Functions
8787 The MIPS-3D Application-Specific Extension (ASE) includes additional
8788 paired-single instructions that are designed to improve the performance
8789 of 3D graphics operations. Support for these instructions is controlled
8790 by the @option{-mips3d} command-line option.
8792 The functions listed below map directly to a particular MIPS-3D
8793 instruction. Please refer to the architecture specification for
8794 more details on what each instruction does.
8797 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
8798 Reduction add (@code{addr.ps}).
8800 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
8801 Reduction multiply (@code{mulr.ps}).
8803 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
8804 Convert paired single to paired word (@code{cvt.pw.ps}).
8806 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
8807 Convert paired word to paired single (@code{cvt.ps.pw}).
8809 @item float __builtin_mips_recip1_s (float)
8810 @itemx double __builtin_mips_recip1_d (double)
8811 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
8812 Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
8814 @item float __builtin_mips_recip2_s (float, float)
8815 @itemx double __builtin_mips_recip2_d (double, double)
8816 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
8817 Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
8819 @item float __builtin_mips_rsqrt1_s (float)
8820 @itemx double __builtin_mips_rsqrt1_d (double)
8821 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
8822 Reduced precision reciprocal square root (sequence step 1)
8823 (@code{rsqrt1.@var{fmt}}).
8825 @item float __builtin_mips_rsqrt2_s (float, float)
8826 @itemx double __builtin_mips_rsqrt2_d (double, double)
8827 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
8828 Reduced precision reciprocal square root (sequence step 2)
8829 (@code{rsqrt2.@var{fmt}}).
8832 The following multi-instruction functions are also available.
8833 In each case, @var{cond} can be any of the 16 floating-point conditions:
8834 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
8835 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
8836 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
8839 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
8840 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
8841 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
8842 @code{bc1t}/@code{bc1f}).
8844 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
8845 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
8850 if (__builtin_mips_cabs_eq_s (a, b))
8856 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8857 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8858 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
8859 @code{bc1t}/@code{bc1f}).
8861 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
8862 and return either the upper or lower half of the result. For example:
8866 if (__builtin_mips_upper_cabs_eq_ps (a, b))
8867 upper_halves_are_equal ();
8869 upper_halves_are_unequal ();
8871 if (__builtin_mips_lower_cabs_eq_ps (a, b))
8872 lower_halves_are_equal ();
8874 lower_halves_are_unequal ();
8877 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8878 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8879 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
8880 @code{movt.ps}/@code{movf.ps}).
8882 The @code{movt} functions return the value @var{x} computed by:
8885 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
8886 mov.ps @var{x},@var{c}
8887 movt.ps @var{x},@var{d},@var{cc}
8890 The @code{movf} functions are similar but use @code{movf.ps} instead
8893 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8894 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8895 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8896 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
8897 Comparison of two paired-single values
8898 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
8899 @code{bc1any2t}/@code{bc1any2f}).
8901 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
8902 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
8903 result is true and the @code{all} forms return true if both results are true.
8908 if (__builtin_mips_any_c_eq_ps (a, b))
8913 if (__builtin_mips_all_c_eq_ps (a, b))
8919 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8920 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8921 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8922 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
8923 Comparison of four paired-single values
8924 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
8925 @code{bc1any4t}/@code{bc1any4f}).
8927 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
8928 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
8929 The @code{any} forms return true if any of the four results are true
8930 and the @code{all} forms return true if all four results are true.
8935 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
8940 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
8947 @node PowerPC AltiVec Built-in Functions
8948 @subsection PowerPC AltiVec Built-in Functions
8950 GCC provides an interface for the PowerPC family of processors to access
8951 the AltiVec operations described in Motorola's AltiVec Programming
8952 Interface Manual. The interface is made available by including
8953 @code{<altivec.h>} and using @option{-maltivec} and
8954 @option{-mabi=altivec}. The interface supports the following vector
8958 vector unsigned char
8962 vector unsigned short
8973 GCC's implementation of the high-level language interface available from
8974 C and C++ code differs from Motorola's documentation in several ways.
8979 A vector constant is a list of constant expressions within curly braces.
8982 A vector initializer requires no cast if the vector constant is of the
8983 same type as the variable it is initializing.
8986 If @code{signed} or @code{unsigned} is omitted, the signedness of the
8987 vector type is the default signedness of the base type. The default
8988 varies depending on the operating system, so a portable program should
8989 always specify the signedness.
8992 Compiling with @option{-maltivec} adds keywords @code{__vector},
8993 @code{__pixel}, and @code{__bool}. Macros @option{vector},
8994 @code{pixel}, and @code{bool} are defined in @code{<altivec.h>} and can
8998 GCC allows using a @code{typedef} name as the type specifier for a
9002 For C, overloaded functions are implemented with macros so the following
9006 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
9009 Since @code{vec_add} is a macro, the vector constant in the example
9010 is treated as four separate arguments. Wrap the entire argument in
9011 parentheses for this to work.
9014 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
9015 Internally, GCC uses built-in functions to achieve the functionality in
9016 the aforementioned header file, but they are not supported and are
9017 subject to change without notice.
9019 The following interfaces are supported for the generic and specific
9020 AltiVec operations and the AltiVec predicates. In cases where there
9021 is a direct mapping between generic and specific operations, only the
9022 generic names are shown here, although the specific operations can also
9025 Arguments that are documented as @code{const int} require literal
9026 integral values within the range required for that operation.
9029 vector signed char vec_abs (vector signed char);
9030 vector signed short vec_abs (vector signed short);
9031 vector signed int vec_abs (vector signed int);
9032 vector float vec_abs (vector float);
9034 vector signed char vec_abss (vector signed char);
9035 vector signed short vec_abss (vector signed short);
9036 vector signed int vec_abss (vector signed int);
9038 vector signed char vec_add (vector bool char, vector signed char);
9039 vector signed char vec_add (vector signed char, vector bool char);
9040 vector signed char vec_add (vector signed char, vector signed char);
9041 vector unsigned char vec_add (vector bool char, vector unsigned char);
9042 vector unsigned char vec_add (vector unsigned char, vector bool char);
9043 vector unsigned char vec_add (vector unsigned char,
9044 vector unsigned char);
9045 vector signed short vec_add (vector bool short, vector signed short);
9046 vector signed short vec_add (vector signed short, vector bool short);
9047 vector signed short vec_add (vector signed short, vector signed short);
9048 vector unsigned short vec_add (vector bool short,
9049 vector unsigned short);
9050 vector unsigned short vec_add (vector unsigned short,
9052 vector unsigned short vec_add (vector unsigned short,
9053 vector unsigned short);
9054 vector signed int vec_add (vector bool int, vector signed int);
9055 vector signed int vec_add (vector signed int, vector bool int);
9056 vector signed int vec_add (vector signed int, vector signed int);
9057 vector unsigned int vec_add (vector bool int, vector unsigned int);
9058 vector unsigned int vec_add (vector unsigned int, vector bool int);
9059 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
9060 vector float vec_add (vector float, vector float);
9062 vector float vec_vaddfp (vector float, vector float);
9064 vector signed int vec_vadduwm (vector bool int, vector signed int);
9065 vector signed int vec_vadduwm (vector signed int, vector bool int);
9066 vector signed int vec_vadduwm (vector signed int, vector signed int);
9067 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
9068 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
9069 vector unsigned int vec_vadduwm (vector unsigned int,
9070 vector unsigned int);
9072 vector signed short vec_vadduhm (vector bool short,
9073 vector signed short);
9074 vector signed short vec_vadduhm (vector signed short,
9076 vector signed short vec_vadduhm (vector signed short,
9077 vector signed short);
9078 vector unsigned short vec_vadduhm (vector bool short,
9079 vector unsigned short);
9080 vector unsigned short vec_vadduhm (vector unsigned short,
9082 vector unsigned short vec_vadduhm (vector unsigned short,
9083 vector unsigned short);
9085 vector signed char vec_vaddubm (vector bool char, vector signed char);
9086 vector signed char vec_vaddubm (vector signed char, vector bool char);
9087 vector signed char vec_vaddubm (vector signed char, vector signed char);
9088 vector unsigned char vec_vaddubm (vector bool char,
9089 vector unsigned char);
9090 vector unsigned char vec_vaddubm (vector unsigned char,
9092 vector unsigned char vec_vaddubm (vector unsigned char,
9093 vector unsigned char);
9095 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
9097 vector unsigned char vec_adds (vector bool char, vector unsigned char);
9098 vector unsigned char vec_adds (vector unsigned char, vector bool char);
9099 vector unsigned char vec_adds (vector unsigned char,
9100 vector unsigned char);
9101 vector signed char vec_adds (vector bool char, vector signed char);
9102 vector signed char vec_adds (vector signed char, vector bool char);
9103 vector signed char vec_adds (vector signed char, vector signed char);
9104 vector unsigned short vec_adds (vector bool short,
9105 vector unsigned short);
9106 vector unsigned short vec_adds (vector unsigned short,
9108 vector unsigned short vec_adds (vector unsigned short,
9109 vector unsigned short);
9110 vector signed short vec_adds (vector bool short, vector signed short);
9111 vector signed short vec_adds (vector signed short, vector bool short);
9112 vector signed short vec_adds (vector signed short, vector signed short);
9113 vector unsigned int vec_adds (vector bool int, vector unsigned int);
9114 vector unsigned int vec_adds (vector unsigned int, vector bool int);
9115 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
9116 vector signed int vec_adds (vector bool int, vector signed int);
9117 vector signed int vec_adds (vector signed int, vector bool int);
9118 vector signed int vec_adds (vector signed int, vector signed int);
9120 vector signed int vec_vaddsws (vector bool int, vector signed int);
9121 vector signed int vec_vaddsws (vector signed int, vector bool int);
9122 vector signed int vec_vaddsws (vector signed int, vector signed int);
9124 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
9125 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
9126 vector unsigned int vec_vadduws (vector unsigned int,
9127 vector unsigned int);
9129 vector signed short vec_vaddshs (vector bool short,
9130 vector signed short);
9131 vector signed short vec_vaddshs (vector signed short,
9133 vector signed short vec_vaddshs (vector signed short,
9134 vector signed short);
9136 vector unsigned short vec_vadduhs (vector bool short,
9137 vector unsigned short);
9138 vector unsigned short vec_vadduhs (vector unsigned short,
9140 vector unsigned short vec_vadduhs (vector unsigned short,
9141 vector unsigned short);
9143 vector signed char vec_vaddsbs (vector bool char, vector signed char);
9144 vector signed char vec_vaddsbs (vector signed char, vector bool char);
9145 vector signed char vec_vaddsbs (vector signed char, vector signed char);
9147 vector unsigned char vec_vaddubs (vector bool char,
9148 vector unsigned char);
9149 vector unsigned char vec_vaddubs (vector unsigned char,
9151 vector unsigned char vec_vaddubs (vector unsigned char,
9152 vector unsigned char);
9154 vector float vec_and (vector float, vector float);
9155 vector float vec_and (vector float, vector bool int);
9156 vector float vec_and (vector bool int, vector float);
9157 vector bool int vec_and (vector bool int, vector bool int);
9158 vector signed int vec_and (vector bool int, vector signed int);
9159 vector signed int vec_and (vector signed int, vector bool int);
9160 vector signed int vec_and (vector signed int, vector signed int);
9161 vector unsigned int vec_and (vector bool int, vector unsigned int);
9162 vector unsigned int vec_and (vector unsigned int, vector bool int);
9163 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
9164 vector bool short vec_and (vector bool short, vector bool short);
9165 vector signed short vec_and (vector bool short, vector signed short);
9166 vector signed short vec_and (vector signed short, vector bool short);
9167 vector signed short vec_and (vector signed short, vector signed short);
9168 vector unsigned short vec_and (vector bool short,
9169 vector unsigned short);
9170 vector unsigned short vec_and (vector unsigned short,
9172 vector unsigned short vec_and (vector unsigned short,
9173 vector unsigned short);
9174 vector signed char vec_and (vector bool char, vector signed char);
9175 vector bool char vec_and (vector bool char, vector bool char);
9176 vector signed char vec_and (vector signed char, vector bool char);
9177 vector signed char vec_and (vector signed char, vector signed char);
9178 vector unsigned char vec_and (vector bool char, vector unsigned char);
9179 vector unsigned char vec_and (vector unsigned char, vector bool char);
9180 vector unsigned char vec_and (vector unsigned char,
9181 vector unsigned char);
9183 vector float vec_andc (vector float, vector float);
9184 vector float vec_andc (vector float, vector bool int);
9185 vector float vec_andc (vector bool int, vector float);
9186 vector bool int vec_andc (vector bool int, vector bool int);
9187 vector signed int vec_andc (vector bool int, vector signed int);
9188 vector signed int vec_andc (vector signed int, vector bool int);
9189 vector signed int vec_andc (vector signed int, vector signed int);
9190 vector unsigned int vec_andc (vector bool int, vector unsigned int);
9191 vector unsigned int vec_andc (vector unsigned int, vector bool int);
9192 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
9193 vector bool short vec_andc (vector bool short, vector bool short);
9194 vector signed short vec_andc (vector bool short, vector signed short);
9195 vector signed short vec_andc (vector signed short, vector bool short);
9196 vector signed short vec_andc (vector signed short, vector signed short);
9197 vector unsigned short vec_andc (vector bool short,
9198 vector unsigned short);
9199 vector unsigned short vec_andc (vector unsigned short,
9201 vector unsigned short vec_andc (vector unsigned short,
9202 vector unsigned short);
9203 vector signed char vec_andc (vector bool char, vector signed char);
9204 vector bool char vec_andc (vector bool char, vector bool char);
9205 vector signed char vec_andc (vector signed char, vector bool char);
9206 vector signed char vec_andc (vector signed char, vector signed char);
9207 vector unsigned char vec_andc (vector bool char, vector unsigned char);
9208 vector unsigned char vec_andc (vector unsigned char, vector bool char);
9209 vector unsigned char vec_andc (vector unsigned char,
9210 vector unsigned char);
9212 vector unsigned char vec_avg (vector unsigned char,
9213 vector unsigned char);
9214 vector signed char vec_avg (vector signed char, vector signed char);
9215 vector unsigned short vec_avg (vector unsigned short,
9216 vector unsigned short);
9217 vector signed short vec_avg (vector signed short, vector signed short);
9218 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
9219 vector signed int vec_avg (vector signed int, vector signed int);
9221 vector signed int vec_vavgsw (vector signed int, vector signed int);
9223 vector unsigned int vec_vavguw (vector unsigned int,
9224 vector unsigned int);
9226 vector signed short vec_vavgsh (vector signed short,
9227 vector signed short);
9229 vector unsigned short vec_vavguh (vector unsigned short,
9230 vector unsigned short);
9232 vector signed char vec_vavgsb (vector signed char, vector signed char);
9234 vector unsigned char vec_vavgub (vector unsigned char,
9235 vector unsigned char);
9237 vector float vec_ceil (vector float);
9239 vector signed int vec_cmpb (vector float, vector float);
9241 vector bool char vec_cmpeq (vector signed char, vector signed char);
9242 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
9243 vector bool short vec_cmpeq (vector signed short, vector signed short);
9244 vector bool short vec_cmpeq (vector unsigned short,
9245 vector unsigned short);
9246 vector bool int vec_cmpeq (vector signed int, vector signed int);
9247 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
9248 vector bool int vec_cmpeq (vector float, vector float);
9250 vector bool int vec_vcmpeqfp (vector float, vector float);
9252 vector bool int vec_vcmpequw (vector signed int, vector signed int);
9253 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
9255 vector bool short vec_vcmpequh (vector signed short,
9256 vector signed short);
9257 vector bool short vec_vcmpequh (vector unsigned short,
9258 vector unsigned short);
9260 vector bool char vec_vcmpequb (vector signed char, vector signed char);
9261 vector bool char vec_vcmpequb (vector unsigned char,
9262 vector unsigned char);
9264 vector bool int vec_cmpge (vector float, vector float);
9266 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
9267 vector bool char vec_cmpgt (vector signed char, vector signed char);
9268 vector bool short vec_cmpgt (vector unsigned short,
9269 vector unsigned short);
9270 vector bool short vec_cmpgt (vector signed short, vector signed short);
9271 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
9272 vector bool int vec_cmpgt (vector signed int, vector signed int);
9273 vector bool int vec_cmpgt (vector float, vector float);
9275 vector bool int vec_vcmpgtfp (vector float, vector float);
9277 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
9279 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
9281 vector bool short vec_vcmpgtsh (vector signed short,
9282 vector signed short);
9284 vector bool short vec_vcmpgtuh (vector unsigned short,
9285 vector unsigned short);
9287 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
9289 vector bool char vec_vcmpgtub (vector unsigned char,
9290 vector unsigned char);
9292 vector bool int vec_cmple (vector float, vector float);
9294 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
9295 vector bool char vec_cmplt (vector signed char, vector signed char);
9296 vector bool short vec_cmplt (vector unsigned short,
9297 vector unsigned short);
9298 vector bool short vec_cmplt (vector signed short, vector signed short);
9299 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
9300 vector bool int vec_cmplt (vector signed int, vector signed int);
9301 vector bool int vec_cmplt (vector float, vector float);
9303 vector float vec_ctf (vector unsigned int, const int);
9304 vector float vec_ctf (vector signed int, const int);
9306 vector float vec_vcfsx (vector signed int, const int);
9308 vector float vec_vcfux (vector unsigned int, const int);
9310 vector signed int vec_cts (vector float, const int);
9312 vector unsigned int vec_ctu (vector float, const int);
9314 void vec_dss (const int);
9316 void vec_dssall (void);
9318 void vec_dst (const vector unsigned char *, int, const int);
9319 void vec_dst (const vector signed char *, int, const int);
9320 void vec_dst (const vector bool char *, int, const int);
9321 void vec_dst (const vector unsigned short *, int, const int);
9322 void vec_dst (const vector signed short *, int, const int);
9323 void vec_dst (const vector bool short *, int, const int);
9324 void vec_dst (const vector pixel *, int, const int);
9325 void vec_dst (const vector unsigned int *, int, const int);
9326 void vec_dst (const vector signed int *, int, const int);
9327 void vec_dst (const vector bool int *, int, const int);
9328 void vec_dst (const vector float *, int, const int);
9329 void vec_dst (const unsigned char *, int, const int);
9330 void vec_dst (const signed char *, int, const int);
9331 void vec_dst (const unsigned short *, int, const int);
9332 void vec_dst (const short *, int, const int);
9333 void vec_dst (const unsigned int *, int, const int);
9334 void vec_dst (const int *, int, const int);
9335 void vec_dst (const unsigned long *, int, const int);
9336 void vec_dst (const long *, int, const int);
9337 void vec_dst (const float *, int, const int);
9339 void vec_dstst (const vector unsigned char *, int, const int);
9340 void vec_dstst (const vector signed char *, int, const int);
9341 void vec_dstst (const vector bool char *, int, const int);
9342 void vec_dstst (const vector unsigned short *, int, const int);
9343 void vec_dstst (const vector signed short *, int, const int);
9344 void vec_dstst (const vector bool short *, int, const int);
9345 void vec_dstst (const vector pixel *, int, const int);
9346 void vec_dstst (const vector unsigned int *, int, const int);
9347 void vec_dstst (const vector signed int *, int, const int);
9348 void vec_dstst (const vector bool int *, int, const int);
9349 void vec_dstst (const vector float *, int, const int);
9350 void vec_dstst (const unsigned char *, int, const int);
9351 void vec_dstst (const signed char *, int, const int);
9352 void vec_dstst (const unsigned short *, int, const int);
9353 void vec_dstst (const short *, int, const int);
9354 void vec_dstst (const unsigned int *, int, const int);
9355 void vec_dstst (const int *, int, const int);
9356 void vec_dstst (const unsigned long *, int, const int);
9357 void vec_dstst (const long *, int, const int);
9358 void vec_dstst (const float *, int, const int);
9360 void vec_dststt (const vector unsigned char *, int, const int);
9361 void vec_dststt (const vector signed char *, int, const int);
9362 void vec_dststt (const vector bool char *, int, const int);
9363 void vec_dststt (const vector unsigned short *, int, const int);
9364 void vec_dststt (const vector signed short *, int, const int);
9365 void vec_dststt (const vector bool short *, int, const int);
9366 void vec_dststt (const vector pixel *, int, const int);
9367 void vec_dststt (const vector unsigned int *, int, const int);
9368 void vec_dststt (const vector signed int *, int, const int);
9369 void vec_dststt (const vector bool int *, int, const int);
9370 void vec_dststt (const vector float *, int, const int);
9371 void vec_dststt (const unsigned char *, int, const int);
9372 void vec_dststt (const signed char *, int, const int);
9373 void vec_dststt (const unsigned short *, int, const int);
9374 void vec_dststt (const short *, int, const int);
9375 void vec_dststt (const unsigned int *, int, const int);
9376 void vec_dststt (const int *, int, const int);
9377 void vec_dststt (const unsigned long *, int, const int);
9378 void vec_dststt (const long *, int, const int);
9379 void vec_dststt (const float *, int, const int);
9381 void vec_dstt (const vector unsigned char *, int, const int);
9382 void vec_dstt (const vector signed char *, int, const int);
9383 void vec_dstt (const vector bool char *, int, const int);
9384 void vec_dstt (const vector unsigned short *, int, const int);
9385 void vec_dstt (const vector signed short *, int, const int);
9386 void vec_dstt (const vector bool short *, int, const int);
9387 void vec_dstt (const vector pixel *, int, const int);
9388 void vec_dstt (const vector unsigned int *, int, const int);
9389 void vec_dstt (const vector signed int *, int, const int);
9390 void vec_dstt (const vector bool int *, int, const int);
9391 void vec_dstt (const vector float *, int, const int);
9392 void vec_dstt (const unsigned char *, int, const int);
9393 void vec_dstt (const signed char *, int, const int);
9394 void vec_dstt (const unsigned short *, int, const int);
9395 void vec_dstt (const short *, int, const int);
9396 void vec_dstt (const unsigned int *, int, const int);
9397 void vec_dstt (const int *, int, const int);
9398 void vec_dstt (const unsigned long *, int, const int);
9399 void vec_dstt (const long *, int, const int);
9400 void vec_dstt (const float *, int, const int);
9402 vector float vec_expte (vector float);
9404 vector float vec_floor (vector float);
9406 vector float vec_ld (int, const vector float *);
9407 vector float vec_ld (int, const float *);
9408 vector bool int vec_ld (int, const vector bool int *);
9409 vector signed int vec_ld (int, const vector signed int *);
9410 vector signed int vec_ld (int, const int *);
9411 vector signed int vec_ld (int, const long *);
9412 vector unsigned int vec_ld (int, const vector unsigned int *);
9413 vector unsigned int vec_ld (int, const unsigned int *);
9414 vector unsigned int vec_ld (int, const unsigned long *);
9415 vector bool short vec_ld (int, const vector bool short *);
9416 vector pixel vec_ld (int, const vector pixel *);
9417 vector signed short vec_ld (int, const vector signed short *);
9418 vector signed short vec_ld (int, const short *);
9419 vector unsigned short vec_ld (int, const vector unsigned short *);
9420 vector unsigned short vec_ld (int, const unsigned short *);
9421 vector bool char vec_ld (int, const vector bool char *);
9422 vector signed char vec_ld (int, const vector signed char *);
9423 vector signed char vec_ld (int, const signed char *);
9424 vector unsigned char vec_ld (int, const vector unsigned char *);
9425 vector unsigned char vec_ld (int, const unsigned char *);
9427 vector signed char vec_lde (int, const signed char *);
9428 vector unsigned char vec_lde (int, const unsigned char *);
9429 vector signed short vec_lde (int, const short *);
9430 vector unsigned short vec_lde (int, const unsigned short *);
9431 vector float vec_lde (int, const float *);
9432 vector signed int vec_lde (int, const int *);
9433 vector unsigned int vec_lde (int, const unsigned int *);
9434 vector signed int vec_lde (int, const long *);
9435 vector unsigned int vec_lde (int, const unsigned long *);
9437 vector float vec_lvewx (int, float *);
9438 vector signed int vec_lvewx (int, int *);
9439 vector unsigned int vec_lvewx (int, unsigned int *);
9440 vector signed int vec_lvewx (int, long *);
9441 vector unsigned int vec_lvewx (int, unsigned long *);
9443 vector signed short vec_lvehx (int, short *);
9444 vector unsigned short vec_lvehx (int, unsigned short *);
9446 vector signed char vec_lvebx (int, char *);
9447 vector unsigned char vec_lvebx (int, unsigned char *);
9449 vector float vec_ldl (int, const vector float *);
9450 vector float vec_ldl (int, const float *);
9451 vector bool int vec_ldl (int, const vector bool int *);
9452 vector signed int vec_ldl (int, const vector signed int *);
9453 vector signed int vec_ldl (int, const int *);
9454 vector signed int vec_ldl (int, const long *);
9455 vector unsigned int vec_ldl (int, const vector unsigned int *);
9456 vector unsigned int vec_ldl (int, const unsigned int *);
9457 vector unsigned int vec_ldl (int, const unsigned long *);
9458 vector bool short vec_ldl (int, const vector bool short *);
9459 vector pixel vec_ldl (int, const vector pixel *);
9460 vector signed short vec_ldl (int, const vector signed short *);
9461 vector signed short vec_ldl (int, const short *);
9462 vector unsigned short vec_ldl (int, const vector unsigned short *);
9463 vector unsigned short vec_ldl (int, const unsigned short *);
9464 vector bool char vec_ldl (int, const vector bool char *);
9465 vector signed char vec_ldl (int, const vector signed char *);
9466 vector signed char vec_ldl (int, const signed char *);
9467 vector unsigned char vec_ldl (int, const vector unsigned char *);
9468 vector unsigned char vec_ldl (int, const unsigned char *);
9470 vector float vec_loge (vector float);
9472 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
9473 vector unsigned char vec_lvsl (int, const volatile signed char *);
9474 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
9475 vector unsigned char vec_lvsl (int, const volatile short *);
9476 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
9477 vector unsigned char vec_lvsl (int, const volatile int *);
9478 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
9479 vector unsigned char vec_lvsl (int, const volatile long *);
9480 vector unsigned char vec_lvsl (int, const volatile float *);
9482 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
9483 vector unsigned char vec_lvsr (int, const volatile signed char *);
9484 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
9485 vector unsigned char vec_lvsr (int, const volatile short *);
9486 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
9487 vector unsigned char vec_lvsr (int, const volatile int *);
9488 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
9489 vector unsigned char vec_lvsr (int, const volatile long *);
9490 vector unsigned char vec_lvsr (int, const volatile float *);
9492 vector float vec_madd (vector float, vector float, vector float);
9494 vector signed short vec_madds (vector signed short,
9495 vector signed short,
9496 vector signed short);
9498 vector unsigned char vec_max (vector bool char, vector unsigned char);
9499 vector unsigned char vec_max (vector unsigned char, vector bool char);
9500 vector unsigned char vec_max (vector unsigned char,
9501 vector unsigned char);
9502 vector signed char vec_max (vector bool char, vector signed char);
9503 vector signed char vec_max (vector signed char, vector bool char);
9504 vector signed char vec_max (vector signed char, vector signed char);
9505 vector unsigned short vec_max (vector bool short,
9506 vector unsigned short);
9507 vector unsigned short vec_max (vector unsigned short,
9509 vector unsigned short vec_max (vector unsigned short,
9510 vector unsigned short);
9511 vector signed short vec_max (vector bool short, vector signed short);
9512 vector signed short vec_max (vector signed short, vector bool short);
9513 vector signed short vec_max (vector signed short, vector signed short);
9514 vector unsigned int vec_max (vector bool int, vector unsigned int);
9515 vector unsigned int vec_max (vector unsigned int, vector bool int);
9516 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
9517 vector signed int vec_max (vector bool int, vector signed int);
9518 vector signed int vec_max (vector signed int, vector bool int);
9519 vector signed int vec_max (vector signed int, vector signed int);
9520 vector float vec_max (vector float, vector float);
9522 vector float vec_vmaxfp (vector float, vector float);
9524 vector signed int vec_vmaxsw (vector bool int, vector signed int);
9525 vector signed int vec_vmaxsw (vector signed int, vector bool int);
9526 vector signed int vec_vmaxsw (vector signed int, vector signed int);
9528 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
9529 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
9530 vector unsigned int vec_vmaxuw (vector unsigned int,
9531 vector unsigned int);
9533 vector signed short vec_vmaxsh (vector bool short, vector signed short);
9534 vector signed short vec_vmaxsh (vector signed short, vector bool short);
9535 vector signed short vec_vmaxsh (vector signed short,
9536 vector signed short);
9538 vector unsigned short vec_vmaxuh (vector bool short,
9539 vector unsigned short);
9540 vector unsigned short vec_vmaxuh (vector unsigned short,
9542 vector unsigned short vec_vmaxuh (vector unsigned short,
9543 vector unsigned short);
9545 vector signed char vec_vmaxsb (vector bool char, vector signed char);
9546 vector signed char vec_vmaxsb (vector signed char, vector bool char);
9547 vector signed char vec_vmaxsb (vector signed char, vector signed char);
9549 vector unsigned char vec_vmaxub (vector bool char,
9550 vector unsigned char);
9551 vector unsigned char vec_vmaxub (vector unsigned char,
9553 vector unsigned char vec_vmaxub (vector unsigned char,
9554 vector unsigned char);
9556 vector bool char vec_mergeh (vector bool char, vector bool char);
9557 vector signed char vec_mergeh (vector signed char, vector signed char);
9558 vector unsigned char vec_mergeh (vector unsigned char,
9559 vector unsigned char);
9560 vector bool short vec_mergeh (vector bool short, vector bool short);
9561 vector pixel vec_mergeh (vector pixel, vector pixel);
9562 vector signed short vec_mergeh (vector signed short,
9563 vector signed short);
9564 vector unsigned short vec_mergeh (vector unsigned short,
9565 vector unsigned short);
9566 vector float vec_mergeh (vector float, vector float);
9567 vector bool int vec_mergeh (vector bool int, vector bool int);
9568 vector signed int vec_mergeh (vector signed int, vector signed int);
9569 vector unsigned int vec_mergeh (vector unsigned int,
9570 vector unsigned int);
9572 vector float vec_vmrghw (vector float, vector float);
9573 vector bool int vec_vmrghw (vector bool int, vector bool int);
9574 vector signed int vec_vmrghw (vector signed int, vector signed int);
9575 vector unsigned int vec_vmrghw (vector unsigned int,
9576 vector unsigned int);
9578 vector bool short vec_vmrghh (vector bool short, vector bool short);
9579 vector signed short vec_vmrghh (vector signed short,
9580 vector signed short);
9581 vector unsigned short vec_vmrghh (vector unsigned short,
9582 vector unsigned short);
9583 vector pixel vec_vmrghh (vector pixel, vector pixel);
9585 vector bool char vec_vmrghb (vector bool char, vector bool char);
9586 vector signed char vec_vmrghb (vector signed char, vector signed char);
9587 vector unsigned char vec_vmrghb (vector unsigned char,
9588 vector unsigned char);
9590 vector bool char vec_mergel (vector bool char, vector bool char);
9591 vector signed char vec_mergel (vector signed char, vector signed char);
9592 vector unsigned char vec_mergel (vector unsigned char,
9593 vector unsigned char);
9594 vector bool short vec_mergel (vector bool short, vector bool short);
9595 vector pixel vec_mergel (vector pixel, vector pixel);
9596 vector signed short vec_mergel (vector signed short,
9597 vector signed short);
9598 vector unsigned short vec_mergel (vector unsigned short,
9599 vector unsigned short);
9600 vector float vec_mergel (vector float, vector float);
9601 vector bool int vec_mergel (vector bool int, vector bool int);
9602 vector signed int vec_mergel (vector signed int, vector signed int);
9603 vector unsigned int vec_mergel (vector unsigned int,
9604 vector unsigned int);
9606 vector float vec_vmrglw (vector float, vector float);
9607 vector signed int vec_vmrglw (vector signed int, vector signed int);
9608 vector unsigned int vec_vmrglw (vector unsigned int,
9609 vector unsigned int);
9610 vector bool int vec_vmrglw (vector bool int, vector bool int);
9612 vector bool short vec_vmrglh (vector bool short, vector bool short);
9613 vector signed short vec_vmrglh (vector signed short,
9614 vector signed short);
9615 vector unsigned short vec_vmrglh (vector unsigned short,
9616 vector unsigned short);
9617 vector pixel vec_vmrglh (vector pixel, vector pixel);
9619 vector bool char vec_vmrglb (vector bool char, vector bool char);
9620 vector signed char vec_vmrglb (vector signed char, vector signed char);
9621 vector unsigned char vec_vmrglb (vector unsigned char,
9622 vector unsigned char);
9624 vector unsigned short vec_mfvscr (void);
9626 vector unsigned char vec_min (vector bool char, vector unsigned char);
9627 vector unsigned char vec_min (vector unsigned char, vector bool char);
9628 vector unsigned char vec_min (vector unsigned char,
9629 vector unsigned char);
9630 vector signed char vec_min (vector bool char, vector signed char);
9631 vector signed char vec_min (vector signed char, vector bool char);
9632 vector signed char vec_min (vector signed char, vector signed char);
9633 vector unsigned short vec_min (vector bool short,
9634 vector unsigned short);
9635 vector unsigned short vec_min (vector unsigned short,
9637 vector unsigned short vec_min (vector unsigned short,
9638 vector unsigned short);
9639 vector signed short vec_min (vector bool short, vector signed short);
9640 vector signed short vec_min (vector signed short, vector bool short);
9641 vector signed short vec_min (vector signed short, vector signed short);
9642 vector unsigned int vec_min (vector bool int, vector unsigned int);
9643 vector unsigned int vec_min (vector unsigned int, vector bool int);
9644 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
9645 vector signed int vec_min (vector bool int, vector signed int);
9646 vector signed int vec_min (vector signed int, vector bool int);
9647 vector signed int vec_min (vector signed int, vector signed int);
9648 vector float vec_min (vector float, vector float);
9650 vector float vec_vminfp (vector float, vector float);
9652 vector signed int vec_vminsw (vector bool int, vector signed int);
9653 vector signed int vec_vminsw (vector signed int, vector bool int);
9654 vector signed int vec_vminsw (vector signed int, vector signed int);
9656 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
9657 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
9658 vector unsigned int vec_vminuw (vector unsigned int,
9659 vector unsigned int);
9661 vector signed short vec_vminsh (vector bool short, vector signed short);
9662 vector signed short vec_vminsh (vector signed short, vector bool short);
9663 vector signed short vec_vminsh (vector signed short,
9664 vector signed short);
9666 vector unsigned short vec_vminuh (vector bool short,
9667 vector unsigned short);
9668 vector unsigned short vec_vminuh (vector unsigned short,
9670 vector unsigned short vec_vminuh (vector unsigned short,
9671 vector unsigned short);
9673 vector signed char vec_vminsb (vector bool char, vector signed char);
9674 vector signed char vec_vminsb (vector signed char, vector bool char);
9675 vector signed char vec_vminsb (vector signed char, vector signed char);
9677 vector unsigned char vec_vminub (vector bool char,
9678 vector unsigned char);
9679 vector unsigned char vec_vminub (vector unsigned char,
9681 vector unsigned char vec_vminub (vector unsigned char,
9682 vector unsigned char);
9684 vector signed short vec_mladd (vector signed short,
9685 vector signed short,
9686 vector signed short);
9687 vector signed short vec_mladd (vector signed short,
9688 vector unsigned short,
9689 vector unsigned short);
9690 vector signed short vec_mladd (vector unsigned short,
9691 vector signed short,
9692 vector signed short);
9693 vector unsigned short vec_mladd (vector unsigned short,
9694 vector unsigned short,
9695 vector unsigned short);
9697 vector signed short vec_mradds (vector signed short,
9698 vector signed short,
9699 vector signed short);
9701 vector unsigned int vec_msum (vector unsigned char,
9702 vector unsigned char,
9703 vector unsigned int);
9704 vector signed int vec_msum (vector signed char,
9705 vector unsigned char,
9707 vector unsigned int vec_msum (vector unsigned short,
9708 vector unsigned short,
9709 vector unsigned int);
9710 vector signed int vec_msum (vector signed short,
9711 vector signed short,
9714 vector signed int vec_vmsumshm (vector signed short,
9715 vector signed short,
9718 vector unsigned int vec_vmsumuhm (vector unsigned short,
9719 vector unsigned short,
9720 vector unsigned int);
9722 vector signed int vec_vmsummbm (vector signed char,
9723 vector unsigned char,
9726 vector unsigned int vec_vmsumubm (vector unsigned char,
9727 vector unsigned char,
9728 vector unsigned int);
9730 vector unsigned int vec_msums (vector unsigned short,
9731 vector unsigned short,
9732 vector unsigned int);
9733 vector signed int vec_msums (vector signed short,
9734 vector signed short,
9737 vector signed int vec_vmsumshs (vector signed short,
9738 vector signed short,
9741 vector unsigned int vec_vmsumuhs (vector unsigned short,
9742 vector unsigned short,
9743 vector unsigned int);
9745 void vec_mtvscr (vector signed int);
9746 void vec_mtvscr (vector unsigned int);
9747 void vec_mtvscr (vector bool int);
9748 void vec_mtvscr (vector signed short);
9749 void vec_mtvscr (vector unsigned short);
9750 void vec_mtvscr (vector bool short);
9751 void vec_mtvscr (vector pixel);
9752 void vec_mtvscr (vector signed char);
9753 void vec_mtvscr (vector unsigned char);
9754 void vec_mtvscr (vector bool char);
9756 vector unsigned short vec_mule (vector unsigned char,
9757 vector unsigned char);
9758 vector signed short vec_mule (vector signed char,
9759 vector signed char);
9760 vector unsigned int vec_mule (vector unsigned short,
9761 vector unsigned short);
9762 vector signed int vec_mule (vector signed short, vector signed short);
9764 vector signed int vec_vmulesh (vector signed short,
9765 vector signed short);
9767 vector unsigned int vec_vmuleuh (vector unsigned short,
9768 vector unsigned short);
9770 vector signed short vec_vmulesb (vector signed char,
9771 vector signed char);
9773 vector unsigned short vec_vmuleub (vector unsigned char,
9774 vector unsigned char);
9776 vector unsigned short vec_mulo (vector unsigned char,
9777 vector unsigned char);
9778 vector signed short vec_mulo (vector signed char, vector signed char);
9779 vector unsigned int vec_mulo (vector unsigned short,
9780 vector unsigned short);
9781 vector signed int vec_mulo (vector signed short, vector signed short);
9783 vector signed int vec_vmulosh (vector signed short,
9784 vector signed short);
9786 vector unsigned int vec_vmulouh (vector unsigned short,
9787 vector unsigned short);
9789 vector signed short vec_vmulosb (vector signed char,
9790 vector signed char);
9792 vector unsigned short vec_vmuloub (vector unsigned char,
9793 vector unsigned char);
9795 vector float vec_nmsub (vector float, vector float, vector float);
9797 vector float vec_nor (vector float, vector float);
9798 vector signed int vec_nor (vector signed int, vector signed int);
9799 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
9800 vector bool int vec_nor (vector bool int, vector bool int);
9801 vector signed short vec_nor (vector signed short, vector signed short);
9802 vector unsigned short vec_nor (vector unsigned short,
9803 vector unsigned short);
9804 vector bool short vec_nor (vector bool short, vector bool short);
9805 vector signed char vec_nor (vector signed char, vector signed char);
9806 vector unsigned char vec_nor (vector unsigned char,
9807 vector unsigned char);
9808 vector bool char vec_nor (vector bool char, vector bool char);
9810 vector float vec_or (vector float, vector float);
9811 vector float vec_or (vector float, vector bool int);
9812 vector float vec_or (vector bool int, vector float);
9813 vector bool int vec_or (vector bool int, vector bool int);
9814 vector signed int vec_or (vector bool int, vector signed int);
9815 vector signed int vec_or (vector signed int, vector bool int);
9816 vector signed int vec_or (vector signed int, vector signed int);
9817 vector unsigned int vec_or (vector bool int, vector unsigned int);
9818 vector unsigned int vec_or (vector unsigned int, vector bool int);
9819 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
9820 vector bool short vec_or (vector bool short, vector bool short);
9821 vector signed short vec_or (vector bool short, vector signed short);
9822 vector signed short vec_or (vector signed short, vector bool short);
9823 vector signed short vec_or (vector signed short, vector signed short);
9824 vector unsigned short vec_or (vector bool short, vector unsigned short);
9825 vector unsigned short vec_or (vector unsigned short, vector bool short);
9826 vector unsigned short vec_or (vector unsigned short,
9827 vector unsigned short);
9828 vector signed char vec_or (vector bool char, vector signed char);
9829 vector bool char vec_or (vector bool char, vector bool char);
9830 vector signed char vec_or (vector signed char, vector bool char);
9831 vector signed char vec_or (vector signed char, vector signed char);
9832 vector unsigned char vec_or (vector bool char, vector unsigned char);
9833 vector unsigned char vec_or (vector unsigned char, vector bool char);
9834 vector unsigned char vec_or (vector unsigned char,
9835 vector unsigned char);
9837 vector signed char vec_pack (vector signed short, vector signed short);
9838 vector unsigned char vec_pack (vector unsigned short,
9839 vector unsigned short);
9840 vector bool char vec_pack (vector bool short, vector bool short);
9841 vector signed short vec_pack (vector signed int, vector signed int);
9842 vector unsigned short vec_pack (vector unsigned int,
9843 vector unsigned int);
9844 vector bool short vec_pack (vector bool int, vector bool int);
9846 vector bool short vec_vpkuwum (vector bool int, vector bool int);
9847 vector signed short vec_vpkuwum (vector signed int, vector signed int);
9848 vector unsigned short vec_vpkuwum (vector unsigned int,
9849 vector unsigned int);
9851 vector bool char vec_vpkuhum (vector bool short, vector bool short);
9852 vector signed char vec_vpkuhum (vector signed short,
9853 vector signed short);
9854 vector unsigned char vec_vpkuhum (vector unsigned short,
9855 vector unsigned short);
9857 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
9859 vector unsigned char vec_packs (vector unsigned short,
9860 vector unsigned short);
9861 vector signed char vec_packs (vector signed short, vector signed short);
9862 vector unsigned short vec_packs (vector unsigned int,
9863 vector unsigned int);
9864 vector signed short vec_packs (vector signed int, vector signed int);
9866 vector signed short vec_vpkswss (vector signed int, vector signed int);
9868 vector unsigned short vec_vpkuwus (vector unsigned int,
9869 vector unsigned int);
9871 vector signed char vec_vpkshss (vector signed short,
9872 vector signed short);
9874 vector unsigned char vec_vpkuhus (vector unsigned short,
9875 vector unsigned short);
9877 vector unsigned char vec_packsu (vector unsigned short,
9878 vector unsigned short);
9879 vector unsigned char vec_packsu (vector signed short,
9880 vector signed short);
9881 vector unsigned short vec_packsu (vector unsigned int,
9882 vector unsigned int);
9883 vector unsigned short vec_packsu (vector signed int, vector signed int);
9885 vector unsigned short vec_vpkswus (vector signed int,
9888 vector unsigned char vec_vpkshus (vector signed short,
9889 vector signed short);
9891 vector float vec_perm (vector float,
9893 vector unsigned char);
9894 vector signed int vec_perm (vector signed int,
9896 vector unsigned char);
9897 vector unsigned int vec_perm (vector unsigned int,
9898 vector unsigned int,
9899 vector unsigned char);
9900 vector bool int vec_perm (vector bool int,
9902 vector unsigned char);
9903 vector signed short vec_perm (vector signed short,
9904 vector signed short,
9905 vector unsigned char);
9906 vector unsigned short vec_perm (vector unsigned short,
9907 vector unsigned short,
9908 vector unsigned char);
9909 vector bool short vec_perm (vector bool short,
9911 vector unsigned char);
9912 vector pixel vec_perm (vector pixel,
9914 vector unsigned char);
9915 vector signed char vec_perm (vector signed char,
9917 vector unsigned char);
9918 vector unsigned char vec_perm (vector unsigned char,
9919 vector unsigned char,
9920 vector unsigned char);
9921 vector bool char vec_perm (vector bool char,
9923 vector unsigned char);
9925 vector float vec_re (vector float);
9927 vector signed char vec_rl (vector signed char,
9928 vector unsigned char);
9929 vector unsigned char vec_rl (vector unsigned char,
9930 vector unsigned char);
9931 vector signed short vec_rl (vector signed short, vector unsigned short);
9932 vector unsigned short vec_rl (vector unsigned short,
9933 vector unsigned short);
9934 vector signed int vec_rl (vector signed int, vector unsigned int);
9935 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
9937 vector signed int vec_vrlw (vector signed int, vector unsigned int);
9938 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
9940 vector signed short vec_vrlh (vector signed short,
9941 vector unsigned short);
9942 vector unsigned short vec_vrlh (vector unsigned short,
9943 vector unsigned short);
9945 vector signed char vec_vrlb (vector signed char, vector unsigned char);
9946 vector unsigned char vec_vrlb (vector unsigned char,
9947 vector unsigned char);
9949 vector float vec_round (vector float);
9951 vector float vec_rsqrte (vector float);
9953 vector float vec_sel (vector float, vector float, vector bool int);
9954 vector float vec_sel (vector float, vector float, vector unsigned int);
9955 vector signed int vec_sel (vector signed int,
9958 vector signed int vec_sel (vector signed int,
9960 vector unsigned int);
9961 vector unsigned int vec_sel (vector unsigned int,
9962 vector unsigned int,
9964 vector unsigned int vec_sel (vector unsigned int,
9965 vector unsigned int,
9966 vector unsigned int);
9967 vector bool int vec_sel (vector bool int,
9970 vector bool int vec_sel (vector bool int,
9972 vector unsigned int);
9973 vector signed short vec_sel (vector signed short,
9974 vector signed short,
9976 vector signed short vec_sel (vector signed short,
9977 vector signed short,
9978 vector unsigned short);
9979 vector unsigned short vec_sel (vector unsigned short,
9980 vector unsigned short,
9982 vector unsigned short vec_sel (vector unsigned short,
9983 vector unsigned short,
9984 vector unsigned short);
9985 vector bool short vec_sel (vector bool short,
9988 vector bool short vec_sel (vector bool short,
9990 vector unsigned short);
9991 vector signed char vec_sel (vector signed char,
9994 vector signed char vec_sel (vector signed char,
9996 vector unsigned char);
9997 vector unsigned char vec_sel (vector unsigned char,
9998 vector unsigned char,
10000 vector unsigned char vec_sel (vector unsigned char,
10001 vector unsigned char,
10002 vector unsigned char);
10003 vector bool char vec_sel (vector bool char,
10006 vector bool char vec_sel (vector bool char,
10008 vector unsigned char);
10010 vector signed char vec_sl (vector signed char,
10011 vector unsigned char);
10012 vector unsigned char vec_sl (vector unsigned char,
10013 vector unsigned char);
10014 vector signed short vec_sl (vector signed short, vector unsigned short);
10015 vector unsigned short vec_sl (vector unsigned short,
10016 vector unsigned short);
10017 vector signed int vec_sl (vector signed int, vector unsigned int);
10018 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
10020 vector signed int vec_vslw (vector signed int, vector unsigned int);
10021 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
10023 vector signed short vec_vslh (vector signed short,
10024 vector unsigned short);
10025 vector unsigned short vec_vslh (vector unsigned short,
10026 vector unsigned short);
10028 vector signed char vec_vslb (vector signed char, vector unsigned char);
10029 vector unsigned char vec_vslb (vector unsigned char,
10030 vector unsigned char);
10032 vector float vec_sld (vector float, vector float, const int);
10033 vector signed int vec_sld (vector signed int,
10036 vector unsigned int vec_sld (vector unsigned int,
10037 vector unsigned int,
10039 vector bool int vec_sld (vector bool int,
10042 vector signed short vec_sld (vector signed short,
10043 vector signed short,
10045 vector unsigned short vec_sld (vector unsigned short,
10046 vector unsigned short,
10048 vector bool short vec_sld (vector bool short,
10051 vector pixel vec_sld (vector pixel,
10054 vector signed char vec_sld (vector signed char,
10055 vector signed char,
10057 vector unsigned char vec_sld (vector unsigned char,
10058 vector unsigned char,
10060 vector bool char vec_sld (vector bool char,
10064 vector signed int vec_sll (vector signed int,
10065 vector unsigned int);
10066 vector signed int vec_sll (vector signed int,
10067 vector unsigned short);
10068 vector signed int vec_sll (vector signed int,
10069 vector unsigned char);
10070 vector unsigned int vec_sll (vector unsigned int,
10071 vector unsigned int);
10072 vector unsigned int vec_sll (vector unsigned int,
10073 vector unsigned short);
10074 vector unsigned int vec_sll (vector unsigned int,
10075 vector unsigned char);
10076 vector bool int vec_sll (vector bool int,
10077 vector unsigned int);
10078 vector bool int vec_sll (vector bool int,
10079 vector unsigned short);
10080 vector bool int vec_sll (vector bool int,
10081 vector unsigned char);
10082 vector signed short vec_sll (vector signed short,
10083 vector unsigned int);
10084 vector signed short vec_sll (vector signed short,
10085 vector unsigned short);
10086 vector signed short vec_sll (vector signed short,
10087 vector unsigned char);
10088 vector unsigned short vec_sll (vector unsigned short,
10089 vector unsigned int);
10090 vector unsigned short vec_sll (vector unsigned short,
10091 vector unsigned short);
10092 vector unsigned short vec_sll (vector unsigned short,
10093 vector unsigned char);
10094 vector bool short vec_sll (vector bool short, vector unsigned int);
10095 vector bool short vec_sll (vector bool short, vector unsigned short);
10096 vector bool short vec_sll (vector bool short, vector unsigned char);
10097 vector pixel vec_sll (vector pixel, vector unsigned int);
10098 vector pixel vec_sll (vector pixel, vector unsigned short);
10099 vector pixel vec_sll (vector pixel, vector unsigned char);
10100 vector signed char vec_sll (vector signed char, vector unsigned int);
10101 vector signed char vec_sll (vector signed char, vector unsigned short);
10102 vector signed char vec_sll (vector signed char, vector unsigned char);
10103 vector unsigned char vec_sll (vector unsigned char,
10104 vector unsigned int);
10105 vector unsigned char vec_sll (vector unsigned char,
10106 vector unsigned short);
10107 vector unsigned char vec_sll (vector unsigned char,
10108 vector unsigned char);
10109 vector bool char vec_sll (vector bool char, vector unsigned int);
10110 vector bool char vec_sll (vector bool char, vector unsigned short);
10111 vector bool char vec_sll (vector bool char, vector unsigned char);
10113 vector float vec_slo (vector float, vector signed char);
10114 vector float vec_slo (vector float, vector unsigned char);
10115 vector signed int vec_slo (vector signed int, vector signed char);
10116 vector signed int vec_slo (vector signed int, vector unsigned char);
10117 vector unsigned int vec_slo (vector unsigned int, vector signed char);
10118 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
10119 vector signed short vec_slo (vector signed short, vector signed char);
10120 vector signed short vec_slo (vector signed short, vector unsigned char);
10121 vector unsigned short vec_slo (vector unsigned short,
10122 vector signed char);
10123 vector unsigned short vec_slo (vector unsigned short,
10124 vector unsigned char);
10125 vector pixel vec_slo (vector pixel, vector signed char);
10126 vector pixel vec_slo (vector pixel, vector unsigned char);
10127 vector signed char vec_slo (vector signed char, vector signed char);
10128 vector signed char vec_slo (vector signed char, vector unsigned char);
10129 vector unsigned char vec_slo (vector unsigned char, vector signed char);
10130 vector unsigned char vec_slo (vector unsigned char,
10131 vector unsigned char);
10133 vector signed char vec_splat (vector signed char, const int);
10134 vector unsigned char vec_splat (vector unsigned char, const int);
10135 vector bool char vec_splat (vector bool char, const int);
10136 vector signed short vec_splat (vector signed short, const int);
10137 vector unsigned short vec_splat (vector unsigned short, const int);
10138 vector bool short vec_splat (vector bool short, const int);
10139 vector pixel vec_splat (vector pixel, const int);
10140 vector float vec_splat (vector float, const int);
10141 vector signed int vec_splat (vector signed int, const int);
10142 vector unsigned int vec_splat (vector unsigned int, const int);
10143 vector bool int vec_splat (vector bool int, const int);
10145 vector float vec_vspltw (vector float, const int);
10146 vector signed int vec_vspltw (vector signed int, const int);
10147 vector unsigned int vec_vspltw (vector unsigned int, const int);
10148 vector bool int vec_vspltw (vector bool int, const int);
10150 vector bool short vec_vsplth (vector bool short, const int);
10151 vector signed short vec_vsplth (vector signed short, const int);
10152 vector unsigned short vec_vsplth (vector unsigned short, const int);
10153 vector pixel vec_vsplth (vector pixel, const int);
10155 vector signed char vec_vspltb (vector signed char, const int);
10156 vector unsigned char vec_vspltb (vector unsigned char, const int);
10157 vector bool char vec_vspltb (vector bool char, const int);
10159 vector signed char vec_splat_s8 (const int);
10161 vector signed short vec_splat_s16 (const int);
10163 vector signed int vec_splat_s32 (const int);
10165 vector unsigned char vec_splat_u8 (const int);
10167 vector unsigned short vec_splat_u16 (const int);
10169 vector unsigned int vec_splat_u32 (const int);
10171 vector signed char vec_sr (vector signed char, vector unsigned char);
10172 vector unsigned char vec_sr (vector unsigned char,
10173 vector unsigned char);
10174 vector signed short vec_sr (vector signed short,
10175 vector unsigned short);
10176 vector unsigned short vec_sr (vector unsigned short,
10177 vector unsigned short);
10178 vector signed int vec_sr (vector signed int, vector unsigned int);
10179 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
10181 vector signed int vec_vsrw (vector signed int, vector unsigned int);
10182 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
10184 vector signed short vec_vsrh (vector signed short,
10185 vector unsigned short);
10186 vector unsigned short vec_vsrh (vector unsigned short,
10187 vector unsigned short);
10189 vector signed char vec_vsrb (vector signed char, vector unsigned char);
10190 vector unsigned char vec_vsrb (vector unsigned char,
10191 vector unsigned char);
10193 vector signed char vec_sra (vector signed char, vector unsigned char);
10194 vector unsigned char vec_sra (vector unsigned char,
10195 vector unsigned char);
10196 vector signed short vec_sra (vector signed short,
10197 vector unsigned short);
10198 vector unsigned short vec_sra (vector unsigned short,
10199 vector unsigned short);
10200 vector signed int vec_sra (vector signed int, vector unsigned int);
10201 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
10203 vector signed int vec_vsraw (vector signed int, vector unsigned int);
10204 vector unsigned int vec_vsraw (vector unsigned int,
10205 vector unsigned int);
10207 vector signed short vec_vsrah (vector signed short,
10208 vector unsigned short);
10209 vector unsigned short vec_vsrah (vector unsigned short,
10210 vector unsigned short);
10212 vector signed char vec_vsrab (vector signed char, vector unsigned char);
10213 vector unsigned char vec_vsrab (vector unsigned char,
10214 vector unsigned char);
10216 vector signed int vec_srl (vector signed int, vector unsigned int);
10217 vector signed int vec_srl (vector signed int, vector unsigned short);
10218 vector signed int vec_srl (vector signed int, vector unsigned char);
10219 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
10220 vector unsigned int vec_srl (vector unsigned int,
10221 vector unsigned short);
10222 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
10223 vector bool int vec_srl (vector bool int, vector unsigned int);
10224 vector bool int vec_srl (vector bool int, vector unsigned short);
10225 vector bool int vec_srl (vector bool int, vector unsigned char);
10226 vector signed short vec_srl (vector signed short, vector unsigned int);
10227 vector signed short vec_srl (vector signed short,
10228 vector unsigned short);
10229 vector signed short vec_srl (vector signed short, vector unsigned char);
10230 vector unsigned short vec_srl (vector unsigned short,
10231 vector unsigned int);
10232 vector unsigned short vec_srl (vector unsigned short,
10233 vector unsigned short);
10234 vector unsigned short vec_srl (vector unsigned short,
10235 vector unsigned char);
10236 vector bool short vec_srl (vector bool short, vector unsigned int);
10237 vector bool short vec_srl (vector bool short, vector unsigned short);
10238 vector bool short vec_srl (vector bool short, vector unsigned char);
10239 vector pixel vec_srl (vector pixel, vector unsigned int);
10240 vector pixel vec_srl (vector pixel, vector unsigned short);
10241 vector pixel vec_srl (vector pixel, vector unsigned char);
10242 vector signed char vec_srl (vector signed char, vector unsigned int);
10243 vector signed char vec_srl (vector signed char, vector unsigned short);
10244 vector signed char vec_srl (vector signed char, vector unsigned char);
10245 vector unsigned char vec_srl (vector unsigned char,
10246 vector unsigned int);
10247 vector unsigned char vec_srl (vector unsigned char,
10248 vector unsigned short);
10249 vector unsigned char vec_srl (vector unsigned char,
10250 vector unsigned char);
10251 vector bool char vec_srl (vector bool char, vector unsigned int);
10252 vector bool char vec_srl (vector bool char, vector unsigned short);
10253 vector bool char vec_srl (vector bool char, vector unsigned char);
10255 vector float vec_sro (vector float, vector signed char);
10256 vector float vec_sro (vector float, vector unsigned char);
10257 vector signed int vec_sro (vector signed int, vector signed char);
10258 vector signed int vec_sro (vector signed int, vector unsigned char);
10259 vector unsigned int vec_sro (vector unsigned int, vector signed char);
10260 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
10261 vector signed short vec_sro (vector signed short, vector signed char);
10262 vector signed short vec_sro (vector signed short, vector unsigned char);
10263 vector unsigned short vec_sro (vector unsigned short,
10264 vector signed char);
10265 vector unsigned short vec_sro (vector unsigned short,
10266 vector unsigned char);
10267 vector pixel vec_sro (vector pixel, vector signed char);
10268 vector pixel vec_sro (vector pixel, vector unsigned char);
10269 vector signed char vec_sro (vector signed char, vector signed char);
10270 vector signed char vec_sro (vector signed char, vector unsigned char);
10271 vector unsigned char vec_sro (vector unsigned char, vector signed char);
10272 vector unsigned char vec_sro (vector unsigned char,
10273 vector unsigned char);
10275 void vec_st (vector float, int, vector float *);
10276 void vec_st (vector float, int, float *);
10277 void vec_st (vector signed int, int, vector signed int *);
10278 void vec_st (vector signed int, int, int *);
10279 void vec_st (vector unsigned int, int, vector unsigned int *);
10280 void vec_st (vector unsigned int, int, unsigned int *);
10281 void vec_st (vector bool int, int, vector bool int *);
10282 void vec_st (vector bool int, int, unsigned int *);
10283 void vec_st (vector bool int, int, int *);
10284 void vec_st (vector signed short, int, vector signed short *);
10285 void vec_st (vector signed short, int, short *);
10286 void vec_st (vector unsigned short, int, vector unsigned short *);
10287 void vec_st (vector unsigned short, int, unsigned short *);
10288 void vec_st (vector bool short, int, vector bool short *);
10289 void vec_st (vector bool short, int, unsigned short *);
10290 void vec_st (vector pixel, int, vector pixel *);
10291 void vec_st (vector pixel, int, unsigned short *);
10292 void vec_st (vector pixel, int, short *);
10293 void vec_st (vector bool short, int, short *);
10294 void vec_st (vector signed char, int, vector signed char *);
10295 void vec_st (vector signed char, int, signed char *);
10296 void vec_st (vector unsigned char, int, vector unsigned char *);
10297 void vec_st (vector unsigned char, int, unsigned char *);
10298 void vec_st (vector bool char, int, vector bool char *);
10299 void vec_st (vector bool char, int, unsigned char *);
10300 void vec_st (vector bool char, int, signed char *);
10302 void vec_ste (vector signed char, int, signed char *);
10303 void vec_ste (vector unsigned char, int, unsigned char *);
10304 void vec_ste (vector bool char, int, signed char *);
10305 void vec_ste (vector bool char, int, unsigned char *);
10306 void vec_ste (vector signed short, int, short *);
10307 void vec_ste (vector unsigned short, int, unsigned short *);
10308 void vec_ste (vector bool short, int, short *);
10309 void vec_ste (vector bool short, int, unsigned short *);
10310 void vec_ste (vector pixel, int, short *);
10311 void vec_ste (vector pixel, int, unsigned short *);
10312 void vec_ste (vector float, int, float *);
10313 void vec_ste (vector signed int, int, int *);
10314 void vec_ste (vector unsigned int, int, unsigned int *);
10315 void vec_ste (vector bool int, int, int *);
10316 void vec_ste (vector bool int, int, unsigned int *);
10318 void vec_stvewx (vector float, int, float *);
10319 void vec_stvewx (vector signed int, int, int *);
10320 void vec_stvewx (vector unsigned int, int, unsigned int *);
10321 void vec_stvewx (vector bool int, int, int *);
10322 void vec_stvewx (vector bool int, int, unsigned int *);
10324 void vec_stvehx (vector signed short, int, short *);
10325 void vec_stvehx (vector unsigned short, int, unsigned short *);
10326 void vec_stvehx (vector bool short, int, short *);
10327 void vec_stvehx (vector bool short, int, unsigned short *);
10328 void vec_stvehx (vector pixel, int, short *);
10329 void vec_stvehx (vector pixel, int, unsigned short *);
10331 void vec_stvebx (vector signed char, int, signed char *);
10332 void vec_stvebx (vector unsigned char, int, unsigned char *);
10333 void vec_stvebx (vector bool char, int, signed char *);
10334 void vec_stvebx (vector bool char, int, unsigned char *);
10336 void vec_stl (vector float, int, vector float *);
10337 void vec_stl (vector float, int, float *);
10338 void vec_stl (vector signed int, int, vector signed int *);
10339 void vec_stl (vector signed int, int, int *);
10340 void vec_stl (vector unsigned int, int, vector unsigned int *);
10341 void vec_stl (vector unsigned int, int, unsigned int *);
10342 void vec_stl (vector bool int, int, vector bool int *);
10343 void vec_stl (vector bool int, int, unsigned int *);
10344 void vec_stl (vector bool int, int, int *);
10345 void vec_stl (vector signed short, int, vector signed short *);
10346 void vec_stl (vector signed short, int, short *);
10347 void vec_stl (vector unsigned short, int, vector unsigned short *);
10348 void vec_stl (vector unsigned short, int, unsigned short *);
10349 void vec_stl (vector bool short, int, vector bool short *);
10350 void vec_stl (vector bool short, int, unsigned short *);
10351 void vec_stl (vector bool short, int, short *);
10352 void vec_stl (vector pixel, int, vector pixel *);
10353 void vec_stl (vector pixel, int, unsigned short *);
10354 void vec_stl (vector pixel, int, short *);
10355 void vec_stl (vector signed char, int, vector signed char *);
10356 void vec_stl (vector signed char, int, signed char *);
10357 void vec_stl (vector unsigned char, int, vector unsigned char *);
10358 void vec_stl (vector unsigned char, int, unsigned char *);
10359 void vec_stl (vector bool char, int, vector bool char *);
10360 void vec_stl (vector bool char, int, unsigned char *);
10361 void vec_stl (vector bool char, int, signed char *);
10363 vector signed char vec_sub (vector bool char, vector signed char);
10364 vector signed char vec_sub (vector signed char, vector bool char);
10365 vector signed char vec_sub (vector signed char, vector signed char);
10366 vector unsigned char vec_sub (vector bool char, vector unsigned char);
10367 vector unsigned char vec_sub (vector unsigned char, vector bool char);
10368 vector unsigned char vec_sub (vector unsigned char,
10369 vector unsigned char);
10370 vector signed short vec_sub (vector bool short, vector signed short);
10371 vector signed short vec_sub (vector signed short, vector bool short);
10372 vector signed short vec_sub (vector signed short, vector signed short);
10373 vector unsigned short vec_sub (vector bool short,
10374 vector unsigned short);
10375 vector unsigned short vec_sub (vector unsigned short,
10376 vector bool short);
10377 vector unsigned short vec_sub (vector unsigned short,
10378 vector unsigned short);
10379 vector signed int vec_sub (vector bool int, vector signed int);
10380 vector signed int vec_sub (vector signed int, vector bool int);
10381 vector signed int vec_sub (vector signed int, vector signed int);
10382 vector unsigned int vec_sub (vector bool int, vector unsigned int);
10383 vector unsigned int vec_sub (vector unsigned int, vector bool int);
10384 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
10385 vector float vec_sub (vector float, vector float);
10387 vector float vec_vsubfp (vector float, vector float);
10389 vector signed int vec_vsubuwm (vector bool int, vector signed int);
10390 vector signed int vec_vsubuwm (vector signed int, vector bool int);
10391 vector signed int vec_vsubuwm (vector signed int, vector signed int);
10392 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
10393 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
10394 vector unsigned int vec_vsubuwm (vector unsigned int,
10395 vector unsigned int);
10397 vector signed short vec_vsubuhm (vector bool short,
10398 vector signed short);
10399 vector signed short vec_vsubuhm (vector signed short,
10400 vector bool short);
10401 vector signed short vec_vsubuhm (vector signed short,
10402 vector signed short);
10403 vector unsigned short vec_vsubuhm (vector bool short,
10404 vector unsigned short);
10405 vector unsigned short vec_vsubuhm (vector unsigned short,
10406 vector bool short);
10407 vector unsigned short vec_vsubuhm (vector unsigned short,
10408 vector unsigned short);
10410 vector signed char vec_vsububm (vector bool char, vector signed char);
10411 vector signed char vec_vsububm (vector signed char, vector bool char);
10412 vector signed char vec_vsububm (vector signed char, vector signed char);
10413 vector unsigned char vec_vsububm (vector bool char,
10414 vector unsigned char);
10415 vector unsigned char vec_vsububm (vector unsigned char,
10417 vector unsigned char vec_vsububm (vector unsigned char,
10418 vector unsigned char);
10420 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
10422 vector unsigned char vec_subs (vector bool char, vector unsigned char);
10423 vector unsigned char vec_subs (vector unsigned char, vector bool char);
10424 vector unsigned char vec_subs (vector unsigned char,
10425 vector unsigned char);
10426 vector signed char vec_subs (vector bool char, vector signed char);
10427 vector signed char vec_subs (vector signed char, vector bool char);
10428 vector signed char vec_subs (vector signed char, vector signed char);
10429 vector unsigned short vec_subs (vector bool short,
10430 vector unsigned short);
10431 vector unsigned short vec_subs (vector unsigned short,
10432 vector bool short);
10433 vector unsigned short vec_subs (vector unsigned short,
10434 vector unsigned short);
10435 vector signed short vec_subs (vector bool short, vector signed short);
10436 vector signed short vec_subs (vector signed short, vector bool short);
10437 vector signed short vec_subs (vector signed short, vector signed short);
10438 vector unsigned int vec_subs (vector bool int, vector unsigned int);
10439 vector unsigned int vec_subs (vector unsigned int, vector bool int);
10440 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
10441 vector signed int vec_subs (vector bool int, vector signed int);
10442 vector signed int vec_subs (vector signed int, vector bool int);
10443 vector signed int vec_subs (vector signed int, vector signed int);
10445 vector signed int vec_vsubsws (vector bool int, vector signed int);
10446 vector signed int vec_vsubsws (vector signed int, vector bool int);
10447 vector signed int vec_vsubsws (vector signed int, vector signed int);
10449 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
10450 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
10451 vector unsigned int vec_vsubuws (vector unsigned int,
10452 vector unsigned int);
10454 vector signed short vec_vsubshs (vector bool short,
10455 vector signed short);
10456 vector signed short vec_vsubshs (vector signed short,
10457 vector bool short);
10458 vector signed short vec_vsubshs (vector signed short,
10459 vector signed short);
10461 vector unsigned short vec_vsubuhs (vector bool short,
10462 vector unsigned short);
10463 vector unsigned short vec_vsubuhs (vector unsigned short,
10464 vector bool short);
10465 vector unsigned short vec_vsubuhs (vector unsigned short,
10466 vector unsigned short);
10468 vector signed char vec_vsubsbs (vector bool char, vector signed char);
10469 vector signed char vec_vsubsbs (vector signed char, vector bool char);
10470 vector signed char vec_vsubsbs (vector signed char, vector signed char);
10472 vector unsigned char vec_vsububs (vector bool char,
10473 vector unsigned char);
10474 vector unsigned char vec_vsububs (vector unsigned char,
10476 vector unsigned char vec_vsububs (vector unsigned char,
10477 vector unsigned char);
10479 vector unsigned int vec_sum4s (vector unsigned char,
10480 vector unsigned int);
10481 vector signed int vec_sum4s (vector signed char, vector signed int);
10482 vector signed int vec_sum4s (vector signed short, vector signed int);
10484 vector signed int vec_vsum4shs (vector signed short, vector signed int);
10486 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
10488 vector unsigned int vec_vsum4ubs (vector unsigned char,
10489 vector unsigned int);
10491 vector signed int vec_sum2s (vector signed int, vector signed int);
10493 vector signed int vec_sums (vector signed int, vector signed int);
10495 vector float vec_trunc (vector float);
10497 vector signed short vec_unpackh (vector signed char);
10498 vector bool short vec_unpackh (vector bool char);
10499 vector signed int vec_unpackh (vector signed short);
10500 vector bool int vec_unpackh (vector bool short);
10501 vector unsigned int vec_unpackh (vector pixel);
10503 vector bool int vec_vupkhsh (vector bool short);
10504 vector signed int vec_vupkhsh (vector signed short);
10506 vector unsigned int vec_vupkhpx (vector pixel);
10508 vector bool short vec_vupkhsb (vector bool char);
10509 vector signed short vec_vupkhsb (vector signed char);
10511 vector signed short vec_unpackl (vector signed char);
10512 vector bool short vec_unpackl (vector bool char);
10513 vector unsigned int vec_unpackl (vector pixel);
10514 vector signed int vec_unpackl (vector signed short);
10515 vector bool int vec_unpackl (vector bool short);
10517 vector unsigned int vec_vupklpx (vector pixel);
10519 vector bool int vec_vupklsh (vector bool short);
10520 vector signed int vec_vupklsh (vector signed short);
10522 vector bool short vec_vupklsb (vector bool char);
10523 vector signed short vec_vupklsb (vector signed char);
10525 vector float vec_xor (vector float, vector float);
10526 vector float vec_xor (vector float, vector bool int);
10527 vector float vec_xor (vector bool int, vector float);
10528 vector bool int vec_xor (vector bool int, vector bool int);
10529 vector signed int vec_xor (vector bool int, vector signed int);
10530 vector signed int vec_xor (vector signed int, vector bool int);
10531 vector signed int vec_xor (vector signed int, vector signed int);
10532 vector unsigned int vec_xor (vector bool int, vector unsigned int);
10533 vector unsigned int vec_xor (vector unsigned int, vector bool int);
10534 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
10535 vector bool short vec_xor (vector bool short, vector bool short);
10536 vector signed short vec_xor (vector bool short, vector signed short);
10537 vector signed short vec_xor (vector signed short, vector bool short);
10538 vector signed short vec_xor (vector signed short, vector signed short);
10539 vector unsigned short vec_xor (vector bool short,
10540 vector unsigned short);
10541 vector unsigned short vec_xor (vector unsigned short,
10542 vector bool short);
10543 vector unsigned short vec_xor (vector unsigned short,
10544 vector unsigned short);
10545 vector signed char vec_xor (vector bool char, vector signed char);
10546 vector bool char vec_xor (vector bool char, vector bool char);
10547 vector signed char vec_xor (vector signed char, vector bool char);
10548 vector signed char vec_xor (vector signed char, vector signed char);
10549 vector unsigned char vec_xor (vector bool char, vector unsigned char);
10550 vector unsigned char vec_xor (vector unsigned char, vector bool char);
10551 vector unsigned char vec_xor (vector unsigned char,
10552 vector unsigned char);
10554 int vec_all_eq (vector signed char, vector bool char);
10555 int vec_all_eq (vector signed char, vector signed char);
10556 int vec_all_eq (vector unsigned char, vector bool char);
10557 int vec_all_eq (vector unsigned char, vector unsigned char);
10558 int vec_all_eq (vector bool char, vector bool char);
10559 int vec_all_eq (vector bool char, vector unsigned char);
10560 int vec_all_eq (vector bool char, vector signed char);
10561 int vec_all_eq (vector signed short, vector bool short);
10562 int vec_all_eq (vector signed short, vector signed short);
10563 int vec_all_eq (vector unsigned short, vector bool short);
10564 int vec_all_eq (vector unsigned short, vector unsigned short);
10565 int vec_all_eq (vector bool short, vector bool short);
10566 int vec_all_eq (vector bool short, vector unsigned short);
10567 int vec_all_eq (vector bool short, vector signed short);
10568 int vec_all_eq (vector pixel, vector pixel);
10569 int vec_all_eq (vector signed int, vector bool int);
10570 int vec_all_eq (vector signed int, vector signed int);
10571 int vec_all_eq (vector unsigned int, vector bool int);
10572 int vec_all_eq (vector unsigned int, vector unsigned int);
10573 int vec_all_eq (vector bool int, vector bool int);
10574 int vec_all_eq (vector bool int, vector unsigned int);
10575 int vec_all_eq (vector bool int, vector signed int);
10576 int vec_all_eq (vector float, vector float);
10578 int vec_all_ge (vector bool char, vector unsigned char);
10579 int vec_all_ge (vector unsigned char, vector bool char);
10580 int vec_all_ge (vector unsigned char, vector unsigned char);
10581 int vec_all_ge (vector bool char, vector signed char);
10582 int vec_all_ge (vector signed char, vector bool char);
10583 int vec_all_ge (vector signed char, vector signed char);
10584 int vec_all_ge (vector bool short, vector unsigned short);
10585 int vec_all_ge (vector unsigned short, vector bool short);
10586 int vec_all_ge (vector unsigned short, vector unsigned short);
10587 int vec_all_ge (vector signed short, vector signed short);
10588 int vec_all_ge (vector bool short, vector signed short);
10589 int vec_all_ge (vector signed short, vector bool short);
10590 int vec_all_ge (vector bool int, vector unsigned int);
10591 int vec_all_ge (vector unsigned int, vector bool int);
10592 int vec_all_ge (vector unsigned int, vector unsigned int);
10593 int vec_all_ge (vector bool int, vector signed int);
10594 int vec_all_ge (vector signed int, vector bool int);
10595 int vec_all_ge (vector signed int, vector signed int);
10596 int vec_all_ge (vector float, vector float);
10598 int vec_all_gt (vector bool char, vector unsigned char);
10599 int vec_all_gt (vector unsigned char, vector bool char);
10600 int vec_all_gt (vector unsigned char, vector unsigned char);
10601 int vec_all_gt (vector bool char, vector signed char);
10602 int vec_all_gt (vector signed char, vector bool char);
10603 int vec_all_gt (vector signed char, vector signed char);
10604 int vec_all_gt (vector bool short, vector unsigned short);
10605 int vec_all_gt (vector unsigned short, vector bool short);
10606 int vec_all_gt (vector unsigned short, vector unsigned short);
10607 int vec_all_gt (vector bool short, vector signed short);
10608 int vec_all_gt (vector signed short, vector bool short);
10609 int vec_all_gt (vector signed short, vector signed short);
10610 int vec_all_gt (vector bool int, vector unsigned int);
10611 int vec_all_gt (vector unsigned int, vector bool int);
10612 int vec_all_gt (vector unsigned int, vector unsigned int);
10613 int vec_all_gt (vector bool int, vector signed int);
10614 int vec_all_gt (vector signed int, vector bool int);
10615 int vec_all_gt (vector signed int, vector signed int);
10616 int vec_all_gt (vector float, vector float);
10618 int vec_all_in (vector float, vector float);
10620 int vec_all_le (vector bool char, vector unsigned char);
10621 int vec_all_le (vector unsigned char, vector bool char);
10622 int vec_all_le (vector unsigned char, vector unsigned char);
10623 int vec_all_le (vector bool char, vector signed char);
10624 int vec_all_le (vector signed char, vector bool char);
10625 int vec_all_le (vector signed char, vector signed char);
10626 int vec_all_le (vector bool short, vector unsigned short);
10627 int vec_all_le (vector unsigned short, vector bool short);
10628 int vec_all_le (vector unsigned short, vector unsigned short);
10629 int vec_all_le (vector bool short, vector signed short);
10630 int vec_all_le (vector signed short, vector bool short);
10631 int vec_all_le (vector signed short, vector signed short);
10632 int vec_all_le (vector bool int, vector unsigned int);
10633 int vec_all_le (vector unsigned int, vector bool int);
10634 int vec_all_le (vector unsigned int, vector unsigned int);
10635 int vec_all_le (vector bool int, vector signed int);
10636 int vec_all_le (vector signed int, vector bool int);
10637 int vec_all_le (vector signed int, vector signed int);
10638 int vec_all_le (vector float, vector float);
10640 int vec_all_lt (vector bool char, vector unsigned char);
10641 int vec_all_lt (vector unsigned char, vector bool char);
10642 int vec_all_lt (vector unsigned char, vector unsigned char);
10643 int vec_all_lt (vector bool char, vector signed char);
10644 int vec_all_lt (vector signed char, vector bool char);
10645 int vec_all_lt (vector signed char, vector signed char);
10646 int vec_all_lt (vector bool short, vector unsigned short);
10647 int vec_all_lt (vector unsigned short, vector bool short);
10648 int vec_all_lt (vector unsigned short, vector unsigned short);
10649 int vec_all_lt (vector bool short, vector signed short);
10650 int vec_all_lt (vector signed short, vector bool short);
10651 int vec_all_lt (vector signed short, vector signed short);
10652 int vec_all_lt (vector bool int, vector unsigned int);
10653 int vec_all_lt (vector unsigned int, vector bool int);
10654 int vec_all_lt (vector unsigned int, vector unsigned int);
10655 int vec_all_lt (vector bool int, vector signed int);
10656 int vec_all_lt (vector signed int, vector bool int);
10657 int vec_all_lt (vector signed int, vector signed int);
10658 int vec_all_lt (vector float, vector float);
10660 int vec_all_nan (vector float);
10662 int vec_all_ne (vector signed char, vector bool char);
10663 int vec_all_ne (vector signed char, vector signed char);
10664 int vec_all_ne (vector unsigned char, vector bool char);
10665 int vec_all_ne (vector unsigned char, vector unsigned char);
10666 int vec_all_ne (vector bool char, vector bool char);
10667 int vec_all_ne (vector bool char, vector unsigned char);
10668 int vec_all_ne (vector bool char, vector signed char);
10669 int vec_all_ne (vector signed short, vector bool short);
10670 int vec_all_ne (vector signed short, vector signed short);
10671 int vec_all_ne (vector unsigned short, vector bool short);
10672 int vec_all_ne (vector unsigned short, vector unsigned short);
10673 int vec_all_ne (vector bool short, vector bool short);
10674 int vec_all_ne (vector bool short, vector unsigned short);
10675 int vec_all_ne (vector bool short, vector signed short);
10676 int vec_all_ne (vector pixel, vector pixel);
10677 int vec_all_ne (vector signed int, vector bool int);
10678 int vec_all_ne (vector signed int, vector signed int);
10679 int vec_all_ne (vector unsigned int, vector bool int);
10680 int vec_all_ne (vector unsigned int, vector unsigned int);
10681 int vec_all_ne (vector bool int, vector bool int);
10682 int vec_all_ne (vector bool int, vector unsigned int);
10683 int vec_all_ne (vector bool int, vector signed int);
10684 int vec_all_ne (vector float, vector float);
10686 int vec_all_nge (vector float, vector float);
10688 int vec_all_ngt (vector float, vector float);
10690 int vec_all_nle (vector float, vector float);
10692 int vec_all_nlt (vector float, vector float);
10694 int vec_all_numeric (vector float);
10696 int vec_any_eq (vector signed char, vector bool char);
10697 int vec_any_eq (vector signed char, vector signed char);
10698 int vec_any_eq (vector unsigned char, vector bool char);
10699 int vec_any_eq (vector unsigned char, vector unsigned char);
10700 int vec_any_eq (vector bool char, vector bool char);
10701 int vec_any_eq (vector bool char, vector unsigned char);
10702 int vec_any_eq (vector bool char, vector signed char);
10703 int vec_any_eq (vector signed short, vector bool short);
10704 int vec_any_eq (vector signed short, vector signed short);
10705 int vec_any_eq (vector unsigned short, vector bool short);
10706 int vec_any_eq (vector unsigned short, vector unsigned short);
10707 int vec_any_eq (vector bool short, vector bool short);
10708 int vec_any_eq (vector bool short, vector unsigned short);
10709 int vec_any_eq (vector bool short, vector signed short);
10710 int vec_any_eq (vector pixel, vector pixel);
10711 int vec_any_eq (vector signed int, vector bool int);
10712 int vec_any_eq (vector signed int, vector signed int);
10713 int vec_any_eq (vector unsigned int, vector bool int);
10714 int vec_any_eq (vector unsigned int, vector unsigned int);
10715 int vec_any_eq (vector bool int, vector bool int);
10716 int vec_any_eq (vector bool int, vector unsigned int);
10717 int vec_any_eq (vector bool int, vector signed int);
10718 int vec_any_eq (vector float, vector float);
10720 int vec_any_ge (vector signed char, vector bool char);
10721 int vec_any_ge (vector unsigned char, vector bool char);
10722 int vec_any_ge (vector unsigned char, vector unsigned char);
10723 int vec_any_ge (vector signed char, vector signed char);
10724 int vec_any_ge (vector bool char, vector unsigned char);
10725 int vec_any_ge (vector bool char, vector signed char);
10726 int vec_any_ge (vector unsigned short, vector bool short);
10727 int vec_any_ge (vector unsigned short, vector unsigned short);
10728 int vec_any_ge (vector signed short, vector signed short);
10729 int vec_any_ge (vector signed short, vector bool short);
10730 int vec_any_ge (vector bool short, vector unsigned short);
10731 int vec_any_ge (vector bool short, vector signed short);
10732 int vec_any_ge (vector signed int, vector bool int);
10733 int vec_any_ge (vector unsigned int, vector bool int);
10734 int vec_any_ge (vector unsigned int, vector unsigned int);
10735 int vec_any_ge (vector signed int, vector signed int);
10736 int vec_any_ge (vector bool int, vector unsigned int);
10737 int vec_any_ge (vector bool int, vector signed int);
10738 int vec_any_ge (vector float, vector float);
10740 int vec_any_gt (vector bool char, vector unsigned char);
10741 int vec_any_gt (vector unsigned char, vector bool char);
10742 int vec_any_gt (vector unsigned char, vector unsigned char);
10743 int vec_any_gt (vector bool char, vector signed char);
10744 int vec_any_gt (vector signed char, vector bool char);
10745 int vec_any_gt (vector signed char, vector signed char);
10746 int vec_any_gt (vector bool short, vector unsigned short);
10747 int vec_any_gt (vector unsigned short, vector bool short);
10748 int vec_any_gt (vector unsigned short, vector unsigned short);
10749 int vec_any_gt (vector bool short, vector signed short);
10750 int vec_any_gt (vector signed short, vector bool short);
10751 int vec_any_gt (vector signed short, vector signed short);
10752 int vec_any_gt (vector bool int, vector unsigned int);
10753 int vec_any_gt (vector unsigned int, vector bool int);
10754 int vec_any_gt (vector unsigned int, vector unsigned int);
10755 int vec_any_gt (vector bool int, vector signed int);
10756 int vec_any_gt (vector signed int, vector bool int);
10757 int vec_any_gt (vector signed int, vector signed int);
10758 int vec_any_gt (vector float, vector float);
10760 int vec_any_le (vector bool char, vector unsigned char);
10761 int vec_any_le (vector unsigned char, vector bool char);
10762 int vec_any_le (vector unsigned char, vector unsigned char);
10763 int vec_any_le (vector bool char, vector signed char);
10764 int vec_any_le (vector signed char, vector bool char);
10765 int vec_any_le (vector signed char, vector signed char);
10766 int vec_any_le (vector bool short, vector unsigned short);
10767 int vec_any_le (vector unsigned short, vector bool short);
10768 int vec_any_le (vector unsigned short, vector unsigned short);
10769 int vec_any_le (vector bool short, vector signed short);
10770 int vec_any_le (vector signed short, vector bool short);
10771 int vec_any_le (vector signed short, vector signed short);
10772 int vec_any_le (vector bool int, vector unsigned int);
10773 int vec_any_le (vector unsigned int, vector bool int);
10774 int vec_any_le (vector unsigned int, vector unsigned int);
10775 int vec_any_le (vector bool int, vector signed int);
10776 int vec_any_le (vector signed int, vector bool int);
10777 int vec_any_le (vector signed int, vector signed int);
10778 int vec_any_le (vector float, vector float);
10780 int vec_any_lt (vector bool char, vector unsigned char);
10781 int vec_any_lt (vector unsigned char, vector bool char);
10782 int vec_any_lt (vector unsigned char, vector unsigned char);
10783 int vec_any_lt (vector bool char, vector signed char);
10784 int vec_any_lt (vector signed char, vector bool char);
10785 int vec_any_lt (vector signed char, vector signed char);
10786 int vec_any_lt (vector bool short, vector unsigned short);
10787 int vec_any_lt (vector unsigned short, vector bool short);
10788 int vec_any_lt (vector unsigned short, vector unsigned short);
10789 int vec_any_lt (vector bool short, vector signed short);
10790 int vec_any_lt (vector signed short, vector bool short);
10791 int vec_any_lt (vector signed short, vector signed short);
10792 int vec_any_lt (vector bool int, vector unsigned int);
10793 int vec_any_lt (vector unsigned int, vector bool int);
10794 int vec_any_lt (vector unsigned int, vector unsigned int);
10795 int vec_any_lt (vector bool int, vector signed int);
10796 int vec_any_lt (vector signed int, vector bool int);
10797 int vec_any_lt (vector signed int, vector signed int);
10798 int vec_any_lt (vector float, vector float);
10800 int vec_any_nan (vector float);
10802 int vec_any_ne (vector signed char, vector bool char);
10803 int vec_any_ne (vector signed char, vector signed char);
10804 int vec_any_ne (vector unsigned char, vector bool char);
10805 int vec_any_ne (vector unsigned char, vector unsigned char);
10806 int vec_any_ne (vector bool char, vector bool char);
10807 int vec_any_ne (vector bool char, vector unsigned char);
10808 int vec_any_ne (vector bool char, vector signed char);
10809 int vec_any_ne (vector signed short, vector bool short);
10810 int vec_any_ne (vector signed short, vector signed short);
10811 int vec_any_ne (vector unsigned short, vector bool short);
10812 int vec_any_ne (vector unsigned short, vector unsigned short);
10813 int vec_any_ne (vector bool short, vector bool short);
10814 int vec_any_ne (vector bool short, vector unsigned short);
10815 int vec_any_ne (vector bool short, vector signed short);
10816 int vec_any_ne (vector pixel, vector pixel);
10817 int vec_any_ne (vector signed int, vector bool int);
10818 int vec_any_ne (vector signed int, vector signed int);
10819 int vec_any_ne (vector unsigned int, vector bool int);
10820 int vec_any_ne (vector unsigned int, vector unsigned int);
10821 int vec_any_ne (vector bool int, vector bool int);
10822 int vec_any_ne (vector bool int, vector unsigned int);
10823 int vec_any_ne (vector bool int, vector signed int);
10824 int vec_any_ne (vector float, vector float);
10826 int vec_any_nge (vector float, vector float);
10828 int vec_any_ngt (vector float, vector float);
10830 int vec_any_nle (vector float, vector float);
10832 int vec_any_nlt (vector float, vector float);
10834 int vec_any_numeric (vector float);
10836 int vec_any_out (vector float, vector float);
10839 @node SPARC VIS Built-in Functions
10840 @subsection SPARC VIS Built-in Functions
10842 GCC supports SIMD operations on the SPARC using both the generic vector
10843 extensions (@pxref{Vector Extensions}) as well as built-in functions for
10844 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
10845 switch, the VIS extension is exposed as the following built-in functions:
10848 typedef int v2si __attribute__ ((vector_size (8)));
10849 typedef short v4hi __attribute__ ((vector_size (8)));
10850 typedef short v2hi __attribute__ ((vector_size (4)));
10851 typedef char v8qi __attribute__ ((vector_size (8)));
10852 typedef char v4qi __attribute__ ((vector_size (4)));
10854 void * __builtin_vis_alignaddr (void *, long);
10855 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
10856 v2si __builtin_vis_faligndatav2si (v2si, v2si);
10857 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
10858 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
10860 v4hi __builtin_vis_fexpand (v4qi);
10862 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
10863 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
10864 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
10865 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
10866 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
10867 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
10868 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
10870 v4qi __builtin_vis_fpack16 (v4hi);
10871 v8qi __builtin_vis_fpack32 (v2si, v2si);
10872 v2hi __builtin_vis_fpackfix (v2si);
10873 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
10875 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
10878 @node SPU Built-in Functions
10879 @subsection SPU Built-in Functions
10881 GCC provides extensions for the SPU processor as described in the
10882 Sony/Toshiba/IBM SPU Language Extensions Specification, which can be
10883 found at @uref{http://cell.scei.co.jp/} or
10884 @uref{http://www.ibm.com/developerworks/power/cell/}. GCC's
10885 implementation differs in several ways.
10890 The optional extension of specifying vector constants in parentheses is
10894 A vector initializer requires no cast if the vector constant is of the
10895 same type as the variable it is initializing.
10898 If @code{signed} or @code{unsigned} is omitted, the signedness of the
10899 vector type is the default signedness of the base type. The default
10900 varies depending on the operating system, so a portable program should
10901 always specify the signedness.
10904 By default, the keyword @code{__vector} is added. The macro
10905 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
10909 GCC allows using a @code{typedef} name as the type specifier for a
10913 For C, overloaded functions are implemented with macros so the following
10917 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
10920 Since @code{spu_add} is a macro, the vector constant in the example
10921 is treated as four separate arguments. Wrap the entire argument in
10922 parentheses for this to work.
10925 The extended version of @code{__builtin_expect} is not supported.
10929 @emph{Note:} Only the interface described in the aforementioned
10930 specification is supported. Internally, GCC uses built-in functions to
10931 implement the required functionality, but these are not supported and
10932 are subject to change without notice.
10934 @node Target Format Checks
10935 @section Format Checks Specific to Particular Target Machines
10937 For some target machines, GCC supports additional options to the
10939 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
10942 * Solaris Format Checks::
10945 @node Solaris Format Checks
10946 @subsection Solaris Format Checks
10948 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
10949 check. @code{cmn_err} accepts a subset of the standard @code{printf}
10950 conversions, and the two-argument @code{%b} conversion for displaying
10951 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
10954 @section Pragmas Accepted by GCC
10958 GCC supports several types of pragmas, primarily in order to compile
10959 code originally written for other compilers. Note that in general
10960 we do not recommend the use of pragmas; @xref{Function Attributes},
10961 for further explanation.
10966 * RS/6000 and PowerPC Pragmas::
10968 * Solaris Pragmas::
10969 * Symbol-Renaming Pragmas::
10970 * Structure-Packing Pragmas::
10972 * Diagnostic Pragmas::
10973 * Visibility Pragmas::
10974 * Push/Pop Macro Pragmas::
10978 @subsection ARM Pragmas
10980 The ARM target defines pragmas for controlling the default addition of
10981 @code{long_call} and @code{short_call} attributes to functions.
10982 @xref{Function Attributes}, for information about the effects of these
10987 @cindex pragma, long_calls
10988 Set all subsequent functions to have the @code{long_call} attribute.
10990 @item no_long_calls
10991 @cindex pragma, no_long_calls
10992 Set all subsequent functions to have the @code{short_call} attribute.
10994 @item long_calls_off
10995 @cindex pragma, long_calls_off
10996 Do not affect the @code{long_call} or @code{short_call} attributes of
10997 subsequent functions.
11001 @subsection M32C Pragmas
11004 @item memregs @var{number}
11005 @cindex pragma, memregs
11006 Overrides the command line option @code{-memregs=} for the current
11007 file. Use with care! This pragma must be before any function in the
11008 file, and mixing different memregs values in different objects may
11009 make them incompatible. This pragma is useful when a
11010 performance-critical function uses a memreg for temporary values,
11011 as it may allow you to reduce the number of memregs used.
11015 @node RS/6000 and PowerPC Pragmas
11016 @subsection RS/6000 and PowerPC Pragmas
11018 The RS/6000 and PowerPC targets define one pragma for controlling
11019 whether or not the @code{longcall} attribute is added to function
11020 declarations by default. This pragma overrides the @option{-mlongcall}
11021 option, but not the @code{longcall} and @code{shortcall} attributes.
11022 @xref{RS/6000 and PowerPC Options}, for more information about when long
11023 calls are and are not necessary.
11027 @cindex pragma, longcall
11028 Apply the @code{longcall} attribute to all subsequent function
11032 Do not apply the @code{longcall} attribute to subsequent function
11036 @c Describe h8300 pragmas here.
11037 @c Describe sh pragmas here.
11038 @c Describe v850 pragmas here.
11040 @node Darwin Pragmas
11041 @subsection Darwin Pragmas
11043 The following pragmas are available for all architectures running the
11044 Darwin operating system. These are useful for compatibility with other
11048 @item mark @var{tokens}@dots{}
11049 @cindex pragma, mark
11050 This pragma is accepted, but has no effect.
11052 @item options align=@var{alignment}
11053 @cindex pragma, options align
11054 This pragma sets the alignment of fields in structures. The values of
11055 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
11056 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
11057 properly; to restore the previous setting, use @code{reset} for the
11060 @item segment @var{tokens}@dots{}
11061 @cindex pragma, segment
11062 This pragma is accepted, but has no effect.
11064 @item unused (@var{var} [, @var{var}]@dots{})
11065 @cindex pragma, unused
11066 This pragma declares variables to be possibly unused. GCC will not
11067 produce warnings for the listed variables. The effect is similar to
11068 that of the @code{unused} attribute, except that this pragma may appear
11069 anywhere within the variables' scopes.
11072 @node Solaris Pragmas
11073 @subsection Solaris Pragmas
11075 The Solaris target supports @code{#pragma redefine_extname}
11076 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
11077 @code{#pragma} directives for compatibility with the system compiler.
11080 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
11081 @cindex pragma, align
11083 Increase the minimum alignment of each @var{variable} to @var{alignment}.
11084 This is the same as GCC's @code{aligned} attribute @pxref{Variable
11085 Attributes}). Macro expansion occurs on the arguments to this pragma
11086 when compiling C and Objective-C@. It does not currently occur when
11087 compiling C++, but this is a bug which may be fixed in a future
11090 @item fini (@var{function} [, @var{function}]...)
11091 @cindex pragma, fini
11093 This pragma causes each listed @var{function} to be called after
11094 main, or during shared module unloading, by adding a call to the
11095 @code{.fini} section.
11097 @item init (@var{function} [, @var{function}]...)
11098 @cindex pragma, init
11100 This pragma causes each listed @var{function} to be called during
11101 initialization (before @code{main}) or during shared module loading, by
11102 adding a call to the @code{.init} section.
11106 @node Symbol-Renaming Pragmas
11107 @subsection Symbol-Renaming Pragmas
11109 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
11110 supports two @code{#pragma} directives which change the name used in
11111 assembly for a given declaration. These pragmas are only available on
11112 platforms whose system headers need them. To get this effect on all
11113 platforms supported by GCC, use the asm labels extension (@pxref{Asm
11117 @item redefine_extname @var{oldname} @var{newname}
11118 @cindex pragma, redefine_extname
11120 This pragma gives the C function @var{oldname} the assembly symbol
11121 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
11122 will be defined if this pragma is available (currently only on
11125 @item extern_prefix @var{string}
11126 @cindex pragma, extern_prefix
11128 This pragma causes all subsequent external function and variable
11129 declarations to have @var{string} prepended to their assembly symbols.
11130 This effect may be terminated with another @code{extern_prefix} pragma
11131 whose argument is an empty string. The preprocessor macro
11132 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
11133 available (currently only on Tru64 UNIX)@.
11136 These pragmas and the asm labels extension interact in a complicated
11137 manner. Here are some corner cases you may want to be aware of.
11140 @item Both pragmas silently apply only to declarations with external
11141 linkage. Asm labels do not have this restriction.
11143 @item In C++, both pragmas silently apply only to declarations with
11144 ``C'' linkage. Again, asm labels do not have this restriction.
11146 @item If any of the three ways of changing the assembly name of a
11147 declaration is applied to a declaration whose assembly name has
11148 already been determined (either by a previous use of one of these
11149 features, or because the compiler needed the assembly name in order to
11150 generate code), and the new name is different, a warning issues and
11151 the name does not change.
11153 @item The @var{oldname} used by @code{#pragma redefine_extname} is
11154 always the C-language name.
11156 @item If @code{#pragma extern_prefix} is in effect, and a declaration
11157 occurs with an asm label attached, the prefix is silently ignored for
11160 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
11161 apply to the same declaration, whichever triggered first wins, and a
11162 warning issues if they contradict each other. (We would like to have
11163 @code{#pragma redefine_extname} always win, for consistency with asm
11164 labels, but if @code{#pragma extern_prefix} triggers first we have no
11165 way of knowing that that happened.)
11168 @node Structure-Packing Pragmas
11169 @subsection Structure-Packing Pragmas
11171 For compatibility with Microsoft Windows compilers, GCC supports a
11172 set of @code{#pragma} directives which change the maximum alignment of
11173 members of structures (other than zero-width bitfields), unions, and
11174 classes subsequently defined. The @var{n} value below always is required
11175 to be a small power of two and specifies the new alignment in bytes.
11178 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
11179 @item @code{#pragma pack()} sets the alignment to the one that was in
11180 effect when compilation started (see also command line option
11181 @option{-fpack-struct[=<n>]} @pxref{Code Gen Options}).
11182 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
11183 setting on an internal stack and then optionally sets the new alignment.
11184 @item @code{#pragma pack(pop)} restores the alignment setting to the one
11185 saved at the top of the internal stack (and removes that stack entry).
11186 Note that @code{#pragma pack([@var{n}])} does not influence this internal
11187 stack; thus it is possible to have @code{#pragma pack(push)} followed by
11188 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
11189 @code{#pragma pack(pop)}.
11192 Some targets, e.g.@: i386 and powerpc, support the @code{ms_struct}
11193 @code{#pragma} which lays out a structure as the documented
11194 @code{__attribute__ ((ms_struct))}.
11196 @item @code{#pragma ms_struct on} turns on the layout for structures
11198 @item @code{#pragma ms_struct off} turns off the layout for structures
11200 @item @code{#pragma ms_struct reset} goes back to the default layout.
11204 @subsection Weak Pragmas
11206 For compatibility with SVR4, GCC supports a set of @code{#pragma}
11207 directives for declaring symbols to be weak, and defining weak
11211 @item #pragma weak @var{symbol}
11212 @cindex pragma, weak
11213 This pragma declares @var{symbol} to be weak, as if the declaration
11214 had the attribute of the same name. The pragma may appear before
11215 or after the declaration of @var{symbol}, but must appear before
11216 either its first use or its definition. It is not an error for
11217 @var{symbol} to never be defined at all.
11219 @item #pragma weak @var{symbol1} = @var{symbol2}
11220 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
11221 It is an error if @var{symbol2} is not defined in the current
11225 @node Diagnostic Pragmas
11226 @subsection Diagnostic Pragmas
11228 GCC allows the user to selectively enable or disable certain types of
11229 diagnostics, and change the kind of the diagnostic. For example, a
11230 project's policy might require that all sources compile with
11231 @option{-Werror} but certain files might have exceptions allowing
11232 specific types of warnings. Or, a project might selectively enable
11233 diagnostics and treat them as errors depending on which preprocessor
11234 macros are defined.
11237 @item #pragma GCC diagnostic @var{kind} @var{option}
11238 @cindex pragma, diagnostic
11240 Modifies the disposition of a diagnostic. Note that not all
11241 diagnostics are modifiable; at the moment only warnings (normally
11242 controlled by @samp{-W@dots{}}) can be controlled, and not all of them.
11243 Use @option{-fdiagnostics-show-option} to determine which diagnostics
11244 are controllable and which option controls them.
11246 @var{kind} is @samp{error} to treat this diagnostic as an error,
11247 @samp{warning} to treat it like a warning (even if @option{-Werror} is
11248 in effect), or @samp{ignored} if the diagnostic is to be ignored.
11249 @var{option} is a double quoted string which matches the command line
11253 #pragma GCC diagnostic warning "-Wformat"
11254 #pragma GCC diagnostic error "-Wformat"
11255 #pragma GCC diagnostic ignored "-Wformat"
11258 Note that these pragmas override any command line options. Also,
11259 while it is syntactically valid to put these pragmas anywhere in your
11260 sources, the only supported location for them is before any data or
11261 functions are defined. Doing otherwise may result in unpredictable
11262 results depending on how the optimizer manages your sources. If the
11263 same option is listed multiple times, the last one specified is the
11264 one that is in effect. This pragma is not intended to be a general
11265 purpose replacement for command line options, but for implementing
11266 strict control over project policies.
11270 @node Visibility Pragmas
11271 @subsection Visibility Pragmas
11274 @item #pragma GCC visibility push(@var{visibility})
11275 @itemx #pragma GCC visibility pop
11276 @cindex pragma, visibility
11278 This pragma allows the user to set the visibility for multiple
11279 declarations without having to give each a visibility attribute
11280 @xref{Function Attributes}, for more information about visibility and
11281 the attribute syntax.
11283 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
11284 declarations. Class members and template specializations are not
11285 affected; if you want to override the visibility for a particular
11286 member or instantiation, you must use an attribute.
11291 @node Push/Pop Macro Pragmas
11292 @subsection Push/Pop Macro Pragmas
11294 For compatibility with Microsoft Windows compilers, GCC supports
11295 @samp{#pragma push_macro(@var{"macro_name"})}
11296 and @samp{#pragma pop_macro(@var{"macro_name"})}.
11299 @item #pragma push_macro(@var{"macro_name"})
11300 @cindex pragma, push_macro
11301 This pragma saves the value of the macro named as @var{macro_name} to
11302 the top of the stack for this macro.
11304 @item #pragma pop_macro(@var{"macro_name"})
11305 @cindex pragma, pop_macro
11306 This pragma sets the value of the macro named as @var{macro_name} to
11307 the value on top of the stack for this macro. If the stack for
11308 @var{macro_name} is empty, the value of the macro remains unchanged.
11315 #pragma push_macro("X")
11318 #pragma pop_macro("X")
11322 In this example, the definition of X as 1 is saved by @code{#pragma
11323 push_macro} and restored by @code{#pragma pop_macro}.
11325 @node Unnamed Fields
11326 @section Unnamed struct/union fields within structs/unions
11330 For compatibility with other compilers, GCC allows you to define
11331 a structure or union that contains, as fields, structures and unions
11332 without names. For example:
11345 In this example, the user would be able to access members of the unnamed
11346 union with code like @samp{foo.b}. Note that only unnamed structs and
11347 unions are allowed, you may not have, for example, an unnamed
11350 You must never create such structures that cause ambiguous field definitions.
11351 For example, this structure:
11362 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
11363 Such constructs are not supported and must be avoided. In the future,
11364 such constructs may be detected and treated as compilation errors.
11366 @opindex fms-extensions
11367 Unless @option{-fms-extensions} is used, the unnamed field must be a
11368 structure or union definition without a tag (for example, @samp{struct
11369 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
11370 also be a definition with a tag such as @samp{struct foo @{ int a;
11371 @};}, a reference to a previously defined structure or union such as
11372 @samp{struct foo;}, or a reference to a @code{typedef} name for a
11373 previously defined structure or union type.
11376 @section Thread-Local Storage
11377 @cindex Thread-Local Storage
11378 @cindex @acronym{TLS}
11381 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
11382 are allocated such that there is one instance of the variable per extant
11383 thread. The run-time model GCC uses to implement this originates
11384 in the IA-64 processor-specific ABI, but has since been migrated
11385 to other processors as well. It requires significant support from
11386 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
11387 system libraries (@file{libc.so} and @file{libpthread.so}), so it
11388 is not available everywhere.
11390 At the user level, the extension is visible with a new storage
11391 class keyword: @code{__thread}. For example:
11395 extern __thread struct state s;
11396 static __thread char *p;
11399 The @code{__thread} specifier may be used alone, with the @code{extern}
11400 or @code{static} specifiers, but with no other storage class specifier.
11401 When used with @code{extern} or @code{static}, @code{__thread} must appear
11402 immediately after the other storage class specifier.
11404 The @code{__thread} specifier may be applied to any global, file-scoped
11405 static, function-scoped static, or static data member of a class. It may
11406 not be applied to block-scoped automatic or non-static data member.
11408 When the address-of operator is applied to a thread-local variable, it is
11409 evaluated at run-time and returns the address of the current thread's
11410 instance of that variable. An address so obtained may be used by any
11411 thread. When a thread terminates, any pointers to thread-local variables
11412 in that thread become invalid.
11414 No static initialization may refer to the address of a thread-local variable.
11416 In C++, if an initializer is present for a thread-local variable, it must
11417 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
11420 See @uref{http://people.redhat.com/drepper/tls.pdf,
11421 ELF Handling For Thread-Local Storage} for a detailed explanation of
11422 the four thread-local storage addressing models, and how the run-time
11423 is expected to function.
11426 * C99 Thread-Local Edits::
11427 * C++98 Thread-Local Edits::
11430 @node C99 Thread-Local Edits
11431 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
11433 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
11434 that document the exact semantics of the language extension.
11438 @cite{5.1.2 Execution environments}
11440 Add new text after paragraph 1
11443 Within either execution environment, a @dfn{thread} is a flow of
11444 control within a program. It is implementation defined whether
11445 or not there may be more than one thread associated with a program.
11446 It is implementation defined how threads beyond the first are
11447 created, the name and type of the function called at thread
11448 startup, and how threads may be terminated. However, objects
11449 with thread storage duration shall be initialized before thread
11454 @cite{6.2.4 Storage durations of objects}
11456 Add new text before paragraph 3
11459 An object whose identifier is declared with the storage-class
11460 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
11461 Its lifetime is the entire execution of the thread, and its
11462 stored value is initialized only once, prior to thread startup.
11466 @cite{6.4.1 Keywords}
11468 Add @code{__thread}.
11471 @cite{6.7.1 Storage-class specifiers}
11473 Add @code{__thread} to the list of storage class specifiers in
11476 Change paragraph 2 to
11479 With the exception of @code{__thread}, at most one storage-class
11480 specifier may be given [@dots{}]. The @code{__thread} specifier may
11481 be used alone, or immediately following @code{extern} or
11485 Add new text after paragraph 6
11488 The declaration of an identifier for a variable that has
11489 block scope that specifies @code{__thread} shall also
11490 specify either @code{extern} or @code{static}.
11492 The @code{__thread} specifier shall be used only with
11497 @node C++98 Thread-Local Edits
11498 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
11500 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
11501 that document the exact semantics of the language extension.
11505 @b{[intro.execution]}
11507 New text after paragraph 4
11510 A @dfn{thread} is a flow of control within the abstract machine.
11511 It is implementation defined whether or not there may be more than
11515 New text after paragraph 7
11518 It is unspecified whether additional action must be taken to
11519 ensure when and whether side effects are visible to other threads.
11525 Add @code{__thread}.
11528 @b{[basic.start.main]}
11530 Add after paragraph 5
11533 The thread that begins execution at the @code{main} function is called
11534 the @dfn{main thread}. It is implementation defined how functions
11535 beginning threads other than the main thread are designated or typed.
11536 A function so designated, as well as the @code{main} function, is called
11537 a @dfn{thread startup function}. It is implementation defined what
11538 happens if a thread startup function returns. It is implementation
11539 defined what happens to other threads when any thread calls @code{exit}.
11543 @b{[basic.start.init]}
11545 Add after paragraph 4
11548 The storage for an object of thread storage duration shall be
11549 statically initialized before the first statement of the thread startup
11550 function. An object of thread storage duration shall not require
11551 dynamic initialization.
11555 @b{[basic.start.term]}
11557 Add after paragraph 3
11560 The type of an object with thread storage duration shall not have a
11561 non-trivial destructor, nor shall it be an array type whose elements
11562 (directly or indirectly) have non-trivial destructors.
11568 Add ``thread storage duration'' to the list in paragraph 1.
11573 Thread, static, and automatic storage durations are associated with
11574 objects introduced by declarations [@dots{}].
11577 Add @code{__thread} to the list of specifiers in paragraph 3.
11580 @b{[basic.stc.thread]}
11582 New section before @b{[basic.stc.static]}
11585 The keyword @code{__thread} applied to a non-local object gives the
11586 object thread storage duration.
11588 A local variable or class data member declared both @code{static}
11589 and @code{__thread} gives the variable or member thread storage
11594 @b{[basic.stc.static]}
11599 All objects which have neither thread storage duration, dynamic
11600 storage duration nor are local [@dots{}].
11606 Add @code{__thread} to the list in paragraph 1.
11611 With the exception of @code{__thread}, at most one
11612 @var{storage-class-specifier} shall appear in a given
11613 @var{decl-specifier-seq}. The @code{__thread} specifier may
11614 be used alone, or immediately following the @code{extern} or
11615 @code{static} specifiers. [@dots{}]
11618 Add after paragraph 5
11621 The @code{__thread} specifier can be applied only to the names of objects
11622 and to anonymous unions.
11628 Add after paragraph 6
11631 Non-@code{static} members shall not be @code{__thread}.
11635 @node Binary constants
11636 @section Binary constants using the @samp{0b} prefix
11637 @cindex Binary constants using the @samp{0b} prefix
11639 Integer constants can be written as binary constants, consisting of a
11640 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
11641 @samp{0B}. This is particularly useful in environments that operate a
11642 lot on the bit-level (like microcontrollers).
11644 The following statements are identical:
11653 The type of these constants follows the same rules as for octal or
11654 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
11657 @node C++ Extensions
11658 @chapter Extensions to the C++ Language
11659 @cindex extensions, C++ language
11660 @cindex C++ language extensions
11662 The GNU compiler provides these extensions to the C++ language (and you
11663 can also use most of the C language extensions in your C++ programs). If you
11664 want to write code that checks whether these features are available, you can
11665 test for the GNU compiler the same way as for C programs: check for a
11666 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
11667 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
11668 Predefined Macros,cpp,The GNU C Preprocessor}).
11671 * Volatiles:: What constitutes an access to a volatile object.
11672 * Restricted Pointers:: C99 restricted pointers and references.
11673 * Vague Linkage:: Where G++ puts inlines, vtables and such.
11674 * C++ Interface:: You can use a single C++ header file for both
11675 declarations and definitions.
11676 * Template Instantiation:: Methods for ensuring that exactly one copy of
11677 each needed template instantiation is emitted.
11678 * Bound member functions:: You can extract a function pointer to the
11679 method denoted by a @samp{->*} or @samp{.*} expression.
11680 * C++ Attributes:: Variable, function, and type attributes for C++ only.
11681 * Namespace Association:: Strong using-directives for namespace association.
11682 * Type Traits:: Compiler support for type traits
11683 * Java Exceptions:: Tweaking exception handling to work with Java.
11684 * Deprecated Features:: Things will disappear from g++.
11685 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
11689 @section When is a Volatile Object Accessed?
11690 @cindex accessing volatiles
11691 @cindex volatile read
11692 @cindex volatile write
11693 @cindex volatile access
11695 Both the C and C++ standard have the concept of volatile objects. These
11696 are normally accessed by pointers and used for accessing hardware. The
11697 standards encourage compilers to refrain from optimizations concerning
11698 accesses to volatile objects. The C standard leaves it implementation
11699 defined as to what constitutes a volatile access. The C++ standard omits
11700 to specify this, except to say that C++ should behave in a similar manner
11701 to C with respect to volatiles, where possible. The minimum either
11702 standard specifies is that at a sequence point all previous accesses to
11703 volatile objects have stabilized and no subsequent accesses have
11704 occurred. Thus an implementation is free to reorder and combine
11705 volatile accesses which occur between sequence points, but cannot do so
11706 for accesses across a sequence point. The use of volatiles does not
11707 allow you to violate the restriction on updating objects multiple times
11708 within a sequence point.
11710 @xref{Qualifiers implementation, , Volatile qualifier and the C compiler}.
11712 The behavior differs slightly between C and C++ in the non-obvious cases:
11715 volatile int *src = @var{somevalue};
11719 With C, such expressions are rvalues, and GCC interprets this either as a
11720 read of the volatile object being pointed to or only as request to evaluate
11721 the side-effects. The C++ standard specifies that such expressions do not
11722 undergo lvalue to rvalue conversion, and that the type of the dereferenced
11723 object may be incomplete. The C++ standard does not specify explicitly
11724 that it is this lvalue to rvalue conversion which may be responsible for
11725 causing an access. However, there is reason to believe that it is,
11726 because otherwise certain simple expressions become undefined. However,
11727 because it would surprise most programmers, G++ treats dereferencing a
11728 pointer to volatile object of complete type when the value is unused as
11729 GCC would do for an equivalent type in C@. When the object has incomplete
11730 type, G++ issues a warning; if you wish to force an error, you must
11731 force a conversion to rvalue with, for instance, a static cast.
11733 When using a reference to volatile, G++ does not treat equivalent
11734 expressions as accesses to volatiles, but instead issues a warning that
11735 no volatile is accessed. The rationale for this is that otherwise it
11736 becomes difficult to determine where volatile access occur, and not
11737 possible to ignore the return value from functions returning volatile
11738 references. Again, if you wish to force a read, cast the reference to
11741 @node Restricted Pointers
11742 @section Restricting Pointer Aliasing
11743 @cindex restricted pointers
11744 @cindex restricted references
11745 @cindex restricted this pointer
11747 As with the C front end, G++ understands the C99 feature of restricted pointers,
11748 specified with the @code{__restrict__}, or @code{__restrict} type
11749 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
11750 language flag, @code{restrict} is not a keyword in C++.
11752 In addition to allowing restricted pointers, you can specify restricted
11753 references, which indicate that the reference is not aliased in the local
11757 void fn (int *__restrict__ rptr, int &__restrict__ rref)
11764 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
11765 @var{rref} refers to a (different) unaliased integer.
11767 You may also specify whether a member function's @var{this} pointer is
11768 unaliased by using @code{__restrict__} as a member function qualifier.
11771 void T::fn () __restrict__
11778 Within the body of @code{T::fn}, @var{this} will have the effective
11779 definition @code{T *__restrict__ const this}. Notice that the
11780 interpretation of a @code{__restrict__} member function qualifier is
11781 different to that of @code{const} or @code{volatile} qualifier, in that it
11782 is applied to the pointer rather than the object. This is consistent with
11783 other compilers which implement restricted pointers.
11785 As with all outermost parameter qualifiers, @code{__restrict__} is
11786 ignored in function definition matching. This means you only need to
11787 specify @code{__restrict__} in a function definition, rather than
11788 in a function prototype as well.
11790 @node Vague Linkage
11791 @section Vague Linkage
11792 @cindex vague linkage
11794 There are several constructs in C++ which require space in the object
11795 file but are not clearly tied to a single translation unit. We say that
11796 these constructs have ``vague linkage''. Typically such constructs are
11797 emitted wherever they are needed, though sometimes we can be more
11801 @item Inline Functions
11802 Inline functions are typically defined in a header file which can be
11803 included in many different compilations. Hopefully they can usually be
11804 inlined, but sometimes an out-of-line copy is necessary, if the address
11805 of the function is taken or if inlining fails. In general, we emit an
11806 out-of-line copy in all translation units where one is needed. As an
11807 exception, we only emit inline virtual functions with the vtable, since
11808 it will always require a copy.
11810 Local static variables and string constants used in an inline function
11811 are also considered to have vague linkage, since they must be shared
11812 between all inlined and out-of-line instances of the function.
11816 C++ virtual functions are implemented in most compilers using a lookup
11817 table, known as a vtable. The vtable contains pointers to the virtual
11818 functions provided by a class, and each object of the class contains a
11819 pointer to its vtable (or vtables, in some multiple-inheritance
11820 situations). If the class declares any non-inline, non-pure virtual
11821 functions, the first one is chosen as the ``key method'' for the class,
11822 and the vtable is only emitted in the translation unit where the key
11825 @emph{Note:} If the chosen key method is later defined as inline, the
11826 vtable will still be emitted in every translation unit which defines it.
11827 Make sure that any inline virtuals are declared inline in the class
11828 body, even if they are not defined there.
11830 @item type_info objects
11833 C++ requires information about types to be written out in order to
11834 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
11835 For polymorphic classes (classes with virtual functions), the type_info
11836 object is written out along with the vtable so that @samp{dynamic_cast}
11837 can determine the dynamic type of a class object at runtime. For all
11838 other types, we write out the type_info object when it is used: when
11839 applying @samp{typeid} to an expression, throwing an object, or
11840 referring to a type in a catch clause or exception specification.
11842 @item Template Instantiations
11843 Most everything in this section also applies to template instantiations,
11844 but there are other options as well.
11845 @xref{Template Instantiation,,Where's the Template?}.
11849 When used with GNU ld version 2.8 or later on an ELF system such as
11850 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
11851 these constructs will be discarded at link time. This is known as
11854 On targets that don't support COMDAT, but do support weak symbols, GCC
11855 will use them. This way one copy will override all the others, but
11856 the unused copies will still take up space in the executable.
11858 For targets which do not support either COMDAT or weak symbols,
11859 most entities with vague linkage will be emitted as local symbols to
11860 avoid duplicate definition errors from the linker. This will not happen
11861 for local statics in inlines, however, as having multiple copies will
11862 almost certainly break things.
11864 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
11865 another way to control placement of these constructs.
11867 @node C++ Interface
11868 @section #pragma interface and implementation
11870 @cindex interface and implementation headers, C++
11871 @cindex C++ interface and implementation headers
11872 @cindex pragmas, interface and implementation
11874 @code{#pragma interface} and @code{#pragma implementation} provide the
11875 user with a way of explicitly directing the compiler to emit entities
11876 with vague linkage (and debugging information) in a particular
11879 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
11880 most cases, because of COMDAT support and the ``key method'' heuristic
11881 mentioned in @ref{Vague Linkage}. Using them can actually cause your
11882 program to grow due to unnecessary out-of-line copies of inline
11883 functions. Currently (3.4) the only benefit of these
11884 @code{#pragma}s is reduced duplication of debugging information, and
11885 that should be addressed soon on DWARF 2 targets with the use of
11889 @item #pragma interface
11890 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
11891 @kindex #pragma interface
11892 Use this directive in @emph{header files} that define object classes, to save
11893 space in most of the object files that use those classes. Normally,
11894 local copies of certain information (backup copies of inline member
11895 functions, debugging information, and the internal tables that implement
11896 virtual functions) must be kept in each object file that includes class
11897 definitions. You can use this pragma to avoid such duplication. When a
11898 header file containing @samp{#pragma interface} is included in a
11899 compilation, this auxiliary information will not be generated (unless
11900 the main input source file itself uses @samp{#pragma implementation}).
11901 Instead, the object files will contain references to be resolved at link
11904 The second form of this directive is useful for the case where you have
11905 multiple headers with the same name in different directories. If you
11906 use this form, you must specify the same string to @samp{#pragma
11909 @item #pragma implementation
11910 @itemx #pragma implementation "@var{objects}.h"
11911 @kindex #pragma implementation
11912 Use this pragma in a @emph{main input file}, when you want full output from
11913 included header files to be generated (and made globally visible). The
11914 included header file, in turn, should use @samp{#pragma interface}.
11915 Backup copies of inline member functions, debugging information, and the
11916 internal tables used to implement virtual functions are all generated in
11917 implementation files.
11919 @cindex implied @code{#pragma implementation}
11920 @cindex @code{#pragma implementation}, implied
11921 @cindex naming convention, implementation headers
11922 If you use @samp{#pragma implementation} with no argument, it applies to
11923 an include file with the same basename@footnote{A file's @dfn{basename}
11924 was the name stripped of all leading path information and of trailing
11925 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
11926 file. For example, in @file{allclass.cc}, giving just
11927 @samp{#pragma implementation}
11928 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
11930 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
11931 an implementation file whenever you would include it from
11932 @file{allclass.cc} even if you never specified @samp{#pragma
11933 implementation}. This was deemed to be more trouble than it was worth,
11934 however, and disabled.
11936 Use the string argument if you want a single implementation file to
11937 include code from multiple header files. (You must also use
11938 @samp{#include} to include the header file; @samp{#pragma
11939 implementation} only specifies how to use the file---it doesn't actually
11942 There is no way to split up the contents of a single header file into
11943 multiple implementation files.
11946 @cindex inlining and C++ pragmas
11947 @cindex C++ pragmas, effect on inlining
11948 @cindex pragmas in C++, effect on inlining
11949 @samp{#pragma implementation} and @samp{#pragma interface} also have an
11950 effect on function inlining.
11952 If you define a class in a header file marked with @samp{#pragma
11953 interface}, the effect on an inline function defined in that class is
11954 similar to an explicit @code{extern} declaration---the compiler emits
11955 no code at all to define an independent version of the function. Its
11956 definition is used only for inlining with its callers.
11958 @opindex fno-implement-inlines
11959 Conversely, when you include the same header file in a main source file
11960 that declares it as @samp{#pragma implementation}, the compiler emits
11961 code for the function itself; this defines a version of the function
11962 that can be found via pointers (or by callers compiled without
11963 inlining). If all calls to the function can be inlined, you can avoid
11964 emitting the function by compiling with @option{-fno-implement-inlines}.
11965 If any calls were not inlined, you will get linker errors.
11967 @node Template Instantiation
11968 @section Where's the Template?
11969 @cindex template instantiation
11971 C++ templates are the first language feature to require more
11972 intelligence from the environment than one usually finds on a UNIX
11973 system. Somehow the compiler and linker have to make sure that each
11974 template instance occurs exactly once in the executable if it is needed,
11975 and not at all otherwise. There are two basic approaches to this
11976 problem, which are referred to as the Borland model and the Cfront model.
11979 @item Borland model
11980 Borland C++ solved the template instantiation problem by adding the code
11981 equivalent of common blocks to their linker; the compiler emits template
11982 instances in each translation unit that uses them, and the linker
11983 collapses them together. The advantage of this model is that the linker
11984 only has to consider the object files themselves; there is no external
11985 complexity to worry about. This disadvantage is that compilation time
11986 is increased because the template code is being compiled repeatedly.
11987 Code written for this model tends to include definitions of all
11988 templates in the header file, since they must be seen to be
11992 The AT&T C++ translator, Cfront, solved the template instantiation
11993 problem by creating the notion of a template repository, an
11994 automatically maintained place where template instances are stored. A
11995 more modern version of the repository works as follows: As individual
11996 object files are built, the compiler places any template definitions and
11997 instantiations encountered in the repository. At link time, the link
11998 wrapper adds in the objects in the repository and compiles any needed
11999 instances that were not previously emitted. The advantages of this
12000 model are more optimal compilation speed and the ability to use the
12001 system linker; to implement the Borland model a compiler vendor also
12002 needs to replace the linker. The disadvantages are vastly increased
12003 complexity, and thus potential for error; for some code this can be
12004 just as transparent, but in practice it can been very difficult to build
12005 multiple programs in one directory and one program in multiple
12006 directories. Code written for this model tends to separate definitions
12007 of non-inline member templates into a separate file, which should be
12008 compiled separately.
12011 When used with GNU ld version 2.8 or later on an ELF system such as
12012 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
12013 Borland model. On other systems, G++ implements neither automatic
12016 A future version of G++ will support a hybrid model whereby the compiler
12017 will emit any instantiations for which the template definition is
12018 included in the compile, and store template definitions and
12019 instantiation context information into the object file for the rest.
12020 The link wrapper will extract that information as necessary and invoke
12021 the compiler to produce the remaining instantiations. The linker will
12022 then combine duplicate instantiations.
12024 In the mean time, you have the following options for dealing with
12025 template instantiations:
12030 Compile your template-using code with @option{-frepo}. The compiler will
12031 generate files with the extension @samp{.rpo} listing all of the
12032 template instantiations used in the corresponding object files which
12033 could be instantiated there; the link wrapper, @samp{collect2}, will
12034 then update the @samp{.rpo} files to tell the compiler where to place
12035 those instantiations and rebuild any affected object files. The
12036 link-time overhead is negligible after the first pass, as the compiler
12037 will continue to place the instantiations in the same files.
12039 This is your best option for application code written for the Borland
12040 model, as it will just work. Code written for the Cfront model will
12041 need to be modified so that the template definitions are available at
12042 one or more points of instantiation; usually this is as simple as adding
12043 @code{#include <tmethods.cc>} to the end of each template header.
12045 For library code, if you want the library to provide all of the template
12046 instantiations it needs, just try to link all of its object files
12047 together; the link will fail, but cause the instantiations to be
12048 generated as a side effect. Be warned, however, that this may cause
12049 conflicts if multiple libraries try to provide the same instantiations.
12050 For greater control, use explicit instantiation as described in the next
12054 @opindex fno-implicit-templates
12055 Compile your code with @option{-fno-implicit-templates} to disable the
12056 implicit generation of template instances, and explicitly instantiate
12057 all the ones you use. This approach requires more knowledge of exactly
12058 which instances you need than do the others, but it's less
12059 mysterious and allows greater control. You can scatter the explicit
12060 instantiations throughout your program, perhaps putting them in the
12061 translation units where the instances are used or the translation units
12062 that define the templates themselves; you can put all of the explicit
12063 instantiations you need into one big file; or you can create small files
12070 template class Foo<int>;
12071 template ostream& operator <<
12072 (ostream&, const Foo<int>&);
12075 for each of the instances you need, and create a template instantiation
12076 library from those.
12078 If you are using Cfront-model code, you can probably get away with not
12079 using @option{-fno-implicit-templates} when compiling files that don't
12080 @samp{#include} the member template definitions.
12082 If you use one big file to do the instantiations, you may want to
12083 compile it without @option{-fno-implicit-templates} so you get all of the
12084 instances required by your explicit instantiations (but not by any
12085 other files) without having to specify them as well.
12087 G++ has extended the template instantiation syntax given in the ISO
12088 standard to allow forward declaration of explicit instantiations
12089 (with @code{extern}), instantiation of the compiler support data for a
12090 template class (i.e.@: the vtable) without instantiating any of its
12091 members (with @code{inline}), and instantiation of only the static data
12092 members of a template class, without the support data or member
12093 functions (with (@code{static}):
12096 extern template int max (int, int);
12097 inline template class Foo<int>;
12098 static template class Foo<int>;
12102 Do nothing. Pretend G++ does implement automatic instantiation
12103 management. Code written for the Borland model will work fine, but
12104 each translation unit will contain instances of each of the templates it
12105 uses. In a large program, this can lead to an unacceptable amount of code
12109 @node Bound member functions
12110 @section Extracting the function pointer from a bound pointer to member function
12112 @cindex pointer to member function
12113 @cindex bound pointer to member function
12115 In C++, pointer to member functions (PMFs) are implemented using a wide
12116 pointer of sorts to handle all the possible call mechanisms; the PMF
12117 needs to store information about how to adjust the @samp{this} pointer,
12118 and if the function pointed to is virtual, where to find the vtable, and
12119 where in the vtable to look for the member function. If you are using
12120 PMFs in an inner loop, you should really reconsider that decision. If
12121 that is not an option, you can extract the pointer to the function that
12122 would be called for a given object/PMF pair and call it directly inside
12123 the inner loop, to save a bit of time.
12125 Note that you will still be paying the penalty for the call through a
12126 function pointer; on most modern architectures, such a call defeats the
12127 branch prediction features of the CPU@. This is also true of normal
12128 virtual function calls.
12130 The syntax for this extension is
12134 extern int (A::*fp)();
12135 typedef int (*fptr)(A *);
12137 fptr p = (fptr)(a.*fp);
12140 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
12141 no object is needed to obtain the address of the function. They can be
12142 converted to function pointers directly:
12145 fptr p1 = (fptr)(&A::foo);
12148 @opindex Wno-pmf-conversions
12149 You must specify @option{-Wno-pmf-conversions} to use this extension.
12151 @node C++ Attributes
12152 @section C++-Specific Variable, Function, and Type Attributes
12154 Some attributes only make sense for C++ programs.
12157 @item init_priority (@var{priority})
12158 @cindex init_priority attribute
12161 In Standard C++, objects defined at namespace scope are guaranteed to be
12162 initialized in an order in strict accordance with that of their definitions
12163 @emph{in a given translation unit}. No guarantee is made for initializations
12164 across translation units. However, GNU C++ allows users to control the
12165 order of initialization of objects defined at namespace scope with the
12166 @code{init_priority} attribute by specifying a relative @var{priority},
12167 a constant integral expression currently bounded between 101 and 65535
12168 inclusive. Lower numbers indicate a higher priority.
12170 In the following example, @code{A} would normally be created before
12171 @code{B}, but the @code{init_priority} attribute has reversed that order:
12174 Some_Class A __attribute__ ((init_priority (2000)));
12175 Some_Class B __attribute__ ((init_priority (543)));
12179 Note that the particular values of @var{priority} do not matter; only their
12182 @item java_interface
12183 @cindex java_interface attribute
12185 This type attribute informs C++ that the class is a Java interface. It may
12186 only be applied to classes declared within an @code{extern "Java"} block.
12187 Calls to methods declared in this interface will be dispatched using GCJ's
12188 interface table mechanism, instead of regular virtual table dispatch.
12192 See also @xref{Namespace Association}.
12194 @node Namespace Association
12195 @section Namespace Association
12197 @strong{Caution:} The semantics of this extension are not fully
12198 defined. Users should refrain from using this extension as its
12199 semantics may change subtly over time. It is possible that this
12200 extension will be removed in future versions of G++.
12202 A using-directive with @code{__attribute ((strong))} is stronger
12203 than a normal using-directive in two ways:
12207 Templates from the used namespace can be specialized and explicitly
12208 instantiated as though they were members of the using namespace.
12211 The using namespace is considered an associated namespace of all
12212 templates in the used namespace for purposes of argument-dependent
12216 The used namespace must be nested within the using namespace so that
12217 normal unqualified lookup works properly.
12219 This is useful for composing a namespace transparently from
12220 implementation namespaces. For example:
12225 template <class T> struct A @{ @};
12227 using namespace debug __attribute ((__strong__));
12228 template <> struct A<int> @{ @}; // @r{ok to specialize}
12230 template <class T> void f (A<T>);
12235 f (std::A<float>()); // @r{lookup finds} std::f
12241 @section Type Traits
12243 The C++ front-end implements syntactic extensions that allow to
12244 determine at compile time various characteristics of a type (or of a
12248 @item __has_nothrow_assign (type)
12249 If @code{type} is const qualified or is a reference type then the trait is
12250 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
12251 is true, else if @code{type} is a cv class or union type with copy assignment
12252 operators that are known not to throw an exception then the trait is true,
12253 else it is false. Requires: @code{type} shall be a complete type, an array
12254 type of unknown bound, or is a @code{void} type.
12256 @item __has_nothrow_copy (type)
12257 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
12258 @code{type} is a cv class or union type with copy constructors that
12259 are known not to throw an exception then the trait is true, else it is false.
12260 Requires: @code{type} shall be a complete type, an array type of
12261 unknown bound, or is a @code{void} type.
12263 @item __has_nothrow_constructor (type)
12264 If @code{__has_trivial_constructor (type)} is true then the trait is
12265 true, else if @code{type} is a cv class or union type (or array
12266 thereof) with a default constructor that is known not to throw an
12267 exception then the trait is true, else it is false. Requires:
12268 @code{type} shall be a complete type, an array type of unknown bound,
12269 or is a @code{void} type.
12271 @item __has_trivial_assign (type)
12272 If @code{type} is const qualified or is a reference type then the trait is
12273 false. Otherwise if @code{__is_pod (type)} is true then the trait is
12274 true, else if @code{type} is a cv class or union type with a trivial
12275 copy assignment ([class.copy]) then the trait is true, else it is
12276 false. Requires: @code{type} shall be a complete type, an array type
12277 of unknown bound, or is a @code{void} type.
12279 @item __has_trivial_copy (type)
12280 If @code{__is_pod (type)} is true or @code{type} is a reference type
12281 then the trait is true, else if @code{type} is a cv class or union type
12282 with a trivial copy constructor ([class.copy]) then the trait
12283 is true, else it is false. Requires: @code{type} shall be a complete
12284 type, an array type of unknown bound, or is a @code{void} type.
12286 @item __has_trivial_constructor (type)
12287 If @code{__is_pod (type)} is true then the trait is true, else if
12288 @code{type} is a cv class or union type (or array thereof) with a
12289 trivial default constructor ([class.ctor]) then the trait is true,
12290 else it is false. Requires: @code{type} shall be a complete type, an
12291 array type of unknown bound, or is a @code{void} type.
12293 @item __has_trivial_destructor (type)
12294 If @code{__is_pod (type)} is true or @code{type} is a reference type then
12295 the trait is true, else if @code{type} is a cv class or union type (or
12296 array thereof) with a trivial destructor ([class.dtor]) then the trait
12297 is true, else it is false. Requires: @code{type} shall be a complete
12298 type, an array type of unknown bound, or is a @code{void} type.
12300 @item __has_virtual_destructor (type)
12301 If @code{type} is a class type with a virtual destructor
12302 ([class.dtor]) then the trait is true, else it is false. Requires:
12303 @code{type} shall be a complete type, an array type of unknown bound,
12304 or is a @code{void} type.
12306 @item __is_abstract (type)
12307 If @code{type} is an abstract class ([class.abstract]) then the trait
12308 is true, else it is false. Requires: @code{type} shall be a complete
12309 type, an array type of unknown bound, or is a @code{void} type.
12311 @item __is_base_of (base_type, derived_type)
12312 If @code{base_type} is a base class of @code{derived_type}
12313 ([class.derived]) then the trait is true, otherwise it is false.
12314 Top-level cv qualifications of @code{base_type} and
12315 @code{derived_type} are ignored. For the purposes of this trait, a
12316 class type is considered is own base. Requires: if @code{__is_class
12317 (base_type)} and @code{__is_class (derived_type)} are true and
12318 @code{base_type} and @code{derived_type} are not the same type
12319 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
12320 type. Diagnostic is produced if this requirement is not met.
12322 @item __is_class (type)
12323 If @code{type} is a cv class type, and not a union type
12324 ([basic.compound]) the the trait is true, else it is false.
12326 @item __is_empty (type)
12327 If @code{__is_class (type)} is false then the trait is false.
12328 Otherwise @code{type} is considered empty if and only if: @code{type}
12329 has no non-static data members, or all non-static data members, if
12330 any, are bit-fields of lenght 0, and @code{type} has no virtual
12331 members, and @code{type} has no virtual base classes, and @code{type}
12332 has no base classes @code{base_type} for which
12333 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
12334 be a complete type, an array type of unknown bound, or is a
12337 @item __is_enum (type)
12338 If @code{type} is a cv enumeration type ([basic.compound]) the the trait is
12339 true, else it is false.
12341 @item __is_pod (type)
12342 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
12343 else it is false. Requires: @code{type} shall be a complete type,
12344 an array type of unknown bound, or is a @code{void} type.
12346 @item __is_polymorphic (type)
12347 If @code{type} is a polymorphic class ([class.virtual]) then the trait
12348 is true, else it is false. Requires: @code{type} shall be a complete
12349 type, an array type of unknown bound, or is a @code{void} type.
12351 @item __is_union (type)
12352 If @code{type} is a cv union type ([basic.compound]) the the trait is
12353 true, else it is false.
12357 @node Java Exceptions
12358 @section Java Exceptions
12360 The Java language uses a slightly different exception handling model
12361 from C++. Normally, GNU C++ will automatically detect when you are
12362 writing C++ code that uses Java exceptions, and handle them
12363 appropriately. However, if C++ code only needs to execute destructors
12364 when Java exceptions are thrown through it, GCC will guess incorrectly.
12365 Sample problematic code is:
12368 struct S @{ ~S(); @};
12369 extern void bar(); // @r{is written in Java, and may throw exceptions}
12378 The usual effect of an incorrect guess is a link failure, complaining of
12379 a missing routine called @samp{__gxx_personality_v0}.
12381 You can inform the compiler that Java exceptions are to be used in a
12382 translation unit, irrespective of what it might think, by writing
12383 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
12384 @samp{#pragma} must appear before any functions that throw or catch
12385 exceptions, or run destructors when exceptions are thrown through them.
12387 You cannot mix Java and C++ exceptions in the same translation unit. It
12388 is believed to be safe to throw a C++ exception from one file through
12389 another file compiled for the Java exception model, or vice versa, but
12390 there may be bugs in this area.
12392 @node Deprecated Features
12393 @section Deprecated Features
12395 In the past, the GNU C++ compiler was extended to experiment with new
12396 features, at a time when the C++ language was still evolving. Now that
12397 the C++ standard is complete, some of those features are superseded by
12398 superior alternatives. Using the old features might cause a warning in
12399 some cases that the feature will be dropped in the future. In other
12400 cases, the feature might be gone already.
12402 While the list below is not exhaustive, it documents some of the options
12403 that are now deprecated:
12406 @item -fexternal-templates
12407 @itemx -falt-external-templates
12408 These are two of the many ways for G++ to implement template
12409 instantiation. @xref{Template Instantiation}. The C++ standard clearly
12410 defines how template definitions have to be organized across
12411 implementation units. G++ has an implicit instantiation mechanism that
12412 should work just fine for standard-conforming code.
12414 @item -fstrict-prototype
12415 @itemx -fno-strict-prototype
12416 Previously it was possible to use an empty prototype parameter list to
12417 indicate an unspecified number of parameters (like C), rather than no
12418 parameters, as C++ demands. This feature has been removed, except where
12419 it is required for backwards compatibility @xref{Backwards Compatibility}.
12422 G++ allows a virtual function returning @samp{void *} to be overridden
12423 by one returning a different pointer type. This extension to the
12424 covariant return type rules is now deprecated and will be removed from a
12427 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
12428 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
12429 and are now removed from G++. Code using these operators should be
12430 modified to use @code{std::min} and @code{std::max} instead.
12432 The named return value extension has been deprecated, and is now
12435 The use of initializer lists with new expressions has been deprecated,
12436 and is now removed from G++.
12438 Floating and complex non-type template parameters have been deprecated,
12439 and are now removed from G++.
12441 The implicit typename extension has been deprecated and is now
12444 The use of default arguments in function pointers, function typedefs
12445 and other places where they are not permitted by the standard is
12446 deprecated and will be removed from a future version of G++.
12448 G++ allows floating-point literals to appear in integral constant expressions,
12449 e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} }
12450 This extension is deprecated and will be removed from a future version.
12452 G++ allows static data members of const floating-point type to be declared
12453 with an initializer in a class definition. The standard only allows
12454 initializers for static members of const integral types and const
12455 enumeration types so this extension has been deprecated and will be removed
12456 from a future version.
12458 @node Backwards Compatibility
12459 @section Backwards Compatibility
12460 @cindex Backwards Compatibility
12461 @cindex ARM [Annotated C++ Reference Manual]
12463 Now that there is a definitive ISO standard C++, G++ has a specification
12464 to adhere to. The C++ language evolved over time, and features that
12465 used to be acceptable in previous drafts of the standard, such as the ARM
12466 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
12467 compilation of C++ written to such drafts, G++ contains some backwards
12468 compatibilities. @emph{All such backwards compatibility features are
12469 liable to disappear in future versions of G++.} They should be considered
12470 deprecated @xref{Deprecated Features}.
12474 If a variable is declared at for scope, it used to remain in scope until
12475 the end of the scope which contained the for statement (rather than just
12476 within the for scope). G++ retains this, but issues a warning, if such a
12477 variable is accessed outside the for scope.
12479 @item Implicit C language
12480 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
12481 scope to set the language. On such systems, all header files are
12482 implicitly scoped inside a C language scope. Also, an empty prototype
12483 @code{()} will be treated as an unspecified number of arguments, rather
12484 than no arguments, as C++ demands.