1 @c Copyright (C) 1988, 1989, 1992, 1993, 1994, 1996, 1998, 1999, 2000, 2001,
2 @c 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009
3 @c Free Software Foundation, Inc.
5 @c This is part of the GCC manual.
6 @c For copying conditions, see the file gcc.texi.
9 @chapter Extensions to the C Language Family
10 @cindex extensions, C language
11 @cindex C language extensions
14 GNU C provides several language features not found in ISO standard C@.
15 (The @option{-pedantic} option directs GCC to print a warning message if
16 any of these features is used.) To test for the availability of these
17 features in conditional compilation, check for a predefined macro
18 @code{__GNUC__}, which is always defined under GCC@.
20 These extensions are available in C and Objective-C@. Most of them are
21 also available in C++. @xref{C++ Extensions,,Extensions to the
22 C++ Language}, for extensions that apply @emph{only} to C++.
24 Some features that are in ISO C99 but not C89 or C++ are also, as
25 extensions, accepted by GCC in C89 mode and in C++.
28 * Statement Exprs:: Putting statements and declarations inside expressions.
29 * Local Labels:: Labels local to a block.
30 * Labels as Values:: Getting pointers to labels, and computed gotos.
31 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
32 * Constructing Calls:: Dispatching a call to another function.
33 * Typeof:: @code{typeof}: referring to the type of an expression.
34 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
35 * Long Long:: Double-word integers---@code{long long int}.
36 * Complex:: Data types for complex numbers.
37 * Floating Types:: Additional Floating Types.
38 * Half-Precision:: Half-Precision Floating Point.
39 * Decimal Float:: Decimal Floating Types.
40 * Hex Floats:: Hexadecimal floating-point constants.
41 * Fixed-Point:: Fixed-Point Types.
42 * Named Address Spaces::Named address spaces.
43 * Zero Length:: Zero-length arrays.
44 * Variable Length:: Arrays whose length is computed at run time.
45 * Empty Structures:: Structures with no members.
46 * Variadic Macros:: Macros with a variable number of arguments.
47 * Escaped Newlines:: Slightly looser rules for escaped newlines.
48 * Subscripting:: Any array can be subscripted, even if not an lvalue.
49 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
50 * Initializers:: Non-constant initializers.
51 * Compound Literals:: Compound literals give structures, unions
53 * Designated Inits:: Labeling elements of initializers.
54 * Cast to Union:: Casting to union type from any member of the union.
55 * Case Ranges:: `case 1 ... 9' and such.
56 * Mixed Declarations:: Mixing declarations and code.
57 * Function Attributes:: Declaring that functions have no side effects,
58 or that they can never return.
59 * Attribute Syntax:: Formal syntax for attributes.
60 * Function Prototypes:: Prototype declarations and old-style definitions.
61 * C++ Comments:: C++ comments are recognized.
62 * Dollar Signs:: Dollar sign is allowed in identifiers.
63 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
64 * Variable Attributes:: Specifying attributes of variables.
65 * Type Attributes:: Specifying attributes of types.
66 * Alignment:: Inquiring about the alignment of a type or variable.
67 * Inline:: Defining inline functions (as fast as macros).
68 * Extended Asm:: Assembler instructions with C expressions as operands.
69 (With them you can define ``built-in'' functions.)
70 * Constraints:: Constraints for asm operands
71 * Asm Labels:: Specifying the assembler name to use for a C symbol.
72 * Explicit Reg Vars:: Defining variables residing in specified registers.
73 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
74 * Incomplete Enums:: @code{enum foo;}, with details to follow.
75 * Function Names:: Printable strings which are the name of the current
77 * Return Address:: Getting the return or frame address of a function.
78 * Vector Extensions:: Using vector instructions through built-in functions.
79 * Offsetof:: Special syntax for implementing @code{offsetof}.
80 * Atomic Builtins:: Built-in functions for atomic memory access.
81 * Object Size Checking:: Built-in functions for limited buffer overflow
83 * Other Builtins:: Other built-in functions.
84 * Target Builtins:: Built-in functions specific to particular targets.
85 * Target Format Checks:: Format checks specific to particular targets.
86 * Pragmas:: Pragmas accepted by GCC.
87 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
88 * Thread-Local:: Per-thread variables.
89 * Binary constants:: Binary constants using the @samp{0b} prefix.
93 @section Statements and Declarations in Expressions
94 @cindex statements inside expressions
95 @cindex declarations inside expressions
96 @cindex expressions containing statements
97 @cindex macros, statements in expressions
99 @c the above section title wrapped and causes an underfull hbox.. i
100 @c changed it from "within" to "in". --mew 4feb93
101 A compound statement enclosed in parentheses may appear as an expression
102 in GNU C@. This allows you to use loops, switches, and local variables
103 within an expression.
105 Recall that a compound statement is a sequence of statements surrounded
106 by braces; in this construct, parentheses go around the braces. For
110 (@{ int y = foo (); int z;
117 is a valid (though slightly more complex than necessary) expression
118 for the absolute value of @code{foo ()}.
120 The last thing in the compound statement should be an expression
121 followed by a semicolon; the value of this subexpression serves as the
122 value of the entire construct. (If you use some other kind of statement
123 last within the braces, the construct has type @code{void}, and thus
124 effectively no value.)
126 This feature is especially useful in making macro definitions ``safe'' (so
127 that they evaluate each operand exactly once). For example, the
128 ``maximum'' function is commonly defined as a macro in standard C as
132 #define max(a,b) ((a) > (b) ? (a) : (b))
136 @cindex side effects, macro argument
137 But this definition computes either @var{a} or @var{b} twice, with bad
138 results if the operand has side effects. In GNU C, if you know the
139 type of the operands (here taken as @code{int}), you can define
140 the macro safely as follows:
143 #define maxint(a,b) \
144 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
147 Embedded statements are not allowed in constant expressions, such as
148 the value of an enumeration constant, the width of a bit-field, or
149 the initial value of a static variable.
151 If you don't know the type of the operand, you can still do this, but you
152 must use @code{typeof} (@pxref{Typeof}).
154 In G++, the result value of a statement expression undergoes array and
155 function pointer decay, and is returned by value to the enclosing
156 expression. For instance, if @code{A} is a class, then
165 will construct a temporary @code{A} object to hold the result of the
166 statement expression, and that will be used to invoke @code{Foo}.
167 Therefore the @code{this} pointer observed by @code{Foo} will not be the
170 Any temporaries created within a statement within a statement expression
171 will be destroyed at the statement's end. This makes statement
172 expressions inside macros slightly different from function calls. In
173 the latter case temporaries introduced during argument evaluation will
174 be destroyed at the end of the statement that includes the function
175 call. In the statement expression case they will be destroyed during
176 the statement expression. For instance,
179 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
180 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
190 will have different places where temporaries are destroyed. For the
191 @code{macro} case, the temporary @code{X} will be destroyed just after
192 the initialization of @code{b}. In the @code{function} case that
193 temporary will be destroyed when the function returns.
195 These considerations mean that it is probably a bad idea to use
196 statement-expressions of this form in header files that are designed to
197 work with C++. (Note that some versions of the GNU C Library contained
198 header files using statement-expression that lead to precisely this
201 Jumping into a statement expression with @code{goto} or using a
202 @code{switch} statement outside the statement expression with a
203 @code{case} or @code{default} label inside the statement expression is
204 not permitted. Jumping into a statement expression with a computed
205 @code{goto} (@pxref{Labels as Values}) yields undefined behavior.
206 Jumping out of a statement expression is permitted, but if the
207 statement expression is part of a larger expression then it is
208 unspecified which other subexpressions of that expression have been
209 evaluated except where the language definition requires certain
210 subexpressions to be evaluated before or after the statement
211 expression. In any case, as with a function call the evaluation of a
212 statement expression is not interleaved with the evaluation of other
213 parts of the containing expression. For example,
216 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
220 will call @code{foo} and @code{bar1} and will not call @code{baz} but
221 may or may not call @code{bar2}. If @code{bar2} is called, it will be
222 called after @code{foo} and before @code{bar1}
225 @section Locally Declared Labels
227 @cindex macros, local labels
229 GCC allows you to declare @dfn{local labels} in any nested block
230 scope. A local label is just like an ordinary label, but you can
231 only reference it (with a @code{goto} statement, or by taking its
232 address) within the block in which it was declared.
234 A local label declaration looks like this:
237 __label__ @var{label};
244 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
247 Local label declarations must come at the beginning of the block,
248 before any ordinary declarations or statements.
250 The label declaration defines the label @emph{name}, but does not define
251 the label itself. You must do this in the usual way, with
252 @code{@var{label}:}, within the statements of the statement expression.
254 The local label feature is useful for complex macros. If a macro
255 contains nested loops, a @code{goto} can be useful for breaking out of
256 them. However, an ordinary label whose scope is the whole function
257 cannot be used: if the macro can be expanded several times in one
258 function, the label will be multiply defined in that function. A
259 local label avoids this problem. For example:
262 #define SEARCH(value, array, target) \
265 typeof (target) _SEARCH_target = (target); \
266 typeof (*(array)) *_SEARCH_array = (array); \
269 for (i = 0; i < max; i++) \
270 for (j = 0; j < max; j++) \
271 if (_SEARCH_array[i][j] == _SEARCH_target) \
272 @{ (value) = i; goto found; @} \
278 This could also be written using a statement-expression:
281 #define SEARCH(array, target) \
284 typeof (target) _SEARCH_target = (target); \
285 typeof (*(array)) *_SEARCH_array = (array); \
288 for (i = 0; i < max; i++) \
289 for (j = 0; j < max; j++) \
290 if (_SEARCH_array[i][j] == _SEARCH_target) \
291 @{ value = i; goto found; @} \
298 Local label declarations also make the labels they declare visible to
299 nested functions, if there are any. @xref{Nested Functions}, for details.
301 @node Labels as Values
302 @section Labels as Values
303 @cindex labels as values
304 @cindex computed gotos
305 @cindex goto with computed label
306 @cindex address of a label
308 You can get the address of a label defined in the current function
309 (or a containing function) with the unary operator @samp{&&}. The
310 value has type @code{void *}. This value is a constant and can be used
311 wherever a constant of that type is valid. For example:
319 To use these values, you need to be able to jump to one. This is done
320 with the computed goto statement@footnote{The analogous feature in
321 Fortran is called an assigned goto, but that name seems inappropriate in
322 C, where one can do more than simply store label addresses in label
323 variables.}, @code{goto *@var{exp};}. For example,
330 Any expression of type @code{void *} is allowed.
332 One way of using these constants is in initializing a static array that
333 will serve as a jump table:
336 static void *array[] = @{ &&foo, &&bar, &&hack @};
339 Then you can select a label with indexing, like this:
346 Note that this does not check whether the subscript is in bounds---array
347 indexing in C never does that.
349 Such an array of label values serves a purpose much like that of the
350 @code{switch} statement. The @code{switch} statement is cleaner, so
351 use that rather than an array unless the problem does not fit a
352 @code{switch} statement very well.
354 Another use of label values is in an interpreter for threaded code.
355 The labels within the interpreter function can be stored in the
356 threaded code for super-fast dispatching.
358 You may not use this mechanism to jump to code in a different function.
359 If you do that, totally unpredictable things will happen. The best way to
360 avoid this is to store the label address only in automatic variables and
361 never pass it as an argument.
363 An alternate way to write the above example is
366 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
368 goto *(&&foo + array[i]);
372 This is more friendly to code living in shared libraries, as it reduces
373 the number of dynamic relocations that are needed, and by consequence,
374 allows the data to be read-only.
376 The @code{&&foo} expressions for the same label might have different
377 values if the containing function is inlined or cloned. If a program
378 relies on them being always the same,
379 @code{__attribute__((__noinline__,__noclone__))} should be used to
380 prevent inlining and cloning. If @code{&&foo} is used in a static
381 variable initializer, inlining and cloning is forbidden.
383 @node Nested Functions
384 @section Nested Functions
385 @cindex nested functions
386 @cindex downward funargs
389 A @dfn{nested function} is a function defined inside another function.
390 (Nested functions are not supported for GNU C++.) The nested function's
391 name is local to the block where it is defined. For example, here we
392 define a nested function named @code{square}, and call it twice:
396 foo (double a, double b)
398 double square (double z) @{ return z * z; @}
400 return square (a) + square (b);
405 The nested function can access all the variables of the containing
406 function that are visible at the point of its definition. This is
407 called @dfn{lexical scoping}. For example, here we show a nested
408 function which uses an inherited variable named @code{offset}:
412 bar (int *array, int offset, int size)
414 int access (int *array, int index)
415 @{ return array[index + offset]; @}
418 for (i = 0; i < size; i++)
419 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
424 Nested function definitions are permitted within functions in the places
425 where variable definitions are allowed; that is, in any block, mixed
426 with the other declarations and statements in the block.
428 It is possible to call the nested function from outside the scope of its
429 name by storing its address or passing the address to another function:
432 hack (int *array, int size)
434 void store (int index, int value)
435 @{ array[index] = value; @}
437 intermediate (store, size);
441 Here, the function @code{intermediate} receives the address of
442 @code{store} as an argument. If @code{intermediate} calls @code{store},
443 the arguments given to @code{store} are used to store into @code{array}.
444 But this technique works only so long as the containing function
445 (@code{hack}, in this example) does not exit.
447 If you try to call the nested function through its address after the
448 containing function has exited, all hell will break loose. If you try
449 to call it after a containing scope level has exited, and if it refers
450 to some of the variables that are no longer in scope, you may be lucky,
451 but it's not wise to take the risk. If, however, the nested function
452 does not refer to anything that has gone out of scope, you should be
455 GCC implements taking the address of a nested function using a technique
456 called @dfn{trampolines}. This technique was described in
457 @cite{Lexical Closures for C++} (Thomas M. Breuel, USENIX
458 C++ Conference Proceedings, October 17-21, 1988).
460 A nested function can jump to a label inherited from a containing
461 function, provided the label was explicitly declared in the containing
462 function (@pxref{Local Labels}). Such a jump returns instantly to the
463 containing function, exiting the nested function which did the
464 @code{goto} and any intermediate functions as well. Here is an example:
468 bar (int *array, int offset, int size)
471 int access (int *array, int index)
475 return array[index + offset];
479 for (i = 0; i < size; i++)
480 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
484 /* @r{Control comes here from @code{access}
485 if it detects an error.} */
492 A nested function always has no linkage. Declaring one with
493 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
494 before its definition, use @code{auto} (which is otherwise meaningless
495 for function declarations).
498 bar (int *array, int offset, int size)
501 auto int access (int *, int);
503 int access (int *array, int index)
507 return array[index + offset];
513 @node Constructing Calls
514 @section Constructing Function Calls
515 @cindex constructing calls
516 @cindex forwarding calls
518 Using the built-in functions described below, you can record
519 the arguments a function received, and call another function
520 with the same arguments, without knowing the number or types
523 You can also record the return value of that function call,
524 and later return that value, without knowing what data type
525 the function tried to return (as long as your caller expects
528 However, these built-in functions may interact badly with some
529 sophisticated features or other extensions of the language. It
530 is, therefore, not recommended to use them outside very simple
531 functions acting as mere forwarders for their arguments.
533 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
534 This built-in function returns a pointer to data
535 describing how to perform a call with the same arguments as were passed
536 to the current function.
538 The function saves the arg pointer register, structure value address,
539 and all registers that might be used to pass arguments to a function
540 into a block of memory allocated on the stack. Then it returns the
541 address of that block.
544 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
545 This built-in function invokes @var{function}
546 with a copy of the parameters described by @var{arguments}
549 The value of @var{arguments} should be the value returned by
550 @code{__builtin_apply_args}. The argument @var{size} specifies the size
551 of the stack argument data, in bytes.
553 This function returns a pointer to data describing
554 how to return whatever value was returned by @var{function}. The data
555 is saved in a block of memory allocated on the stack.
557 It is not always simple to compute the proper value for @var{size}. The
558 value is used by @code{__builtin_apply} to compute the amount of data
559 that should be pushed on the stack and copied from the incoming argument
563 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
564 This built-in function returns the value described by @var{result} from
565 the containing function. You should specify, for @var{result}, a value
566 returned by @code{__builtin_apply}.
569 @deftypefn {Built-in Function} __builtin_va_arg_pack ()
570 This built-in function represents all anonymous arguments of an inline
571 function. It can be used only in inline functions which will be always
572 inlined, never compiled as a separate function, such as those using
573 @code{__attribute__ ((__always_inline__))} or
574 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
575 It must be only passed as last argument to some other function
576 with variable arguments. This is useful for writing small wrapper
577 inlines for variable argument functions, when using preprocessor
578 macros is undesirable. For example:
580 extern int myprintf (FILE *f, const char *format, ...);
581 extern inline __attribute__ ((__gnu_inline__)) int
582 myprintf (FILE *f, const char *format, ...)
584 int r = fprintf (f, "myprintf: ");
587 int s = fprintf (f, format, __builtin_va_arg_pack ());
595 @deftypefn {Built-in Function} __builtin_va_arg_pack_len ()
596 This built-in function returns the number of anonymous arguments of
597 an inline function. It can be used only in inline functions which
598 will be always inlined, never compiled as a separate function, such
599 as those using @code{__attribute__ ((__always_inline__))} or
600 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
601 For example following will do link or runtime checking of open
602 arguments for optimized code:
605 extern inline __attribute__((__gnu_inline__)) int
606 myopen (const char *path, int oflag, ...)
608 if (__builtin_va_arg_pack_len () > 1)
609 warn_open_too_many_arguments ();
611 if (__builtin_constant_p (oflag))
613 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1)
615 warn_open_missing_mode ();
616 return __open_2 (path, oflag);
618 return open (path, oflag, __builtin_va_arg_pack ());
621 if (__builtin_va_arg_pack_len () < 1)
622 return __open_2 (path, oflag);
624 return open (path, oflag, __builtin_va_arg_pack ());
631 @section Referring to a Type with @code{typeof}
634 @cindex macros, types of arguments
636 Another way to refer to the type of an expression is with @code{typeof}.
637 The syntax of using of this keyword looks like @code{sizeof}, but the
638 construct acts semantically like a type name defined with @code{typedef}.
640 There are two ways of writing the argument to @code{typeof}: with an
641 expression or with a type. Here is an example with an expression:
648 This assumes that @code{x} is an array of pointers to functions;
649 the type described is that of the values of the functions.
651 Here is an example with a typename as the argument:
658 Here the type described is that of pointers to @code{int}.
660 If you are writing a header file that must work when included in ISO C
661 programs, write @code{__typeof__} instead of @code{typeof}.
662 @xref{Alternate Keywords}.
664 A @code{typeof}-construct can be used anywhere a typedef name could be
665 used. For example, you can use it in a declaration, in a cast, or inside
666 of @code{sizeof} or @code{typeof}.
668 The operand of @code{typeof} is evaluated for its side effects if and
669 only if it is an expression of variably modified type or the name of
672 @code{typeof} is often useful in conjunction with the
673 statements-within-expressions feature. Here is how the two together can
674 be used to define a safe ``maximum'' macro that operates on any
675 arithmetic type and evaluates each of its arguments exactly once:
679 (@{ typeof (a) _a = (a); \
680 typeof (b) _b = (b); \
681 _a > _b ? _a : _b; @})
684 @cindex underscores in variables in macros
685 @cindex @samp{_} in variables in macros
686 @cindex local variables in macros
687 @cindex variables, local, in macros
688 @cindex macros, local variables in
690 The reason for using names that start with underscores for the local
691 variables is to avoid conflicts with variable names that occur within the
692 expressions that are substituted for @code{a} and @code{b}. Eventually we
693 hope to design a new form of declaration syntax that allows you to declare
694 variables whose scopes start only after their initializers; this will be a
695 more reliable way to prevent such conflicts.
698 Some more examples of the use of @code{typeof}:
702 This declares @code{y} with the type of what @code{x} points to.
709 This declares @code{y} as an array of such values.
716 This declares @code{y} as an array of pointers to characters:
719 typeof (typeof (char *)[4]) y;
723 It is equivalent to the following traditional C declaration:
729 To see the meaning of the declaration using @code{typeof}, and why it
730 might be a useful way to write, rewrite it with these macros:
733 #define pointer(T) typeof(T *)
734 #define array(T, N) typeof(T [N])
738 Now the declaration can be rewritten this way:
741 array (pointer (char), 4) y;
745 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
746 pointers to @code{char}.
749 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
750 a more limited extension which permitted one to write
753 typedef @var{T} = @var{expr};
757 with the effect of declaring @var{T} to have the type of the expression
758 @var{expr}. This extension does not work with GCC 3 (versions between
759 3.0 and 3.2 will crash; 3.2.1 and later give an error). Code which
760 relies on it should be rewritten to use @code{typeof}:
763 typedef typeof(@var{expr}) @var{T};
767 This will work with all versions of GCC@.
770 @section Conditionals with Omitted Operands
771 @cindex conditional expressions, extensions
772 @cindex omitted middle-operands
773 @cindex middle-operands, omitted
774 @cindex extensions, @code{?:}
775 @cindex @code{?:} extensions
777 The middle operand in a conditional expression may be omitted. Then
778 if the first operand is nonzero, its value is the value of the conditional
781 Therefore, the expression
788 has the value of @code{x} if that is nonzero; otherwise, the value of
791 This example is perfectly equivalent to
797 @cindex side effect in ?:
798 @cindex ?: side effect
800 In this simple case, the ability to omit the middle operand is not
801 especially useful. When it becomes useful is when the first operand does,
802 or may (if it is a macro argument), contain a side effect. Then repeating
803 the operand in the middle would perform the side effect twice. Omitting
804 the middle operand uses the value already computed without the undesirable
805 effects of recomputing it.
808 @section Double-Word Integers
809 @cindex @code{long long} data types
810 @cindex double-word arithmetic
811 @cindex multiprecision arithmetic
812 @cindex @code{LL} integer suffix
813 @cindex @code{ULL} integer suffix
815 ISO C99 supports data types for integers that are at least 64 bits wide,
816 and as an extension GCC supports them in C89 mode and in C++.
817 Simply write @code{long long int} for a signed integer, or
818 @code{unsigned long long int} for an unsigned integer. To make an
819 integer constant of type @code{long long int}, add the suffix @samp{LL}
820 to the integer. To make an integer constant of type @code{unsigned long
821 long int}, add the suffix @samp{ULL} to the integer.
823 You can use these types in arithmetic like any other integer types.
824 Addition, subtraction, and bitwise boolean operations on these types
825 are open-coded on all types of machines. Multiplication is open-coded
826 if the machine supports fullword-to-doubleword a widening multiply
827 instruction. Division and shifts are open-coded only on machines that
828 provide special support. The operations that are not open-coded use
829 special library routines that come with GCC@.
831 There may be pitfalls when you use @code{long long} types for function
832 arguments, unless you declare function prototypes. If a function
833 expects type @code{int} for its argument, and you pass a value of type
834 @code{long long int}, confusion will result because the caller and the
835 subroutine will disagree about the number of bytes for the argument.
836 Likewise, if the function expects @code{long long int} and you pass
837 @code{int}. The best way to avoid such problems is to use prototypes.
840 @section Complex Numbers
841 @cindex complex numbers
842 @cindex @code{_Complex} keyword
843 @cindex @code{__complex__} keyword
845 ISO C99 supports complex floating data types, and as an extension GCC
846 supports them in C89 mode and in C++, and supports complex integer data
847 types which are not part of ISO C99. You can declare complex types
848 using the keyword @code{_Complex}. As an extension, the older GNU
849 keyword @code{__complex__} is also supported.
851 For example, @samp{_Complex double x;} declares @code{x} as a
852 variable whose real part and imaginary part are both of type
853 @code{double}. @samp{_Complex short int y;} declares @code{y} to
854 have real and imaginary parts of type @code{short int}; this is not
855 likely to be useful, but it shows that the set of complex types is
858 To write a constant with a complex data type, use the suffix @samp{i} or
859 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
860 has type @code{_Complex float} and @code{3i} has type
861 @code{_Complex int}. Such a constant always has a pure imaginary
862 value, but you can form any complex value you like by adding one to a
863 real constant. This is a GNU extension; if you have an ISO C99
864 conforming C library (such as GNU libc), and want to construct complex
865 constants of floating type, you should include @code{<complex.h>} and
866 use the macros @code{I} or @code{_Complex_I} instead.
868 @cindex @code{__real__} keyword
869 @cindex @code{__imag__} keyword
870 To extract the real part of a complex-valued expression @var{exp}, write
871 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
872 extract the imaginary part. This is a GNU extension; for values of
873 floating type, you should use the ISO C99 functions @code{crealf},
874 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
875 @code{cimagl}, declared in @code{<complex.h>} and also provided as
876 built-in functions by GCC@.
878 @cindex complex conjugation
879 The operator @samp{~} performs complex conjugation when used on a value
880 with a complex type. This is a GNU extension; for values of
881 floating type, you should use the ISO C99 functions @code{conjf},
882 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
883 provided as built-in functions by GCC@.
885 GCC can allocate complex automatic variables in a noncontiguous
886 fashion; it's even possible for the real part to be in a register while
887 the imaginary part is on the stack (or vice-versa). Only the DWARF2
888 debug info format can represent this, so use of DWARF2 is recommended.
889 If you are using the stabs debug info format, GCC describes a noncontiguous
890 complex variable as if it were two separate variables of noncomplex type.
891 If the variable's actual name is @code{foo}, the two fictitious
892 variables are named @code{foo$real} and @code{foo$imag}. You can
893 examine and set these two fictitious variables with your debugger.
896 @section Additional Floating Types
897 @cindex additional floating types
898 @cindex @code{__float80} data type
899 @cindex @code{__float128} data type
900 @cindex @code{w} floating point suffix
901 @cindex @code{q} floating point suffix
902 @cindex @code{W} floating point suffix
903 @cindex @code{Q} floating point suffix
905 As an extension, the GNU C compiler supports additional floating
906 types, @code{__float80} and @code{__float128} to support 80bit
907 (@code{XFmode}) and 128 bit (@code{TFmode}) floating types.
908 Support for additional types includes the arithmetic operators:
909 add, subtract, multiply, divide; unary arithmetic operators;
910 relational operators; equality operators; and conversions to and from
911 integer and other floating types. Use a suffix @samp{w} or @samp{W}
912 in a literal constant of type @code{__float80} and @samp{q} or @samp{Q}
913 for @code{_float128}. You can declare complex types using the
914 corresponding internal complex type, @code{XCmode} for @code{__float80}
915 type and @code{TCmode} for @code{__float128} type:
918 typedef _Complex float __attribute__((mode(TC))) _Complex128;
919 typedef _Complex float __attribute__((mode(XC))) _Complex80;
922 Not all targets support additional floating point types. @code{__float80}
923 and @code{__float128} types are supported on i386, x86_64 and ia64 targets.
926 @section Half-Precision Floating Point
927 @cindex half-precision floating point
928 @cindex @code{__fp16} data type
930 On ARM targets, GCC supports half-precision (16-bit) floating point via
931 the @code{__fp16} type. You must enable this type explicitly
932 with the @option{-mfp16-format} command-line option in order to use it.
934 ARM supports two incompatible representations for half-precision
935 floating-point values. You must choose one of the representations and
936 use it consistently in your program.
938 Specifying @option{-mfp16-format=ieee} selects the IEEE 754-2008 format.
939 This format can represent normalized values in the range of @math{2^{-14}} to 65504.
940 There are 11 bits of significand precision, approximately 3
943 Specifying @option{-mfp16-format=alternative} selects the ARM
944 alternative format. This representation is similar to the IEEE
945 format, but does not support infinities or NaNs. Instead, the range
946 of exponents is extended, so that this format can represent normalized
947 values in the range of @math{2^{-14}} to 131008.
949 The @code{__fp16} type is a storage format only. For purposes
950 of arithmetic and other operations, @code{__fp16} values in C or C++
951 expressions are automatically promoted to @code{float}. In addition,
952 you cannot declare a function with a return value or parameters
953 of type @code{__fp16}.
955 Note that conversions from @code{double} to @code{__fp16}
956 involve an intermediate conversion to @code{float}. Because
957 of rounding, this can sometimes produce a different result than a
960 ARM provides hardware support for conversions between
961 @code{__fp16} and @code{float} values
962 as an extension to VFP and NEON (Advanced SIMD). GCC generates
963 code using the instructions provided by this extension if you compile
964 with the options @option{-mfpu=neon-fp16 -mfloat-abi=softfp},
965 in addition to the @option{-mfp16-format} option to select
966 a half-precision format.
968 Language-level support for the @code{__fp16} data type is
969 independent of whether GCC generates code using hardware floating-point
970 instructions. In cases where hardware support is not specified, GCC
971 implements conversions between @code{__fp16} and @code{float} values
975 @section Decimal Floating Types
976 @cindex decimal floating types
977 @cindex @code{_Decimal32} data type
978 @cindex @code{_Decimal64} data type
979 @cindex @code{_Decimal128} data type
980 @cindex @code{df} integer suffix
981 @cindex @code{dd} integer suffix
982 @cindex @code{dl} integer suffix
983 @cindex @code{DF} integer suffix
984 @cindex @code{DD} integer suffix
985 @cindex @code{DL} integer suffix
987 As an extension, the GNU C compiler supports decimal floating types as
988 defined in the N1312 draft of ISO/IEC WDTR24732. Support for decimal
989 floating types in GCC will evolve as the draft technical report changes.
990 Calling conventions for any target might also change. Not all targets
991 support decimal floating types.
993 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
994 @code{_Decimal128}. They use a radix of ten, unlike the floating types
995 @code{float}, @code{double}, and @code{long double} whose radix is not
996 specified by the C standard but is usually two.
998 Support for decimal floating types includes the arithmetic operators
999 add, subtract, multiply, divide; unary arithmetic operators;
1000 relational operators; equality operators; and conversions to and from
1001 integer and other floating types. Use a suffix @samp{df} or
1002 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
1003 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
1006 GCC support of decimal float as specified by the draft technical report
1011 When the value of a decimal floating type cannot be represented in the
1012 integer type to which it is being converted, the result is undefined
1013 rather than the result value specified by the draft technical report.
1016 GCC does not provide the C library functionality associated with
1017 @file{math.h}, @file{fenv.h}, @file{stdio.h}, @file{stdlib.h}, and
1018 @file{wchar.h}, which must come from a separate C library implementation.
1019 Because of this the GNU C compiler does not define macro
1020 @code{__STDC_DEC_FP__} to indicate that the implementation conforms to
1021 the technical report.
1024 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
1025 are supported by the DWARF2 debug information format.
1031 ISO C99 supports floating-point numbers written not only in the usual
1032 decimal notation, such as @code{1.55e1}, but also numbers such as
1033 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1034 supports this in C89 mode (except in some cases when strictly
1035 conforming) and in C++. In that format the
1036 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1037 mandatory. The exponent is a decimal number that indicates the power of
1038 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1045 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1046 is the same as @code{1.55e1}.
1048 Unlike for floating-point numbers in the decimal notation the exponent
1049 is always required in the hexadecimal notation. Otherwise the compiler
1050 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1051 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1052 extension for floating-point constants of type @code{float}.
1055 @section Fixed-Point Types
1056 @cindex fixed-point types
1057 @cindex @code{_Fract} data type
1058 @cindex @code{_Accum} data type
1059 @cindex @code{_Sat} data type
1060 @cindex @code{hr} fixed-suffix
1061 @cindex @code{r} fixed-suffix
1062 @cindex @code{lr} fixed-suffix
1063 @cindex @code{llr} fixed-suffix
1064 @cindex @code{uhr} fixed-suffix
1065 @cindex @code{ur} fixed-suffix
1066 @cindex @code{ulr} fixed-suffix
1067 @cindex @code{ullr} fixed-suffix
1068 @cindex @code{hk} fixed-suffix
1069 @cindex @code{k} fixed-suffix
1070 @cindex @code{lk} fixed-suffix
1071 @cindex @code{llk} fixed-suffix
1072 @cindex @code{uhk} fixed-suffix
1073 @cindex @code{uk} fixed-suffix
1074 @cindex @code{ulk} fixed-suffix
1075 @cindex @code{ullk} fixed-suffix
1076 @cindex @code{HR} fixed-suffix
1077 @cindex @code{R} fixed-suffix
1078 @cindex @code{LR} fixed-suffix
1079 @cindex @code{LLR} fixed-suffix
1080 @cindex @code{UHR} fixed-suffix
1081 @cindex @code{UR} fixed-suffix
1082 @cindex @code{ULR} fixed-suffix
1083 @cindex @code{ULLR} fixed-suffix
1084 @cindex @code{HK} fixed-suffix
1085 @cindex @code{K} fixed-suffix
1086 @cindex @code{LK} fixed-suffix
1087 @cindex @code{LLK} fixed-suffix
1088 @cindex @code{UHK} fixed-suffix
1089 @cindex @code{UK} fixed-suffix
1090 @cindex @code{ULK} fixed-suffix
1091 @cindex @code{ULLK} fixed-suffix
1093 As an extension, the GNU C compiler supports fixed-point types as
1094 defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point
1095 types in GCC will evolve as the draft technical report changes.
1096 Calling conventions for any target might also change. Not all targets
1097 support fixed-point types.
1099 The fixed-point types are
1100 @code{short _Fract},
1103 @code{long long _Fract},
1104 @code{unsigned short _Fract},
1105 @code{unsigned _Fract},
1106 @code{unsigned long _Fract},
1107 @code{unsigned long long _Fract},
1108 @code{_Sat short _Fract},
1110 @code{_Sat long _Fract},
1111 @code{_Sat long long _Fract},
1112 @code{_Sat unsigned short _Fract},
1113 @code{_Sat unsigned _Fract},
1114 @code{_Sat unsigned long _Fract},
1115 @code{_Sat unsigned long long _Fract},
1116 @code{short _Accum},
1119 @code{long long _Accum},
1120 @code{unsigned short _Accum},
1121 @code{unsigned _Accum},
1122 @code{unsigned long _Accum},
1123 @code{unsigned long long _Accum},
1124 @code{_Sat short _Accum},
1126 @code{_Sat long _Accum},
1127 @code{_Sat long long _Accum},
1128 @code{_Sat unsigned short _Accum},
1129 @code{_Sat unsigned _Accum},
1130 @code{_Sat unsigned long _Accum},
1131 @code{_Sat unsigned long long _Accum}.
1133 Fixed-point data values contain fractional and optional integral parts.
1134 The format of fixed-point data varies and depends on the target machine.
1136 Support for fixed-point types includes:
1139 prefix and postfix increment and decrement operators (@code{++}, @code{--})
1141 unary arithmetic operators (@code{+}, @code{-}, @code{!})
1143 binary arithmetic operators (@code{+}, @code{-}, @code{*}, @code{/})
1145 binary shift operators (@code{<<}, @code{>>})
1147 relational operators (@code{<}, @code{<=}, @code{>=}, @code{>})
1149 equality operators (@code{==}, @code{!=})
1151 assignment operators (@code{+=}, @code{-=}, @code{*=}, @code{/=},
1152 @code{<<=}, @code{>>=})
1154 conversions to and from integer, floating-point, or fixed-point types
1157 Use a suffix in a fixed-point literal constant:
1159 @item @samp{hr} or @samp{HR} for @code{short _Fract} and
1160 @code{_Sat short _Fract}
1161 @item @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract}
1162 @item @samp{lr} or @samp{LR} for @code{long _Fract} and
1163 @code{_Sat long _Fract}
1164 @item @samp{llr} or @samp{LLR} for @code{long long _Fract} and
1165 @code{_Sat long long _Fract}
1166 @item @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and
1167 @code{_Sat unsigned short _Fract}
1168 @item @samp{ur} or @samp{UR} for @code{unsigned _Fract} and
1169 @code{_Sat unsigned _Fract}
1170 @item @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and
1171 @code{_Sat unsigned long _Fract}
1172 @item @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract}
1173 and @code{_Sat unsigned long long _Fract}
1174 @item @samp{hk} or @samp{HK} for @code{short _Accum} and
1175 @code{_Sat short _Accum}
1176 @item @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum}
1177 @item @samp{lk} or @samp{LK} for @code{long _Accum} and
1178 @code{_Sat long _Accum}
1179 @item @samp{llk} or @samp{LLK} for @code{long long _Accum} and
1180 @code{_Sat long long _Accum}
1181 @item @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and
1182 @code{_Sat unsigned short _Accum}
1183 @item @samp{uk} or @samp{UK} for @code{unsigned _Accum} and
1184 @code{_Sat unsigned _Accum}
1185 @item @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and
1186 @code{_Sat unsigned long _Accum}
1187 @item @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum}
1188 and @code{_Sat unsigned long long _Accum}
1191 GCC support of fixed-point types as specified by the draft technical report
1196 Pragmas to control overflow and rounding behaviors are not implemented.
1199 Fixed-point types are supported by the DWARF2 debug information format.
1201 @node Named Address Spaces
1202 @section Named address spaces
1203 @cindex named address spaces
1205 As an extension, the GNU C compiler supports named address spaces as
1206 defined in the N1275 draft of ISO/IEC DTR 18037. Support for named
1207 address spaces in GCC will evolve as the draft technical report changes.
1208 Calling conventions for any target might also change. At present, only
1209 the SPU target supports other address spaces. On the SPU target, for
1210 example, variables may be declared as belonging to another address space
1211 by qualifying the type with the @code{__ea} address space identifier:
1217 When the variable @code{i} is accessed, the compiler will generate
1218 special code to access this variable. It may use runtime library
1219 support, or generate special machine instructions to access that address
1222 The @code{__ea} identifier may be used exactly like any other C type
1223 qualifier (e.g., @code{const} or @code{volatile}). See the N1275
1224 document for more details.
1227 @section Arrays of Length Zero
1228 @cindex arrays of length zero
1229 @cindex zero-length arrays
1230 @cindex length-zero arrays
1231 @cindex flexible array members
1233 Zero-length arrays are allowed in GNU C@. They are very useful as the
1234 last element of a structure which is really a header for a variable-length
1243 struct line *thisline = (struct line *)
1244 malloc (sizeof (struct line) + this_length);
1245 thisline->length = this_length;
1248 In ISO C90, you would have to give @code{contents} a length of 1, which
1249 means either you waste space or complicate the argument to @code{malloc}.
1251 In ISO C99, you would use a @dfn{flexible array member}, which is
1252 slightly different in syntax and semantics:
1256 Flexible array members are written as @code{contents[]} without
1260 Flexible array members have incomplete type, and so the @code{sizeof}
1261 operator may not be applied. As a quirk of the original implementation
1262 of zero-length arrays, @code{sizeof} evaluates to zero.
1265 Flexible array members may only appear as the last member of a
1266 @code{struct} that is otherwise non-empty.
1269 A structure containing a flexible array member, or a union containing
1270 such a structure (possibly recursively), may not be a member of a
1271 structure or an element of an array. (However, these uses are
1272 permitted by GCC as extensions.)
1275 GCC versions before 3.0 allowed zero-length arrays to be statically
1276 initialized, as if they were flexible arrays. In addition to those
1277 cases that were useful, it also allowed initializations in situations
1278 that would corrupt later data. Non-empty initialization of zero-length
1279 arrays is now treated like any case where there are more initializer
1280 elements than the array holds, in that a suitable warning about "excess
1281 elements in array" is given, and the excess elements (all of them, in
1282 this case) are ignored.
1284 Instead GCC allows static initialization of flexible array members.
1285 This is equivalent to defining a new structure containing the original
1286 structure followed by an array of sufficient size to contain the data.
1287 I.e.@: in the following, @code{f1} is constructed as if it were declared
1293 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1296 struct f1 f1; int data[3];
1297 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1301 The convenience of this extension is that @code{f1} has the desired
1302 type, eliminating the need to consistently refer to @code{f2.f1}.
1304 This has symmetry with normal static arrays, in that an array of
1305 unknown size is also written with @code{[]}.
1307 Of course, this extension only makes sense if the extra data comes at
1308 the end of a top-level object, as otherwise we would be overwriting
1309 data at subsequent offsets. To avoid undue complication and confusion
1310 with initialization of deeply nested arrays, we simply disallow any
1311 non-empty initialization except when the structure is the top-level
1312 object. For example:
1315 struct foo @{ int x; int y[]; @};
1316 struct bar @{ struct foo z; @};
1318 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1319 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1320 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1321 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1324 @node Empty Structures
1325 @section Structures With No Members
1326 @cindex empty structures
1327 @cindex zero-size structures
1329 GCC permits a C structure to have no members:
1336 The structure will have size zero. In C++, empty structures are part
1337 of the language. G++ treats empty structures as if they had a single
1338 member of type @code{char}.
1340 @node Variable Length
1341 @section Arrays of Variable Length
1342 @cindex variable-length arrays
1343 @cindex arrays of variable length
1346 Variable-length automatic arrays are allowed in ISO C99, and as an
1347 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1348 implementation of variable-length arrays does not yet conform in detail
1349 to the ISO C99 standard.) These arrays are
1350 declared like any other automatic arrays, but with a length that is not
1351 a constant expression. The storage is allocated at the point of
1352 declaration and deallocated when the brace-level is exited. For
1357 concat_fopen (char *s1, char *s2, char *mode)
1359 char str[strlen (s1) + strlen (s2) + 1];
1362 return fopen (str, mode);
1366 @cindex scope of a variable length array
1367 @cindex variable-length array scope
1368 @cindex deallocating variable length arrays
1369 Jumping or breaking out of the scope of the array name deallocates the
1370 storage. Jumping into the scope is not allowed; you get an error
1373 @cindex @code{alloca} vs variable-length arrays
1374 You can use the function @code{alloca} to get an effect much like
1375 variable-length arrays. The function @code{alloca} is available in
1376 many other C implementations (but not in all). On the other hand,
1377 variable-length arrays are more elegant.
1379 There are other differences between these two methods. Space allocated
1380 with @code{alloca} exists until the containing @emph{function} returns.
1381 The space for a variable-length array is deallocated as soon as the array
1382 name's scope ends. (If you use both variable-length arrays and
1383 @code{alloca} in the same function, deallocation of a variable-length array
1384 will also deallocate anything more recently allocated with @code{alloca}.)
1386 You can also use variable-length arrays as arguments to functions:
1390 tester (int len, char data[len][len])
1396 The length of an array is computed once when the storage is allocated
1397 and is remembered for the scope of the array in case you access it with
1400 If you want to pass the array first and the length afterward, you can
1401 use a forward declaration in the parameter list---another GNU extension.
1405 tester (int len; char data[len][len], int len)
1411 @cindex parameter forward declaration
1412 The @samp{int len} before the semicolon is a @dfn{parameter forward
1413 declaration}, and it serves the purpose of making the name @code{len}
1414 known when the declaration of @code{data} is parsed.
1416 You can write any number of such parameter forward declarations in the
1417 parameter list. They can be separated by commas or semicolons, but the
1418 last one must end with a semicolon, which is followed by the ``real''
1419 parameter declarations. Each forward declaration must match a ``real''
1420 declaration in parameter name and data type. ISO C99 does not support
1421 parameter forward declarations.
1423 @node Variadic Macros
1424 @section Macros with a Variable Number of Arguments.
1425 @cindex variable number of arguments
1426 @cindex macro with variable arguments
1427 @cindex rest argument (in macro)
1428 @cindex variadic macros
1430 In the ISO C standard of 1999, a macro can be declared to accept a
1431 variable number of arguments much as a function can. The syntax for
1432 defining the macro is similar to that of a function. Here is an
1436 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1439 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1440 such a macro, it represents the zero or more tokens until the closing
1441 parenthesis that ends the invocation, including any commas. This set of
1442 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1443 wherever it appears. See the CPP manual for more information.
1445 GCC has long supported variadic macros, and used a different syntax that
1446 allowed you to give a name to the variable arguments just like any other
1447 argument. Here is an example:
1450 #define debug(format, args...) fprintf (stderr, format, args)
1453 This is in all ways equivalent to the ISO C example above, but arguably
1454 more readable and descriptive.
1456 GNU CPP has two further variadic macro extensions, and permits them to
1457 be used with either of the above forms of macro definition.
1459 In standard C, you are not allowed to leave the variable argument out
1460 entirely; but you are allowed to pass an empty argument. For example,
1461 this invocation is invalid in ISO C, because there is no comma after
1468 GNU CPP permits you to completely omit the variable arguments in this
1469 way. In the above examples, the compiler would complain, though since
1470 the expansion of the macro still has the extra comma after the format
1473 To help solve this problem, CPP behaves specially for variable arguments
1474 used with the token paste operator, @samp{##}. If instead you write
1477 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1480 and if the variable arguments are omitted or empty, the @samp{##}
1481 operator causes the preprocessor to remove the comma before it. If you
1482 do provide some variable arguments in your macro invocation, GNU CPP
1483 does not complain about the paste operation and instead places the
1484 variable arguments after the comma. Just like any other pasted macro
1485 argument, these arguments are not macro expanded.
1487 @node Escaped Newlines
1488 @section Slightly Looser Rules for Escaped Newlines
1489 @cindex escaped newlines
1490 @cindex newlines (escaped)
1492 Recently, the preprocessor has relaxed its treatment of escaped
1493 newlines. Previously, the newline had to immediately follow a
1494 backslash. The current implementation allows whitespace in the form
1495 of spaces, horizontal and vertical tabs, and form feeds between the
1496 backslash and the subsequent newline. The preprocessor issues a
1497 warning, but treats it as a valid escaped newline and combines the two
1498 lines to form a single logical line. This works within comments and
1499 tokens, as well as between tokens. Comments are @emph{not} treated as
1500 whitespace for the purposes of this relaxation, since they have not
1501 yet been replaced with spaces.
1504 @section Non-Lvalue Arrays May Have Subscripts
1505 @cindex subscripting
1506 @cindex arrays, non-lvalue
1508 @cindex subscripting and function values
1509 In ISO C99, arrays that are not lvalues still decay to pointers, and
1510 may be subscripted, although they may not be modified or used after
1511 the next sequence point and the unary @samp{&} operator may not be
1512 applied to them. As an extension, GCC allows such arrays to be
1513 subscripted in C89 mode, though otherwise they do not decay to
1514 pointers outside C99 mode. For example,
1515 this is valid in GNU C though not valid in C89:
1519 struct foo @{int a[4];@};
1525 return f().a[index];
1531 @section Arithmetic on @code{void}- and Function-Pointers
1532 @cindex void pointers, arithmetic
1533 @cindex void, size of pointer to
1534 @cindex function pointers, arithmetic
1535 @cindex function, size of pointer to
1537 In GNU C, addition and subtraction operations are supported on pointers to
1538 @code{void} and on pointers to functions. This is done by treating the
1539 size of a @code{void} or of a function as 1.
1541 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1542 and on function types, and returns 1.
1544 @opindex Wpointer-arith
1545 The option @option{-Wpointer-arith} requests a warning if these extensions
1549 @section Non-Constant Initializers
1550 @cindex initializers, non-constant
1551 @cindex non-constant initializers
1553 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1554 automatic variable are not required to be constant expressions in GNU C@.
1555 Here is an example of an initializer with run-time varying elements:
1558 foo (float f, float g)
1560 float beat_freqs[2] = @{ f-g, f+g @};
1565 @node Compound Literals
1566 @section Compound Literals
1567 @cindex constructor expressions
1568 @cindex initializations in expressions
1569 @cindex structures, constructor expression
1570 @cindex expressions, constructor
1571 @cindex compound literals
1572 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1574 ISO C99 supports compound literals. A compound literal looks like
1575 a cast containing an initializer. Its value is an object of the
1576 type specified in the cast, containing the elements specified in
1577 the initializer; it is an lvalue. As an extension, GCC supports
1578 compound literals in C89 mode and in C++.
1580 Usually, the specified type is a structure. Assume that
1581 @code{struct foo} and @code{structure} are declared as shown:
1584 struct foo @{int a; char b[2];@} structure;
1588 Here is an example of constructing a @code{struct foo} with a compound literal:
1591 structure = ((struct foo) @{x + y, 'a', 0@});
1595 This is equivalent to writing the following:
1599 struct foo temp = @{x + y, 'a', 0@};
1604 You can also construct an array. If all the elements of the compound literal
1605 are (made up of) simple constant expressions, suitable for use in
1606 initializers of objects of static storage duration, then the compound
1607 literal can be coerced to a pointer to its first element and used in
1608 such an initializer, as shown here:
1611 char **foo = (char *[]) @{ "x", "y", "z" @};
1614 Compound literals for scalar types and union types are is
1615 also allowed, but then the compound literal is equivalent
1618 As a GNU extension, GCC allows initialization of objects with static storage
1619 duration by compound literals (which is not possible in ISO C99, because
1620 the initializer is not a constant).
1621 It is handled as if the object was initialized only with the bracket
1622 enclosed list if the types of the compound literal and the object match.
1623 The initializer list of the compound literal must be constant.
1624 If the object being initialized has array type of unknown size, the size is
1625 determined by compound literal size.
1628 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1629 static int y[] = (int []) @{1, 2, 3@};
1630 static int z[] = (int [3]) @{1@};
1634 The above lines are equivalent to the following:
1636 static struct foo x = @{1, 'a', 'b'@};
1637 static int y[] = @{1, 2, 3@};
1638 static int z[] = @{1, 0, 0@};
1641 @node Designated Inits
1642 @section Designated Initializers
1643 @cindex initializers with labeled elements
1644 @cindex labeled elements in initializers
1645 @cindex case labels in initializers
1646 @cindex designated initializers
1648 Standard C89 requires the elements of an initializer to appear in a fixed
1649 order, the same as the order of the elements in the array or structure
1652 In ISO C99 you can give the elements in any order, specifying the array
1653 indices or structure field names they apply to, and GNU C allows this as
1654 an extension in C89 mode as well. This extension is not
1655 implemented in GNU C++.
1657 To specify an array index, write
1658 @samp{[@var{index}] =} before the element value. For example,
1661 int a[6] = @{ [4] = 29, [2] = 15 @};
1668 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1672 The index values must be constant expressions, even if the array being
1673 initialized is automatic.
1675 An alternative syntax for this which has been obsolete since GCC 2.5 but
1676 GCC still accepts is to write @samp{[@var{index}]} before the element
1677 value, with no @samp{=}.
1679 To initialize a range of elements to the same value, write
1680 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1681 extension. For example,
1684 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1688 If the value in it has side-effects, the side-effects will happen only once,
1689 not for each initialized field by the range initializer.
1692 Note that the length of the array is the highest value specified
1695 In a structure initializer, specify the name of a field to initialize
1696 with @samp{.@var{fieldname} =} before the element value. For example,
1697 given the following structure,
1700 struct point @{ int x, y; @};
1704 the following initialization
1707 struct point p = @{ .y = yvalue, .x = xvalue @};
1714 struct point p = @{ xvalue, yvalue @};
1717 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1718 @samp{@var{fieldname}:}, as shown here:
1721 struct point p = @{ y: yvalue, x: xvalue @};
1725 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1726 @dfn{designator}. You can also use a designator (or the obsolete colon
1727 syntax) when initializing a union, to specify which element of the union
1728 should be used. For example,
1731 union foo @{ int i; double d; @};
1733 union foo f = @{ .d = 4 @};
1737 will convert 4 to a @code{double} to store it in the union using
1738 the second element. By contrast, casting 4 to type @code{union foo}
1739 would store it into the union as the integer @code{i}, since it is
1740 an integer. (@xref{Cast to Union}.)
1742 You can combine this technique of naming elements with ordinary C
1743 initialization of successive elements. Each initializer element that
1744 does not have a designator applies to the next consecutive element of the
1745 array or structure. For example,
1748 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1755 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1758 Labeling the elements of an array initializer is especially useful
1759 when the indices are characters or belong to an @code{enum} type.
1764 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1765 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1768 @cindex designator lists
1769 You can also write a series of @samp{.@var{fieldname}} and
1770 @samp{[@var{index}]} designators before an @samp{=} to specify a
1771 nested subobject to initialize; the list is taken relative to the
1772 subobject corresponding to the closest surrounding brace pair. For
1773 example, with the @samp{struct point} declaration above:
1776 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1780 If the same field is initialized multiple times, it will have value from
1781 the last initialization. If any such overridden initialization has
1782 side-effect, it is unspecified whether the side-effect happens or not.
1783 Currently, GCC will discard them and issue a warning.
1786 @section Case Ranges
1788 @cindex ranges in case statements
1790 You can specify a range of consecutive values in a single @code{case} label,
1794 case @var{low} ... @var{high}:
1798 This has the same effect as the proper number of individual @code{case}
1799 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1801 This feature is especially useful for ranges of ASCII character codes:
1807 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1808 it may be parsed wrong when you use it with integer values. For example,
1823 @section Cast to a Union Type
1824 @cindex cast to a union
1825 @cindex union, casting to a
1827 A cast to union type is similar to other casts, except that the type
1828 specified is a union type. You can specify the type either with
1829 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1830 a constructor though, not a cast, and hence does not yield an lvalue like
1831 normal casts. (@xref{Compound Literals}.)
1833 The types that may be cast to the union type are those of the members
1834 of the union. Thus, given the following union and variables:
1837 union foo @{ int i; double d; @};
1843 both @code{x} and @code{y} can be cast to type @code{union foo}.
1845 Using the cast as the right-hand side of an assignment to a variable of
1846 union type is equivalent to storing in a member of the union:
1851 u = (union foo) x @equiv{} u.i = x
1852 u = (union foo) y @equiv{} u.d = y
1855 You can also use the union cast as a function argument:
1858 void hack (union foo);
1860 hack ((union foo) x);
1863 @node Mixed Declarations
1864 @section Mixed Declarations and Code
1865 @cindex mixed declarations and code
1866 @cindex declarations, mixed with code
1867 @cindex code, mixed with declarations
1869 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1870 within compound statements. As an extension, GCC also allows this in
1871 C89 mode. For example, you could do:
1880 Each identifier is visible from where it is declared until the end of
1881 the enclosing block.
1883 @node Function Attributes
1884 @section Declaring Attributes of Functions
1885 @cindex function attributes
1886 @cindex declaring attributes of functions
1887 @cindex functions that never return
1888 @cindex functions that return more than once
1889 @cindex functions that have no side effects
1890 @cindex functions in arbitrary sections
1891 @cindex functions that behave like malloc
1892 @cindex @code{volatile} applied to function
1893 @cindex @code{const} applied to function
1894 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1895 @cindex functions with non-null pointer arguments
1896 @cindex functions that are passed arguments in registers on the 386
1897 @cindex functions that pop the argument stack on the 386
1898 @cindex functions that do not pop the argument stack on the 386
1899 @cindex functions that have different compilation options on the 386
1900 @cindex functions that have different optimization options
1902 In GNU C, you declare certain things about functions called in your program
1903 which help the compiler optimize function calls and check your code more
1906 The keyword @code{__attribute__} allows you to specify special
1907 attributes when making a declaration. This keyword is followed by an
1908 attribute specification inside double parentheses. The following
1909 attributes are currently defined for functions on all targets:
1910 @code{aligned}, @code{alloc_size}, @code{noreturn},
1911 @code{returns_twice}, @code{noinline}, @code{noclone},
1912 @code{always_inline}, @code{flatten}, @code{pure}, @code{const},
1913 @code{nothrow}, @code{sentinel}, @code{format}, @code{format_arg},
1914 @code{no_instrument_function}, @code{section}, @code{constructor},
1915 @code{destructor}, @code{used}, @code{unused}, @code{deprecated},
1916 @code{weak}, @code{malloc}, @code{alias}, @code{warn_unused_result},
1917 @code{nonnull}, @code{gnu_inline}, @code{externally_visible},
1918 @code{hot}, @code{cold}, @code{artificial}, @code{error} and
1919 @code{warning}. Several other attributes are defined for functions on
1920 particular target systems. Other attributes, including @code{section}
1921 are supported for variables declarations (@pxref{Variable Attributes})
1922 and for types (@pxref{Type Attributes}).
1924 You may also specify attributes with @samp{__} preceding and following
1925 each keyword. This allows you to use them in header files without
1926 being concerned about a possible macro of the same name. For example,
1927 you may use @code{__noreturn__} instead of @code{noreturn}.
1929 @xref{Attribute Syntax}, for details of the exact syntax for using
1933 @c Keep this table alphabetized by attribute name. Treat _ as space.
1935 @item alias ("@var{target}")
1936 @cindex @code{alias} attribute
1937 The @code{alias} attribute causes the declaration to be emitted as an
1938 alias for another symbol, which must be specified. For instance,
1941 void __f () @{ /* @r{Do something.} */; @}
1942 void f () __attribute__ ((weak, alias ("__f")));
1945 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
1946 mangled name for the target must be used. It is an error if @samp{__f}
1947 is not defined in the same translation unit.
1949 Not all target machines support this attribute.
1951 @item aligned (@var{alignment})
1952 @cindex @code{aligned} attribute
1953 This attribute specifies a minimum alignment for the function,
1956 You cannot use this attribute to decrease the alignment of a function,
1957 only to increase it. However, when you explicitly specify a function
1958 alignment this will override the effect of the
1959 @option{-falign-functions} (@pxref{Optimize Options}) option for this
1962 Note that the effectiveness of @code{aligned} attributes may be
1963 limited by inherent limitations in your linker. On many systems, the
1964 linker is only able to arrange for functions to be aligned up to a
1965 certain maximum alignment. (For some linkers, the maximum supported
1966 alignment may be very very small.) See your linker documentation for
1967 further information.
1969 The @code{aligned} attribute can also be used for variables and fields
1970 (@pxref{Variable Attributes}.)
1973 @cindex @code{alloc_size} attribute
1974 The @code{alloc_size} attribute is used to tell the compiler that the
1975 function return value points to memory, where the size is given by
1976 one or two of the functions parameters. GCC uses this
1977 information to improve the correctness of @code{__builtin_object_size}.
1979 The function parameter(s) denoting the allocated size are specified by
1980 one or two integer arguments supplied to the attribute. The allocated size
1981 is either the value of the single function argument specified or the product
1982 of the two function arguments specified. Argument numbering starts at
1988 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
1989 void my_realloc(void*, size_t) __attribute__((alloc_size(2)))
1992 declares that my_calloc will return memory of the size given by
1993 the product of parameter 1 and 2 and that my_realloc will return memory
1994 of the size given by parameter 2.
1997 @cindex @code{always_inline} function attribute
1998 Generally, functions are not inlined unless optimization is specified.
1999 For functions declared inline, this attribute inlines the function even
2000 if no optimization level was specified.
2003 @cindex @code{gnu_inline} function attribute
2004 This attribute should be used with a function which is also declared
2005 with the @code{inline} keyword. It directs GCC to treat the function
2006 as if it were defined in gnu89 mode even when compiling in C99 or
2009 If the function is declared @code{extern}, then this definition of the
2010 function is used only for inlining. In no case is the function
2011 compiled as a standalone function, not even if you take its address
2012 explicitly. Such an address becomes an external reference, as if you
2013 had only declared the function, and had not defined it. This has
2014 almost the effect of a macro. The way to use this is to put a
2015 function definition in a header file with this attribute, and put
2016 another copy of the function, without @code{extern}, in a library
2017 file. The definition in the header file will cause most calls to the
2018 function to be inlined. If any uses of the function remain, they will
2019 refer to the single copy in the library. Note that the two
2020 definitions of the functions need not be precisely the same, although
2021 if they do not have the same effect your program may behave oddly.
2023 In C, if the function is neither @code{extern} nor @code{static}, then
2024 the function is compiled as a standalone function, as well as being
2025 inlined where possible.
2027 This is how GCC traditionally handled functions declared
2028 @code{inline}. Since ISO C99 specifies a different semantics for
2029 @code{inline}, this function attribute is provided as a transition
2030 measure and as a useful feature in its own right. This attribute is
2031 available in GCC 4.1.3 and later. It is available if either of the
2032 preprocessor macros @code{__GNUC_GNU_INLINE__} or
2033 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
2034 Function is As Fast As a Macro}.
2036 In C++, this attribute does not depend on @code{extern} in any way,
2037 but it still requires the @code{inline} keyword to enable its special
2041 @cindex @code{artificial} function attribute
2042 This attribute is useful for small inline wrappers which if possible
2043 should appear during debugging as a unit, depending on the debug
2044 info format it will either mean marking the function as artificial
2045 or using the caller location for all instructions within the inlined
2049 @cindex interrupt handler functions
2050 When added to an interrupt handler with the M32C port, causes the
2051 prologue and epilogue to use bank switching to preserve the registers
2052 rather than saving them on the stack.
2055 @cindex @code{flatten} function attribute
2056 Generally, inlining into a function is limited. For a function marked with
2057 this attribute, every call inside this function will be inlined, if possible.
2058 Whether the function itself is considered for inlining depends on its size and
2059 the current inlining parameters.
2061 @item error ("@var{message}")
2062 @cindex @code{error} function attribute
2063 If this attribute is used on a function declaration and a call to such a function
2064 is not eliminated through dead code elimination or other optimizations, an error
2065 which will include @var{message} will be diagnosed. This is useful
2066 for compile time checking, especially together with @code{__builtin_constant_p}
2067 and inline functions where checking the inline function arguments is not
2068 possible through @code{extern char [(condition) ? 1 : -1];} tricks.
2069 While it is possible to leave the function undefined and thus invoke
2070 a link failure, when using this attribute the problem will be diagnosed
2071 earlier and with exact location of the call even in presence of inline
2072 functions or when not emitting debugging information.
2074 @item warning ("@var{message}")
2075 @cindex @code{warning} function attribute
2076 If this attribute is used on a function declaration and a call to such a function
2077 is not eliminated through dead code elimination or other optimizations, a warning
2078 which will include @var{message} will be diagnosed. This is useful
2079 for compile time checking, especially together with @code{__builtin_constant_p}
2080 and inline functions. While it is possible to define the function with
2081 a message in @code{.gnu.warning*} section, when using this attribute the problem
2082 will be diagnosed earlier and with exact location of the call even in presence
2083 of inline functions or when not emitting debugging information.
2086 @cindex functions that do pop the argument stack on the 386
2088 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2089 assume that the calling function will pop off the stack space used to
2090 pass arguments. This is
2091 useful to override the effects of the @option{-mrtd} switch.
2094 @cindex @code{const} function attribute
2095 Many functions do not examine any values except their arguments, and
2096 have no effects except the return value. Basically this is just slightly
2097 more strict class than the @code{pure} attribute below, since function is not
2098 allowed to read global memory.
2100 @cindex pointer arguments
2101 Note that a function that has pointer arguments and examines the data
2102 pointed to must @emph{not} be declared @code{const}. Likewise, a
2103 function that calls a non-@code{const} function usually must not be
2104 @code{const}. It does not make sense for a @code{const} function to
2107 The attribute @code{const} is not implemented in GCC versions earlier
2108 than 2.5. An alternative way to declare that a function has no side
2109 effects, which works in the current version and in some older versions,
2113 typedef int intfn ();
2115 extern const intfn square;
2118 This approach does not work in GNU C++ from 2.6.0 on, since the language
2119 specifies that the @samp{const} must be attached to the return value.
2123 @itemx constructor (@var{priority})
2124 @itemx destructor (@var{priority})
2125 @cindex @code{constructor} function attribute
2126 @cindex @code{destructor} function attribute
2127 The @code{constructor} attribute causes the function to be called
2128 automatically before execution enters @code{main ()}. Similarly, the
2129 @code{destructor} attribute causes the function to be called
2130 automatically after @code{main ()} has completed or @code{exit ()} has
2131 been called. Functions with these attributes are useful for
2132 initializing data that will be used implicitly during the execution of
2135 You may provide an optional integer priority to control the order in
2136 which constructor and destructor functions are run. A constructor
2137 with a smaller priority number runs before a constructor with a larger
2138 priority number; the opposite relationship holds for destructors. So,
2139 if you have a constructor that allocates a resource and a destructor
2140 that deallocates the same resource, both functions typically have the
2141 same priority. The priorities for constructor and destructor
2142 functions are the same as those specified for namespace-scope C++
2143 objects (@pxref{C++ Attributes}).
2145 These attributes are not currently implemented for Objective-C@.
2148 @itemx deprecated (@var{msg})
2149 @cindex @code{deprecated} attribute.
2150 The @code{deprecated} attribute results in a warning if the function
2151 is used anywhere in the source file. This is useful when identifying
2152 functions that are expected to be removed in a future version of a
2153 program. The warning also includes the location of the declaration
2154 of the deprecated function, to enable users to easily find further
2155 information about why the function is deprecated, or what they should
2156 do instead. Note that the warnings only occurs for uses:
2159 int old_fn () __attribute__ ((deprecated));
2161 int (*fn_ptr)() = old_fn;
2164 results in a warning on line 3 but not line 2. The optional msg
2165 argument, which must be a string, will be printed in the warning if
2168 The @code{deprecated} attribute can also be used for variables and
2169 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2172 @cindex @code{disinterrupt} attribute
2173 On MeP targets, this attribute causes the compiler to emit
2174 instructions to disable interrupts for the duration of the given
2178 @cindex @code{__declspec(dllexport)}
2179 On Microsoft Windows targets and Symbian OS targets the
2180 @code{dllexport} attribute causes the compiler to provide a global
2181 pointer to a pointer in a DLL, so that it can be referenced with the
2182 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
2183 name is formed by combining @code{_imp__} and the function or variable
2186 You can use @code{__declspec(dllexport)} as a synonym for
2187 @code{__attribute__ ((dllexport))} for compatibility with other
2190 On systems that support the @code{visibility} attribute, this
2191 attribute also implies ``default'' visibility. It is an error to
2192 explicitly specify any other visibility.
2194 Currently, the @code{dllexport} attribute is ignored for inlined
2195 functions, unless the @option{-fkeep-inline-functions} flag has been
2196 used. The attribute is also ignored for undefined symbols.
2198 When applied to C++ classes, the attribute marks defined non-inlined
2199 member functions and static data members as exports. Static consts
2200 initialized in-class are not marked unless they are also defined
2203 For Microsoft Windows targets there are alternative methods for
2204 including the symbol in the DLL's export table such as using a
2205 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
2206 the @option{--export-all} linker flag.
2209 @cindex @code{__declspec(dllimport)}
2210 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
2211 attribute causes the compiler to reference a function or variable via
2212 a global pointer to a pointer that is set up by the DLL exporting the
2213 symbol. The attribute implies @code{extern}. On Microsoft Windows
2214 targets, the pointer name is formed by combining @code{_imp__} and the
2215 function or variable name.
2217 You can use @code{__declspec(dllimport)} as a synonym for
2218 @code{__attribute__ ((dllimport))} for compatibility with other
2221 On systems that support the @code{visibility} attribute, this
2222 attribute also implies ``default'' visibility. It is an error to
2223 explicitly specify any other visibility.
2225 Currently, the attribute is ignored for inlined functions. If the
2226 attribute is applied to a symbol @emph{definition}, an error is reported.
2227 If a symbol previously declared @code{dllimport} is later defined, the
2228 attribute is ignored in subsequent references, and a warning is emitted.
2229 The attribute is also overridden by a subsequent declaration as
2232 When applied to C++ classes, the attribute marks non-inlined
2233 member functions and static data members as imports. However, the
2234 attribute is ignored for virtual methods to allow creation of vtables
2237 On the SH Symbian OS target the @code{dllimport} attribute also has
2238 another affect---it can cause the vtable and run-time type information
2239 for a class to be exported. This happens when the class has a
2240 dllimport'ed constructor or a non-inline, non-pure virtual function
2241 and, for either of those two conditions, the class also has an inline
2242 constructor or destructor and has a key function that is defined in
2243 the current translation unit.
2245 For Microsoft Windows based targets the use of the @code{dllimport}
2246 attribute on functions is not necessary, but provides a small
2247 performance benefit by eliminating a thunk in the DLL@. The use of the
2248 @code{dllimport} attribute on imported variables was required on older
2249 versions of the GNU linker, but can now be avoided by passing the
2250 @option{--enable-auto-import} switch to the GNU linker. As with
2251 functions, using the attribute for a variable eliminates a thunk in
2254 One drawback to using this attribute is that a pointer to a
2255 @emph{variable} marked as @code{dllimport} cannot be used as a constant
2256 address. However, a pointer to a @emph{function} with the
2257 @code{dllimport} attribute can be used as a constant initializer; in
2258 this case, the address of a stub function in the import lib is
2259 referenced. On Microsoft Windows targets, the attribute can be disabled
2260 for functions by setting the @option{-mnop-fun-dllimport} flag.
2263 @cindex eight bit data on the H8/300, H8/300H, and H8S
2264 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2265 variable should be placed into the eight bit data section.
2266 The compiler will generate more efficient code for certain operations
2267 on data in the eight bit data area. Note the eight bit data area is limited to
2270 You must use GAS and GLD from GNU binutils version 2.7 or later for
2271 this attribute to work correctly.
2274 @cindex exception handler functions on the RX processor
2275 Use this attribute on the RX to indicate that the specified function
2276 is an exception handler. The compiler will generate function entry and
2277 exit sequences suitable for use in an exception handler when this
2278 attribute is present.
2280 @item exception_handler
2281 @cindex exception handler functions on the Blackfin processor
2282 Use this attribute on the Blackfin to indicate that the specified function
2283 is an exception handler. The compiler will generate function entry and
2284 exit sequences suitable for use in an exception handler when this
2285 attribute is present.
2287 @item externally_visible
2288 @cindex @code{externally_visible} attribute.
2289 This attribute, attached to a global variable or function, nullifies
2290 the effect of the @option{-fwhole-program} command-line option, so the
2291 object remains visible outside the current compilation unit.
2294 @cindex functions which handle memory bank switching
2295 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
2296 use a calling convention that takes care of switching memory banks when
2297 entering and leaving a function. This calling convention is also the
2298 default when using the @option{-mlong-calls} option.
2300 On 68HC12 the compiler will use the @code{call} and @code{rtc} instructions
2301 to call and return from a function.
2303 On 68HC11 the compiler will generate a sequence of instructions
2304 to invoke a board-specific routine to switch the memory bank and call the
2305 real function. The board-specific routine simulates a @code{call}.
2306 At the end of a function, it will jump to a board-specific routine
2307 instead of using @code{rts}. The board-specific return routine simulates
2310 On MeP targets this causes the compiler to use a calling convention
2311 which assumes the called function is too far away for the built-in
2314 @item fast_interrupt
2315 @cindex interrupt handler functions
2316 Use this attribute on the M32C and RX ports to indicate that the specified
2317 function is a fast interrupt handler. This is just like the
2318 @code{interrupt} attribute, except that @code{freit} is used to return
2319 instead of @code{reit}.
2322 @cindex functions that pop the argument stack on the 386
2323 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2324 pass the first argument (if of integral type) in the register ECX and
2325 the second argument (if of integral type) in the register EDX@. Subsequent
2326 and other typed arguments are passed on the stack. The called function will
2327 pop the arguments off the stack. If the number of arguments is variable all
2328 arguments are pushed on the stack.
2330 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2331 @cindex @code{format} function attribute
2333 The @code{format} attribute specifies that a function takes @code{printf},
2334 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2335 should be type-checked against a format string. For example, the
2340 my_printf (void *my_object, const char *my_format, ...)
2341 __attribute__ ((format (printf, 2, 3)));
2345 causes the compiler to check the arguments in calls to @code{my_printf}
2346 for consistency with the @code{printf} style format string argument
2349 The parameter @var{archetype} determines how the format string is
2350 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime},
2351 @code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or
2352 @code{strfmon}. (You can also use @code{__printf__},
2353 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On
2354 MinGW targets, @code{ms_printf}, @code{ms_scanf}, and
2355 @code{ms_strftime} are also present.
2356 @var{archtype} values such as @code{printf} refer to the formats accepted
2357 by the system's C run-time library, while @code{gnu_} values always refer
2358 to the formats accepted by the GNU C Library. On Microsoft Windows
2359 targets, @code{ms_} values refer to the formats accepted by the
2360 @file{msvcrt.dll} library.
2361 The parameter @var{string-index}
2362 specifies which argument is the format string argument (starting
2363 from 1), while @var{first-to-check} is the number of the first
2364 argument to check against the format string. For functions
2365 where the arguments are not available to be checked (such as
2366 @code{vprintf}), specify the third parameter as zero. In this case the
2367 compiler only checks the format string for consistency. For
2368 @code{strftime} formats, the third parameter is required to be zero.
2369 Since non-static C++ methods have an implicit @code{this} argument, the
2370 arguments of such methods should be counted from two, not one, when
2371 giving values for @var{string-index} and @var{first-to-check}.
2373 In the example above, the format string (@code{my_format}) is the second
2374 argument of the function @code{my_print}, and the arguments to check
2375 start with the third argument, so the correct parameters for the format
2376 attribute are 2 and 3.
2378 @opindex ffreestanding
2379 @opindex fno-builtin
2380 The @code{format} attribute allows you to identify your own functions
2381 which take format strings as arguments, so that GCC can check the
2382 calls to these functions for errors. The compiler always (unless
2383 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2384 for the standard library functions @code{printf}, @code{fprintf},
2385 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2386 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2387 warnings are requested (using @option{-Wformat}), so there is no need to
2388 modify the header file @file{stdio.h}. In C99 mode, the functions
2389 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2390 @code{vsscanf} are also checked. Except in strictly conforming C
2391 standard modes, the X/Open function @code{strfmon} is also checked as
2392 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2393 @xref{C Dialect Options,,Options Controlling C Dialect}.
2395 The target may provide additional types of format checks.
2396 @xref{Target Format Checks,,Format Checks Specific to Particular
2399 @item format_arg (@var{string-index})
2400 @cindex @code{format_arg} function attribute
2401 @opindex Wformat-nonliteral
2402 The @code{format_arg} attribute specifies that a function takes a format
2403 string for a @code{printf}, @code{scanf}, @code{strftime} or
2404 @code{strfmon} style function and modifies it (for example, to translate
2405 it into another language), so the result can be passed to a
2406 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2407 function (with the remaining arguments to the format function the same
2408 as they would have been for the unmodified string). For example, the
2413 my_dgettext (char *my_domain, const char *my_format)
2414 __attribute__ ((format_arg (2)));
2418 causes the compiler to check the arguments in calls to a @code{printf},
2419 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2420 format string argument is a call to the @code{my_dgettext} function, for
2421 consistency with the format string argument @code{my_format}. If the
2422 @code{format_arg} attribute had not been specified, all the compiler
2423 could tell in such calls to format functions would be that the format
2424 string argument is not constant; this would generate a warning when
2425 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2426 without the attribute.
2428 The parameter @var{string-index} specifies which argument is the format
2429 string argument (starting from one). Since non-static C++ methods have
2430 an implicit @code{this} argument, the arguments of such methods should
2431 be counted from two.
2433 The @code{format-arg} attribute allows you to identify your own
2434 functions which modify format strings, so that GCC can check the
2435 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2436 type function whose operands are a call to one of your own function.
2437 The compiler always treats @code{gettext}, @code{dgettext}, and
2438 @code{dcgettext} in this manner except when strict ISO C support is
2439 requested by @option{-ansi} or an appropriate @option{-std} option, or
2440 @option{-ffreestanding} or @option{-fno-builtin}
2441 is used. @xref{C Dialect Options,,Options
2442 Controlling C Dialect}.
2444 @item function_vector
2445 @cindex calling functions through the function vector on H8/300, M16C, M32C and SH2A processors
2446 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2447 function should be called through the function vector. Calling a
2448 function through the function vector will reduce code size, however;
2449 the function vector has a limited size (maximum 128 entries on the H8/300
2450 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
2452 In SH2A target, this attribute declares a function to be called using the
2453 TBR relative addressing mode. The argument to this attribute is the entry
2454 number of the same function in a vector table containing all the TBR
2455 relative addressable functions. For the successful jump, register TBR
2456 should contain the start address of this TBR relative vector table.
2457 In the startup routine of the user application, user needs to care of this
2458 TBR register initialization. The TBR relative vector table can have at
2459 max 256 function entries. The jumps to these functions will be generated
2460 using a SH2A specific, non delayed branch instruction JSR/N @@(disp8,TBR).
2461 You must use GAS and GLD from GNU binutils version 2.7 or later for
2462 this attribute to work correctly.
2464 Please refer the example of M16C target, to see the use of this
2465 attribute while declaring a function,
2467 In an application, for a function being called once, this attribute will
2468 save at least 8 bytes of code; and if other successive calls are being
2469 made to the same function, it will save 2 bytes of code per each of these
2472 On M16C/M32C targets, the @code{function_vector} attribute declares a
2473 special page subroutine call function. Use of this attribute reduces
2474 the code size by 2 bytes for each call generated to the
2475 subroutine. The argument to the attribute is the vector number entry
2476 from the special page vector table which contains the 16 low-order
2477 bits of the subroutine's entry address. Each vector table has special
2478 page number (18 to 255) which are used in @code{jsrs} instruction.
2479 Jump addresses of the routines are generated by adding 0x0F0000 (in
2480 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the 2
2481 byte addresses set in the vector table. Therefore you need to ensure
2482 that all the special page vector routines should get mapped within the
2483 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
2486 In the following example 2 bytes will be saved for each call to
2487 function @code{foo}.
2490 void foo (void) __attribute__((function_vector(0x18)));
2501 If functions are defined in one file and are called in another file,
2502 then be sure to write this declaration in both files.
2504 This attribute is ignored for R8C target.
2507 @cindex interrupt handler functions
2508 Use this attribute on the ARM, AVR, CRX, M32C, M32R/D, m68k, MeP, MIPS,
2509 RX and Xstormy16 ports to indicate that the specified function is an
2510 interrupt handler. The compiler will generate function entry and exit
2511 sequences suitable for use in an interrupt handler when this attribute
2514 Note, interrupt handlers for the Blackfin, H8/300, H8/300H, H8S, and
2515 SH processors can be specified via the @code{interrupt_handler} attribute.
2517 Note, on the AVR, interrupts will be enabled inside the function.
2519 Note, for the ARM, you can specify the kind of interrupt to be handled by
2520 adding an optional parameter to the interrupt attribute like this:
2523 void f () __attribute__ ((interrupt ("IRQ")));
2526 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2528 On ARMv7-M the interrupt type is ignored, and the attribute means the function
2529 may be called with a word aligned stack pointer.
2531 On MIPS targets, you can use the following attributes to modify the behavior
2532 of an interrupt handler:
2534 @item use_shadow_register_set
2535 @cindex @code{use_shadow_register_set} attribute
2536 Assume that the handler uses a shadow register set, instead of
2537 the main general-purpose registers.
2539 @item keep_interrupts_masked
2540 @cindex @code{keep_interrupts_masked} attribute
2541 Keep interrupts masked for the whole function. Without this attribute,
2542 GCC tries to reenable interrupts for as much of the function as it can.
2544 @item use_debug_exception_return
2545 @cindex @code{use_debug_exception_return} attribute
2546 Return using the @code{deret} instruction. Interrupt handlers that don't
2547 have this attribute return using @code{eret} instead.
2550 You can use any combination of these attributes, as shown below:
2552 void __attribute__ ((interrupt)) v0 ();
2553 void __attribute__ ((interrupt, use_shadow_register_set)) v1 ();
2554 void __attribute__ ((interrupt, keep_interrupts_masked)) v2 ();
2555 void __attribute__ ((interrupt, use_debug_exception_return)) v3 ();
2556 void __attribute__ ((interrupt, use_shadow_register_set,
2557 keep_interrupts_masked)) v4 ();
2558 void __attribute__ ((interrupt, use_shadow_register_set,
2559 use_debug_exception_return)) v5 ();
2560 void __attribute__ ((interrupt, keep_interrupts_masked,
2561 use_debug_exception_return)) v6 ();
2562 void __attribute__ ((interrupt, use_shadow_register_set,
2563 keep_interrupts_masked,
2564 use_debug_exception_return)) v7 ();
2567 @item interrupt_handler
2568 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
2569 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
2570 indicate that the specified function is an interrupt handler. The compiler
2571 will generate function entry and exit sequences suitable for use in an
2572 interrupt handler when this attribute is present.
2574 @item interrupt_thread
2575 @cindex interrupt thread functions on fido
2576 Use this attribute on fido, a subarchitecture of the m68k, to indicate
2577 that the specified function is an interrupt handler that is designed
2578 to run as a thread. The compiler omits generate prologue/epilogue
2579 sequences and replaces the return instruction with a @code{sleep}
2580 instruction. This attribute is available only on fido.
2583 @cindex interrupt service routines on ARM
2584 Use this attribute on ARM to write Interrupt Service Routines. This is an
2585 alias to the @code{interrupt} attribute above.
2588 @cindex User stack pointer in interrupts on the Blackfin
2589 When used together with @code{interrupt_handler}, @code{exception_handler}
2590 or @code{nmi_handler}, code will be generated to load the stack pointer
2591 from the USP register in the function prologue.
2594 @cindex @code{l1_text} function attribute
2595 This attribute specifies a function to be placed into L1 Instruction
2596 SRAM@. The function will be put into a specific section named @code{.l1.text}.
2597 With @option{-mfdpic}, function calls with a such function as the callee
2598 or caller will use inlined PLT.
2601 @cindex @code{l2} function attribute
2602 On the Blackfin, this attribute specifies a function to be placed into L2
2603 SRAM. The function will be put into a specific section named
2604 @code{.l1.text}. With @option{-mfdpic}, callers of such functions will use
2607 @item long_call/short_call
2608 @cindex indirect calls on ARM
2609 This attribute specifies how a particular function is called on
2610 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2611 command line switch and @code{#pragma long_calls} settings. The
2612 @code{long_call} attribute indicates that the function might be far
2613 away from the call site and require a different (more expensive)
2614 calling sequence. The @code{short_call} attribute always places
2615 the offset to the function from the call site into the @samp{BL}
2616 instruction directly.
2618 @item longcall/shortcall
2619 @cindex functions called via pointer on the RS/6000 and PowerPC
2620 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute
2621 indicates that the function might be far away from the call site and
2622 require a different (more expensive) calling sequence. The
2623 @code{shortcall} attribute indicates that the function is always close
2624 enough for the shorter calling sequence to be used. These attributes
2625 override both the @option{-mlongcall} switch and, on the RS/6000 and
2626 PowerPC, the @code{#pragma longcall} setting.
2628 @xref{RS/6000 and PowerPC Options}, for more information on whether long
2629 calls are necessary.
2631 @item long_call/near/far
2632 @cindex indirect calls on MIPS
2633 These attributes specify how a particular function is called on MIPS@.
2634 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
2635 command-line switch. The @code{long_call} and @code{far} attributes are
2636 synonyms, and cause the compiler to always call
2637 the function by first loading its address into a register, and then using
2638 the contents of that register. The @code{near} attribute has the opposite
2639 effect; it specifies that non-PIC calls should be made using the more
2640 efficient @code{jal} instruction.
2643 @cindex @code{malloc} attribute
2644 The @code{malloc} attribute is used to tell the compiler that a function
2645 may be treated as if any non-@code{NULL} pointer it returns cannot
2646 alias any other pointer valid when the function returns.
2647 This will often improve optimization.
2648 Standard functions with this property include @code{malloc} and
2649 @code{calloc}. @code{realloc}-like functions have this property as
2650 long as the old pointer is never referred to (including comparing it
2651 to the new pointer) after the function returns a non-@code{NULL}
2654 @item mips16/nomips16
2655 @cindex @code{mips16} attribute
2656 @cindex @code{nomips16} attribute
2658 On MIPS targets, you can use the @code{mips16} and @code{nomips16}
2659 function attributes to locally select or turn off MIPS16 code generation.
2660 A function with the @code{mips16} attribute is emitted as MIPS16 code,
2661 while MIPS16 code generation is disabled for functions with the
2662 @code{nomips16} attribute. These attributes override the
2663 @option{-mips16} and @option{-mno-mips16} options on the command line
2664 (@pxref{MIPS Options}).
2666 When compiling files containing mixed MIPS16 and non-MIPS16 code, the
2667 preprocessor symbol @code{__mips16} reflects the setting on the command line,
2668 not that within individual functions. Mixed MIPS16 and non-MIPS16 code
2669 may interact badly with some GCC extensions such as @code{__builtin_apply}
2670 (@pxref{Constructing Calls}).
2672 @item model (@var{model-name})
2673 @cindex function addressability on the M32R/D
2674 @cindex variable addressability on the IA-64
2676 On the M32R/D, use this attribute to set the addressability of an
2677 object, and of the code generated for a function. The identifier
2678 @var{model-name} is one of @code{small}, @code{medium}, or
2679 @code{large}, representing each of the code models.
2681 Small model objects live in the lower 16MB of memory (so that their
2682 addresses can be loaded with the @code{ld24} instruction), and are
2683 callable with the @code{bl} instruction.
2685 Medium model objects may live anywhere in the 32-bit address space (the
2686 compiler will generate @code{seth/add3} instructions to load their addresses),
2687 and are callable with the @code{bl} instruction.
2689 Large model objects may live anywhere in the 32-bit address space (the
2690 compiler will generate @code{seth/add3} instructions to load their addresses),
2691 and may not be reachable with the @code{bl} instruction (the compiler will
2692 generate the much slower @code{seth/add3/jl} instruction sequence).
2694 On IA-64, use this attribute to set the addressability of an object.
2695 At present, the only supported identifier for @var{model-name} is
2696 @code{small}, indicating addressability via ``small'' (22-bit)
2697 addresses (so that their addresses can be loaded with the @code{addl}
2698 instruction). Caveat: such addressing is by definition not position
2699 independent and hence this attribute must not be used for objects
2700 defined by shared libraries.
2702 @item ms_abi/sysv_abi
2703 @cindex @code{ms_abi} attribute
2704 @cindex @code{sysv_abi} attribute
2706 On 64-bit x86_64-*-* targets, you can use an ABI attribute to indicate
2707 which calling convention should be used for a function. The @code{ms_abi}
2708 attribute tells the compiler to use the Microsoft ABI, while the
2709 @code{sysv_abi} attribute tells the compiler to use the ABI used on
2710 GNU/Linux and other systems. The default is to use the Microsoft ABI
2711 when targeting Windows. On all other systems, the default is the AMD ABI.
2713 Note, This feature is currently sorried out for Windows targets trying to
2715 @item ms_hook_prologue
2716 @cindex @code{ms_hook_prologue} attribute
2718 On 32 bit i[34567]86-*-* targets, you can use this function attribute to make
2719 gcc generate the "hot-patching" function prologue used in Win32 API
2720 functions in Microsoft Windows XP Service Pack 2 and newer. This requires
2721 support for the swap suffix in the assembler. (GNU Binutils 2.19.51 or later)
2724 @cindex function without a prologue/epilogue code
2725 Use this attribute on the ARM, AVR, IP2K, RX and SPU ports to indicate that
2726 the specified function does not need prologue/epilogue sequences generated by
2727 the compiler. It is up to the programmer to provide these sequences. The
2728 only statements that can be safely included in naked functions are
2729 @code{asm} statements that do not have operands. All other statements,
2730 including declarations of local variables, @code{if} statements, and so
2731 forth, should be avoided. Naked functions should be used to implement the
2732 body of an assembly function, while allowing the compiler to construct
2733 the requisite function declaration for the assembler.
2736 @cindex functions which do not handle memory bank switching on 68HC11/68HC12
2737 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
2738 use the normal calling convention based on @code{jsr} and @code{rts}.
2739 This attribute can be used to cancel the effect of the @option{-mlong-calls}
2742 On MeP targets this attribute causes the compiler to assume the called
2743 function is close enough to use the normal calling convention,
2744 overriding the @code{-mtf} command line option.
2747 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
2748 Use this attribute together with @code{interrupt_handler},
2749 @code{exception_handler} or @code{nmi_handler} to indicate that the function
2750 entry code should enable nested interrupts or exceptions.
2753 @cindex NMI handler functions on the Blackfin processor
2754 Use this attribute on the Blackfin to indicate that the specified function
2755 is an NMI handler. The compiler will generate function entry and
2756 exit sequences suitable for use in an NMI handler when this
2757 attribute is present.
2759 @item no_instrument_function
2760 @cindex @code{no_instrument_function} function attribute
2761 @opindex finstrument-functions
2762 If @option{-finstrument-functions} is given, profiling function calls will
2763 be generated at entry and exit of most user-compiled functions.
2764 Functions with this attribute will not be so instrumented.
2767 @cindex @code{noinline} function attribute
2768 This function attribute prevents a function from being considered for
2770 @c Don't enumerate the optimizations by name here; we try to be
2771 @c future-compatible with this mechanism.
2772 If the function does not have side-effects, there are optimizations
2773 other than inlining that causes function calls to be optimized away,
2774 although the function call is live. To keep such calls from being
2779 (@pxref{Extended Asm}) in the called function, to serve as a special
2783 @cindex @code{noclone} function attribute
2784 This function attribute prevents a function from being considered for
2785 cloning - a mechanism which produces specialized copies of functions
2786 and which is (currently) performed by interprocedural constant
2789 @item nonnull (@var{arg-index}, @dots{})
2790 @cindex @code{nonnull} function attribute
2791 The @code{nonnull} attribute specifies that some function parameters should
2792 be non-null pointers. For instance, the declaration:
2796 my_memcpy (void *dest, const void *src, size_t len)
2797 __attribute__((nonnull (1, 2)));
2801 causes the compiler to check that, in calls to @code{my_memcpy},
2802 arguments @var{dest} and @var{src} are non-null. If the compiler
2803 determines that a null pointer is passed in an argument slot marked
2804 as non-null, and the @option{-Wnonnull} option is enabled, a warning
2805 is issued. The compiler may also choose to make optimizations based
2806 on the knowledge that certain function arguments will not be null.
2808 If no argument index list is given to the @code{nonnull} attribute,
2809 all pointer arguments are marked as non-null. To illustrate, the
2810 following declaration is equivalent to the previous example:
2814 my_memcpy (void *dest, const void *src, size_t len)
2815 __attribute__((nonnull));
2819 @cindex @code{noreturn} function attribute
2820 A few standard library functions, such as @code{abort} and @code{exit},
2821 cannot return. GCC knows this automatically. Some programs define
2822 their own functions that never return. You can declare them
2823 @code{noreturn} to tell the compiler this fact. For example,
2827 void fatal () __attribute__ ((noreturn));
2830 fatal (/* @r{@dots{}} */)
2832 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
2838 The @code{noreturn} keyword tells the compiler to assume that
2839 @code{fatal} cannot return. It can then optimize without regard to what
2840 would happen if @code{fatal} ever did return. This makes slightly
2841 better code. More importantly, it helps avoid spurious warnings of
2842 uninitialized variables.
2844 The @code{noreturn} keyword does not affect the exceptional path when that
2845 applies: a @code{noreturn}-marked function may still return to the caller
2846 by throwing an exception or calling @code{longjmp}.
2848 Do not assume that registers saved by the calling function are
2849 restored before calling the @code{noreturn} function.
2851 It does not make sense for a @code{noreturn} function to have a return
2852 type other than @code{void}.
2854 The attribute @code{noreturn} is not implemented in GCC versions
2855 earlier than 2.5. An alternative way to declare that a function does
2856 not return, which works in the current version and in some older
2857 versions, is as follows:
2860 typedef void voidfn ();
2862 volatile voidfn fatal;
2865 This approach does not work in GNU C++.
2868 @cindex @code{nothrow} function attribute
2869 The @code{nothrow} attribute is used to inform the compiler that a
2870 function cannot throw an exception. For example, most functions in
2871 the standard C library can be guaranteed not to throw an exception
2872 with the notable exceptions of @code{qsort} and @code{bsearch} that
2873 take function pointer arguments. The @code{nothrow} attribute is not
2874 implemented in GCC versions earlier than 3.3.
2877 @cindex @code{optimize} function attribute
2878 The @code{optimize} attribute is used to specify that a function is to
2879 be compiled with different optimization options than specified on the
2880 command line. Arguments can either be numbers or strings. Numbers
2881 are assumed to be an optimization level. Strings that begin with
2882 @code{O} are assumed to be an optimization option, while other options
2883 are assumed to be used with a @code{-f} prefix. You can also use the
2884 @samp{#pragma GCC optimize} pragma to set the optimization options
2885 that affect more than one function.
2886 @xref{Function Specific Option Pragmas}, for details about the
2887 @samp{#pragma GCC optimize} pragma.
2889 This can be used for instance to have frequently executed functions
2890 compiled with more aggressive optimization options that produce faster
2891 and larger code, while other functions can be called with less
2895 @cindex @code{pcs} function attribute
2897 The @code{pcs} attribute can be used to control the calling convention
2898 used for a function on ARM. The attribute takes an argument that specifies
2899 the calling convention to use.
2901 When compiling using the AAPCS ABI (or a variant of that) then valid
2902 values for the argument are @code{"aapcs"} and @code{"aapcs-vfp"}. In
2903 order to use a variant other than @code{"aapcs"} then the compiler must
2904 be permitted to use the appropriate co-processor registers (i.e., the
2905 VFP registers must be available in order to use @code{"aapcs-vfp"}).
2909 /* Argument passed in r0, and result returned in r0+r1. */
2910 double f2d (float) __attribute__((pcs("aapcs")));
2913 Variadic functions always use the @code{"aapcs"} calling convention and
2914 the compiler will reject attempts to specify an alternative.
2917 @cindex @code{pure} function attribute
2918 Many functions have no effects except the return value and their
2919 return value depends only on the parameters and/or global variables.
2920 Such a function can be subject
2921 to common subexpression elimination and loop optimization just as an
2922 arithmetic operator would be. These functions should be declared
2923 with the attribute @code{pure}. For example,
2926 int square (int) __attribute__ ((pure));
2930 says that the hypothetical function @code{square} is safe to call
2931 fewer times than the program says.
2933 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2934 Interesting non-pure functions are functions with infinite loops or those
2935 depending on volatile memory or other system resource, that may change between
2936 two consecutive calls (such as @code{feof} in a multithreading environment).
2938 The attribute @code{pure} is not implemented in GCC versions earlier
2942 @cindex @code{hot} function attribute
2943 The @code{hot} attribute is used to inform the compiler that a function is a
2944 hot spot of the compiled program. The function is optimized more aggressively
2945 and on many target it is placed into special subsection of the text section so
2946 all hot functions appears close together improving locality.
2948 When profile feedback is available, via @option{-fprofile-use}, hot functions
2949 are automatically detected and this attribute is ignored.
2951 The @code{hot} attribute is not implemented in GCC versions earlier
2955 @cindex @code{cold} function attribute
2956 The @code{cold} attribute is used to inform the compiler that a function is
2957 unlikely executed. The function is optimized for size rather than speed and on
2958 many targets it is placed into special subsection of the text section so all
2959 cold functions appears close together improving code locality of non-cold parts
2960 of program. The paths leading to call of cold functions within code are marked
2961 as unlikely by the branch prediction mechanism. It is thus useful to mark
2962 functions used to handle unlikely conditions, such as @code{perror}, as cold to
2963 improve optimization of hot functions that do call marked functions in rare
2966 When profile feedback is available, via @option{-fprofile-use}, hot functions
2967 are automatically detected and this attribute is ignored.
2969 The @code{cold} attribute is not implemented in GCC versions earlier than 4.3.
2971 @item regparm (@var{number})
2972 @cindex @code{regparm} attribute
2973 @cindex functions that are passed arguments in registers on the 386
2974 On the Intel 386, the @code{regparm} attribute causes the compiler to
2975 pass arguments number one to @var{number} if they are of integral type
2976 in registers EAX, EDX, and ECX instead of on the stack. Functions that
2977 take a variable number of arguments will continue to be passed all of their
2978 arguments on the stack.
2980 Beware that on some ELF systems this attribute is unsuitable for
2981 global functions in shared libraries with lazy binding (which is the
2982 default). Lazy binding will send the first call via resolving code in
2983 the loader, which might assume EAX, EDX and ECX can be clobbered, as
2984 per the standard calling conventions. Solaris 8 is affected by this.
2985 GNU systems with GLIBC 2.1 or higher, and FreeBSD, are believed to be
2986 safe since the loaders there save EAX, EDX and ECX. (Lazy binding can be
2987 disabled with the linker or the loader if desired, to avoid the
2991 @cindex @code{sseregparm} attribute
2992 On the Intel 386 with SSE support, the @code{sseregparm} attribute
2993 causes the compiler to pass up to 3 floating point arguments in
2994 SSE registers instead of on the stack. Functions that take a
2995 variable number of arguments will continue to pass all of their
2996 floating point arguments on the stack.
2998 @item force_align_arg_pointer
2999 @cindex @code{force_align_arg_pointer} attribute
3000 On the Intel x86, the @code{force_align_arg_pointer} attribute may be
3001 applied to individual function definitions, generating an alternate
3002 prologue and epilogue that realigns the runtime stack if necessary.
3003 This supports mixing legacy codes that run with a 4-byte aligned stack
3004 with modern codes that keep a 16-byte stack for SSE compatibility.
3007 @cindex @code{resbank} attribute
3008 On the SH2A target, this attribute enables the high-speed register
3009 saving and restoration using a register bank for @code{interrupt_handler}
3010 routines. Saving to the bank is performed automatically after the CPU
3011 accepts an interrupt that uses a register bank.
3013 The nineteen 32-bit registers comprising general register R0 to R14,
3014 control register GBR, and system registers MACH, MACL, and PR and the
3015 vector table address offset are saved into a register bank. Register
3016 banks are stacked in first-in last-out (FILO) sequence. Restoration
3017 from the bank is executed by issuing a RESBANK instruction.
3020 @cindex @code{returns_twice} attribute
3021 The @code{returns_twice} attribute tells the compiler that a function may
3022 return more than one time. The compiler will ensure that all registers
3023 are dead before calling such a function and will emit a warning about
3024 the variables that may be clobbered after the second return from the
3025 function. Examples of such functions are @code{setjmp} and @code{vfork}.
3026 The @code{longjmp}-like counterpart of such function, if any, might need
3027 to be marked with the @code{noreturn} attribute.
3030 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
3031 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
3032 all registers except the stack pointer should be saved in the prologue
3033 regardless of whether they are used or not.
3035 @item section ("@var{section-name}")
3036 @cindex @code{section} function attribute
3037 Normally, the compiler places the code it generates in the @code{text} section.
3038 Sometimes, however, you need additional sections, or you need certain
3039 particular functions to appear in special sections. The @code{section}
3040 attribute specifies that a function lives in a particular section.
3041 For example, the declaration:
3044 extern void foobar (void) __attribute__ ((section ("bar")));
3048 puts the function @code{foobar} in the @code{bar} section.
3050 Some file formats do not support arbitrary sections so the @code{section}
3051 attribute is not available on all platforms.
3052 If you need to map the entire contents of a module to a particular
3053 section, consider using the facilities of the linker instead.
3056 @cindex @code{sentinel} function attribute
3057 This function attribute ensures that a parameter in a function call is
3058 an explicit @code{NULL}. The attribute is only valid on variadic
3059 functions. By default, the sentinel is located at position zero, the
3060 last parameter of the function call. If an optional integer position
3061 argument P is supplied to the attribute, the sentinel must be located at
3062 position P counting backwards from the end of the argument list.
3065 __attribute__ ((sentinel))
3067 __attribute__ ((sentinel(0)))
3070 The attribute is automatically set with a position of 0 for the built-in
3071 functions @code{execl} and @code{execlp}. The built-in function
3072 @code{execle} has the attribute set with a position of 1.
3074 A valid @code{NULL} in this context is defined as zero with any pointer
3075 type. If your system defines the @code{NULL} macro with an integer type
3076 then you need to add an explicit cast. GCC replaces @code{stddef.h}
3077 with a copy that redefines NULL appropriately.
3079 The warnings for missing or incorrect sentinels are enabled with
3083 See long_call/short_call.
3086 See longcall/shortcall.
3089 @cindex signal handler functions on the AVR processors
3090 Use this attribute on the AVR to indicate that the specified
3091 function is a signal handler. The compiler will generate function
3092 entry and exit sequences suitable for use in a signal handler when this
3093 attribute is present. Interrupts will be disabled inside the function.
3096 Use this attribute on the SH to indicate an @code{interrupt_handler}
3097 function should switch to an alternate stack. It expects a string
3098 argument that names a global variable holding the address of the
3103 void f () __attribute__ ((interrupt_handler,
3104 sp_switch ("alt_stack")));
3108 @cindex functions that pop the argument stack on the 386
3109 On the Intel 386, the @code{stdcall} attribute causes the compiler to
3110 assume that the called function will pop off the stack space used to
3111 pass arguments, unless it takes a variable number of arguments.
3113 @item syscall_linkage
3114 @cindex @code{syscall_linkage} attribute
3115 This attribute is used to modify the IA64 calling convention by marking
3116 all input registers as live at all function exits. This makes it possible
3117 to restart a system call after an interrupt without having to save/restore
3118 the input registers. This also prevents kernel data from leaking into
3122 @cindex @code{target} function attribute
3123 The @code{target} attribute is used to specify that a function is to
3124 be compiled with different target options than specified on the
3125 command line. This can be used for instance to have functions
3126 compiled with a different ISA (instruction set architecture) than the
3127 default. You can also use the @samp{#pragma GCC target} pragma to set
3128 more than one function to be compiled with specific target options.
3129 @xref{Function Specific Option Pragmas}, for details about the
3130 @samp{#pragma GCC target} pragma.
3132 For instance on a 386, you could compile one function with
3133 @code{target("sse4.1,arch=core2")} and another with
3134 @code{target("sse4a,arch=amdfam10")} that would be equivalent to
3135 compiling the first function with @option{-msse4.1} and
3136 @option{-march=core2} options, and the second function with
3137 @option{-msse4a} and @option{-march=amdfam10} options. It is up to the
3138 user to make sure that a function is only invoked on a machine that
3139 supports the particular ISA it was compiled for (for example by using
3140 @code{cpuid} on 386 to determine what feature bits and architecture
3144 int core2_func (void) __attribute__ ((__target__ ("arch=core2")));
3145 int sse3_func (void) __attribute__ ((__target__ ("sse3")));
3148 On the 386, the following options are allowed:
3153 @cindex @code{target("abm")} attribute
3154 Enable/disable the generation of the advanced bit instructions.
3158 @cindex @code{target("aes")} attribute
3159 Enable/disable the generation of the AES instructions.
3163 @cindex @code{target("mmx")} attribute
3164 Enable/disable the generation of the MMX instructions.
3168 @cindex @code{target("pclmul")} attribute
3169 Enable/disable the generation of the PCLMUL instructions.
3173 @cindex @code{target("popcnt")} attribute
3174 Enable/disable the generation of the POPCNT instruction.
3178 @cindex @code{target("sse")} attribute
3179 Enable/disable the generation of the SSE instructions.
3183 @cindex @code{target("sse2")} attribute
3184 Enable/disable the generation of the SSE2 instructions.
3188 @cindex @code{target("sse3")} attribute
3189 Enable/disable the generation of the SSE3 instructions.
3193 @cindex @code{target("sse4")} attribute
3194 Enable/disable the generation of the SSE4 instructions (both SSE4.1
3199 @cindex @code{target("sse4.1")} attribute
3200 Enable/disable the generation of the sse4.1 instructions.
3204 @cindex @code{target("sse4.2")} attribute
3205 Enable/disable the generation of the sse4.2 instructions.
3209 @cindex @code{target("sse4a")} attribute
3210 Enable/disable the generation of the SSE4A instructions.
3214 @cindex @code{target("fma4")} attribute
3215 Enable/disable the generation of the FMA4 instructions.
3219 @cindex @code{target("ssse3")} attribute
3220 Enable/disable the generation of the SSSE3 instructions.
3224 @cindex @code{target("cld")} attribute
3225 Enable/disable the generation of the CLD before string moves.
3227 @item fancy-math-387
3228 @itemx no-fancy-math-387
3229 @cindex @code{target("fancy-math-387")} attribute
3230 Enable/disable the generation of the @code{sin}, @code{cos}, and
3231 @code{sqrt} instructions on the 387 floating point unit.
3234 @itemx no-fused-madd
3235 @cindex @code{target("fused-madd")} attribute
3236 Enable/disable the generation of the fused multiply/add instructions.
3240 @cindex @code{target("ieee-fp")} attribute
3241 Enable/disable the generation of floating point that depends on IEEE arithmetic.
3243 @item inline-all-stringops
3244 @itemx no-inline-all-stringops
3245 @cindex @code{target("inline-all-stringops")} attribute
3246 Enable/disable inlining of string operations.
3248 @item inline-stringops-dynamically
3249 @itemx no-inline-stringops-dynamically
3250 @cindex @code{target("inline-stringops-dynamically")} attribute
3251 Enable/disable the generation of the inline code to do small string
3252 operations and calling the library routines for large operations.
3254 @item align-stringops
3255 @itemx no-align-stringops
3256 @cindex @code{target("align-stringops")} attribute
3257 Do/do not align destination of inlined string operations.
3261 @cindex @code{target("recip")} attribute
3262 Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS
3263 instructions followed an additional Newton-Raphson step instead of
3264 doing a floating point division.
3266 @item arch=@var{ARCH}
3267 @cindex @code{target("arch=@var{ARCH}")} attribute
3268 Specify the architecture to generate code for in compiling the function.
3270 @item tune=@var{TUNE}
3271 @cindex @code{target("tune=@var{TUNE}")} attribute
3272 Specify the architecture to tune for in compiling the function.
3274 @item fpmath=@var{FPMATH}
3275 @cindex @code{target("fpmath=@var{FPMATH}")} attribute
3276 Specify which floating point unit to use. The
3277 @code{target("fpmath=sse,387")} option must be specified as
3278 @code{target("fpmath=sse+387")} because the comma would separate
3282 On the 386, you can use either multiple strings to specify multiple
3283 options, or you can separate the option with a comma (@code{,}).
3285 On the 386, the inliner will not inline a function that has different
3286 target options than the caller, unless the callee has a subset of the
3287 target options of the caller. For example a function declared with
3288 @code{target("sse3")} can inline a function with
3289 @code{target("sse2")}, since @code{-msse3} implies @code{-msse2}.
3291 The @code{target} attribute is not implemented in GCC versions earlier
3292 than 4.4, and at present only the 386 uses it.
3295 @cindex tiny data section on the H8/300H and H8S
3296 Use this attribute on the H8/300H and H8S to indicate that the specified
3297 variable should be placed into the tiny data section.
3298 The compiler will generate more efficient code for loads and stores
3299 on data in the tiny data section. Note the tiny data area is limited to
3300 slightly under 32kbytes of data.
3303 Use this attribute on the SH for an @code{interrupt_handler} to return using
3304 @code{trapa} instead of @code{rte}. This attribute expects an integer
3305 argument specifying the trap number to be used.
3308 @cindex @code{unused} attribute.
3309 This attribute, attached to a function, means that the function is meant
3310 to be possibly unused. GCC will not produce a warning for this
3314 @cindex @code{used} attribute.
3315 This attribute, attached to a function, means that code must be emitted
3316 for the function even if it appears that the function is not referenced.
3317 This is useful, for example, when the function is referenced only in
3321 @cindex @code{version_id} attribute
3322 This IA64 HP-UX attribute, attached to a global variable or function, renames a
3323 symbol to contain a version string, thus allowing for function level
3324 versioning. HP-UX system header files may use version level functioning
3325 for some system calls.
3328 extern int foo () __attribute__((version_id ("20040821")));
3331 Calls to @var{foo} will be mapped to calls to @var{foo@{20040821@}}.
3333 @item visibility ("@var{visibility_type}")
3334 @cindex @code{visibility} attribute
3335 This attribute affects the linkage of the declaration to which it is attached.
3336 There are four supported @var{visibility_type} values: default,
3337 hidden, protected or internal visibility.
3340 void __attribute__ ((visibility ("protected")))
3341 f () @{ /* @r{Do something.} */; @}
3342 int i __attribute__ ((visibility ("hidden")));
3345 The possible values of @var{visibility_type} correspond to the
3346 visibility settings in the ELF gABI.
3349 @c keep this list of visibilities in alphabetical order.
3352 Default visibility is the normal case for the object file format.
3353 This value is available for the visibility attribute to override other
3354 options that may change the assumed visibility of entities.
3356 On ELF, default visibility means that the declaration is visible to other
3357 modules and, in shared libraries, means that the declared entity may be
3360 On Darwin, default visibility means that the declaration is visible to
3363 Default visibility corresponds to ``external linkage'' in the language.
3366 Hidden visibility indicates that the entity declared will have a new
3367 form of linkage, which we'll call ``hidden linkage''. Two
3368 declarations of an object with hidden linkage refer to the same object
3369 if they are in the same shared object.
3372 Internal visibility is like hidden visibility, but with additional
3373 processor specific semantics. Unless otherwise specified by the
3374 psABI, GCC defines internal visibility to mean that a function is
3375 @emph{never} called from another module. Compare this with hidden
3376 functions which, while they cannot be referenced directly by other
3377 modules, can be referenced indirectly via function pointers. By
3378 indicating that a function cannot be called from outside the module,
3379 GCC may for instance omit the load of a PIC register since it is known
3380 that the calling function loaded the correct value.
3383 Protected visibility is like default visibility except that it
3384 indicates that references within the defining module will bind to the
3385 definition in that module. That is, the declared entity cannot be
3386 overridden by another module.
3390 All visibilities are supported on many, but not all, ELF targets
3391 (supported when the assembler supports the @samp{.visibility}
3392 pseudo-op). Default visibility is supported everywhere. Hidden
3393 visibility is supported on Darwin targets.
3395 The visibility attribute should be applied only to declarations which
3396 would otherwise have external linkage. The attribute should be applied
3397 consistently, so that the same entity should not be declared with
3398 different settings of the attribute.
3400 In C++, the visibility attribute applies to types as well as functions
3401 and objects, because in C++ types have linkage. A class must not have
3402 greater visibility than its non-static data member types and bases,
3403 and class members default to the visibility of their class. Also, a
3404 declaration without explicit visibility is limited to the visibility
3407 In C++, you can mark member functions and static member variables of a
3408 class with the visibility attribute. This is useful if you know a
3409 particular method or static member variable should only be used from
3410 one shared object; then you can mark it hidden while the rest of the
3411 class has default visibility. Care must be taken to avoid breaking
3412 the One Definition Rule; for example, it is usually not useful to mark
3413 an inline method as hidden without marking the whole class as hidden.
3415 A C++ namespace declaration can also have the visibility attribute.
3416 This attribute applies only to the particular namespace body, not to
3417 other definitions of the same namespace; it is equivalent to using
3418 @samp{#pragma GCC visibility} before and after the namespace
3419 definition (@pxref{Visibility Pragmas}).
3421 In C++, if a template argument has limited visibility, this
3422 restriction is implicitly propagated to the template instantiation.
3423 Otherwise, template instantiations and specializations default to the
3424 visibility of their template.
3426 If both the template and enclosing class have explicit visibility, the
3427 visibility from the template is used.
3430 @cindex @code{vliw} attribute
3431 On MeP, the @code{vliw} attribute tells the compiler to emit
3432 instructions in VLIW mode instead of core mode. Note that this
3433 attribute is not allowed unless a VLIW coprocessor has been configured
3434 and enabled through command line options.
3436 @item warn_unused_result
3437 @cindex @code{warn_unused_result} attribute
3438 The @code{warn_unused_result} attribute causes a warning to be emitted
3439 if a caller of the function with this attribute does not use its
3440 return value. This is useful for functions where not checking
3441 the result is either a security problem or always a bug, such as
3445 int fn () __attribute__ ((warn_unused_result));
3448 if (fn () < 0) return -1;
3454 results in warning on line 5.
3457 @cindex @code{weak} attribute
3458 The @code{weak} attribute causes the declaration to be emitted as a weak
3459 symbol rather than a global. This is primarily useful in defining
3460 library functions which can be overridden in user code, though it can
3461 also be used with non-function declarations. Weak symbols are supported
3462 for ELF targets, and also for a.out targets when using the GNU assembler
3466 @itemx weakref ("@var{target}")
3467 @cindex @code{weakref} attribute
3468 The @code{weakref} attribute marks a declaration as a weak reference.
3469 Without arguments, it should be accompanied by an @code{alias} attribute
3470 naming the target symbol. Optionally, the @var{target} may be given as
3471 an argument to @code{weakref} itself. In either case, @code{weakref}
3472 implicitly marks the declaration as @code{weak}. Without a
3473 @var{target}, given as an argument to @code{weakref} or to @code{alias},
3474 @code{weakref} is equivalent to @code{weak}.
3477 static int x() __attribute__ ((weakref ("y")));
3478 /* is equivalent to... */
3479 static int x() __attribute__ ((weak, weakref, alias ("y")));
3481 static int x() __attribute__ ((weakref));
3482 static int x() __attribute__ ((alias ("y")));
3485 A weak reference is an alias that does not by itself require a
3486 definition to be given for the target symbol. If the target symbol is
3487 only referenced through weak references, then the becomes a @code{weak}
3488 undefined symbol. If it is directly referenced, however, then such
3489 strong references prevail, and a definition will be required for the
3490 symbol, not necessarily in the same translation unit.
3492 The effect is equivalent to moving all references to the alias to a
3493 separate translation unit, renaming the alias to the aliased symbol,
3494 declaring it as weak, compiling the two separate translation units and
3495 performing a reloadable link on them.
3497 At present, a declaration to which @code{weakref} is attached can
3498 only be @code{static}.
3502 You can specify multiple attributes in a declaration by separating them
3503 by commas within the double parentheses or by immediately following an
3504 attribute declaration with another attribute declaration.
3506 @cindex @code{#pragma}, reason for not using
3507 @cindex pragma, reason for not using
3508 Some people object to the @code{__attribute__} feature, suggesting that
3509 ISO C's @code{#pragma} should be used instead. At the time
3510 @code{__attribute__} was designed, there were two reasons for not doing
3515 It is impossible to generate @code{#pragma} commands from a macro.
3518 There is no telling what the same @code{#pragma} might mean in another
3522 These two reasons applied to almost any application that might have been
3523 proposed for @code{#pragma}. It was basically a mistake to use
3524 @code{#pragma} for @emph{anything}.
3526 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
3527 to be generated from macros. In addition, a @code{#pragma GCC}
3528 namespace is now in use for GCC-specific pragmas. However, it has been
3529 found convenient to use @code{__attribute__} to achieve a natural
3530 attachment of attributes to their corresponding declarations, whereas
3531 @code{#pragma GCC} is of use for constructs that do not naturally form
3532 part of the grammar. @xref{Other Directives,,Miscellaneous
3533 Preprocessing Directives, cpp, The GNU C Preprocessor}.
3535 @node Attribute Syntax
3536 @section Attribute Syntax
3537 @cindex attribute syntax
3539 This section describes the syntax with which @code{__attribute__} may be
3540 used, and the constructs to which attribute specifiers bind, for the C
3541 language. Some details may vary for C++ and Objective-C@. Because of
3542 infelicities in the grammar for attributes, some forms described here
3543 may not be successfully parsed in all cases.
3545 There are some problems with the semantics of attributes in C++. For
3546 example, there are no manglings for attributes, although they may affect
3547 code generation, so problems may arise when attributed types are used in
3548 conjunction with templates or overloading. Similarly, @code{typeid}
3549 does not distinguish between types with different attributes. Support
3550 for attributes in C++ may be restricted in future to attributes on
3551 declarations only, but not on nested declarators.
3553 @xref{Function Attributes}, for details of the semantics of attributes
3554 applying to functions. @xref{Variable Attributes}, for details of the
3555 semantics of attributes applying to variables. @xref{Type Attributes},
3556 for details of the semantics of attributes applying to structure, union
3557 and enumerated types.
3559 An @dfn{attribute specifier} is of the form
3560 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
3561 is a possibly empty comma-separated sequence of @dfn{attributes}, where
3562 each attribute is one of the following:
3566 Empty. Empty attributes are ignored.
3569 A word (which may be an identifier such as @code{unused}, or a reserved
3570 word such as @code{const}).
3573 A word, followed by, in parentheses, parameters for the attribute.
3574 These parameters take one of the following forms:
3578 An identifier. For example, @code{mode} attributes use this form.
3581 An identifier followed by a comma and a non-empty comma-separated list
3582 of expressions. For example, @code{format} attributes use this form.
3585 A possibly empty comma-separated list of expressions. For example,
3586 @code{format_arg} attributes use this form with the list being a single
3587 integer constant expression, and @code{alias} attributes use this form
3588 with the list being a single string constant.
3592 An @dfn{attribute specifier list} is a sequence of one or more attribute
3593 specifiers, not separated by any other tokens.
3595 In GNU C, an attribute specifier list may appear after the colon following a
3596 label, other than a @code{case} or @code{default} label. The only
3597 attribute it makes sense to use after a label is @code{unused}. This
3598 feature is intended for code generated by programs which contains labels
3599 that may be unused but which is compiled with @option{-Wall}. It would
3600 not normally be appropriate to use in it human-written code, though it
3601 could be useful in cases where the code that jumps to the label is
3602 contained within an @code{#ifdef} conditional. GNU C++ only permits
3603 attributes on labels if the attribute specifier is immediately
3604 followed by a semicolon (i.e., the label applies to an empty
3605 statement). If the semicolon is missing, C++ label attributes are
3606 ambiguous, as it is permissible for a declaration, which could begin
3607 with an attribute list, to be labelled in C++. Declarations cannot be
3608 labelled in C90 or C99, so the ambiguity does not arise there.
3610 An attribute specifier list may appear as part of a @code{struct},
3611 @code{union} or @code{enum} specifier. It may go either immediately
3612 after the @code{struct}, @code{union} or @code{enum} keyword, or after
3613 the closing brace. The former syntax is preferred.
3614 Where attribute specifiers follow the closing brace, they are considered
3615 to relate to the structure, union or enumerated type defined, not to any
3616 enclosing declaration the type specifier appears in, and the type
3617 defined is not complete until after the attribute specifiers.
3618 @c Otherwise, there would be the following problems: a shift/reduce
3619 @c conflict between attributes binding the struct/union/enum and
3620 @c binding to the list of specifiers/qualifiers; and "aligned"
3621 @c attributes could use sizeof for the structure, but the size could be
3622 @c changed later by "packed" attributes.
3624 Otherwise, an attribute specifier appears as part of a declaration,
3625 counting declarations of unnamed parameters and type names, and relates
3626 to that declaration (which may be nested in another declaration, for
3627 example in the case of a parameter declaration), or to a particular declarator
3628 within a declaration. Where an
3629 attribute specifier is applied to a parameter declared as a function or
3630 an array, it should apply to the function or array rather than the
3631 pointer to which the parameter is implicitly converted, but this is not
3632 yet correctly implemented.
3634 Any list of specifiers and qualifiers at the start of a declaration may
3635 contain attribute specifiers, whether or not such a list may in that
3636 context contain storage class specifiers. (Some attributes, however,
3637 are essentially in the nature of storage class specifiers, and only make
3638 sense where storage class specifiers may be used; for example,
3639 @code{section}.) There is one necessary limitation to this syntax: the
3640 first old-style parameter declaration in a function definition cannot
3641 begin with an attribute specifier, because such an attribute applies to
3642 the function instead by syntax described below (which, however, is not
3643 yet implemented in this case). In some other cases, attribute
3644 specifiers are permitted by this grammar but not yet supported by the
3645 compiler. All attribute specifiers in this place relate to the
3646 declaration as a whole. In the obsolescent usage where a type of
3647 @code{int} is implied by the absence of type specifiers, such a list of
3648 specifiers and qualifiers may be an attribute specifier list with no
3649 other specifiers or qualifiers.
3651 At present, the first parameter in a function prototype must have some
3652 type specifier which is not an attribute specifier; this resolves an
3653 ambiguity in the interpretation of @code{void f(int
3654 (__attribute__((foo)) x))}, but is subject to change. At present, if
3655 the parentheses of a function declarator contain only attributes then
3656 those attributes are ignored, rather than yielding an error or warning
3657 or implying a single parameter of type int, but this is subject to
3660 An attribute specifier list may appear immediately before a declarator
3661 (other than the first) in a comma-separated list of declarators in a
3662 declaration of more than one identifier using a single list of
3663 specifiers and qualifiers. Such attribute specifiers apply
3664 only to the identifier before whose declarator they appear. For
3668 __attribute__((noreturn)) void d0 (void),
3669 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
3674 the @code{noreturn} attribute applies to all the functions
3675 declared; the @code{format} attribute only applies to @code{d1}.
3677 An attribute specifier list may appear immediately before the comma,
3678 @code{=} or semicolon terminating the declaration of an identifier other
3679 than a function definition. Such attribute specifiers apply
3680 to the declared object or function. Where an
3681 assembler name for an object or function is specified (@pxref{Asm
3682 Labels}), the attribute must follow the @code{asm}
3685 An attribute specifier list may, in future, be permitted to appear after
3686 the declarator in a function definition (before any old-style parameter
3687 declarations or the function body).
3689 Attribute specifiers may be mixed with type qualifiers appearing inside
3690 the @code{[]} of a parameter array declarator, in the C99 construct by
3691 which such qualifiers are applied to the pointer to which the array is
3692 implicitly converted. Such attribute specifiers apply to the pointer,
3693 not to the array, but at present this is not implemented and they are
3696 An attribute specifier list may appear at the start of a nested
3697 declarator. At present, there are some limitations in this usage: the
3698 attributes correctly apply to the declarator, but for most individual
3699 attributes the semantics this implies are not implemented.
3700 When attribute specifiers follow the @code{*} of a pointer
3701 declarator, they may be mixed with any type qualifiers present.
3702 The following describes the formal semantics of this syntax. It will make the
3703 most sense if you are familiar with the formal specification of
3704 declarators in the ISO C standard.
3706 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
3707 D1}, where @code{T} contains declaration specifiers that specify a type
3708 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
3709 contains an identifier @var{ident}. The type specified for @var{ident}
3710 for derived declarators whose type does not include an attribute
3711 specifier is as in the ISO C standard.
3713 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
3714 and the declaration @code{T D} specifies the type
3715 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
3716 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
3717 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
3719 If @code{D1} has the form @code{*
3720 @var{type-qualifier-and-attribute-specifier-list} D}, and the
3721 declaration @code{T D} specifies the type
3722 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
3723 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
3724 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
3730 void (__attribute__((noreturn)) ****f) (void);
3734 specifies the type ``pointer to pointer to pointer to pointer to
3735 non-returning function returning @code{void}''. As another example,
3738 char *__attribute__((aligned(8))) *f;
3742 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
3743 Note again that this does not work with most attributes; for example,
3744 the usage of @samp{aligned} and @samp{noreturn} attributes given above
3745 is not yet supported.
3747 For compatibility with existing code written for compiler versions that
3748 did not implement attributes on nested declarators, some laxity is
3749 allowed in the placing of attributes. If an attribute that only applies
3750 to types is applied to a declaration, it will be treated as applying to
3751 the type of that declaration. If an attribute that only applies to
3752 declarations is applied to the type of a declaration, it will be treated
3753 as applying to that declaration; and, for compatibility with code
3754 placing the attributes immediately before the identifier declared, such
3755 an attribute applied to a function return type will be treated as
3756 applying to the function type, and such an attribute applied to an array
3757 element type will be treated as applying to the array type. If an
3758 attribute that only applies to function types is applied to a
3759 pointer-to-function type, it will be treated as applying to the pointer
3760 target type; if such an attribute is applied to a function return type
3761 that is not a pointer-to-function type, it will be treated as applying
3762 to the function type.
3764 @node Function Prototypes
3765 @section Prototypes and Old-Style Function Definitions
3766 @cindex function prototype declarations
3767 @cindex old-style function definitions
3768 @cindex promotion of formal parameters
3770 GNU C extends ISO C to allow a function prototype to override a later
3771 old-style non-prototype definition. Consider the following example:
3774 /* @r{Use prototypes unless the compiler is old-fashioned.} */
3781 /* @r{Prototype function declaration.} */
3782 int isroot P((uid_t));
3784 /* @r{Old-style function definition.} */
3786 isroot (x) /* @r{??? lossage here ???} */
3793 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
3794 not allow this example, because subword arguments in old-style
3795 non-prototype definitions are promoted. Therefore in this example the
3796 function definition's argument is really an @code{int}, which does not
3797 match the prototype argument type of @code{short}.
3799 This restriction of ISO C makes it hard to write code that is portable
3800 to traditional C compilers, because the programmer does not know
3801 whether the @code{uid_t} type is @code{short}, @code{int}, or
3802 @code{long}. Therefore, in cases like these GNU C allows a prototype
3803 to override a later old-style definition. More precisely, in GNU C, a
3804 function prototype argument type overrides the argument type specified
3805 by a later old-style definition if the former type is the same as the
3806 latter type before promotion. Thus in GNU C the above example is
3807 equivalent to the following:
3820 GNU C++ does not support old-style function definitions, so this
3821 extension is irrelevant.
3824 @section C++ Style Comments
3826 @cindex C++ comments
3827 @cindex comments, C++ style
3829 In GNU C, you may use C++ style comments, which start with @samp{//} and
3830 continue until the end of the line. Many other C implementations allow
3831 such comments, and they are included in the 1999 C standard. However,
3832 C++ style comments are not recognized if you specify an @option{-std}
3833 option specifying a version of ISO C before C99, or @option{-ansi}
3834 (equivalent to @option{-std=c89}).
3837 @section Dollar Signs in Identifier Names
3839 @cindex dollar signs in identifier names
3840 @cindex identifier names, dollar signs in
3842 In GNU C, you may normally use dollar signs in identifier names.
3843 This is because many traditional C implementations allow such identifiers.
3844 However, dollar signs in identifiers are not supported on a few target
3845 machines, typically because the target assembler does not allow them.
3847 @node Character Escapes
3848 @section The Character @key{ESC} in Constants
3850 You can use the sequence @samp{\e} in a string or character constant to
3851 stand for the ASCII character @key{ESC}.
3854 @section Inquiring on Alignment of Types or Variables
3856 @cindex type alignment
3857 @cindex variable alignment
3859 The keyword @code{__alignof__} allows you to inquire about how an object
3860 is aligned, or the minimum alignment usually required by a type. Its
3861 syntax is just like @code{sizeof}.
3863 For example, if the target machine requires a @code{double} value to be
3864 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
3865 This is true on many RISC machines. On more traditional machine
3866 designs, @code{__alignof__ (double)} is 4 or even 2.
3868 Some machines never actually require alignment; they allow reference to any
3869 data type even at an odd address. For these machines, @code{__alignof__}
3870 reports the smallest alignment that GCC will give the data type, usually as
3871 mandated by the target ABI.
3873 If the operand of @code{__alignof__} is an lvalue rather than a type,
3874 its value is the required alignment for its type, taking into account
3875 any minimum alignment specified with GCC's @code{__attribute__}
3876 extension (@pxref{Variable Attributes}). For example, after this
3880 struct foo @{ int x; char y; @} foo1;
3884 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
3885 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
3887 It is an error to ask for the alignment of an incomplete type.
3889 @node Variable Attributes
3890 @section Specifying Attributes of Variables
3891 @cindex attribute of variables
3892 @cindex variable attributes
3894 The keyword @code{__attribute__} allows you to specify special
3895 attributes of variables or structure fields. This keyword is followed
3896 by an attribute specification inside double parentheses. Some
3897 attributes are currently defined generically for variables.
3898 Other attributes are defined for variables on particular target
3899 systems. Other attributes are available for functions
3900 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
3901 Other front ends might define more attributes
3902 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
3904 You may also specify attributes with @samp{__} preceding and following
3905 each keyword. This allows you to use them in header files without
3906 being concerned about a possible macro of the same name. For example,
3907 you may use @code{__aligned__} instead of @code{aligned}.
3909 @xref{Attribute Syntax}, for details of the exact syntax for using
3913 @cindex @code{aligned} attribute
3914 @item aligned (@var{alignment})
3915 This attribute specifies a minimum alignment for the variable or
3916 structure field, measured in bytes. For example, the declaration:
3919 int x __attribute__ ((aligned (16))) = 0;
3923 causes the compiler to allocate the global variable @code{x} on a
3924 16-byte boundary. On a 68040, this could be used in conjunction with
3925 an @code{asm} expression to access the @code{move16} instruction which
3926 requires 16-byte aligned operands.
3928 You can also specify the alignment of structure fields. For example, to
3929 create a double-word aligned @code{int} pair, you could write:
3932 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
3936 This is an alternative to creating a union with a @code{double} member
3937 that forces the union to be double-word aligned.
3939 As in the preceding examples, you can explicitly specify the alignment
3940 (in bytes) that you wish the compiler to use for a given variable or
3941 structure field. Alternatively, you can leave out the alignment factor
3942 and just ask the compiler to align a variable or field to the
3943 default alignment for the target architecture you are compiling for.
3944 The default alignment is sufficient for all scalar types, but may not be
3945 enough for all vector types on a target which supports vector operations.
3946 The default alignment is fixed for a particular target ABI.
3948 Gcc also provides a target specific macro @code{__BIGGEST_ALIGNMENT__},
3949 which is the largest alignment ever used for any data type on the
3950 target machine you are compiling for. For example, you could write:
3953 short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__)));
3956 The compiler automatically sets the alignment for the declared
3957 variable or field to @code{__BIGGEST_ALIGNMENT__}. Doing this can
3958 often make copy operations more efficient, because the compiler can
3959 use whatever instructions copy the biggest chunks of memory when
3960 performing copies to or from the variables or fields that you have
3961 aligned this way. Note that the value of @code{__BIGGEST_ALIGNMENT__}
3962 may change depending on command line options.
3964 When used on a struct, or struct member, the @code{aligned} attribute can
3965 only increase the alignment; in order to decrease it, the @code{packed}
3966 attribute must be specified as well. When used as part of a typedef, the
3967 @code{aligned} attribute can both increase and decrease alignment, and
3968 specifying the @code{packed} attribute will generate a warning.
3970 Note that the effectiveness of @code{aligned} attributes may be limited
3971 by inherent limitations in your linker. On many systems, the linker is
3972 only able to arrange for variables to be aligned up to a certain maximum
3973 alignment. (For some linkers, the maximum supported alignment may
3974 be very very small.) If your linker is only able to align variables
3975 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3976 in an @code{__attribute__} will still only provide you with 8 byte
3977 alignment. See your linker documentation for further information.
3979 The @code{aligned} attribute can also be used for functions
3980 (@pxref{Function Attributes}.)
3982 @item cleanup (@var{cleanup_function})
3983 @cindex @code{cleanup} attribute
3984 The @code{cleanup} attribute runs a function when the variable goes
3985 out of scope. This attribute can only be applied to auto function
3986 scope variables; it may not be applied to parameters or variables
3987 with static storage duration. The function must take one parameter,
3988 a pointer to a type compatible with the variable. The return value
3989 of the function (if any) is ignored.
3991 If @option{-fexceptions} is enabled, then @var{cleanup_function}
3992 will be run during the stack unwinding that happens during the
3993 processing of the exception. Note that the @code{cleanup} attribute
3994 does not allow the exception to be caught, only to perform an action.
3995 It is undefined what happens if @var{cleanup_function} does not
4000 @cindex @code{common} attribute
4001 @cindex @code{nocommon} attribute
4004 The @code{common} attribute requests GCC to place a variable in
4005 ``common'' storage. The @code{nocommon} attribute requests the
4006 opposite---to allocate space for it directly.
4008 These attributes override the default chosen by the
4009 @option{-fno-common} and @option{-fcommon} flags respectively.
4012 @itemx deprecated (@var{msg})
4013 @cindex @code{deprecated} attribute
4014 The @code{deprecated} attribute results in a warning if the variable
4015 is used anywhere in the source file. This is useful when identifying
4016 variables that are expected to be removed in a future version of a
4017 program. The warning also includes the location of the declaration
4018 of the deprecated variable, to enable users to easily find further
4019 information about why the variable is deprecated, or what they should
4020 do instead. Note that the warning only occurs for uses:
4023 extern int old_var __attribute__ ((deprecated));
4025 int new_fn () @{ return old_var; @}
4028 results in a warning on line 3 but not line 2. The optional msg
4029 argument, which must be a string, will be printed in the warning if
4032 The @code{deprecated} attribute can also be used for functions and
4033 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
4035 @item mode (@var{mode})
4036 @cindex @code{mode} attribute
4037 This attribute specifies the data type for the declaration---whichever
4038 type corresponds to the mode @var{mode}. This in effect lets you
4039 request an integer or floating point type according to its width.
4041 You may also specify a mode of @samp{byte} or @samp{__byte__} to
4042 indicate the mode corresponding to a one-byte integer, @samp{word} or
4043 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
4044 or @samp{__pointer__} for the mode used to represent pointers.
4047 @cindex @code{packed} attribute
4048 The @code{packed} attribute specifies that a variable or structure field
4049 should have the smallest possible alignment---one byte for a variable,
4050 and one bit for a field, unless you specify a larger value with the
4051 @code{aligned} attribute.
4053 Here is a structure in which the field @code{x} is packed, so that it
4054 immediately follows @code{a}:
4060 int x[2] __attribute__ ((packed));
4064 @emph{Note:} The 4.1, 4.2 and 4.3 series of GCC ignore the
4065 @code{packed} attribute on bit-fields of type @code{char}. This has
4066 been fixed in GCC 4.4 but the change can lead to differences in the
4067 structure layout. See the documentation of
4068 @option{-Wpacked-bitfield-compat} for more information.
4070 @item section ("@var{section-name}")
4071 @cindex @code{section} variable attribute
4072 Normally, the compiler places the objects it generates in sections like
4073 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
4074 or you need certain particular variables to appear in special sections,
4075 for example to map to special hardware. The @code{section}
4076 attribute specifies that a variable (or function) lives in a particular
4077 section. For example, this small program uses several specific section names:
4080 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
4081 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
4082 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
4083 int init_data __attribute__ ((section ("INITDATA")));
4087 /* @r{Initialize stack pointer} */
4088 init_sp (stack + sizeof (stack));
4090 /* @r{Initialize initialized data} */
4091 memcpy (&init_data, &data, &edata - &data);
4093 /* @r{Turn on the serial ports} */
4100 Use the @code{section} attribute with
4101 @emph{global} variables and not @emph{local} variables,
4102 as shown in the example.
4104 You may use the @code{section} attribute with initialized or
4105 uninitialized global variables but the linker requires
4106 each object be defined once, with the exception that uninitialized
4107 variables tentatively go in the @code{common} (or @code{bss}) section
4108 and can be multiply ``defined''. Using the @code{section} attribute
4109 will change what section the variable goes into and may cause the
4110 linker to issue an error if an uninitialized variable has multiple
4111 definitions. You can force a variable to be initialized with the
4112 @option{-fno-common} flag or the @code{nocommon} attribute.
4114 Some file formats do not support arbitrary sections so the @code{section}
4115 attribute is not available on all platforms.
4116 If you need to map the entire contents of a module to a particular
4117 section, consider using the facilities of the linker instead.
4120 @cindex @code{shared} variable attribute
4121 On Microsoft Windows, in addition to putting variable definitions in a named
4122 section, the section can also be shared among all running copies of an
4123 executable or DLL@. For example, this small program defines shared data
4124 by putting it in a named section @code{shared} and marking the section
4128 int foo __attribute__((section ("shared"), shared)) = 0;
4133 /* @r{Read and write foo. All running
4134 copies see the same value.} */
4140 You may only use the @code{shared} attribute along with @code{section}
4141 attribute with a fully initialized global definition because of the way
4142 linkers work. See @code{section} attribute for more information.
4144 The @code{shared} attribute is only available on Microsoft Windows@.
4146 @item tls_model ("@var{tls_model}")
4147 @cindex @code{tls_model} attribute
4148 The @code{tls_model} attribute sets thread-local storage model
4149 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
4150 overriding @option{-ftls-model=} command line switch on a per-variable
4152 The @var{tls_model} argument should be one of @code{global-dynamic},
4153 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
4155 Not all targets support this attribute.
4158 This attribute, attached to a variable, means that the variable is meant
4159 to be possibly unused. GCC will not produce a warning for this
4163 This attribute, attached to a variable, means that the variable must be
4164 emitted even if it appears that the variable is not referenced.
4166 @item vector_size (@var{bytes})
4167 This attribute specifies the vector size for the variable, measured in
4168 bytes. For example, the declaration:
4171 int foo __attribute__ ((vector_size (16)));
4175 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
4176 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
4177 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
4179 This attribute is only applicable to integral and float scalars,
4180 although arrays, pointers, and function return values are allowed in
4181 conjunction with this construct.
4183 Aggregates with this attribute are invalid, even if they are of the same
4184 size as a corresponding scalar. For example, the declaration:
4187 struct S @{ int a; @};
4188 struct S __attribute__ ((vector_size (16))) foo;
4192 is invalid even if the size of the structure is the same as the size of
4196 The @code{selectany} attribute causes an initialized global variable to
4197 have link-once semantics. When multiple definitions of the variable are
4198 encountered by the linker, the first is selected and the remainder are
4199 discarded. Following usage by the Microsoft compiler, the linker is told
4200 @emph{not} to warn about size or content differences of the multiple
4203 Although the primary usage of this attribute is for POD types, the
4204 attribute can also be applied to global C++ objects that are initialized
4205 by a constructor. In this case, the static initialization and destruction
4206 code for the object is emitted in each translation defining the object,
4207 but the calls to the constructor and destructor are protected by a
4208 link-once guard variable.
4210 The @code{selectany} attribute is only available on Microsoft Windows
4211 targets. You can use @code{__declspec (selectany)} as a synonym for
4212 @code{__attribute__ ((selectany))} for compatibility with other
4216 The @code{weak} attribute is described in @ref{Function Attributes}.
4219 The @code{dllimport} attribute is described in @ref{Function Attributes}.
4222 The @code{dllexport} attribute is described in @ref{Function Attributes}.
4226 @subsection Blackfin Variable Attributes
4228 Three attributes are currently defined for the Blackfin.
4234 @cindex @code{l1_data} variable attribute
4235 @cindex @code{l1_data_A} variable attribute
4236 @cindex @code{l1_data_B} variable attribute
4237 Use these attributes on the Blackfin to place the variable into L1 Data SRAM.
4238 Variables with @code{l1_data} attribute will be put into the specific section
4239 named @code{.l1.data}. Those with @code{l1_data_A} attribute will be put into
4240 the specific section named @code{.l1.data.A}. Those with @code{l1_data_B}
4241 attribute will be put into the specific section named @code{.l1.data.B}.
4244 @cindex @code{l2} variable attribute
4245 Use this attribute on the Blackfin to place the variable into L2 SRAM.
4246 Variables with @code{l2} attribute will be put into the specific section
4247 named @code{.l2.data}.
4250 @subsection M32R/D Variable Attributes
4252 One attribute is currently defined for the M32R/D@.
4255 @item model (@var{model-name})
4256 @cindex variable addressability on the M32R/D
4257 Use this attribute on the M32R/D to set the addressability of an object.
4258 The identifier @var{model-name} is one of @code{small}, @code{medium},
4259 or @code{large}, representing each of the code models.
4261 Small model objects live in the lower 16MB of memory (so that their
4262 addresses can be loaded with the @code{ld24} instruction).
4264 Medium and large model objects may live anywhere in the 32-bit address space
4265 (the compiler will generate @code{seth/add3} instructions to load their
4269 @anchor{MeP Variable Attributes}
4270 @subsection MeP Variable Attributes
4272 The MeP target has a number of addressing modes and busses. The
4273 @code{near} space spans the standard memory space's first 16 megabytes
4274 (24 bits). The @code{far} space spans the entire 32-bit memory space.
4275 The @code{based} space is a 128 byte region in the memory space which
4276 is addressed relative to the @code{$tp} register. The @code{tiny}
4277 space is a 65536 byte region relative to the @code{$gp} register. In
4278 addition to these memory regions, the MeP target has a separate 16-bit
4279 control bus which is specified with @code{cb} attributes.
4284 Any variable with the @code{based} attribute will be assigned to the
4285 @code{.based} section, and will be accessed with relative to the
4286 @code{$tp} register.
4289 Likewise, the @code{tiny} attribute assigned variables to the
4290 @code{.tiny} section, relative to the @code{$gp} register.
4293 Variables with the @code{near} attribute are assumed to have addresses
4294 that fit in a 24-bit addressing mode. This is the default for large
4295 variables (@code{-mtiny=4} is the default) but this attribute can
4296 override @code{-mtiny=} for small variables, or override @code{-ml}.
4299 Variables with the @code{far} attribute are addressed using a full
4300 32-bit address. Since this covers the entire memory space, this
4301 allows modules to make no assumptions about where variables might be
4305 @item io (@var{addr})
4306 Variables with the @code{io} attribute are used to address
4307 memory-mapped peripherals. If an address is specified, the variable
4308 is assigned that address, else it is not assigned an address (it is
4309 assumed some other module will assign an address). Example:
4312 int timer_count __attribute__((io(0x123)));
4316 @item cb (@var{addr})
4317 Variables with the @code{cb} attribute are used to access the control
4318 bus, using special instructions. @code{addr} indicates the control bus
4322 int cpu_clock __attribute__((cb(0x123)));
4327 @anchor{i386 Variable Attributes}
4328 @subsection i386 Variable Attributes
4330 Two attributes are currently defined for i386 configurations:
4331 @code{ms_struct} and @code{gcc_struct}
4336 @cindex @code{ms_struct} attribute
4337 @cindex @code{gcc_struct} attribute
4339 If @code{packed} is used on a structure, or if bit-fields are used
4340 it may be that the Microsoft ABI packs them differently
4341 than GCC would normally pack them. Particularly when moving packed
4342 data between functions compiled with GCC and the native Microsoft compiler
4343 (either via function call or as data in a file), it may be necessary to access
4346 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
4347 compilers to match the native Microsoft compiler.
4349 The Microsoft structure layout algorithm is fairly simple with the exception
4350 of the bitfield packing:
4352 The padding and alignment of members of structures and whether a bit field
4353 can straddle a storage-unit boundary
4356 @item Structure members are stored sequentially in the order in which they are
4357 declared: the first member has the lowest memory address and the last member
4360 @item Every data object has an alignment-requirement. The alignment-requirement
4361 for all data except structures, unions, and arrays is either the size of the
4362 object or the current packing size (specified with either the aligned attribute
4363 or the pack pragma), whichever is less. For structures, unions, and arrays,
4364 the alignment-requirement is the largest alignment-requirement of its members.
4365 Every object is allocated an offset so that:
4367 offset % alignment-requirement == 0
4369 @item Adjacent bit fields are packed into the same 1-, 2-, or 4-byte allocation
4370 unit if the integral types are the same size and if the next bit field fits
4371 into the current allocation unit without crossing the boundary imposed by the
4372 common alignment requirements of the bit fields.
4375 Handling of zero-length bitfields:
4377 MSVC interprets zero-length bitfields in the following ways:
4380 @item If a zero-length bitfield is inserted between two bitfields that would
4381 normally be coalesced, the bitfields will not be coalesced.
4388 unsigned long bf_1 : 12;
4390 unsigned long bf_2 : 12;
4394 The size of @code{t1} would be 8 bytes with the zero-length bitfield. If the
4395 zero-length bitfield were removed, @code{t1}'s size would be 4 bytes.
4397 @item If a zero-length bitfield is inserted after a bitfield, @code{foo}, and the
4398 alignment of the zero-length bitfield is greater than the member that follows it,
4399 @code{bar}, @code{bar} will be aligned as the type of the zero-length bitfield.
4419 For @code{t2}, @code{bar} will be placed at offset 2, rather than offset 1.
4420 Accordingly, the size of @code{t2} will be 4. For @code{t3}, the zero-length
4421 bitfield will not affect the alignment of @code{bar} or, as a result, the size
4424 Taking this into account, it is important to note the following:
4427 @item If a zero-length bitfield follows a normal bitfield, the type of the
4428 zero-length bitfield may affect the alignment of the structure as whole. For
4429 example, @code{t2} has a size of 4 bytes, since the zero-length bitfield follows a
4430 normal bitfield, and is of type short.
4432 @item Even if a zero-length bitfield is not followed by a normal bitfield, it may
4433 still affect the alignment of the structure:
4443 Here, @code{t4} will take up 4 bytes.
4446 @item Zero-length bitfields following non-bitfield members are ignored:
4457 Here, @code{t5} will take up 2 bytes.
4461 @subsection PowerPC Variable Attributes
4463 Three attributes currently are defined for PowerPC configurations:
4464 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
4466 For full documentation of the struct attributes please see the
4467 documentation in @ref{i386 Variable Attributes}.
4469 For documentation of @code{altivec} attribute please see the
4470 documentation in @ref{PowerPC Type Attributes}.
4472 @subsection SPU Variable Attributes
4474 The SPU supports the @code{spu_vector} attribute for variables. For
4475 documentation of this attribute please see the documentation in
4476 @ref{SPU Type Attributes}.
4478 @subsection Xstormy16 Variable Attributes
4480 One attribute is currently defined for xstormy16 configurations:
4485 @cindex @code{below100} attribute
4487 If a variable has the @code{below100} attribute (@code{BELOW100} is
4488 allowed also), GCC will place the variable in the first 0x100 bytes of
4489 memory and use special opcodes to access it. Such variables will be
4490 placed in either the @code{.bss_below100} section or the
4491 @code{.data_below100} section.
4495 @subsection AVR Variable Attributes
4499 @cindex @code{progmem} variable attribute
4500 The @code{progmem} attribute is used on the AVR to place data in the Program
4501 Memory address space. The AVR is a Harvard Architecture processor and data
4502 normally resides in the Data Memory address space.
4505 @node Type Attributes
4506 @section Specifying Attributes of Types
4507 @cindex attribute of types
4508 @cindex type attributes
4510 The keyword @code{__attribute__} allows you to specify special
4511 attributes of @code{struct} and @code{union} types when you define
4512 such types. This keyword is followed by an attribute specification
4513 inside double parentheses. Seven attributes are currently defined for
4514 types: @code{aligned}, @code{packed}, @code{transparent_union},
4515 @code{unused}, @code{deprecated}, @code{visibility}, and
4516 @code{may_alias}. Other attributes are defined for functions
4517 (@pxref{Function Attributes}) and for variables (@pxref{Variable
4520 You may also specify any one of these attributes with @samp{__}
4521 preceding and following its keyword. This allows you to use these
4522 attributes in header files without being concerned about a possible
4523 macro of the same name. For example, you may use @code{__aligned__}
4524 instead of @code{aligned}.
4526 You may specify type attributes in an enum, struct or union type
4527 declaration or definition, or for other types in a @code{typedef}
4530 For an enum, struct or union type, you may specify attributes either
4531 between the enum, struct or union tag and the name of the type, or
4532 just past the closing curly brace of the @emph{definition}. The
4533 former syntax is preferred.
4535 @xref{Attribute Syntax}, for details of the exact syntax for using
4539 @cindex @code{aligned} attribute
4540 @item aligned (@var{alignment})
4541 This attribute specifies a minimum alignment (in bytes) for variables
4542 of the specified type. For example, the declarations:
4545 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
4546 typedef int more_aligned_int __attribute__ ((aligned (8)));
4550 force the compiler to insure (as far as it can) that each variable whose
4551 type is @code{struct S} or @code{more_aligned_int} will be allocated and
4552 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
4553 variables of type @code{struct S} aligned to 8-byte boundaries allows
4554 the compiler to use the @code{ldd} and @code{std} (doubleword load and
4555 store) instructions when copying one variable of type @code{struct S} to
4556 another, thus improving run-time efficiency.
4558 Note that the alignment of any given @code{struct} or @code{union} type
4559 is required by the ISO C standard to be at least a perfect multiple of
4560 the lowest common multiple of the alignments of all of the members of
4561 the @code{struct} or @code{union} in question. This means that you @emph{can}
4562 effectively adjust the alignment of a @code{struct} or @code{union}
4563 type by attaching an @code{aligned} attribute to any one of the members
4564 of such a type, but the notation illustrated in the example above is a
4565 more obvious, intuitive, and readable way to request the compiler to
4566 adjust the alignment of an entire @code{struct} or @code{union} type.
4568 As in the preceding example, you can explicitly specify the alignment
4569 (in bytes) that you wish the compiler to use for a given @code{struct}
4570 or @code{union} type. Alternatively, you can leave out the alignment factor
4571 and just ask the compiler to align a type to the maximum
4572 useful alignment for the target machine you are compiling for. For
4573 example, you could write:
4576 struct S @{ short f[3]; @} __attribute__ ((aligned));
4579 Whenever you leave out the alignment factor in an @code{aligned}
4580 attribute specification, the compiler automatically sets the alignment
4581 for the type to the largest alignment which is ever used for any data
4582 type on the target machine you are compiling for. Doing this can often
4583 make copy operations more efficient, because the compiler can use
4584 whatever instructions copy the biggest chunks of memory when performing
4585 copies to or from the variables which have types that you have aligned
4588 In the example above, if the size of each @code{short} is 2 bytes, then
4589 the size of the entire @code{struct S} type is 6 bytes. The smallest
4590 power of two which is greater than or equal to that is 8, so the
4591 compiler sets the alignment for the entire @code{struct S} type to 8
4594 Note that although you can ask the compiler to select a time-efficient
4595 alignment for a given type and then declare only individual stand-alone
4596 objects of that type, the compiler's ability to select a time-efficient
4597 alignment is primarily useful only when you plan to create arrays of
4598 variables having the relevant (efficiently aligned) type. If you
4599 declare or use arrays of variables of an efficiently-aligned type, then
4600 it is likely that your program will also be doing pointer arithmetic (or
4601 subscripting, which amounts to the same thing) on pointers to the
4602 relevant type, and the code that the compiler generates for these
4603 pointer arithmetic operations will often be more efficient for
4604 efficiently-aligned types than for other types.
4606 The @code{aligned} attribute can only increase the alignment; but you
4607 can decrease it by specifying @code{packed} as well. See below.
4609 Note that the effectiveness of @code{aligned} attributes may be limited
4610 by inherent limitations in your linker. On many systems, the linker is
4611 only able to arrange for variables to be aligned up to a certain maximum
4612 alignment. (For some linkers, the maximum supported alignment may
4613 be very very small.) If your linker is only able to align variables
4614 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
4615 in an @code{__attribute__} will still only provide you with 8 byte
4616 alignment. See your linker documentation for further information.
4619 This attribute, attached to @code{struct} or @code{union} type
4620 definition, specifies that each member (other than zero-width bitfields)
4621 of the structure or union is placed to minimize the memory required. When
4622 attached to an @code{enum} definition, it indicates that the smallest
4623 integral type should be used.
4625 @opindex fshort-enums
4626 Specifying this attribute for @code{struct} and @code{union} types is
4627 equivalent to specifying the @code{packed} attribute on each of the
4628 structure or union members. Specifying the @option{-fshort-enums}
4629 flag on the line is equivalent to specifying the @code{packed}
4630 attribute on all @code{enum} definitions.
4632 In the following example @code{struct my_packed_struct}'s members are
4633 packed closely together, but the internal layout of its @code{s} member
4634 is not packed---to do that, @code{struct my_unpacked_struct} would need to
4638 struct my_unpacked_struct
4644 struct __attribute__ ((__packed__)) my_packed_struct
4648 struct my_unpacked_struct s;
4652 You may only specify this attribute on the definition of an @code{enum},
4653 @code{struct} or @code{union}, not on a @code{typedef} which does not
4654 also define the enumerated type, structure or union.
4656 @item transparent_union
4657 This attribute, attached to a @code{union} type definition, indicates
4658 that any function parameter having that union type causes calls to that
4659 function to be treated in a special way.
4661 First, the argument corresponding to a transparent union type can be of
4662 any type in the union; no cast is required. Also, if the union contains
4663 a pointer type, the corresponding argument can be a null pointer
4664 constant or a void pointer expression; and if the union contains a void
4665 pointer type, the corresponding argument can be any pointer expression.
4666 If the union member type is a pointer, qualifiers like @code{const} on
4667 the referenced type must be respected, just as with normal pointer
4670 Second, the argument is passed to the function using the calling
4671 conventions of the first member of the transparent union, not the calling
4672 conventions of the union itself. All members of the union must have the
4673 same machine representation; this is necessary for this argument passing
4676 Transparent unions are designed for library functions that have multiple
4677 interfaces for compatibility reasons. For example, suppose the
4678 @code{wait} function must accept either a value of type @code{int *} to
4679 comply with Posix, or a value of type @code{union wait *} to comply with
4680 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
4681 @code{wait} would accept both kinds of arguments, but it would also
4682 accept any other pointer type and this would make argument type checking
4683 less useful. Instead, @code{<sys/wait.h>} might define the interface
4687 typedef union __attribute__ ((__transparent_union__))
4691 @} wait_status_ptr_t;
4693 pid_t wait (wait_status_ptr_t);
4696 This interface allows either @code{int *} or @code{union wait *}
4697 arguments to be passed, using the @code{int *} calling convention.
4698 The program can call @code{wait} with arguments of either type:
4701 int w1 () @{ int w; return wait (&w); @}
4702 int w2 () @{ union wait w; return wait (&w); @}
4705 With this interface, @code{wait}'s implementation might look like this:
4708 pid_t wait (wait_status_ptr_t p)
4710 return waitpid (-1, p.__ip, 0);
4715 When attached to a type (including a @code{union} or a @code{struct}),
4716 this attribute means that variables of that type are meant to appear
4717 possibly unused. GCC will not produce a warning for any variables of
4718 that type, even if the variable appears to do nothing. This is often
4719 the case with lock or thread classes, which are usually defined and then
4720 not referenced, but contain constructors and destructors that have
4721 nontrivial bookkeeping functions.
4724 @itemx deprecated (@var{msg})
4725 The @code{deprecated} attribute results in a warning if the type
4726 is used anywhere in the source file. This is useful when identifying
4727 types that are expected to be removed in a future version of a program.
4728 If possible, the warning also includes the location of the declaration
4729 of the deprecated type, to enable users to easily find further
4730 information about why the type is deprecated, or what they should do
4731 instead. Note that the warnings only occur for uses and then only
4732 if the type is being applied to an identifier that itself is not being
4733 declared as deprecated.
4736 typedef int T1 __attribute__ ((deprecated));
4740 typedef T1 T3 __attribute__ ((deprecated));
4741 T3 z __attribute__ ((deprecated));
4744 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
4745 warning is issued for line 4 because T2 is not explicitly
4746 deprecated. Line 5 has no warning because T3 is explicitly
4747 deprecated. Similarly for line 6. The optional msg
4748 argument, which must be a string, will be printed in the warning if
4751 The @code{deprecated} attribute can also be used for functions and
4752 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
4755 Accesses through pointers to types with this attribute are not subject
4756 to type-based alias analysis, but are instead assumed to be able to alias
4757 any other type of objects. In the context of 6.5/7 an lvalue expression
4758 dereferencing such a pointer is treated like having a character type.
4759 See @option{-fstrict-aliasing} for more information on aliasing issues.
4760 This extension exists to support some vector APIs, in which pointers to
4761 one vector type are permitted to alias pointers to a different vector type.
4763 Note that an object of a type with this attribute does not have any
4769 typedef short __attribute__((__may_alias__)) short_a;
4775 short_a *b = (short_a *) &a;
4779 if (a == 0x12345678)
4786 If you replaced @code{short_a} with @code{short} in the variable
4787 declaration, the above program would abort when compiled with
4788 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
4789 above in recent GCC versions.
4792 In C++, attribute visibility (@pxref{Function Attributes}) can also be
4793 applied to class, struct, union and enum types. Unlike other type
4794 attributes, the attribute must appear between the initial keyword and
4795 the name of the type; it cannot appear after the body of the type.
4797 Note that the type visibility is applied to vague linkage entities
4798 associated with the class (vtable, typeinfo node, etc.). In
4799 particular, if a class is thrown as an exception in one shared object
4800 and caught in another, the class must have default visibility.
4801 Otherwise the two shared objects will be unable to use the same
4802 typeinfo node and exception handling will break.
4806 @subsection ARM Type Attributes
4808 On those ARM targets that support @code{dllimport} (such as Symbian
4809 OS), you can use the @code{notshared} attribute to indicate that the
4810 virtual table and other similar data for a class should not be
4811 exported from a DLL@. For example:
4814 class __declspec(notshared) C @{
4816 __declspec(dllimport) C();
4820 __declspec(dllexport)
4824 In this code, @code{C::C} is exported from the current DLL, but the
4825 virtual table for @code{C} is not exported. (You can use
4826 @code{__attribute__} instead of @code{__declspec} if you prefer, but
4827 most Symbian OS code uses @code{__declspec}.)
4829 @anchor{MeP Type Attributes}
4830 @subsection MeP Type Attributes
4832 Many of the MeP variable attributes may be applied to types as well.
4833 Specifically, the @code{based}, @code{tiny}, @code{near}, and
4834 @code{far} attributes may be applied to either. The @code{io} and
4835 @code{cb} attributes may not be applied to types.
4837 @anchor{i386 Type Attributes}
4838 @subsection i386 Type Attributes
4840 Two attributes are currently defined for i386 configurations:
4841 @code{ms_struct} and @code{gcc_struct}.
4847 @cindex @code{ms_struct}
4848 @cindex @code{gcc_struct}
4850 If @code{packed} is used on a structure, or if bit-fields are used
4851 it may be that the Microsoft ABI packs them differently
4852 than GCC would normally pack them. Particularly when moving packed
4853 data between functions compiled with GCC and the native Microsoft compiler
4854 (either via function call or as data in a file), it may be necessary to access
4857 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
4858 compilers to match the native Microsoft compiler.
4861 To specify multiple attributes, separate them by commas within the
4862 double parentheses: for example, @samp{__attribute__ ((aligned (16),
4865 @anchor{PowerPC Type Attributes}
4866 @subsection PowerPC Type Attributes
4868 Three attributes currently are defined for PowerPC configurations:
4869 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
4871 For full documentation of the @code{ms_struct} and @code{gcc_struct}
4872 attributes please see the documentation in @ref{i386 Type Attributes}.
4874 The @code{altivec} attribute allows one to declare AltiVec vector data
4875 types supported by the AltiVec Programming Interface Manual. The
4876 attribute requires an argument to specify one of three vector types:
4877 @code{vector__}, @code{pixel__} (always followed by unsigned short),
4878 and @code{bool__} (always followed by unsigned).
4881 __attribute__((altivec(vector__)))
4882 __attribute__((altivec(pixel__))) unsigned short
4883 __attribute__((altivec(bool__))) unsigned
4886 These attributes mainly are intended to support the @code{__vector},
4887 @code{__pixel}, and @code{__bool} AltiVec keywords.
4889 @anchor{SPU Type Attributes}
4890 @subsection SPU Type Attributes
4892 The SPU supports the @code{spu_vector} attribute for types. This attribute
4893 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
4894 Language Extensions Specification. It is intended to support the
4895 @code{__vector} keyword.
4899 @section An Inline Function is As Fast As a Macro
4900 @cindex inline functions
4901 @cindex integrating function code
4903 @cindex macros, inline alternative
4905 By declaring a function inline, you can direct GCC to make
4906 calls to that function faster. One way GCC can achieve this is to
4907 integrate that function's code into the code for its callers. This
4908 makes execution faster by eliminating the function-call overhead; in
4909 addition, if any of the actual argument values are constant, their
4910 known values may permit simplifications at compile time so that not
4911 all of the inline function's code needs to be included. The effect on
4912 code size is less predictable; object code may be larger or smaller
4913 with function inlining, depending on the particular case. You can
4914 also direct GCC to try to integrate all ``simple enough'' functions
4915 into their callers with the option @option{-finline-functions}.
4917 GCC implements three different semantics of declaring a function
4918 inline. One is available with @option{-std=gnu89} or
4919 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
4920 on all inline declarations, another when @option{-std=c99} or
4921 @option{-std=gnu99} (without @option{-fgnu89-inline}), and the third
4922 is used when compiling C++.
4924 To declare a function inline, use the @code{inline} keyword in its
4925 declaration, like this:
4935 If you are writing a header file to be included in ISO C89 programs, write
4936 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
4938 The three types of inlining behave similarly in two important cases:
4939 when the @code{inline} keyword is used on a @code{static} function,
4940 like the example above, and when a function is first declared without
4941 using the @code{inline} keyword and then is defined with
4942 @code{inline}, like this:
4945 extern int inc (int *a);
4953 In both of these common cases, the program behaves the same as if you
4954 had not used the @code{inline} keyword, except for its speed.
4956 @cindex inline functions, omission of
4957 @opindex fkeep-inline-functions
4958 When a function is both inline and @code{static}, if all calls to the
4959 function are integrated into the caller, and the function's address is
4960 never used, then the function's own assembler code is never referenced.
4961 In this case, GCC does not actually output assembler code for the
4962 function, unless you specify the option @option{-fkeep-inline-functions}.
4963 Some calls cannot be integrated for various reasons (in particular,
4964 calls that precede the function's definition cannot be integrated, and
4965 neither can recursive calls within the definition). If there is a
4966 nonintegrated call, then the function is compiled to assembler code as
4967 usual. The function must also be compiled as usual if the program
4968 refers to its address, because that can't be inlined.
4971 Note that certain usages in a function definition can make it unsuitable
4972 for inline substitution. Among these usages are: use of varargs, use of
4973 alloca, use of variable sized data types (@pxref{Variable Length}),
4974 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
4975 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
4976 will warn when a function marked @code{inline} could not be substituted,
4977 and will give the reason for the failure.
4979 @cindex automatic @code{inline} for C++ member fns
4980 @cindex @code{inline} automatic for C++ member fns
4981 @cindex member fns, automatically @code{inline}
4982 @cindex C++ member fns, automatically @code{inline}
4983 @opindex fno-default-inline
4984 As required by ISO C++, GCC considers member functions defined within
4985 the body of a class to be marked inline even if they are
4986 not explicitly declared with the @code{inline} keyword. You can
4987 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
4988 Options,,Options Controlling C++ Dialect}.
4990 GCC does not inline any functions when not optimizing unless you specify
4991 the @samp{always_inline} attribute for the function, like this:
4994 /* @r{Prototype.} */
4995 inline void foo (const char) __attribute__((always_inline));
4998 The remainder of this section is specific to GNU C89 inlining.
5000 @cindex non-static inline function
5001 When an inline function is not @code{static}, then the compiler must assume
5002 that there may be calls from other source files; since a global symbol can
5003 be defined only once in any program, the function must not be defined in
5004 the other source files, so the calls therein cannot be integrated.
5005 Therefore, a non-@code{static} inline function is always compiled on its
5006 own in the usual fashion.
5008 If you specify both @code{inline} and @code{extern} in the function
5009 definition, then the definition is used only for inlining. In no case
5010 is the function compiled on its own, not even if you refer to its
5011 address explicitly. Such an address becomes an external reference, as
5012 if you had only declared the function, and had not defined it.
5014 This combination of @code{inline} and @code{extern} has almost the
5015 effect of a macro. The way to use it is to put a function definition in
5016 a header file with these keywords, and put another copy of the
5017 definition (lacking @code{inline} and @code{extern}) in a library file.
5018 The definition in the header file will cause most calls to the function
5019 to be inlined. If any uses of the function remain, they will refer to
5020 the single copy in the library.
5023 @section Assembler Instructions with C Expression Operands
5024 @cindex extended @code{asm}
5025 @cindex @code{asm} expressions
5026 @cindex assembler instructions
5029 In an assembler instruction using @code{asm}, you can specify the
5030 operands of the instruction using C expressions. This means you need not
5031 guess which registers or memory locations will contain the data you want
5034 You must specify an assembler instruction template much like what
5035 appears in a machine description, plus an operand constraint string for
5038 For example, here is how to use the 68881's @code{fsinx} instruction:
5041 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
5045 Here @code{angle} is the C expression for the input operand while
5046 @code{result} is that of the output operand. Each has @samp{"f"} as its
5047 operand constraint, saying that a floating point register is required.
5048 The @samp{=} in @samp{=f} indicates that the operand is an output; all
5049 output operands' constraints must use @samp{=}. The constraints use the
5050 same language used in the machine description (@pxref{Constraints}).
5052 Each operand is described by an operand-constraint string followed by
5053 the C expression in parentheses. A colon separates the assembler
5054 template from the first output operand and another separates the last
5055 output operand from the first input, if any. Commas separate the
5056 operands within each group. The total number of operands is currently
5057 limited to 30; this limitation may be lifted in some future version of
5060 If there are no output operands but there are input operands, you must
5061 place two consecutive colons surrounding the place where the output
5064 As of GCC version 3.1, it is also possible to specify input and output
5065 operands using symbolic names which can be referenced within the
5066 assembler code. These names are specified inside square brackets
5067 preceding the constraint string, and can be referenced inside the
5068 assembler code using @code{%[@var{name}]} instead of a percentage sign
5069 followed by the operand number. Using named operands the above example
5073 asm ("fsinx %[angle],%[output]"
5074 : [output] "=f" (result)
5075 : [angle] "f" (angle));
5079 Note that the symbolic operand names have no relation whatsoever to
5080 other C identifiers. You may use any name you like, even those of
5081 existing C symbols, but you must ensure that no two operands within the same
5082 assembler construct use the same symbolic name.
5084 Output operand expressions must be lvalues; the compiler can check this.
5085 The input operands need not be lvalues. The compiler cannot check
5086 whether the operands have data types that are reasonable for the
5087 instruction being executed. It does not parse the assembler instruction
5088 template and does not know what it means or even whether it is valid
5089 assembler input. The extended @code{asm} feature is most often used for
5090 machine instructions the compiler itself does not know exist. If
5091 the output expression cannot be directly addressed (for example, it is a
5092 bit-field), your constraint must allow a register. In that case, GCC
5093 will use the register as the output of the @code{asm}, and then store
5094 that register into the output.
5096 The ordinary output operands must be write-only; GCC will assume that
5097 the values in these operands before the instruction are dead and need
5098 not be generated. Extended asm supports input-output or read-write
5099 operands. Use the constraint character @samp{+} to indicate such an
5100 operand and list it with the output operands. You should only use
5101 read-write operands when the constraints for the operand (or the
5102 operand in which only some of the bits are to be changed) allow a
5105 You may, as an alternative, logically split its function into two
5106 separate operands, one input operand and one write-only output
5107 operand. The connection between them is expressed by constraints
5108 which say they need to be in the same location when the instruction
5109 executes. You can use the same C expression for both operands, or
5110 different expressions. For example, here we write the (fictitious)
5111 @samp{combine} instruction with @code{bar} as its read-only source
5112 operand and @code{foo} as its read-write destination:
5115 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
5119 The constraint @samp{"0"} for operand 1 says that it must occupy the
5120 same location as operand 0. A number in constraint is allowed only in
5121 an input operand and it must refer to an output operand.
5123 Only a number in the constraint can guarantee that one operand will be in
5124 the same place as another. The mere fact that @code{foo} is the value
5125 of both operands is not enough to guarantee that they will be in the
5126 same place in the generated assembler code. The following would not
5130 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
5133 Various optimizations or reloading could cause operands 0 and 1 to be in
5134 different registers; GCC knows no reason not to do so. For example, the
5135 compiler might find a copy of the value of @code{foo} in one register and
5136 use it for operand 1, but generate the output operand 0 in a different
5137 register (copying it afterward to @code{foo}'s own address). Of course,
5138 since the register for operand 1 is not even mentioned in the assembler
5139 code, the result will not work, but GCC can't tell that.
5141 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
5142 the operand number for a matching constraint. For example:
5145 asm ("cmoveq %1,%2,%[result]"
5146 : [result] "=r"(result)
5147 : "r" (test), "r"(new), "[result]"(old));
5150 Sometimes you need to make an @code{asm} operand be a specific register,
5151 but there's no matching constraint letter for that register @emph{by
5152 itself}. To force the operand into that register, use a local variable
5153 for the operand and specify the register in the variable declaration.
5154 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
5155 register constraint letter that matches the register:
5158 register int *p1 asm ("r0") = @dots{};
5159 register int *p2 asm ("r1") = @dots{};
5160 register int *result asm ("r0");
5161 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
5164 @anchor{Example of asm with clobbered asm reg}
5165 In the above example, beware that a register that is call-clobbered by
5166 the target ABI will be overwritten by any function call in the
5167 assignment, including library calls for arithmetic operators.
5168 Also a register may be clobbered when generating some operations,
5169 like variable shift, memory copy or memory move on x86.
5170 Assuming it is a call-clobbered register, this may happen to @code{r0}
5171 above by the assignment to @code{p2}. If you have to use such a
5172 register, use temporary variables for expressions between the register
5177 register int *p1 asm ("r0") = @dots{};
5178 register int *p2 asm ("r1") = t1;
5179 register int *result asm ("r0");
5180 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
5183 Some instructions clobber specific hard registers. To describe this,
5184 write a third colon after the input operands, followed by the names of
5185 the clobbered hard registers (given as strings). Here is a realistic
5186 example for the VAX:
5189 asm volatile ("movc3 %0,%1,%2"
5190 : /* @r{no outputs} */
5191 : "g" (from), "g" (to), "g" (count)
5192 : "r0", "r1", "r2", "r3", "r4", "r5");
5195 You may not write a clobber description in a way that overlaps with an
5196 input or output operand. For example, you may not have an operand
5197 describing a register class with one member if you mention that register
5198 in the clobber list. Variables declared to live in specific registers
5199 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
5200 have no part mentioned in the clobber description.
5201 There is no way for you to specify that an input
5202 operand is modified without also specifying it as an output
5203 operand. Note that if all the output operands you specify are for this
5204 purpose (and hence unused), you will then also need to specify
5205 @code{volatile} for the @code{asm} construct, as described below, to
5206 prevent GCC from deleting the @code{asm} statement as unused.
5208 If you refer to a particular hardware register from the assembler code,
5209 you will probably have to list the register after the third colon to
5210 tell the compiler the register's value is modified. In some assemblers,
5211 the register names begin with @samp{%}; to produce one @samp{%} in the
5212 assembler code, you must write @samp{%%} in the input.
5214 If your assembler instruction can alter the condition code register, add
5215 @samp{cc} to the list of clobbered registers. GCC on some machines
5216 represents the condition codes as a specific hardware register;
5217 @samp{cc} serves to name this register. On other machines, the
5218 condition code is handled differently, and specifying @samp{cc} has no
5219 effect. But it is valid no matter what the machine.
5221 If your assembler instructions access memory in an unpredictable
5222 fashion, add @samp{memory} to the list of clobbered registers. This
5223 will cause GCC to not keep memory values cached in registers across the
5224 assembler instruction and not optimize stores or loads to that memory.
5225 You will also want to add the @code{volatile} keyword if the memory
5226 affected is not listed in the inputs or outputs of the @code{asm}, as
5227 the @samp{memory} clobber does not count as a side-effect of the
5228 @code{asm}. If you know how large the accessed memory is, you can add
5229 it as input or output but if this is not known, you should add
5230 @samp{memory}. As an example, if you access ten bytes of a string, you
5231 can use a memory input like:
5234 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
5237 Note that in the following example the memory input is necessary,
5238 otherwise GCC might optimize the store to @code{x} away:
5245 asm ("magic stuff accessing an 'int' pointed to by '%1'"
5246 "=&d" (r) : "a" (y), "m" (*y));
5251 You can put multiple assembler instructions together in a single
5252 @code{asm} template, separated by the characters normally used in assembly
5253 code for the system. A combination that works in most places is a newline
5254 to break the line, plus a tab character to move to the instruction field
5255 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
5256 assembler allows semicolons as a line-breaking character. Note that some
5257 assembler dialects use semicolons to start a comment.
5258 The input operands are guaranteed not to use any of the clobbered
5259 registers, and neither will the output operands' addresses, so you can
5260 read and write the clobbered registers as many times as you like. Here
5261 is an example of multiple instructions in a template; it assumes the
5262 subroutine @code{_foo} accepts arguments in registers 9 and 10:
5265 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
5267 : "g" (from), "g" (to)
5271 Unless an output operand has the @samp{&} constraint modifier, GCC
5272 may allocate it in the same register as an unrelated input operand, on
5273 the assumption the inputs are consumed before the outputs are produced.
5274 This assumption may be false if the assembler code actually consists of
5275 more than one instruction. In such a case, use @samp{&} for each output
5276 operand that may not overlap an input. @xref{Modifiers}.
5278 If you want to test the condition code produced by an assembler
5279 instruction, you must include a branch and a label in the @code{asm}
5280 construct, as follows:
5283 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
5289 This assumes your assembler supports local labels, as the GNU assembler
5290 and most Unix assemblers do.
5292 Speaking of labels, jumps from one @code{asm} to another are not
5293 supported. The compiler's optimizers do not know about these jumps, and
5294 therefore they cannot take account of them when deciding how to
5295 optimize. @xref{Extended asm with goto}.
5297 @cindex macros containing @code{asm}
5298 Usually the most convenient way to use these @code{asm} instructions is to
5299 encapsulate them in macros that look like functions. For example,
5303 (@{ double __value, __arg = (x); \
5304 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
5309 Here the variable @code{__arg} is used to make sure that the instruction
5310 operates on a proper @code{double} value, and to accept only those
5311 arguments @code{x} which can convert automatically to a @code{double}.
5313 Another way to make sure the instruction operates on the correct data
5314 type is to use a cast in the @code{asm}. This is different from using a
5315 variable @code{__arg} in that it converts more different types. For
5316 example, if the desired type were @code{int}, casting the argument to
5317 @code{int} would accept a pointer with no complaint, while assigning the
5318 argument to an @code{int} variable named @code{__arg} would warn about
5319 using a pointer unless the caller explicitly casts it.
5321 If an @code{asm} has output operands, GCC assumes for optimization
5322 purposes the instruction has no side effects except to change the output
5323 operands. This does not mean instructions with a side effect cannot be
5324 used, but you must be careful, because the compiler may eliminate them
5325 if the output operands aren't used, or move them out of loops, or
5326 replace two with one if they constitute a common subexpression. Also,
5327 if your instruction does have a side effect on a variable that otherwise
5328 appears not to change, the old value of the variable may be reused later
5329 if it happens to be found in a register.
5331 You can prevent an @code{asm} instruction from being deleted
5332 by writing the keyword @code{volatile} after
5333 the @code{asm}. For example:
5336 #define get_and_set_priority(new) \
5338 asm volatile ("get_and_set_priority %0, %1" \
5339 : "=g" (__old) : "g" (new)); \
5344 The @code{volatile} keyword indicates that the instruction has
5345 important side-effects. GCC will not delete a volatile @code{asm} if
5346 it is reachable. (The instruction can still be deleted if GCC can
5347 prove that control-flow will never reach the location of the
5348 instruction.) Note that even a volatile @code{asm} instruction
5349 can be moved relative to other code, including across jump
5350 instructions. For example, on many targets there is a system
5351 register which can be set to control the rounding mode of
5352 floating point operations. You might try
5353 setting it with a volatile @code{asm}, like this PowerPC example:
5356 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
5361 This will not work reliably, as the compiler may move the addition back
5362 before the volatile @code{asm}. To make it work you need to add an
5363 artificial dependency to the @code{asm} referencing a variable in the code
5364 you don't want moved, for example:
5367 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
5371 Similarly, you can't expect a
5372 sequence of volatile @code{asm} instructions to remain perfectly
5373 consecutive. If you want consecutive output, use a single @code{asm}.
5374 Also, GCC will perform some optimizations across a volatile @code{asm}
5375 instruction; GCC does not ``forget everything'' when it encounters
5376 a volatile @code{asm} instruction the way some other compilers do.
5378 An @code{asm} instruction without any output operands will be treated
5379 identically to a volatile @code{asm} instruction.
5381 It is a natural idea to look for a way to give access to the condition
5382 code left by the assembler instruction. However, when we attempted to
5383 implement this, we found no way to make it work reliably. The problem
5384 is that output operands might need reloading, which would result in
5385 additional following ``store'' instructions. On most machines, these
5386 instructions would alter the condition code before there was time to
5387 test it. This problem doesn't arise for ordinary ``test'' and
5388 ``compare'' instructions because they don't have any output operands.
5390 For reasons similar to those described above, it is not possible to give
5391 an assembler instruction access to the condition code left by previous
5394 @anchor{Extended asm with goto}
5395 As of GCC version 4.5, @code{asm goto} may be used to have the assembly
5396 jump to one or more C labels. In this form, a fifth section after the
5397 clobber list contains a list of all C labels to which the assembly may jump.
5398 Each label operand is implicitly self-named. The @code{asm} is also assumed
5399 to fall through to the next statement.
5401 This form of @code{asm} is restricted to not have outputs. This is due
5402 to a internal restriction in the compiler that control transfer instructions
5403 cannot have outputs. This restriction on @code{asm goto} may be lifted
5404 in some future version of the compiler. In the mean time, @code{asm goto}
5405 may include a memory clobber, and so leave outputs in memory.
5411 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
5412 : : "r"(x), "r"(&y) : "r5", "memory" : error);
5419 In this (inefficient) example, the @code{frob} instruction sets the
5420 carry bit to indicate an error. The @code{jc} instruction detects
5421 this and branches to the @code{error} label. Finally, the output
5422 of the @code{frob} instruction (@code{%r5}) is stored into the memory
5423 for variable @code{y}, which is later read by the @code{return} statement.
5429 asm goto ("mfsr %%r1, 123; jmp %%r1;"
5430 ".pushsection doit_table;"
5431 ".long %l0, %l1, %l2, %l3;"
5433 : : : "r1" : label1, label2, label3, label4);
5434 __builtin_unreachable ();
5449 In this (also inefficient) example, the @code{mfsr} instruction reads
5450 an address from some out-of-band machine register, and the following
5451 @code{jmp} instruction branches to that address. The address read by
5452 the @code{mfsr} instruction is assumed to have been previously set via
5453 some application-specific mechanism to be one of the four values stored
5454 in the @code{doit_table} section. Finally, the @code{asm} is followed
5455 by a call to @code{__builtin_unreachable} to indicate that the @code{asm}
5456 does not in fact fall through.
5459 #define TRACE1(NUM) \
5461 asm goto ("0: nop;" \
5462 ".pushsection trace_table;" \
5465 : : : : trace#NUM); \
5466 if (0) @{ trace#NUM: trace(); @} \
5468 #define TRACE TRACE1(__COUNTER__)
5471 In this example (which in fact inspired the @code{asm goto} feature)
5472 we want on rare occasions to call the @code{trace} function; on other
5473 occasions we'd like to keep the overhead to the absolute minimum.
5474 The normal code path consists of a single @code{nop} instruction.
5475 However, we record the address of this @code{nop} together with the
5476 address of a label that calls the @code{trace} function. This allows
5477 the @code{nop} instruction to be patched at runtime to be an
5478 unconditional branch to the stored label. It is assumed that an
5479 optimizing compiler will move the labeled block out of line, to
5480 optimize the fall through path from the @code{asm}.
5482 If you are writing a header file that should be includable in ISO C
5483 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
5486 @subsection Size of an @code{asm}
5488 Some targets require that GCC track the size of each instruction used in
5489 order to generate correct code. Because the final length of an
5490 @code{asm} is only known by the assembler, GCC must make an estimate as
5491 to how big it will be. The estimate is formed by counting the number of
5492 statements in the pattern of the @code{asm} and multiplying that by the
5493 length of the longest instruction on that processor. Statements in the
5494 @code{asm} are identified by newline characters and whatever statement
5495 separator characters are supported by the assembler; on most processors
5496 this is the `@code{;}' character.
5498 Normally, GCC's estimate is perfectly adequate to ensure that correct
5499 code is generated, but it is possible to confuse the compiler if you use
5500 pseudo instructions or assembler macros that expand into multiple real
5501 instructions or if you use assembler directives that expand to more
5502 space in the object file than would be needed for a single instruction.
5503 If this happens then the assembler will produce a diagnostic saying that
5504 a label is unreachable.
5506 @subsection i386 floating point asm operands
5508 There are several rules on the usage of stack-like regs in
5509 asm_operands insns. These rules apply only to the operands that are
5514 Given a set of input regs that die in an asm_operands, it is
5515 necessary to know which are implicitly popped by the asm, and
5516 which must be explicitly popped by gcc.
5518 An input reg that is implicitly popped by the asm must be
5519 explicitly clobbered, unless it is constrained to match an
5523 For any input reg that is implicitly popped by an asm, it is
5524 necessary to know how to adjust the stack to compensate for the pop.
5525 If any non-popped input is closer to the top of the reg-stack than
5526 the implicitly popped reg, it would not be possible to know what the
5527 stack looked like---it's not clear how the rest of the stack ``slides
5530 All implicitly popped input regs must be closer to the top of
5531 the reg-stack than any input that is not implicitly popped.
5533 It is possible that if an input dies in an insn, reload might
5534 use the input reg for an output reload. Consider this example:
5537 asm ("foo" : "=t" (a) : "f" (b));
5540 This asm says that input B is not popped by the asm, and that
5541 the asm pushes a result onto the reg-stack, i.e., the stack is one
5542 deeper after the asm than it was before. But, it is possible that
5543 reload will think that it can use the same reg for both the input and
5544 the output, if input B dies in this insn.
5546 If any input operand uses the @code{f} constraint, all output reg
5547 constraints must use the @code{&} earlyclobber.
5549 The asm above would be written as
5552 asm ("foo" : "=&t" (a) : "f" (b));
5556 Some operands need to be in particular places on the stack. All
5557 output operands fall in this category---there is no other way to
5558 know which regs the outputs appear in unless the user indicates
5559 this in the constraints.
5561 Output operands must specifically indicate which reg an output
5562 appears in after an asm. @code{=f} is not allowed: the operand
5563 constraints must select a class with a single reg.
5566 Output operands may not be ``inserted'' between existing stack regs.
5567 Since no 387 opcode uses a read/write operand, all output operands
5568 are dead before the asm_operands, and are pushed by the asm_operands.
5569 It makes no sense to push anywhere but the top of the reg-stack.
5571 Output operands must start at the top of the reg-stack: output
5572 operands may not ``skip'' a reg.
5575 Some asm statements may need extra stack space for internal
5576 calculations. This can be guaranteed by clobbering stack registers
5577 unrelated to the inputs and outputs.
5581 Here are a couple of reasonable asms to want to write. This asm
5582 takes one input, which is internally popped, and produces two outputs.
5585 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
5588 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
5589 and replaces them with one output. The user must code the @code{st(1)}
5590 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
5593 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
5599 @section Controlling Names Used in Assembler Code
5600 @cindex assembler names for identifiers
5601 @cindex names used in assembler code
5602 @cindex identifiers, names in assembler code
5604 You can specify the name to be used in the assembler code for a C
5605 function or variable by writing the @code{asm} (or @code{__asm__})
5606 keyword after the declarator as follows:
5609 int foo asm ("myfoo") = 2;
5613 This specifies that the name to be used for the variable @code{foo} in
5614 the assembler code should be @samp{myfoo} rather than the usual
5617 On systems where an underscore is normally prepended to the name of a C
5618 function or variable, this feature allows you to define names for the
5619 linker that do not start with an underscore.
5621 It does not make sense to use this feature with a non-static local
5622 variable since such variables do not have assembler names. If you are
5623 trying to put the variable in a particular register, see @ref{Explicit
5624 Reg Vars}. GCC presently accepts such code with a warning, but will
5625 probably be changed to issue an error, rather than a warning, in the
5628 You cannot use @code{asm} in this way in a function @emph{definition}; but
5629 you can get the same effect by writing a declaration for the function
5630 before its definition and putting @code{asm} there, like this:
5633 extern func () asm ("FUNC");
5640 It is up to you to make sure that the assembler names you choose do not
5641 conflict with any other assembler symbols. Also, you must not use a
5642 register name; that would produce completely invalid assembler code. GCC
5643 does not as yet have the ability to store static variables in registers.
5644 Perhaps that will be added.
5646 @node Explicit Reg Vars
5647 @section Variables in Specified Registers
5648 @cindex explicit register variables
5649 @cindex variables in specified registers
5650 @cindex specified registers
5651 @cindex registers, global allocation
5653 GNU C allows you to put a few global variables into specified hardware
5654 registers. You can also specify the register in which an ordinary
5655 register variable should be allocated.
5659 Global register variables reserve registers throughout the program.
5660 This may be useful in programs such as programming language
5661 interpreters which have a couple of global variables that are accessed
5665 Local register variables in specific registers do not reserve the
5666 registers, except at the point where they are used as input or output
5667 operands in an @code{asm} statement and the @code{asm} statement itself is
5668 not deleted. The compiler's data flow analysis is capable of determining
5669 where the specified registers contain live values, and where they are
5670 available for other uses. Stores into local register variables may be deleted
5671 when they appear to be dead according to dataflow analysis. References
5672 to local register variables may be deleted or moved or simplified.
5674 These local variables are sometimes convenient for use with the extended
5675 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
5676 output of the assembler instruction directly into a particular register.
5677 (This will work provided the register you specify fits the constraints
5678 specified for that operand in the @code{asm}.)
5686 @node Global Reg Vars
5687 @subsection Defining Global Register Variables
5688 @cindex global register variables
5689 @cindex registers, global variables in
5691 You can define a global register variable in GNU C like this:
5694 register int *foo asm ("a5");
5698 Here @code{a5} is the name of the register which should be used. Choose a
5699 register which is normally saved and restored by function calls on your
5700 machine, so that library routines will not clobber it.
5702 Naturally the register name is cpu-dependent, so you would need to
5703 conditionalize your program according to cpu type. The register
5704 @code{a5} would be a good choice on a 68000 for a variable of pointer
5705 type. On machines with register windows, be sure to choose a ``global''
5706 register that is not affected magically by the function call mechanism.
5708 In addition, operating systems on one type of cpu may differ in how they
5709 name the registers; then you would need additional conditionals. For
5710 example, some 68000 operating systems call this register @code{%a5}.
5712 Eventually there may be a way of asking the compiler to choose a register
5713 automatically, but first we need to figure out how it should choose and
5714 how to enable you to guide the choice. No solution is evident.
5716 Defining a global register variable in a certain register reserves that
5717 register entirely for this use, at least within the current compilation.
5718 The register will not be allocated for any other purpose in the functions
5719 in the current compilation. The register will not be saved and restored by
5720 these functions. Stores into this register are never deleted even if they
5721 would appear to be dead, but references may be deleted or moved or
5724 It is not safe to access the global register variables from signal
5725 handlers, or from more than one thread of control, because the system
5726 library routines may temporarily use the register for other things (unless
5727 you recompile them specially for the task at hand).
5729 @cindex @code{qsort}, and global register variables
5730 It is not safe for one function that uses a global register variable to
5731 call another such function @code{foo} by way of a third function
5732 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
5733 different source file in which the variable wasn't declared). This is
5734 because @code{lose} might save the register and put some other value there.
5735 For example, you can't expect a global register variable to be available in
5736 the comparison-function that you pass to @code{qsort}, since @code{qsort}
5737 might have put something else in that register. (If you are prepared to
5738 recompile @code{qsort} with the same global register variable, you can
5739 solve this problem.)
5741 If you want to recompile @code{qsort} or other source files which do not
5742 actually use your global register variable, so that they will not use that
5743 register for any other purpose, then it suffices to specify the compiler
5744 option @option{-ffixed-@var{reg}}. You need not actually add a global
5745 register declaration to their source code.
5747 A function which can alter the value of a global register variable cannot
5748 safely be called from a function compiled without this variable, because it
5749 could clobber the value the caller expects to find there on return.
5750 Therefore, the function which is the entry point into the part of the
5751 program that uses the global register variable must explicitly save and
5752 restore the value which belongs to its caller.
5754 @cindex register variable after @code{longjmp}
5755 @cindex global register after @code{longjmp}
5756 @cindex value after @code{longjmp}
5759 On most machines, @code{longjmp} will restore to each global register
5760 variable the value it had at the time of the @code{setjmp}. On some
5761 machines, however, @code{longjmp} will not change the value of global
5762 register variables. To be portable, the function that called @code{setjmp}
5763 should make other arrangements to save the values of the global register
5764 variables, and to restore them in a @code{longjmp}. This way, the same
5765 thing will happen regardless of what @code{longjmp} does.
5767 All global register variable declarations must precede all function
5768 definitions. If such a declaration could appear after function
5769 definitions, the declaration would be too late to prevent the register from
5770 being used for other purposes in the preceding functions.
5772 Global register variables may not have initial values, because an
5773 executable file has no means to supply initial contents for a register.
5775 On the SPARC, there are reports that g3 @dots{} g7 are suitable
5776 registers, but certain library functions, such as @code{getwd}, as well
5777 as the subroutines for division and remainder, modify g3 and g4. g1 and
5778 g2 are local temporaries.
5780 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
5781 Of course, it will not do to use more than a few of those.
5783 @node Local Reg Vars
5784 @subsection Specifying Registers for Local Variables
5785 @cindex local variables, specifying registers
5786 @cindex specifying registers for local variables
5787 @cindex registers for local variables
5789 You can define a local register variable with a specified register
5793 register int *foo asm ("a5");
5797 Here @code{a5} is the name of the register which should be used. Note
5798 that this is the same syntax used for defining global register
5799 variables, but for a local variable it would appear within a function.
5801 Naturally the register name is cpu-dependent, but this is not a
5802 problem, since specific registers are most often useful with explicit
5803 assembler instructions (@pxref{Extended Asm}). Both of these things
5804 generally require that you conditionalize your program according to
5807 In addition, operating systems on one type of cpu may differ in how they
5808 name the registers; then you would need additional conditionals. For
5809 example, some 68000 operating systems call this register @code{%a5}.
5811 Defining such a register variable does not reserve the register; it
5812 remains available for other uses in places where flow control determines
5813 the variable's value is not live.
5815 This option does not guarantee that GCC will generate code that has
5816 this variable in the register you specify at all times. You may not
5817 code an explicit reference to this register in the @emph{assembler
5818 instruction template} part of an @code{asm} statement and assume it will
5819 always refer to this variable. However, using the variable as an
5820 @code{asm} @emph{operand} guarantees that the specified register is used
5823 Stores into local register variables may be deleted when they appear to be dead
5824 according to dataflow analysis. References to local register variables may
5825 be deleted or moved or simplified.
5827 As for global register variables, it's recommended that you choose a
5828 register which is normally saved and restored by function calls on
5829 your machine, so that library routines will not clobber it. A common
5830 pitfall is to initialize multiple call-clobbered registers with
5831 arbitrary expressions, where a function call or library call for an
5832 arithmetic operator will overwrite a register value from a previous
5833 assignment, for example @code{r0} below:
5835 register int *p1 asm ("r0") = @dots{};
5836 register int *p2 asm ("r1") = @dots{};
5838 In those cases, a solution is to use a temporary variable for
5839 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
5841 @node Alternate Keywords
5842 @section Alternate Keywords
5843 @cindex alternate keywords
5844 @cindex keywords, alternate
5846 @option{-ansi} and the various @option{-std} options disable certain
5847 keywords. This causes trouble when you want to use GNU C extensions, or
5848 a general-purpose header file that should be usable by all programs,
5849 including ISO C programs. The keywords @code{asm}, @code{typeof} and
5850 @code{inline} are not available in programs compiled with
5851 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
5852 program compiled with @option{-std=c99}). The ISO C99 keyword
5853 @code{restrict} is only available when @option{-std=gnu99} (which will
5854 eventually be the default) or @option{-std=c99} (or the equivalent
5855 @option{-std=iso9899:1999}) is used.
5857 The way to solve these problems is to put @samp{__} at the beginning and
5858 end of each problematical keyword. For example, use @code{__asm__}
5859 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
5861 Other C compilers won't accept these alternative keywords; if you want to
5862 compile with another compiler, you can define the alternate keywords as
5863 macros to replace them with the customary keywords. It looks like this:
5871 @findex __extension__
5873 @option{-pedantic} and other options cause warnings for many GNU C extensions.
5875 prevent such warnings within one expression by writing
5876 @code{__extension__} before the expression. @code{__extension__} has no
5877 effect aside from this.
5879 @node Incomplete Enums
5880 @section Incomplete @code{enum} Types
5882 You can define an @code{enum} tag without specifying its possible values.
5883 This results in an incomplete type, much like what you get if you write
5884 @code{struct foo} without describing the elements. A later declaration
5885 which does specify the possible values completes the type.
5887 You can't allocate variables or storage using the type while it is
5888 incomplete. However, you can work with pointers to that type.
5890 This extension may not be very useful, but it makes the handling of
5891 @code{enum} more consistent with the way @code{struct} and @code{union}
5894 This extension is not supported by GNU C++.
5896 @node Function Names
5897 @section Function Names as Strings
5898 @cindex @code{__func__} identifier
5899 @cindex @code{__FUNCTION__} identifier
5900 @cindex @code{__PRETTY_FUNCTION__} identifier
5902 GCC provides three magic variables which hold the name of the current
5903 function, as a string. The first of these is @code{__func__}, which
5904 is part of the C99 standard:
5906 The identifier @code{__func__} is implicitly declared by the translator
5907 as if, immediately following the opening brace of each function
5908 definition, the declaration
5911 static const char __func__[] = "function-name";
5915 appeared, where function-name is the name of the lexically-enclosing
5916 function. This name is the unadorned name of the function.
5918 @code{__FUNCTION__} is another name for @code{__func__}. Older
5919 versions of GCC recognize only this name. However, it is not
5920 standardized. For maximum portability, we recommend you use
5921 @code{__func__}, but provide a fallback definition with the
5925 #if __STDC_VERSION__ < 199901L
5927 # define __func__ __FUNCTION__
5929 # define __func__ "<unknown>"
5934 In C, @code{__PRETTY_FUNCTION__} is yet another name for
5935 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
5936 the type signature of the function as well as its bare name. For
5937 example, this program:
5941 extern int printf (char *, ...);
5948 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
5949 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
5967 __PRETTY_FUNCTION__ = void a::sub(int)
5970 These identifiers are not preprocessor macros. In GCC 3.3 and
5971 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
5972 were treated as string literals; they could be used to initialize
5973 @code{char} arrays, and they could be concatenated with other string
5974 literals. GCC 3.4 and later treat them as variables, like
5975 @code{__func__}. In C++, @code{__FUNCTION__} and
5976 @code{__PRETTY_FUNCTION__} have always been variables.
5978 @node Return Address
5979 @section Getting the Return or Frame Address of a Function
5981 These functions may be used to get information about the callers of a
5984 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
5985 This function returns the return address of the current function, or of
5986 one of its callers. The @var{level} argument is number of frames to
5987 scan up the call stack. A value of @code{0} yields the return address
5988 of the current function, a value of @code{1} yields the return address
5989 of the caller of the current function, and so forth. When inlining
5990 the expected behavior is that the function will return the address of
5991 the function that will be returned to. To work around this behavior use
5992 the @code{noinline} function attribute.
5994 The @var{level} argument must be a constant integer.
5996 On some machines it may be impossible to determine the return address of
5997 any function other than the current one; in such cases, or when the top
5998 of the stack has been reached, this function will return @code{0} or a
5999 random value. In addition, @code{__builtin_frame_address} may be used
6000 to determine if the top of the stack has been reached.
6002 Additional post-processing of the returned value may be needed, see
6003 @code{__builtin_extract_return_address}.
6005 This function should only be used with a nonzero argument for debugging
6009 @deftypefn {Built-in Function} {void *} __builtin_extract_return_address (void *@var{addr})
6010 The address as returned by @code{__builtin_return_address} may have to be fed
6011 through this function to get the actual encoded address. For example, on the
6012 31-bit S/390 platform the highest bit has to be masked out, or on SPARC
6013 platforms an offset has to be added for the true next instruction to be
6016 If no fixup is needed, this function simply passes through @var{addr}.
6019 @deftypefn {Built-in Function} {void *} __builtin_frob_return_address (void *@var{addr})
6020 This function does the reverse of @code{__builtin_extract_return_address}.
6023 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
6024 This function is similar to @code{__builtin_return_address}, but it
6025 returns the address of the function frame rather than the return address
6026 of the function. Calling @code{__builtin_frame_address} with a value of
6027 @code{0} yields the frame address of the current function, a value of
6028 @code{1} yields the frame address of the caller of the current function,
6031 The frame is the area on the stack which holds local variables and saved
6032 registers. The frame address is normally the address of the first word
6033 pushed on to the stack by the function. However, the exact definition
6034 depends upon the processor and the calling convention. If the processor
6035 has a dedicated frame pointer register, and the function has a frame,
6036 then @code{__builtin_frame_address} will return the value of the frame
6039 On some machines it may be impossible to determine the frame address of
6040 any function other than the current one; in such cases, or when the top
6041 of the stack has been reached, this function will return @code{0} if
6042 the first frame pointer is properly initialized by the startup code.
6044 This function should only be used with a nonzero argument for debugging
6048 @node Vector Extensions
6049 @section Using vector instructions through built-in functions
6051 On some targets, the instruction set contains SIMD vector instructions that
6052 operate on multiple values contained in one large register at the same time.
6053 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
6056 The first step in using these extensions is to provide the necessary data
6057 types. This should be done using an appropriate @code{typedef}:
6060 typedef int v4si __attribute__ ((vector_size (16)));
6063 The @code{int} type specifies the base type, while the attribute specifies
6064 the vector size for the variable, measured in bytes. For example, the
6065 declaration above causes the compiler to set the mode for the @code{v4si}
6066 type to be 16 bytes wide and divided into @code{int} sized units. For
6067 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
6068 corresponding mode of @code{foo} will be @acronym{V4SI}.
6070 The @code{vector_size} attribute is only applicable to integral and
6071 float scalars, although arrays, pointers, and function return values
6072 are allowed in conjunction with this construct.
6074 All the basic integer types can be used as base types, both as signed
6075 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
6076 @code{long long}. In addition, @code{float} and @code{double} can be
6077 used to build floating-point vector types.
6079 Specifying a combination that is not valid for the current architecture
6080 will cause GCC to synthesize the instructions using a narrower mode.
6081 For example, if you specify a variable of type @code{V4SI} and your
6082 architecture does not allow for this specific SIMD type, GCC will
6083 produce code that uses 4 @code{SIs}.
6085 The types defined in this manner can be used with a subset of normal C
6086 operations. Currently, GCC will allow using the following operators
6087 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~, %}@.
6089 The operations behave like C++ @code{valarrays}. Addition is defined as
6090 the addition of the corresponding elements of the operands. For
6091 example, in the code below, each of the 4 elements in @var{a} will be
6092 added to the corresponding 4 elements in @var{b} and the resulting
6093 vector will be stored in @var{c}.
6096 typedef int v4si __attribute__ ((vector_size (16)));
6103 Subtraction, multiplication, division, and the logical operations
6104 operate in a similar manner. Likewise, the result of using the unary
6105 minus or complement operators on a vector type is a vector whose
6106 elements are the negative or complemented values of the corresponding
6107 elements in the operand.
6109 You can declare variables and use them in function calls and returns, as
6110 well as in assignments and some casts. You can specify a vector type as
6111 a return type for a function. Vector types can also be used as function
6112 arguments. It is possible to cast from one vector type to another,
6113 provided they are of the same size (in fact, you can also cast vectors
6114 to and from other datatypes of the same size).
6116 You cannot operate between vectors of different lengths or different
6117 signedness without a cast.
6119 A port that supports hardware vector operations, usually provides a set
6120 of built-in functions that can be used to operate on vectors. For
6121 example, a function to add two vectors and multiply the result by a
6122 third could look like this:
6125 v4si f (v4si a, v4si b, v4si c)
6127 v4si tmp = __builtin_addv4si (a, b);
6128 return __builtin_mulv4si (tmp, c);
6135 @findex __builtin_offsetof
6137 GCC implements for both C and C++ a syntactic extension to implement
6138 the @code{offsetof} macro.
6142 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
6144 offsetof_member_designator:
6146 | offsetof_member_designator "." @code{identifier}
6147 | offsetof_member_designator "[" @code{expr} "]"
6150 This extension is sufficient such that
6153 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
6156 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
6157 may be dependent. In either case, @var{member} may consist of a single
6158 identifier, or a sequence of member accesses and array references.
6160 @node Atomic Builtins
6161 @section Built-in functions for atomic memory access
6163 The following builtins are intended to be compatible with those described
6164 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
6165 section 7.4. As such, they depart from the normal GCC practice of using
6166 the ``__builtin_'' prefix, and further that they are overloaded such that
6167 they work on multiple types.
6169 The definition given in the Intel documentation allows only for the use of
6170 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
6171 counterparts. GCC will allow any integral scalar or pointer type that is
6172 1, 2, 4 or 8 bytes in length.
6174 Not all operations are supported by all target processors. If a particular
6175 operation cannot be implemented on the target processor, a warning will be
6176 generated and a call an external function will be generated. The external
6177 function will carry the same name as the builtin, with an additional suffix
6178 @samp{_@var{n}} where @var{n} is the size of the data type.
6180 @c ??? Should we have a mechanism to suppress this warning? This is almost
6181 @c useful for implementing the operation under the control of an external
6184 In most cases, these builtins are considered a @dfn{full barrier}. That is,
6185 no memory operand will be moved across the operation, either forward or
6186 backward. Further, instructions will be issued as necessary to prevent the
6187 processor from speculating loads across the operation and from queuing stores
6188 after the operation.
6190 All of the routines are described in the Intel documentation to take
6191 ``an optional list of variables protected by the memory barrier''. It's
6192 not clear what is meant by that; it could mean that @emph{only} the
6193 following variables are protected, or it could mean that these variables
6194 should in addition be protected. At present GCC ignores this list and
6195 protects all variables which are globally accessible. If in the future
6196 we make some use of this list, an empty list will continue to mean all
6197 globally accessible variables.
6200 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
6201 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
6202 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
6203 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
6204 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
6205 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
6206 @findex __sync_fetch_and_add
6207 @findex __sync_fetch_and_sub
6208 @findex __sync_fetch_and_or
6209 @findex __sync_fetch_and_and
6210 @findex __sync_fetch_and_xor
6211 @findex __sync_fetch_and_nand
6212 These builtins perform the operation suggested by the name, and
6213 returns the value that had previously been in memory. That is,
6216 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
6217 @{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; @} // nand
6220 @emph{Note:} GCC 4.4 and later implement @code{__sync_fetch_and_nand}
6221 builtin as @code{*ptr = ~(tmp & value)} instead of @code{*ptr = ~tmp & value}.
6223 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
6224 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
6225 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
6226 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
6227 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
6228 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
6229 @findex __sync_add_and_fetch
6230 @findex __sync_sub_and_fetch
6231 @findex __sync_or_and_fetch
6232 @findex __sync_and_and_fetch
6233 @findex __sync_xor_and_fetch
6234 @findex __sync_nand_and_fetch
6235 These builtins perform the operation suggested by the name, and
6236 return the new value. That is,
6239 @{ *ptr @var{op}= value; return *ptr; @}
6240 @{ *ptr = ~(*ptr & value); return *ptr; @} // nand
6243 @emph{Note:} GCC 4.4 and later implement @code{__sync_nand_and_fetch}
6244 builtin as @code{*ptr = ~(*ptr & value)} instead of
6245 @code{*ptr = ~*ptr & value}.
6247 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
6248 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval @var{type} newval, ...)
6249 @findex __sync_bool_compare_and_swap
6250 @findex __sync_val_compare_and_swap
6251 These builtins perform an atomic compare and swap. That is, if the current
6252 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
6255 The ``bool'' version returns true if the comparison is successful and
6256 @var{newval} was written. The ``val'' version returns the contents
6257 of @code{*@var{ptr}} before the operation.
6259 @item __sync_synchronize (...)
6260 @findex __sync_synchronize
6261 This builtin issues a full memory barrier.
6263 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
6264 @findex __sync_lock_test_and_set
6265 This builtin, as described by Intel, is not a traditional test-and-set
6266 operation, but rather an atomic exchange operation. It writes @var{value}
6267 into @code{*@var{ptr}}, and returns the previous contents of
6270 Many targets have only minimal support for such locks, and do not support
6271 a full exchange operation. In this case, a target may support reduced
6272 functionality here by which the @emph{only} valid value to store is the
6273 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
6274 is implementation defined.
6276 This builtin is not a full barrier, but rather an @dfn{acquire barrier}.
6277 This means that references after the builtin cannot move to (or be
6278 speculated to) before the builtin, but previous memory stores may not
6279 be globally visible yet, and previous memory loads may not yet be
6282 @item void __sync_lock_release (@var{type} *ptr, ...)
6283 @findex __sync_lock_release
6284 This builtin releases the lock acquired by @code{__sync_lock_test_and_set}.
6285 Normally this means writing the constant 0 to @code{*@var{ptr}}.
6287 This builtin is not a full barrier, but rather a @dfn{release barrier}.
6288 This means that all previous memory stores are globally visible, and all
6289 previous memory loads have been satisfied, but following memory reads
6290 are not prevented from being speculated to before the barrier.
6293 @node Object Size Checking
6294 @section Object Size Checking Builtins
6295 @findex __builtin_object_size
6296 @findex __builtin___memcpy_chk
6297 @findex __builtin___mempcpy_chk
6298 @findex __builtin___memmove_chk
6299 @findex __builtin___memset_chk
6300 @findex __builtin___strcpy_chk
6301 @findex __builtin___stpcpy_chk
6302 @findex __builtin___strncpy_chk
6303 @findex __builtin___strcat_chk
6304 @findex __builtin___strncat_chk
6305 @findex __builtin___sprintf_chk
6306 @findex __builtin___snprintf_chk
6307 @findex __builtin___vsprintf_chk
6308 @findex __builtin___vsnprintf_chk
6309 @findex __builtin___printf_chk
6310 @findex __builtin___vprintf_chk
6311 @findex __builtin___fprintf_chk
6312 @findex __builtin___vfprintf_chk
6314 GCC implements a limited buffer overflow protection mechanism
6315 that can prevent some buffer overflow attacks.
6317 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
6318 is a built-in construct that returns a constant number of bytes from
6319 @var{ptr} to the end of the object @var{ptr} pointer points to
6320 (if known at compile time). @code{__builtin_object_size} never evaluates
6321 its arguments for side-effects. If there are any side-effects in them, it
6322 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
6323 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
6324 point to and all of them are known at compile time, the returned number
6325 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
6326 0 and minimum if nonzero. If it is not possible to determine which objects
6327 @var{ptr} points to at compile time, @code{__builtin_object_size} should
6328 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
6329 for @var{type} 2 or 3.
6331 @var{type} is an integer constant from 0 to 3. If the least significant
6332 bit is clear, objects are whole variables, if it is set, a closest
6333 surrounding subobject is considered the object a pointer points to.
6334 The second bit determines if maximum or minimum of remaining bytes
6338 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
6339 char *p = &var.buf1[1], *q = &var.b;
6341 /* Here the object p points to is var. */
6342 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
6343 /* The subobject p points to is var.buf1. */
6344 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
6345 /* The object q points to is var. */
6346 assert (__builtin_object_size (q, 0)
6347 == (char *) (&var + 1) - (char *) &var.b);
6348 /* The subobject q points to is var.b. */
6349 assert (__builtin_object_size (q, 1) == sizeof (var.b));
6353 There are built-in functions added for many common string operation
6354 functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk}
6355 built-in is provided. This built-in has an additional last argument,
6356 which is the number of bytes remaining in object the @var{dest}
6357 argument points to or @code{(size_t) -1} if the size is not known.
6359 The built-in functions are optimized into the normal string functions
6360 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
6361 it is known at compile time that the destination object will not
6362 be overflown. If the compiler can determine at compile time the
6363 object will be always overflown, it issues a warning.
6365 The intended use can be e.g.
6369 #define bos0(dest) __builtin_object_size (dest, 0)
6370 #define memcpy(dest, src, n) \
6371 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
6375 /* It is unknown what object p points to, so this is optimized
6376 into plain memcpy - no checking is possible. */
6377 memcpy (p, "abcde", n);
6378 /* Destination is known and length too. It is known at compile
6379 time there will be no overflow. */
6380 memcpy (&buf[5], "abcde", 5);
6381 /* Destination is known, but the length is not known at compile time.
6382 This will result in __memcpy_chk call that can check for overflow
6384 memcpy (&buf[5], "abcde", n);
6385 /* Destination is known and it is known at compile time there will
6386 be overflow. There will be a warning and __memcpy_chk call that
6387 will abort the program at runtime. */
6388 memcpy (&buf[6], "abcde", 5);
6391 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
6392 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
6393 @code{strcat} and @code{strncat}.
6395 There are also checking built-in functions for formatted output functions.
6397 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
6398 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
6399 const char *fmt, ...);
6400 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
6402 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
6403 const char *fmt, va_list ap);
6406 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
6407 etc.@: functions and can contain implementation specific flags on what
6408 additional security measures the checking function might take, such as
6409 handling @code{%n} differently.
6411 The @var{os} argument is the object size @var{s} points to, like in the
6412 other built-in functions. There is a small difference in the behavior
6413 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
6414 optimized into the non-checking functions only if @var{flag} is 0, otherwise
6415 the checking function is called with @var{os} argument set to
6418 In addition to this, there are checking built-in functions
6419 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
6420 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
6421 These have just one additional argument, @var{flag}, right before
6422 format string @var{fmt}. If the compiler is able to optimize them to
6423 @code{fputc} etc.@: functions, it will, otherwise the checking function
6424 should be called and the @var{flag} argument passed to it.
6426 @node Other Builtins
6427 @section Other built-in functions provided by GCC
6428 @cindex built-in functions
6429 @findex __builtin_fpclassify
6430 @findex __builtin_isfinite
6431 @findex __builtin_isnormal
6432 @findex __builtin_isgreater
6433 @findex __builtin_isgreaterequal
6434 @findex __builtin_isinf_sign
6435 @findex __builtin_isless
6436 @findex __builtin_islessequal
6437 @findex __builtin_islessgreater
6438 @findex __builtin_isunordered
6439 @findex __builtin_powi
6440 @findex __builtin_powif
6441 @findex __builtin_powil
6599 @findex fprintf_unlocked
6601 @findex fputs_unlocked
6718 @findex printf_unlocked
6750 @findex significandf
6751 @findex significandl
6822 GCC provides a large number of built-in functions other than the ones
6823 mentioned above. Some of these are for internal use in the processing
6824 of exceptions or variable-length argument lists and will not be
6825 documented here because they may change from time to time; we do not
6826 recommend general use of these functions.
6828 The remaining functions are provided for optimization purposes.
6830 @opindex fno-builtin
6831 GCC includes built-in versions of many of the functions in the standard
6832 C library. The versions prefixed with @code{__builtin_} will always be
6833 treated as having the same meaning as the C library function even if you
6834 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
6835 Many of these functions are only optimized in certain cases; if they are
6836 not optimized in a particular case, a call to the library function will
6841 Outside strict ISO C mode (@option{-ansi}, @option{-std=c89} or
6842 @option{-std=c99}), the functions
6843 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
6844 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
6845 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
6846 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
6847 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
6848 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
6849 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
6850 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
6851 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
6852 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
6853 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
6854 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
6855 @code{signbitd64}, @code{signbitd128}, @code{significandf},
6856 @code{significandl}, @code{significand}, @code{sincosf},
6857 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
6858 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
6859 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
6860 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
6862 may be handled as built-in functions.
6863 All these functions have corresponding versions
6864 prefixed with @code{__builtin_}, which may be used even in strict C89
6867 The ISO C99 functions
6868 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
6869 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
6870 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
6871 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
6872 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
6873 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
6874 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
6875 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
6876 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
6877 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
6878 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
6879 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
6880 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
6881 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
6882 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
6883 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
6884 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
6885 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
6886 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
6887 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
6888 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
6889 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
6890 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
6891 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
6892 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
6893 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
6894 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
6895 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
6896 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
6897 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
6898 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
6899 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
6900 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
6901 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
6902 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
6903 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
6904 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
6905 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
6906 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
6907 are handled as built-in functions
6908 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
6910 There are also built-in versions of the ISO C99 functions
6911 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
6912 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
6913 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
6914 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
6915 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
6916 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
6917 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
6918 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
6919 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
6920 that are recognized in any mode since ISO C90 reserves these names for
6921 the purpose to which ISO C99 puts them. All these functions have
6922 corresponding versions prefixed with @code{__builtin_}.
6924 The ISO C94 functions
6925 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
6926 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
6927 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
6929 are handled as built-in functions
6930 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c89}).
6932 The ISO C90 functions
6933 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
6934 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
6935 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
6936 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
6937 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
6938 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
6939 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
6940 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
6941 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
6942 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
6943 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
6944 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
6945 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
6946 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
6947 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
6948 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
6949 are all recognized as built-in functions unless
6950 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
6951 is specified for an individual function). All of these functions have
6952 corresponding versions prefixed with @code{__builtin_}.
6954 GCC provides built-in versions of the ISO C99 floating point comparison
6955 macros that avoid raising exceptions for unordered operands. They have
6956 the same names as the standard macros ( @code{isgreater},
6957 @code{isgreaterequal}, @code{isless}, @code{islessequal},
6958 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
6959 prefixed. We intend for a library implementor to be able to simply
6960 @code{#define} each standard macro to its built-in equivalent.
6961 In the same fashion, GCC provides @code{fpclassify}, @code{isfinite},
6962 @code{isinf_sign} and @code{isnormal} built-ins used with
6963 @code{__builtin_} prefixed. The @code{isinf} and @code{isnan}
6964 builtins appear both with and without the @code{__builtin_} prefix.
6966 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
6968 You can use the built-in function @code{__builtin_types_compatible_p} to
6969 determine whether two types are the same.
6971 This built-in function returns 1 if the unqualified versions of the
6972 types @var{type1} and @var{type2} (which are types, not expressions) are
6973 compatible, 0 otherwise. The result of this built-in function can be
6974 used in integer constant expressions.
6976 This built-in function ignores top level qualifiers (e.g., @code{const},
6977 @code{volatile}). For example, @code{int} is equivalent to @code{const
6980 The type @code{int[]} and @code{int[5]} are compatible. On the other
6981 hand, @code{int} and @code{char *} are not compatible, even if the size
6982 of their types, on the particular architecture are the same. Also, the
6983 amount of pointer indirection is taken into account when determining
6984 similarity. Consequently, @code{short *} is not similar to
6985 @code{short **}. Furthermore, two types that are typedefed are
6986 considered compatible if their underlying types are compatible.
6988 An @code{enum} type is not considered to be compatible with another
6989 @code{enum} type even if both are compatible with the same integer
6990 type; this is what the C standard specifies.
6991 For example, @code{enum @{foo, bar@}} is not similar to
6992 @code{enum @{hot, dog@}}.
6994 You would typically use this function in code whose execution varies
6995 depending on the arguments' types. For example:
7000 typeof (x) tmp = (x); \
7001 if (__builtin_types_compatible_p (typeof (x), long double)) \
7002 tmp = foo_long_double (tmp); \
7003 else if (__builtin_types_compatible_p (typeof (x), double)) \
7004 tmp = foo_double (tmp); \
7005 else if (__builtin_types_compatible_p (typeof (x), float)) \
7006 tmp = foo_float (tmp); \
7013 @emph{Note:} This construct is only available for C@.
7017 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
7019 You can use the built-in function @code{__builtin_choose_expr} to
7020 evaluate code depending on the value of a constant expression. This
7021 built-in function returns @var{exp1} if @var{const_exp}, which is an
7022 integer constant expression, is nonzero. Otherwise it returns 0.
7024 This built-in function is analogous to the @samp{? :} operator in C,
7025 except that the expression returned has its type unaltered by promotion
7026 rules. Also, the built-in function does not evaluate the expression
7027 that was not chosen. For example, if @var{const_exp} evaluates to true,
7028 @var{exp2} is not evaluated even if it has side-effects.
7030 This built-in function can return an lvalue if the chosen argument is an
7033 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
7034 type. Similarly, if @var{exp2} is returned, its return type is the same
7041 __builtin_choose_expr ( \
7042 __builtin_types_compatible_p (typeof (x), double), \
7044 __builtin_choose_expr ( \
7045 __builtin_types_compatible_p (typeof (x), float), \
7047 /* @r{The void expression results in a compile-time error} \
7048 @r{when assigning the result to something.} */ \
7052 @emph{Note:} This construct is only available for C@. Furthermore, the
7053 unused expression (@var{exp1} or @var{exp2} depending on the value of
7054 @var{const_exp}) may still generate syntax errors. This may change in
7059 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
7060 You can use the built-in function @code{__builtin_constant_p} to
7061 determine if a value is known to be constant at compile-time and hence
7062 that GCC can perform constant-folding on expressions involving that
7063 value. The argument of the function is the value to test. The function
7064 returns the integer 1 if the argument is known to be a compile-time
7065 constant and 0 if it is not known to be a compile-time constant. A
7066 return of 0 does not indicate that the value is @emph{not} a constant,
7067 but merely that GCC cannot prove it is a constant with the specified
7068 value of the @option{-O} option.
7070 You would typically use this function in an embedded application where
7071 memory was a critical resource. If you have some complex calculation,
7072 you may want it to be folded if it involves constants, but need to call
7073 a function if it does not. For example:
7076 #define Scale_Value(X) \
7077 (__builtin_constant_p (X) \
7078 ? ((X) * SCALE + OFFSET) : Scale (X))
7081 You may use this built-in function in either a macro or an inline
7082 function. However, if you use it in an inlined function and pass an
7083 argument of the function as the argument to the built-in, GCC will
7084 never return 1 when you call the inline function with a string constant
7085 or compound literal (@pxref{Compound Literals}) and will not return 1
7086 when you pass a constant numeric value to the inline function unless you
7087 specify the @option{-O} option.
7089 You may also use @code{__builtin_constant_p} in initializers for static
7090 data. For instance, you can write
7093 static const int table[] = @{
7094 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
7100 This is an acceptable initializer even if @var{EXPRESSION} is not a
7101 constant expression, including the case where
7102 @code{__builtin_constant_p} returns 1 because @var{EXPRESSION} can be
7103 folded to a constant but @var{EXPRESSION} contains operands that would
7104 not otherwise be permitted in a static initializer (for example,
7105 @code{0 && foo ()}). GCC must be more conservative about evaluating the
7106 built-in in this case, because it has no opportunity to perform
7109 Previous versions of GCC did not accept this built-in in data
7110 initializers. The earliest version where it is completely safe is
7114 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
7115 @opindex fprofile-arcs
7116 You may use @code{__builtin_expect} to provide the compiler with
7117 branch prediction information. In general, you should prefer to
7118 use actual profile feedback for this (@option{-fprofile-arcs}), as
7119 programmers are notoriously bad at predicting how their programs
7120 actually perform. However, there are applications in which this
7121 data is hard to collect.
7123 The return value is the value of @var{exp}, which should be an integral
7124 expression. The semantics of the built-in are that it is expected that
7125 @var{exp} == @var{c}. For example:
7128 if (__builtin_expect (x, 0))
7133 would indicate that we do not expect to call @code{foo}, since
7134 we expect @code{x} to be zero. Since you are limited to integral
7135 expressions for @var{exp}, you should use constructions such as
7138 if (__builtin_expect (ptr != NULL, 1))
7143 when testing pointer or floating-point values.
7146 @deftypefn {Built-in Function} void __builtin_trap (void)
7147 This function causes the program to exit abnormally. GCC implements
7148 this function by using a target-dependent mechanism (such as
7149 intentionally executing an illegal instruction) or by calling
7150 @code{abort}. The mechanism used may vary from release to release so
7151 you should not rely on any particular implementation.
7154 @deftypefn {Built-in Function} void __builtin_unreachable (void)
7155 If control flow reaches the point of the @code{__builtin_unreachable},
7156 the program is undefined. It is useful in situations where the
7157 compiler cannot deduce the unreachability of the code.
7159 One such case is immediately following an @code{asm} statement that
7160 will either never terminate, or one that transfers control elsewhere
7161 and never returns. In this example, without the
7162 @code{__builtin_unreachable}, GCC would issue a warning that control
7163 reaches the end of a non-void function. It would also generate code
7164 to return after the @code{asm}.
7167 int f (int c, int v)
7175 asm("jmp error_handler");
7176 __builtin_unreachable ();
7181 Because the @code{asm} statement unconditionally transfers control out
7182 of the function, control will never reach the end of the function
7183 body. The @code{__builtin_unreachable} is in fact unreachable and
7184 communicates this fact to the compiler.
7186 Another use for @code{__builtin_unreachable} is following a call a
7187 function that never returns but that is not declared
7188 @code{__attribute__((noreturn))}, as in this example:
7191 void function_that_never_returns (void);
7201 function_that_never_returns ();
7202 __builtin_unreachable ();
7209 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
7210 This function is used to flush the processor's instruction cache for
7211 the region of memory between @var{begin} inclusive and @var{end}
7212 exclusive. Some targets require that the instruction cache be
7213 flushed, after modifying memory containing code, in order to obtain
7214 deterministic behavior.
7216 If the target does not require instruction cache flushes,
7217 @code{__builtin___clear_cache} has no effect. Otherwise either
7218 instructions are emitted in-line to clear the instruction cache or a
7219 call to the @code{__clear_cache} function in libgcc is made.
7222 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
7223 This function is used to minimize cache-miss latency by moving data into
7224 a cache before it is accessed.
7225 You can insert calls to @code{__builtin_prefetch} into code for which
7226 you know addresses of data in memory that is likely to be accessed soon.
7227 If the target supports them, data prefetch instructions will be generated.
7228 If the prefetch is done early enough before the access then the data will
7229 be in the cache by the time it is accessed.
7231 The value of @var{addr} is the address of the memory to prefetch.
7232 There are two optional arguments, @var{rw} and @var{locality}.
7233 The value of @var{rw} is a compile-time constant one or zero; one
7234 means that the prefetch is preparing for a write to the memory address
7235 and zero, the default, means that the prefetch is preparing for a read.
7236 The value @var{locality} must be a compile-time constant integer between
7237 zero and three. A value of zero means that the data has no temporal
7238 locality, so it need not be left in the cache after the access. A value
7239 of three means that the data has a high degree of temporal locality and
7240 should be left in all levels of cache possible. Values of one and two
7241 mean, respectively, a low or moderate degree of temporal locality. The
7245 for (i = 0; i < n; i++)
7248 __builtin_prefetch (&a[i+j], 1, 1);
7249 __builtin_prefetch (&b[i+j], 0, 1);
7254 Data prefetch does not generate faults if @var{addr} is invalid, but
7255 the address expression itself must be valid. For example, a prefetch
7256 of @code{p->next} will not fault if @code{p->next} is not a valid
7257 address, but evaluation will fault if @code{p} is not a valid address.
7259 If the target does not support data prefetch, the address expression
7260 is evaluated if it includes side effects but no other code is generated
7261 and GCC does not issue a warning.
7264 @deftypefn {Built-in Function} double __builtin_huge_val (void)
7265 Returns a positive infinity, if supported by the floating-point format,
7266 else @code{DBL_MAX}. This function is suitable for implementing the
7267 ISO C macro @code{HUGE_VAL}.
7270 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
7271 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
7274 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
7275 Similar to @code{__builtin_huge_val}, except the return
7276 type is @code{long double}.
7279 @deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...)
7280 This built-in implements the C99 fpclassify functionality. The first
7281 five int arguments should be the target library's notion of the
7282 possible FP classes and are used for return values. They must be
7283 constant values and they must appear in this order: @code{FP_NAN},
7284 @code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and
7285 @code{FP_ZERO}. The ellipsis is for exactly one floating point value
7286 to classify. GCC treats the last argument as type-generic, which
7287 means it does not do default promotion from float to double.
7290 @deftypefn {Built-in Function} double __builtin_inf (void)
7291 Similar to @code{__builtin_huge_val}, except a warning is generated
7292 if the target floating-point format does not support infinities.
7295 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
7296 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
7299 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
7300 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
7303 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
7304 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
7307 @deftypefn {Built-in Function} float __builtin_inff (void)
7308 Similar to @code{__builtin_inf}, except the return type is @code{float}.
7309 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
7312 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
7313 Similar to @code{__builtin_inf}, except the return
7314 type is @code{long double}.
7317 @deftypefn {Built-in Function} int __builtin_isinf_sign (...)
7318 Similar to @code{isinf}, except the return value will be negative for
7319 an argument of @code{-Inf}. Note while the parameter list is an
7320 ellipsis, this function only accepts exactly one floating point
7321 argument. GCC treats this parameter as type-generic, which means it
7322 does not do default promotion from float to double.
7325 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
7326 This is an implementation of the ISO C99 function @code{nan}.
7328 Since ISO C99 defines this function in terms of @code{strtod}, which we
7329 do not implement, a description of the parsing is in order. The string
7330 is parsed as by @code{strtol}; that is, the base is recognized by
7331 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
7332 in the significand such that the least significant bit of the number
7333 is at the least significant bit of the significand. The number is
7334 truncated to fit the significand field provided. The significand is
7335 forced to be a quiet NaN@.
7337 This function, if given a string literal all of which would have been
7338 consumed by strtol, is evaluated early enough that it is considered a
7339 compile-time constant.
7342 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
7343 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
7346 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
7347 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
7350 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
7351 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
7354 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
7355 Similar to @code{__builtin_nan}, except the return type is @code{float}.
7358 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
7359 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
7362 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
7363 Similar to @code{__builtin_nan}, except the significand is forced
7364 to be a signaling NaN@. The @code{nans} function is proposed by
7365 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
7368 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
7369 Similar to @code{__builtin_nans}, except the return type is @code{float}.
7372 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
7373 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
7376 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
7377 Returns one plus the index of the least significant 1-bit of @var{x}, or
7378 if @var{x} is zero, returns zero.
7381 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
7382 Returns the number of leading 0-bits in @var{x}, starting at the most
7383 significant bit position. If @var{x} is 0, the result is undefined.
7386 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
7387 Returns the number of trailing 0-bits in @var{x}, starting at the least
7388 significant bit position. If @var{x} is 0, the result is undefined.
7391 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
7392 Returns the number of 1-bits in @var{x}.
7395 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
7396 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
7400 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
7401 Similar to @code{__builtin_ffs}, except the argument type is
7402 @code{unsigned long}.
7405 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
7406 Similar to @code{__builtin_clz}, except the argument type is
7407 @code{unsigned long}.
7410 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
7411 Similar to @code{__builtin_ctz}, except the argument type is
7412 @code{unsigned long}.
7415 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
7416 Similar to @code{__builtin_popcount}, except the argument type is
7417 @code{unsigned long}.
7420 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
7421 Similar to @code{__builtin_parity}, except the argument type is
7422 @code{unsigned long}.
7425 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
7426 Similar to @code{__builtin_ffs}, except the argument type is
7427 @code{unsigned long long}.
7430 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
7431 Similar to @code{__builtin_clz}, except the argument type is
7432 @code{unsigned long long}.
7435 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
7436 Similar to @code{__builtin_ctz}, except the argument type is
7437 @code{unsigned long long}.
7440 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
7441 Similar to @code{__builtin_popcount}, except the argument type is
7442 @code{unsigned long long}.
7445 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
7446 Similar to @code{__builtin_parity}, except the argument type is
7447 @code{unsigned long long}.
7450 @deftypefn {Built-in Function} double __builtin_powi (double, int)
7451 Returns the first argument raised to the power of the second. Unlike the
7452 @code{pow} function no guarantees about precision and rounding are made.
7455 @deftypefn {Built-in Function} float __builtin_powif (float, int)
7456 Similar to @code{__builtin_powi}, except the argument and return types
7460 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
7461 Similar to @code{__builtin_powi}, except the argument and return types
7462 are @code{long double}.
7465 @deftypefn {Built-in Function} int32_t __builtin_bswap32 (int32_t x)
7466 Returns @var{x} with the order of the bytes reversed; for example,
7467 @code{0xaabbccdd} becomes @code{0xddccbbaa}. Byte here always means
7471 @deftypefn {Built-in Function} int64_t __builtin_bswap64 (int64_t x)
7472 Similar to @code{__builtin_bswap32}, except the argument and return types
7476 @node Target Builtins
7477 @section Built-in Functions Specific to Particular Target Machines
7479 On some target machines, GCC supports many built-in functions specific
7480 to those machines. Generally these generate calls to specific machine
7481 instructions, but allow the compiler to schedule those calls.
7484 * Alpha Built-in Functions::
7485 * ARM iWMMXt Built-in Functions::
7486 * ARM NEON Intrinsics::
7487 * Blackfin Built-in Functions::
7488 * FR-V Built-in Functions::
7489 * X86 Built-in Functions::
7490 * MIPS DSP Built-in Functions::
7491 * MIPS Paired-Single Support::
7492 * MIPS Loongson Built-in Functions::
7493 * Other MIPS Built-in Functions::
7494 * picoChip Built-in Functions::
7495 * PowerPC AltiVec/VSX Built-in Functions::
7496 * RX Built-in Functions::
7497 * SPARC VIS Built-in Functions::
7498 * SPU Built-in Functions::
7501 @node Alpha Built-in Functions
7502 @subsection Alpha Built-in Functions
7504 These built-in functions are available for the Alpha family of
7505 processors, depending on the command-line switches used.
7507 The following built-in functions are always available. They
7508 all generate the machine instruction that is part of the name.
7511 long __builtin_alpha_implver (void)
7512 long __builtin_alpha_rpcc (void)
7513 long __builtin_alpha_amask (long)
7514 long __builtin_alpha_cmpbge (long, long)
7515 long __builtin_alpha_extbl (long, long)
7516 long __builtin_alpha_extwl (long, long)
7517 long __builtin_alpha_extll (long, long)
7518 long __builtin_alpha_extql (long, long)
7519 long __builtin_alpha_extwh (long, long)
7520 long __builtin_alpha_extlh (long, long)
7521 long __builtin_alpha_extqh (long, long)
7522 long __builtin_alpha_insbl (long, long)
7523 long __builtin_alpha_inswl (long, long)
7524 long __builtin_alpha_insll (long, long)
7525 long __builtin_alpha_insql (long, long)
7526 long __builtin_alpha_inswh (long, long)
7527 long __builtin_alpha_inslh (long, long)
7528 long __builtin_alpha_insqh (long, long)
7529 long __builtin_alpha_mskbl (long, long)
7530 long __builtin_alpha_mskwl (long, long)
7531 long __builtin_alpha_mskll (long, long)
7532 long __builtin_alpha_mskql (long, long)
7533 long __builtin_alpha_mskwh (long, long)
7534 long __builtin_alpha_msklh (long, long)
7535 long __builtin_alpha_mskqh (long, long)
7536 long __builtin_alpha_umulh (long, long)
7537 long __builtin_alpha_zap (long, long)
7538 long __builtin_alpha_zapnot (long, long)
7541 The following built-in functions are always with @option{-mmax}
7542 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
7543 later. They all generate the machine instruction that is part
7547 long __builtin_alpha_pklb (long)
7548 long __builtin_alpha_pkwb (long)
7549 long __builtin_alpha_unpkbl (long)
7550 long __builtin_alpha_unpkbw (long)
7551 long __builtin_alpha_minub8 (long, long)
7552 long __builtin_alpha_minsb8 (long, long)
7553 long __builtin_alpha_minuw4 (long, long)
7554 long __builtin_alpha_minsw4 (long, long)
7555 long __builtin_alpha_maxub8 (long, long)
7556 long __builtin_alpha_maxsb8 (long, long)
7557 long __builtin_alpha_maxuw4 (long, long)
7558 long __builtin_alpha_maxsw4 (long, long)
7559 long __builtin_alpha_perr (long, long)
7562 The following built-in functions are always with @option{-mcix}
7563 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
7564 later. They all generate the machine instruction that is part
7568 long __builtin_alpha_cttz (long)
7569 long __builtin_alpha_ctlz (long)
7570 long __builtin_alpha_ctpop (long)
7573 The following builtins are available on systems that use the OSF/1
7574 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
7575 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
7576 @code{rdval} and @code{wrval}.
7579 void *__builtin_thread_pointer (void)
7580 void __builtin_set_thread_pointer (void *)
7583 @node ARM iWMMXt Built-in Functions
7584 @subsection ARM iWMMXt Built-in Functions
7586 These built-in functions are available for the ARM family of
7587 processors when the @option{-mcpu=iwmmxt} switch is used:
7590 typedef int v2si __attribute__ ((vector_size (8)));
7591 typedef short v4hi __attribute__ ((vector_size (8)));
7592 typedef char v8qi __attribute__ ((vector_size (8)));
7594 int __builtin_arm_getwcx (int)
7595 void __builtin_arm_setwcx (int, int)
7596 int __builtin_arm_textrmsb (v8qi, int)
7597 int __builtin_arm_textrmsh (v4hi, int)
7598 int __builtin_arm_textrmsw (v2si, int)
7599 int __builtin_arm_textrmub (v8qi, int)
7600 int __builtin_arm_textrmuh (v4hi, int)
7601 int __builtin_arm_textrmuw (v2si, int)
7602 v8qi __builtin_arm_tinsrb (v8qi, int)
7603 v4hi __builtin_arm_tinsrh (v4hi, int)
7604 v2si __builtin_arm_tinsrw (v2si, int)
7605 long long __builtin_arm_tmia (long long, int, int)
7606 long long __builtin_arm_tmiabb (long long, int, int)
7607 long long __builtin_arm_tmiabt (long long, int, int)
7608 long long __builtin_arm_tmiaph (long long, int, int)
7609 long long __builtin_arm_tmiatb (long long, int, int)
7610 long long __builtin_arm_tmiatt (long long, int, int)
7611 int __builtin_arm_tmovmskb (v8qi)
7612 int __builtin_arm_tmovmskh (v4hi)
7613 int __builtin_arm_tmovmskw (v2si)
7614 long long __builtin_arm_waccb (v8qi)
7615 long long __builtin_arm_wacch (v4hi)
7616 long long __builtin_arm_waccw (v2si)
7617 v8qi __builtin_arm_waddb (v8qi, v8qi)
7618 v8qi __builtin_arm_waddbss (v8qi, v8qi)
7619 v8qi __builtin_arm_waddbus (v8qi, v8qi)
7620 v4hi __builtin_arm_waddh (v4hi, v4hi)
7621 v4hi __builtin_arm_waddhss (v4hi, v4hi)
7622 v4hi __builtin_arm_waddhus (v4hi, v4hi)
7623 v2si __builtin_arm_waddw (v2si, v2si)
7624 v2si __builtin_arm_waddwss (v2si, v2si)
7625 v2si __builtin_arm_waddwus (v2si, v2si)
7626 v8qi __builtin_arm_walign (v8qi, v8qi, int)
7627 long long __builtin_arm_wand(long long, long long)
7628 long long __builtin_arm_wandn (long long, long long)
7629 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
7630 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
7631 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
7632 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
7633 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
7634 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
7635 v2si __builtin_arm_wcmpeqw (v2si, v2si)
7636 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
7637 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
7638 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
7639 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
7640 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
7641 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
7642 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
7643 long long __builtin_arm_wmacsz (v4hi, v4hi)
7644 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
7645 long long __builtin_arm_wmacuz (v4hi, v4hi)
7646 v4hi __builtin_arm_wmadds (v4hi, v4hi)
7647 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
7648 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
7649 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
7650 v2si __builtin_arm_wmaxsw (v2si, v2si)
7651 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
7652 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
7653 v2si __builtin_arm_wmaxuw (v2si, v2si)
7654 v8qi __builtin_arm_wminsb (v8qi, v8qi)
7655 v4hi __builtin_arm_wminsh (v4hi, v4hi)
7656 v2si __builtin_arm_wminsw (v2si, v2si)
7657 v8qi __builtin_arm_wminub (v8qi, v8qi)
7658 v4hi __builtin_arm_wminuh (v4hi, v4hi)
7659 v2si __builtin_arm_wminuw (v2si, v2si)
7660 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
7661 v4hi __builtin_arm_wmulul (v4hi, v4hi)
7662 v4hi __builtin_arm_wmulum (v4hi, v4hi)
7663 long long __builtin_arm_wor (long long, long long)
7664 v2si __builtin_arm_wpackdss (long long, long long)
7665 v2si __builtin_arm_wpackdus (long long, long long)
7666 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
7667 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
7668 v4hi __builtin_arm_wpackwss (v2si, v2si)
7669 v4hi __builtin_arm_wpackwus (v2si, v2si)
7670 long long __builtin_arm_wrord (long long, long long)
7671 long long __builtin_arm_wrordi (long long, int)
7672 v4hi __builtin_arm_wrorh (v4hi, long long)
7673 v4hi __builtin_arm_wrorhi (v4hi, int)
7674 v2si __builtin_arm_wrorw (v2si, long long)
7675 v2si __builtin_arm_wrorwi (v2si, int)
7676 v2si __builtin_arm_wsadb (v8qi, v8qi)
7677 v2si __builtin_arm_wsadbz (v8qi, v8qi)
7678 v2si __builtin_arm_wsadh (v4hi, v4hi)
7679 v2si __builtin_arm_wsadhz (v4hi, v4hi)
7680 v4hi __builtin_arm_wshufh (v4hi, int)
7681 long long __builtin_arm_wslld (long long, long long)
7682 long long __builtin_arm_wslldi (long long, int)
7683 v4hi __builtin_arm_wsllh (v4hi, long long)
7684 v4hi __builtin_arm_wsllhi (v4hi, int)
7685 v2si __builtin_arm_wsllw (v2si, long long)
7686 v2si __builtin_arm_wsllwi (v2si, int)
7687 long long __builtin_arm_wsrad (long long, long long)
7688 long long __builtin_arm_wsradi (long long, int)
7689 v4hi __builtin_arm_wsrah (v4hi, long long)
7690 v4hi __builtin_arm_wsrahi (v4hi, int)
7691 v2si __builtin_arm_wsraw (v2si, long long)
7692 v2si __builtin_arm_wsrawi (v2si, int)
7693 long long __builtin_arm_wsrld (long long, long long)
7694 long long __builtin_arm_wsrldi (long long, int)
7695 v4hi __builtin_arm_wsrlh (v4hi, long long)
7696 v4hi __builtin_arm_wsrlhi (v4hi, int)
7697 v2si __builtin_arm_wsrlw (v2si, long long)
7698 v2si __builtin_arm_wsrlwi (v2si, int)
7699 v8qi __builtin_arm_wsubb (v8qi, v8qi)
7700 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
7701 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
7702 v4hi __builtin_arm_wsubh (v4hi, v4hi)
7703 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
7704 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
7705 v2si __builtin_arm_wsubw (v2si, v2si)
7706 v2si __builtin_arm_wsubwss (v2si, v2si)
7707 v2si __builtin_arm_wsubwus (v2si, v2si)
7708 v4hi __builtin_arm_wunpckehsb (v8qi)
7709 v2si __builtin_arm_wunpckehsh (v4hi)
7710 long long __builtin_arm_wunpckehsw (v2si)
7711 v4hi __builtin_arm_wunpckehub (v8qi)
7712 v2si __builtin_arm_wunpckehuh (v4hi)
7713 long long __builtin_arm_wunpckehuw (v2si)
7714 v4hi __builtin_arm_wunpckelsb (v8qi)
7715 v2si __builtin_arm_wunpckelsh (v4hi)
7716 long long __builtin_arm_wunpckelsw (v2si)
7717 v4hi __builtin_arm_wunpckelub (v8qi)
7718 v2si __builtin_arm_wunpckeluh (v4hi)
7719 long long __builtin_arm_wunpckeluw (v2si)
7720 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
7721 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
7722 v2si __builtin_arm_wunpckihw (v2si, v2si)
7723 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
7724 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
7725 v2si __builtin_arm_wunpckilw (v2si, v2si)
7726 long long __builtin_arm_wxor (long long, long long)
7727 long long __builtin_arm_wzero ()
7730 @node ARM NEON Intrinsics
7731 @subsection ARM NEON Intrinsics
7733 These built-in intrinsics for the ARM Advanced SIMD extension are available
7734 when the @option{-mfpu=neon} switch is used:
7736 @include arm-neon-intrinsics.texi
7738 @node Blackfin Built-in Functions
7739 @subsection Blackfin Built-in Functions
7741 Currently, there are two Blackfin-specific built-in functions. These are
7742 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
7743 using inline assembly; by using these built-in functions the compiler can
7744 automatically add workarounds for hardware errata involving these
7745 instructions. These functions are named as follows:
7748 void __builtin_bfin_csync (void)
7749 void __builtin_bfin_ssync (void)
7752 @node FR-V Built-in Functions
7753 @subsection FR-V Built-in Functions
7755 GCC provides many FR-V-specific built-in functions. In general,
7756 these functions are intended to be compatible with those described
7757 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
7758 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
7759 @code{__MBTOHE}, the gcc forms of which pass 128-bit values by
7760 pointer rather than by value.
7762 Most of the functions are named after specific FR-V instructions.
7763 Such functions are said to be ``directly mapped'' and are summarized
7764 here in tabular form.
7768 * Directly-mapped Integer Functions::
7769 * Directly-mapped Media Functions::
7770 * Raw read/write Functions::
7771 * Other Built-in Functions::
7774 @node Argument Types
7775 @subsubsection Argument Types
7777 The arguments to the built-in functions can be divided into three groups:
7778 register numbers, compile-time constants and run-time values. In order
7779 to make this classification clear at a glance, the arguments and return
7780 values are given the following pseudo types:
7782 @multitable @columnfractions .20 .30 .15 .35
7783 @item Pseudo type @tab Real C type @tab Constant? @tab Description
7784 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
7785 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
7786 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
7787 @item @code{uw2} @tab @code{unsigned long long} @tab No
7788 @tab an unsigned doubleword
7789 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
7790 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
7791 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
7792 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
7795 These pseudo types are not defined by GCC, they are simply a notational
7796 convenience used in this manual.
7798 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
7799 and @code{sw2} are evaluated at run time. They correspond to
7800 register operands in the underlying FR-V instructions.
7802 @code{const} arguments represent immediate operands in the underlying
7803 FR-V instructions. They must be compile-time constants.
7805 @code{acc} arguments are evaluated at compile time and specify the number
7806 of an accumulator register. For example, an @code{acc} argument of 2
7807 will select the ACC2 register.
7809 @code{iacc} arguments are similar to @code{acc} arguments but specify the
7810 number of an IACC register. See @pxref{Other Built-in Functions}
7813 @node Directly-mapped Integer Functions
7814 @subsubsection Directly-mapped Integer Functions
7816 The functions listed below map directly to FR-V I-type instructions.
7818 @multitable @columnfractions .45 .32 .23
7819 @item Function prototype @tab Example usage @tab Assembly output
7820 @item @code{sw1 __ADDSS (sw1, sw1)}
7821 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
7822 @tab @code{ADDSS @var{a},@var{b},@var{c}}
7823 @item @code{sw1 __SCAN (sw1, sw1)}
7824 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
7825 @tab @code{SCAN @var{a},@var{b},@var{c}}
7826 @item @code{sw1 __SCUTSS (sw1)}
7827 @tab @code{@var{b} = __SCUTSS (@var{a})}
7828 @tab @code{SCUTSS @var{a},@var{b}}
7829 @item @code{sw1 __SLASS (sw1, sw1)}
7830 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
7831 @tab @code{SLASS @var{a},@var{b},@var{c}}
7832 @item @code{void __SMASS (sw1, sw1)}
7833 @tab @code{__SMASS (@var{a}, @var{b})}
7834 @tab @code{SMASS @var{a},@var{b}}
7835 @item @code{void __SMSSS (sw1, sw1)}
7836 @tab @code{__SMSSS (@var{a}, @var{b})}
7837 @tab @code{SMSSS @var{a},@var{b}}
7838 @item @code{void __SMU (sw1, sw1)}
7839 @tab @code{__SMU (@var{a}, @var{b})}
7840 @tab @code{SMU @var{a},@var{b}}
7841 @item @code{sw2 __SMUL (sw1, sw1)}
7842 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
7843 @tab @code{SMUL @var{a},@var{b},@var{c}}
7844 @item @code{sw1 __SUBSS (sw1, sw1)}
7845 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
7846 @tab @code{SUBSS @var{a},@var{b},@var{c}}
7847 @item @code{uw2 __UMUL (uw1, uw1)}
7848 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
7849 @tab @code{UMUL @var{a},@var{b},@var{c}}
7852 @node Directly-mapped Media Functions
7853 @subsubsection Directly-mapped Media Functions
7855 The functions listed below map directly to FR-V M-type instructions.
7857 @multitable @columnfractions .45 .32 .23
7858 @item Function prototype @tab Example usage @tab Assembly output
7859 @item @code{uw1 __MABSHS (sw1)}
7860 @tab @code{@var{b} = __MABSHS (@var{a})}
7861 @tab @code{MABSHS @var{a},@var{b}}
7862 @item @code{void __MADDACCS (acc, acc)}
7863 @tab @code{__MADDACCS (@var{b}, @var{a})}
7864 @tab @code{MADDACCS @var{a},@var{b}}
7865 @item @code{sw1 __MADDHSS (sw1, sw1)}
7866 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
7867 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
7868 @item @code{uw1 __MADDHUS (uw1, uw1)}
7869 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
7870 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
7871 @item @code{uw1 __MAND (uw1, uw1)}
7872 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
7873 @tab @code{MAND @var{a},@var{b},@var{c}}
7874 @item @code{void __MASACCS (acc, acc)}
7875 @tab @code{__MASACCS (@var{b}, @var{a})}
7876 @tab @code{MASACCS @var{a},@var{b}}
7877 @item @code{uw1 __MAVEH (uw1, uw1)}
7878 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
7879 @tab @code{MAVEH @var{a},@var{b},@var{c}}
7880 @item @code{uw2 __MBTOH (uw1)}
7881 @tab @code{@var{b} = __MBTOH (@var{a})}
7882 @tab @code{MBTOH @var{a},@var{b}}
7883 @item @code{void __MBTOHE (uw1 *, uw1)}
7884 @tab @code{__MBTOHE (&@var{b}, @var{a})}
7885 @tab @code{MBTOHE @var{a},@var{b}}
7886 @item @code{void __MCLRACC (acc)}
7887 @tab @code{__MCLRACC (@var{a})}
7888 @tab @code{MCLRACC @var{a}}
7889 @item @code{void __MCLRACCA (void)}
7890 @tab @code{__MCLRACCA ()}
7891 @tab @code{MCLRACCA}
7892 @item @code{uw1 __Mcop1 (uw1, uw1)}
7893 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
7894 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
7895 @item @code{uw1 __Mcop2 (uw1, uw1)}
7896 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
7897 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
7898 @item @code{uw1 __MCPLHI (uw2, const)}
7899 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
7900 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
7901 @item @code{uw1 __MCPLI (uw2, const)}
7902 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
7903 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
7904 @item @code{void __MCPXIS (acc, sw1, sw1)}
7905 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
7906 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
7907 @item @code{void __MCPXIU (acc, uw1, uw1)}
7908 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
7909 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
7910 @item @code{void __MCPXRS (acc, sw1, sw1)}
7911 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
7912 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
7913 @item @code{void __MCPXRU (acc, uw1, uw1)}
7914 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
7915 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
7916 @item @code{uw1 __MCUT (acc, uw1)}
7917 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
7918 @tab @code{MCUT @var{a},@var{b},@var{c}}
7919 @item @code{uw1 __MCUTSS (acc, sw1)}
7920 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
7921 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
7922 @item @code{void __MDADDACCS (acc, acc)}
7923 @tab @code{__MDADDACCS (@var{b}, @var{a})}
7924 @tab @code{MDADDACCS @var{a},@var{b}}
7925 @item @code{void __MDASACCS (acc, acc)}
7926 @tab @code{__MDASACCS (@var{b}, @var{a})}
7927 @tab @code{MDASACCS @var{a},@var{b}}
7928 @item @code{uw2 __MDCUTSSI (acc, const)}
7929 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
7930 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
7931 @item @code{uw2 __MDPACKH (uw2, uw2)}
7932 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
7933 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
7934 @item @code{uw2 __MDROTLI (uw2, const)}
7935 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
7936 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
7937 @item @code{void __MDSUBACCS (acc, acc)}
7938 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
7939 @tab @code{MDSUBACCS @var{a},@var{b}}
7940 @item @code{void __MDUNPACKH (uw1 *, uw2)}
7941 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
7942 @tab @code{MDUNPACKH @var{a},@var{b}}
7943 @item @code{uw2 __MEXPDHD (uw1, const)}
7944 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
7945 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
7946 @item @code{uw1 __MEXPDHW (uw1, const)}
7947 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
7948 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
7949 @item @code{uw1 __MHDSETH (uw1, const)}
7950 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
7951 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
7952 @item @code{sw1 __MHDSETS (const)}
7953 @tab @code{@var{b} = __MHDSETS (@var{a})}
7954 @tab @code{MHDSETS #@var{a},@var{b}}
7955 @item @code{uw1 __MHSETHIH (uw1, const)}
7956 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
7957 @tab @code{MHSETHIH #@var{a},@var{b}}
7958 @item @code{sw1 __MHSETHIS (sw1, const)}
7959 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
7960 @tab @code{MHSETHIS #@var{a},@var{b}}
7961 @item @code{uw1 __MHSETLOH (uw1, const)}
7962 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
7963 @tab @code{MHSETLOH #@var{a},@var{b}}
7964 @item @code{sw1 __MHSETLOS (sw1, const)}
7965 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
7966 @tab @code{MHSETLOS #@var{a},@var{b}}
7967 @item @code{uw1 __MHTOB (uw2)}
7968 @tab @code{@var{b} = __MHTOB (@var{a})}
7969 @tab @code{MHTOB @var{a},@var{b}}
7970 @item @code{void __MMACHS (acc, sw1, sw1)}
7971 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
7972 @tab @code{MMACHS @var{a},@var{b},@var{c}}
7973 @item @code{void __MMACHU (acc, uw1, uw1)}
7974 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
7975 @tab @code{MMACHU @var{a},@var{b},@var{c}}
7976 @item @code{void __MMRDHS (acc, sw1, sw1)}
7977 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
7978 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
7979 @item @code{void __MMRDHU (acc, uw1, uw1)}
7980 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
7981 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
7982 @item @code{void __MMULHS (acc, sw1, sw1)}
7983 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
7984 @tab @code{MMULHS @var{a},@var{b},@var{c}}
7985 @item @code{void __MMULHU (acc, uw1, uw1)}
7986 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
7987 @tab @code{MMULHU @var{a},@var{b},@var{c}}
7988 @item @code{void __MMULXHS (acc, sw1, sw1)}
7989 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
7990 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
7991 @item @code{void __MMULXHU (acc, uw1, uw1)}
7992 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
7993 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
7994 @item @code{uw1 __MNOT (uw1)}
7995 @tab @code{@var{b} = __MNOT (@var{a})}
7996 @tab @code{MNOT @var{a},@var{b}}
7997 @item @code{uw1 __MOR (uw1, uw1)}
7998 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
7999 @tab @code{MOR @var{a},@var{b},@var{c}}
8000 @item @code{uw1 __MPACKH (uh, uh)}
8001 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
8002 @tab @code{MPACKH @var{a},@var{b},@var{c}}
8003 @item @code{sw2 __MQADDHSS (sw2, sw2)}
8004 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
8005 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
8006 @item @code{uw2 __MQADDHUS (uw2, uw2)}
8007 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
8008 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
8009 @item @code{void __MQCPXIS (acc, sw2, sw2)}
8010 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
8011 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
8012 @item @code{void __MQCPXIU (acc, uw2, uw2)}
8013 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
8014 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
8015 @item @code{void __MQCPXRS (acc, sw2, sw2)}
8016 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
8017 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
8018 @item @code{void __MQCPXRU (acc, uw2, uw2)}
8019 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
8020 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
8021 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
8022 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
8023 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
8024 @item @code{sw2 __MQLMTHS (sw2, sw2)}
8025 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
8026 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
8027 @item @code{void __MQMACHS (acc, sw2, sw2)}
8028 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
8029 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
8030 @item @code{void __MQMACHU (acc, uw2, uw2)}
8031 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
8032 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
8033 @item @code{void __MQMACXHS (acc, sw2, sw2)}
8034 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
8035 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
8036 @item @code{void __MQMULHS (acc, sw2, sw2)}
8037 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
8038 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
8039 @item @code{void __MQMULHU (acc, uw2, uw2)}
8040 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
8041 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
8042 @item @code{void __MQMULXHS (acc, sw2, sw2)}
8043 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
8044 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
8045 @item @code{void __MQMULXHU (acc, uw2, uw2)}
8046 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
8047 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
8048 @item @code{sw2 __MQSATHS (sw2, sw2)}
8049 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
8050 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
8051 @item @code{uw2 __MQSLLHI (uw2, int)}
8052 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
8053 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
8054 @item @code{sw2 __MQSRAHI (sw2, int)}
8055 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
8056 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
8057 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
8058 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
8059 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
8060 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
8061 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
8062 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
8063 @item @code{void __MQXMACHS (acc, sw2, sw2)}
8064 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
8065 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
8066 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
8067 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
8068 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
8069 @item @code{uw1 __MRDACC (acc)}
8070 @tab @code{@var{b} = __MRDACC (@var{a})}
8071 @tab @code{MRDACC @var{a},@var{b}}
8072 @item @code{uw1 __MRDACCG (acc)}
8073 @tab @code{@var{b} = __MRDACCG (@var{a})}
8074 @tab @code{MRDACCG @var{a},@var{b}}
8075 @item @code{uw1 __MROTLI (uw1, const)}
8076 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
8077 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
8078 @item @code{uw1 __MROTRI (uw1, const)}
8079 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
8080 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
8081 @item @code{sw1 __MSATHS (sw1, sw1)}
8082 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
8083 @tab @code{MSATHS @var{a},@var{b},@var{c}}
8084 @item @code{uw1 __MSATHU (uw1, uw1)}
8085 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
8086 @tab @code{MSATHU @var{a},@var{b},@var{c}}
8087 @item @code{uw1 __MSLLHI (uw1, const)}
8088 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
8089 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
8090 @item @code{sw1 __MSRAHI (sw1, const)}
8091 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
8092 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
8093 @item @code{uw1 __MSRLHI (uw1, const)}
8094 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
8095 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
8096 @item @code{void __MSUBACCS (acc, acc)}
8097 @tab @code{__MSUBACCS (@var{b}, @var{a})}
8098 @tab @code{MSUBACCS @var{a},@var{b}}
8099 @item @code{sw1 __MSUBHSS (sw1, sw1)}
8100 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
8101 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
8102 @item @code{uw1 __MSUBHUS (uw1, uw1)}
8103 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
8104 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
8105 @item @code{void __MTRAP (void)}
8106 @tab @code{__MTRAP ()}
8108 @item @code{uw2 __MUNPACKH (uw1)}
8109 @tab @code{@var{b} = __MUNPACKH (@var{a})}
8110 @tab @code{MUNPACKH @var{a},@var{b}}
8111 @item @code{uw1 __MWCUT (uw2, uw1)}
8112 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
8113 @tab @code{MWCUT @var{a},@var{b},@var{c}}
8114 @item @code{void __MWTACC (acc, uw1)}
8115 @tab @code{__MWTACC (@var{b}, @var{a})}
8116 @tab @code{MWTACC @var{a},@var{b}}
8117 @item @code{void __MWTACCG (acc, uw1)}
8118 @tab @code{__MWTACCG (@var{b}, @var{a})}
8119 @tab @code{MWTACCG @var{a},@var{b}}
8120 @item @code{uw1 __MXOR (uw1, uw1)}
8121 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
8122 @tab @code{MXOR @var{a},@var{b},@var{c}}
8125 @node Raw read/write Functions
8126 @subsubsection Raw read/write Functions
8128 This sections describes built-in functions related to read and write
8129 instructions to access memory. These functions generate
8130 @code{membar} instructions to flush the I/O load and stores where
8131 appropriate, as described in Fujitsu's manual described above.
8135 @item unsigned char __builtin_read8 (void *@var{data})
8136 @item unsigned short __builtin_read16 (void *@var{data})
8137 @item unsigned long __builtin_read32 (void *@var{data})
8138 @item unsigned long long __builtin_read64 (void *@var{data})
8140 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
8141 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
8142 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
8143 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
8146 @node Other Built-in Functions
8147 @subsubsection Other Built-in Functions
8149 This section describes built-in functions that are not named after
8150 a specific FR-V instruction.
8153 @item sw2 __IACCreadll (iacc @var{reg})
8154 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
8155 for future expansion and must be 0.
8157 @item sw1 __IACCreadl (iacc @var{reg})
8158 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
8159 Other values of @var{reg} are rejected as invalid.
8161 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
8162 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
8163 is reserved for future expansion and must be 0.
8165 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
8166 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
8167 is 1. Other values of @var{reg} are rejected as invalid.
8169 @item void __data_prefetch0 (const void *@var{x})
8170 Use the @code{dcpl} instruction to load the contents of address @var{x}
8171 into the data cache.
8173 @item void __data_prefetch (const void *@var{x})
8174 Use the @code{nldub} instruction to load the contents of address @var{x}
8175 into the data cache. The instruction will be issued in slot I1@.
8178 @node X86 Built-in Functions
8179 @subsection X86 Built-in Functions
8181 These built-in functions are available for the i386 and x86-64 family
8182 of computers, depending on the command-line switches used.
8184 Note that, if you specify command-line switches such as @option{-msse},
8185 the compiler could use the extended instruction sets even if the built-ins
8186 are not used explicitly in the program. For this reason, applications
8187 which perform runtime CPU detection must compile separate files for each
8188 supported architecture, using the appropriate flags. In particular,
8189 the file containing the CPU detection code should be compiled without
8192 The following machine modes are available for use with MMX built-in functions
8193 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
8194 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
8195 vector of eight 8-bit integers. Some of the built-in functions operate on
8196 MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode.
8198 If 3Dnow extensions are enabled, @code{V2SF} is used as a mode for a vector
8199 of two 32-bit floating point values.
8201 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
8202 floating point values. Some instructions use a vector of four 32-bit
8203 integers, these use @code{V4SI}. Finally, some instructions operate on an
8204 entire vector register, interpreting it as a 128-bit integer, these use mode
8207 In 64-bit mode, the x86-64 family of processors uses additional built-in
8208 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
8209 floating point and @code{TC} 128-bit complex floating point values.
8211 The following floating point built-in functions are available in 64-bit
8212 mode. All of them implement the function that is part of the name.
8215 __float128 __builtin_fabsq (__float128)
8216 __float128 __builtin_copysignq (__float128, __float128)
8219 The following floating point built-in functions are made available in the
8223 @item __float128 __builtin_infq (void)
8224 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
8225 @findex __builtin_infq
8227 @item __float128 __builtin_huge_valq (void)
8228 Similar to @code{__builtin_huge_val}, except the return type is @code{__float128}.
8229 @findex __builtin_huge_valq
8232 The following built-in functions are made available by @option{-mmmx}.
8233 All of them generate the machine instruction that is part of the name.
8236 v8qi __builtin_ia32_paddb (v8qi, v8qi)
8237 v4hi __builtin_ia32_paddw (v4hi, v4hi)
8238 v2si __builtin_ia32_paddd (v2si, v2si)
8239 v8qi __builtin_ia32_psubb (v8qi, v8qi)
8240 v4hi __builtin_ia32_psubw (v4hi, v4hi)
8241 v2si __builtin_ia32_psubd (v2si, v2si)
8242 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
8243 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
8244 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
8245 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
8246 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
8247 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
8248 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
8249 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
8250 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
8251 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
8252 di __builtin_ia32_pand (di, di)
8253 di __builtin_ia32_pandn (di,di)
8254 di __builtin_ia32_por (di, di)
8255 di __builtin_ia32_pxor (di, di)
8256 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
8257 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
8258 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
8259 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
8260 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
8261 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
8262 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
8263 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
8264 v2si __builtin_ia32_punpckhdq (v2si, v2si)
8265 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
8266 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
8267 v2si __builtin_ia32_punpckldq (v2si, v2si)
8268 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
8269 v4hi __builtin_ia32_packssdw (v2si, v2si)
8270 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
8272 v4hi __builtin_ia32_psllw (v4hi, v4hi)
8273 v2si __builtin_ia32_pslld (v2si, v2si)
8274 v1di __builtin_ia32_psllq (v1di, v1di)
8275 v4hi __builtin_ia32_psrlw (v4hi, v4hi)
8276 v2si __builtin_ia32_psrld (v2si, v2si)
8277 v1di __builtin_ia32_psrlq (v1di, v1di)
8278 v4hi __builtin_ia32_psraw (v4hi, v4hi)
8279 v2si __builtin_ia32_psrad (v2si, v2si)
8280 v4hi __builtin_ia32_psllwi (v4hi, int)
8281 v2si __builtin_ia32_pslldi (v2si, int)
8282 v1di __builtin_ia32_psllqi (v1di, int)
8283 v4hi __builtin_ia32_psrlwi (v4hi, int)
8284 v2si __builtin_ia32_psrldi (v2si, int)
8285 v1di __builtin_ia32_psrlqi (v1di, int)
8286 v4hi __builtin_ia32_psrawi (v4hi, int)
8287 v2si __builtin_ia32_psradi (v2si, int)
8291 The following built-in functions are made available either with
8292 @option{-msse}, or with a combination of @option{-m3dnow} and
8293 @option{-march=athlon}. All of them generate the machine
8294 instruction that is part of the name.
8297 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
8298 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
8299 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
8300 v1di __builtin_ia32_psadbw (v8qi, v8qi)
8301 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
8302 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
8303 v8qi __builtin_ia32_pminub (v8qi, v8qi)
8304 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
8305 int __builtin_ia32_pextrw (v4hi, int)
8306 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
8307 int __builtin_ia32_pmovmskb (v8qi)
8308 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
8309 void __builtin_ia32_movntq (di *, di)
8310 void __builtin_ia32_sfence (void)
8313 The following built-in functions are available when @option{-msse} is used.
8314 All of them generate the machine instruction that is part of the name.
8317 int __builtin_ia32_comieq (v4sf, v4sf)
8318 int __builtin_ia32_comineq (v4sf, v4sf)
8319 int __builtin_ia32_comilt (v4sf, v4sf)
8320 int __builtin_ia32_comile (v4sf, v4sf)
8321 int __builtin_ia32_comigt (v4sf, v4sf)
8322 int __builtin_ia32_comige (v4sf, v4sf)
8323 int __builtin_ia32_ucomieq (v4sf, v4sf)
8324 int __builtin_ia32_ucomineq (v4sf, v4sf)
8325 int __builtin_ia32_ucomilt (v4sf, v4sf)
8326 int __builtin_ia32_ucomile (v4sf, v4sf)
8327 int __builtin_ia32_ucomigt (v4sf, v4sf)
8328 int __builtin_ia32_ucomige (v4sf, v4sf)
8329 v4sf __builtin_ia32_addps (v4sf, v4sf)
8330 v4sf __builtin_ia32_subps (v4sf, v4sf)
8331 v4sf __builtin_ia32_mulps (v4sf, v4sf)
8332 v4sf __builtin_ia32_divps (v4sf, v4sf)
8333 v4sf __builtin_ia32_addss (v4sf, v4sf)
8334 v4sf __builtin_ia32_subss (v4sf, v4sf)
8335 v4sf __builtin_ia32_mulss (v4sf, v4sf)
8336 v4sf __builtin_ia32_divss (v4sf, v4sf)
8337 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
8338 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
8339 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
8340 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
8341 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
8342 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
8343 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
8344 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
8345 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
8346 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
8347 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
8348 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
8349 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
8350 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
8351 v4si __builtin_ia32_cmpless (v4sf, v4sf)
8352 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
8353 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
8354 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
8355 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
8356 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
8357 v4sf __builtin_ia32_maxps (v4sf, v4sf)
8358 v4sf __builtin_ia32_maxss (v4sf, v4sf)
8359 v4sf __builtin_ia32_minps (v4sf, v4sf)
8360 v4sf __builtin_ia32_minss (v4sf, v4sf)
8361 v4sf __builtin_ia32_andps (v4sf, v4sf)
8362 v4sf __builtin_ia32_andnps (v4sf, v4sf)
8363 v4sf __builtin_ia32_orps (v4sf, v4sf)
8364 v4sf __builtin_ia32_xorps (v4sf, v4sf)
8365 v4sf __builtin_ia32_movss (v4sf, v4sf)
8366 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
8367 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
8368 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
8369 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
8370 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
8371 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
8372 v2si __builtin_ia32_cvtps2pi (v4sf)
8373 int __builtin_ia32_cvtss2si (v4sf)
8374 v2si __builtin_ia32_cvttps2pi (v4sf)
8375 int __builtin_ia32_cvttss2si (v4sf)
8376 v4sf __builtin_ia32_rcpps (v4sf)
8377 v4sf __builtin_ia32_rsqrtps (v4sf)
8378 v4sf __builtin_ia32_sqrtps (v4sf)
8379 v4sf __builtin_ia32_rcpss (v4sf)
8380 v4sf __builtin_ia32_rsqrtss (v4sf)
8381 v4sf __builtin_ia32_sqrtss (v4sf)
8382 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
8383 void __builtin_ia32_movntps (float *, v4sf)
8384 int __builtin_ia32_movmskps (v4sf)
8387 The following built-in functions are available when @option{-msse} is used.
8390 @item v4sf __builtin_ia32_loadaps (float *)
8391 Generates the @code{movaps} machine instruction as a load from memory.
8392 @item void __builtin_ia32_storeaps (float *, v4sf)
8393 Generates the @code{movaps} machine instruction as a store to memory.
8394 @item v4sf __builtin_ia32_loadups (float *)
8395 Generates the @code{movups} machine instruction as a load from memory.
8396 @item void __builtin_ia32_storeups (float *, v4sf)
8397 Generates the @code{movups} machine instruction as a store to memory.
8398 @item v4sf __builtin_ia32_loadsss (float *)
8399 Generates the @code{movss} machine instruction as a load from memory.
8400 @item void __builtin_ia32_storess (float *, v4sf)
8401 Generates the @code{movss} machine instruction as a store to memory.
8402 @item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)
8403 Generates the @code{movhps} machine instruction as a load from memory.
8404 @item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)
8405 Generates the @code{movlps} machine instruction as a load from memory
8406 @item void __builtin_ia32_storehps (v2sf *, v4sf)
8407 Generates the @code{movhps} machine instruction as a store to memory.
8408 @item void __builtin_ia32_storelps (v2sf *, v4sf)
8409 Generates the @code{movlps} machine instruction as a store to memory.
8412 The following built-in functions are available when @option{-msse2} is used.
8413 All of them generate the machine instruction that is part of the name.
8416 int __builtin_ia32_comisdeq (v2df, v2df)
8417 int __builtin_ia32_comisdlt (v2df, v2df)
8418 int __builtin_ia32_comisdle (v2df, v2df)
8419 int __builtin_ia32_comisdgt (v2df, v2df)
8420 int __builtin_ia32_comisdge (v2df, v2df)
8421 int __builtin_ia32_comisdneq (v2df, v2df)
8422 int __builtin_ia32_ucomisdeq (v2df, v2df)
8423 int __builtin_ia32_ucomisdlt (v2df, v2df)
8424 int __builtin_ia32_ucomisdle (v2df, v2df)
8425 int __builtin_ia32_ucomisdgt (v2df, v2df)
8426 int __builtin_ia32_ucomisdge (v2df, v2df)
8427 int __builtin_ia32_ucomisdneq (v2df, v2df)
8428 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
8429 v2df __builtin_ia32_cmpltpd (v2df, v2df)
8430 v2df __builtin_ia32_cmplepd (v2df, v2df)
8431 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
8432 v2df __builtin_ia32_cmpgepd (v2df, v2df)
8433 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
8434 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
8435 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
8436 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
8437 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
8438 v2df __builtin_ia32_cmpngepd (v2df, v2df)
8439 v2df __builtin_ia32_cmpordpd (v2df, v2df)
8440 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
8441 v2df __builtin_ia32_cmpltsd (v2df, v2df)
8442 v2df __builtin_ia32_cmplesd (v2df, v2df)
8443 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
8444 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
8445 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
8446 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
8447 v2df __builtin_ia32_cmpordsd (v2df, v2df)
8448 v2di __builtin_ia32_paddq (v2di, v2di)
8449 v2di __builtin_ia32_psubq (v2di, v2di)
8450 v2df __builtin_ia32_addpd (v2df, v2df)
8451 v2df __builtin_ia32_subpd (v2df, v2df)
8452 v2df __builtin_ia32_mulpd (v2df, v2df)
8453 v2df __builtin_ia32_divpd (v2df, v2df)
8454 v2df __builtin_ia32_addsd (v2df, v2df)
8455 v2df __builtin_ia32_subsd (v2df, v2df)
8456 v2df __builtin_ia32_mulsd (v2df, v2df)
8457 v2df __builtin_ia32_divsd (v2df, v2df)
8458 v2df __builtin_ia32_minpd (v2df, v2df)
8459 v2df __builtin_ia32_maxpd (v2df, v2df)
8460 v2df __builtin_ia32_minsd (v2df, v2df)
8461 v2df __builtin_ia32_maxsd (v2df, v2df)
8462 v2df __builtin_ia32_andpd (v2df, v2df)
8463 v2df __builtin_ia32_andnpd (v2df, v2df)
8464 v2df __builtin_ia32_orpd (v2df, v2df)
8465 v2df __builtin_ia32_xorpd (v2df, v2df)
8466 v2df __builtin_ia32_movsd (v2df, v2df)
8467 v2df __builtin_ia32_unpckhpd (v2df, v2df)
8468 v2df __builtin_ia32_unpcklpd (v2df, v2df)
8469 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
8470 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
8471 v4si __builtin_ia32_paddd128 (v4si, v4si)
8472 v2di __builtin_ia32_paddq128 (v2di, v2di)
8473 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
8474 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
8475 v4si __builtin_ia32_psubd128 (v4si, v4si)
8476 v2di __builtin_ia32_psubq128 (v2di, v2di)
8477 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
8478 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
8479 v2di __builtin_ia32_pand128 (v2di, v2di)
8480 v2di __builtin_ia32_pandn128 (v2di, v2di)
8481 v2di __builtin_ia32_por128 (v2di, v2di)
8482 v2di __builtin_ia32_pxor128 (v2di, v2di)
8483 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
8484 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
8485 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
8486 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
8487 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
8488 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
8489 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
8490 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
8491 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
8492 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
8493 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
8494 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
8495 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
8496 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
8497 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
8498 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
8499 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
8500 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
8501 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
8502 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
8503 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi)
8504 v8hi __builtin_ia32_packssdw128 (v4si, v4si)
8505 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi)
8506 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
8507 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
8508 v2df __builtin_ia32_loadupd (double *)
8509 void __builtin_ia32_storeupd (double *, v2df)
8510 v2df __builtin_ia32_loadhpd (v2df, double const *)
8511 v2df __builtin_ia32_loadlpd (v2df, double const *)
8512 int __builtin_ia32_movmskpd (v2df)
8513 int __builtin_ia32_pmovmskb128 (v16qi)
8514 void __builtin_ia32_movnti (int *, int)
8515 void __builtin_ia32_movntpd (double *, v2df)
8516 void __builtin_ia32_movntdq (v2df *, v2df)
8517 v4si __builtin_ia32_pshufd (v4si, int)
8518 v8hi __builtin_ia32_pshuflw (v8hi, int)
8519 v8hi __builtin_ia32_pshufhw (v8hi, int)
8520 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
8521 v2df __builtin_ia32_sqrtpd (v2df)
8522 v2df __builtin_ia32_sqrtsd (v2df)
8523 v2df __builtin_ia32_shufpd (v2df, v2df, int)
8524 v2df __builtin_ia32_cvtdq2pd (v4si)
8525 v4sf __builtin_ia32_cvtdq2ps (v4si)
8526 v4si __builtin_ia32_cvtpd2dq (v2df)
8527 v2si __builtin_ia32_cvtpd2pi (v2df)
8528 v4sf __builtin_ia32_cvtpd2ps (v2df)
8529 v4si __builtin_ia32_cvttpd2dq (v2df)
8530 v2si __builtin_ia32_cvttpd2pi (v2df)
8531 v2df __builtin_ia32_cvtpi2pd (v2si)
8532 int __builtin_ia32_cvtsd2si (v2df)
8533 int __builtin_ia32_cvttsd2si (v2df)
8534 long long __builtin_ia32_cvtsd2si64 (v2df)
8535 long long __builtin_ia32_cvttsd2si64 (v2df)
8536 v4si __builtin_ia32_cvtps2dq (v4sf)
8537 v2df __builtin_ia32_cvtps2pd (v4sf)
8538 v4si __builtin_ia32_cvttps2dq (v4sf)
8539 v2df __builtin_ia32_cvtsi2sd (v2df, int)
8540 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
8541 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
8542 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
8543 void __builtin_ia32_clflush (const void *)
8544 void __builtin_ia32_lfence (void)
8545 void __builtin_ia32_mfence (void)
8546 v16qi __builtin_ia32_loaddqu (const char *)
8547 void __builtin_ia32_storedqu (char *, v16qi)
8548 v1di __builtin_ia32_pmuludq (v2si, v2si)
8549 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
8550 v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
8551 v4si __builtin_ia32_pslld128 (v4si, v4si)
8552 v2di __builtin_ia32_psllq128 (v2di, v2di)
8553 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
8554 v4si __builtin_ia32_psrld128 (v4si, v4si)
8555 v2di __builtin_ia32_psrlq128 (v2di, v2di)
8556 v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
8557 v4si __builtin_ia32_psrad128 (v4si, v4si)
8558 v2di __builtin_ia32_pslldqi128 (v2di, int)
8559 v8hi __builtin_ia32_psllwi128 (v8hi, int)
8560 v4si __builtin_ia32_pslldi128 (v4si, int)
8561 v2di __builtin_ia32_psllqi128 (v2di, int)
8562 v2di __builtin_ia32_psrldqi128 (v2di, int)
8563 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
8564 v4si __builtin_ia32_psrldi128 (v4si, int)
8565 v2di __builtin_ia32_psrlqi128 (v2di, int)
8566 v8hi __builtin_ia32_psrawi128 (v8hi, int)
8567 v4si __builtin_ia32_psradi128 (v4si, int)
8568 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
8569 v2di __builtin_ia32_movq128 (v2di)
8572 The following built-in functions are available when @option{-msse3} is used.
8573 All of them generate the machine instruction that is part of the name.
8576 v2df __builtin_ia32_addsubpd (v2df, v2df)
8577 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
8578 v2df __builtin_ia32_haddpd (v2df, v2df)
8579 v4sf __builtin_ia32_haddps (v4sf, v4sf)
8580 v2df __builtin_ia32_hsubpd (v2df, v2df)
8581 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
8582 v16qi __builtin_ia32_lddqu (char const *)
8583 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
8584 v2df __builtin_ia32_movddup (v2df)
8585 v4sf __builtin_ia32_movshdup (v4sf)
8586 v4sf __builtin_ia32_movsldup (v4sf)
8587 void __builtin_ia32_mwait (unsigned int, unsigned int)
8590 The following built-in functions are available when @option{-msse3} is used.
8593 @item v2df __builtin_ia32_loadddup (double const *)
8594 Generates the @code{movddup} machine instruction as a load from memory.
8597 The following built-in functions are available when @option{-mssse3} is used.
8598 All of them generate the machine instruction that is part of the name
8602 v2si __builtin_ia32_phaddd (v2si, v2si)
8603 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
8604 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
8605 v2si __builtin_ia32_phsubd (v2si, v2si)
8606 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
8607 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
8608 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi)
8609 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
8610 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
8611 v8qi __builtin_ia32_psignb (v8qi, v8qi)
8612 v2si __builtin_ia32_psignd (v2si, v2si)
8613 v4hi __builtin_ia32_psignw (v4hi, v4hi)
8614 v1di __builtin_ia32_palignr (v1di, v1di, int)
8615 v8qi __builtin_ia32_pabsb (v8qi)
8616 v2si __builtin_ia32_pabsd (v2si)
8617 v4hi __builtin_ia32_pabsw (v4hi)
8620 The following built-in functions are available when @option{-mssse3} is used.
8621 All of them generate the machine instruction that is part of the name
8625 v4si __builtin_ia32_phaddd128 (v4si, v4si)
8626 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
8627 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
8628 v4si __builtin_ia32_phsubd128 (v4si, v4si)
8629 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
8630 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
8631 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
8632 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
8633 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
8634 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
8635 v4si __builtin_ia32_psignd128 (v4si, v4si)
8636 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
8637 v2di __builtin_ia32_palignr128 (v2di, v2di, int)
8638 v16qi __builtin_ia32_pabsb128 (v16qi)
8639 v4si __builtin_ia32_pabsd128 (v4si)
8640 v8hi __builtin_ia32_pabsw128 (v8hi)
8643 The following built-in functions are available when @option{-msse4.1} is
8644 used. All of them generate the machine instruction that is part of the
8648 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
8649 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
8650 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
8651 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
8652 v2df __builtin_ia32_dppd (v2df, v2df, const int)
8653 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
8654 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
8655 v2di __builtin_ia32_movntdqa (v2di *);
8656 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
8657 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
8658 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
8659 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
8660 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
8661 v8hi __builtin_ia32_phminposuw128 (v8hi)
8662 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
8663 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
8664 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
8665 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
8666 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
8667 v4si __builtin_ia32_pminsd128 (v4si, v4si)
8668 v4si __builtin_ia32_pminud128 (v4si, v4si)
8669 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
8670 v4si __builtin_ia32_pmovsxbd128 (v16qi)
8671 v2di __builtin_ia32_pmovsxbq128 (v16qi)
8672 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
8673 v2di __builtin_ia32_pmovsxdq128 (v4si)
8674 v4si __builtin_ia32_pmovsxwd128 (v8hi)
8675 v2di __builtin_ia32_pmovsxwq128 (v8hi)
8676 v4si __builtin_ia32_pmovzxbd128 (v16qi)
8677 v2di __builtin_ia32_pmovzxbq128 (v16qi)
8678 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
8679 v2di __builtin_ia32_pmovzxdq128 (v4si)
8680 v4si __builtin_ia32_pmovzxwd128 (v8hi)
8681 v2di __builtin_ia32_pmovzxwq128 (v8hi)
8682 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
8683 v4si __builtin_ia32_pmulld128 (v4si, v4si)
8684 int __builtin_ia32_ptestc128 (v2di, v2di)
8685 int __builtin_ia32_ptestnzc128 (v2di, v2di)
8686 int __builtin_ia32_ptestz128 (v2di, v2di)
8687 v2df __builtin_ia32_roundpd (v2df, const int)
8688 v4sf __builtin_ia32_roundps (v4sf, const int)
8689 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
8690 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
8693 The following built-in functions are available when @option{-msse4.1} is
8697 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
8698 Generates the @code{insertps} machine instruction.
8699 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
8700 Generates the @code{pextrb} machine instruction.
8701 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
8702 Generates the @code{pinsrb} machine instruction.
8703 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
8704 Generates the @code{pinsrd} machine instruction.
8705 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
8706 Generates the @code{pinsrq} machine instruction in 64bit mode.
8709 The following built-in functions are changed to generate new SSE4.1
8710 instructions when @option{-msse4.1} is used.
8713 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
8714 Generates the @code{extractps} machine instruction.
8715 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
8716 Generates the @code{pextrd} machine instruction.
8717 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
8718 Generates the @code{pextrq} machine instruction in 64bit mode.
8721 The following built-in functions are available when @option{-msse4.2} is
8722 used. All of them generate the machine instruction that is part of the
8726 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
8727 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
8728 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
8729 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
8730 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
8731 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
8732 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
8733 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
8734 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
8735 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
8736 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
8737 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
8738 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
8739 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
8740 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
8743 The following built-in functions are available when @option{-msse4.2} is
8747 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
8748 Generates the @code{crc32b} machine instruction.
8749 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
8750 Generates the @code{crc32w} machine instruction.
8751 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
8752 Generates the @code{crc32l} machine instruction.
8753 @item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)
8754 Generates the @code{crc32q} machine instruction.
8757 The following built-in functions are changed to generate new SSE4.2
8758 instructions when @option{-msse4.2} is used.
8761 @item int __builtin_popcount (unsigned int)
8762 Generates the @code{popcntl} machine instruction.
8763 @item int __builtin_popcountl (unsigned long)
8764 Generates the @code{popcntl} or @code{popcntq} machine instruction,
8765 depending on the size of @code{unsigned long}.
8766 @item int __builtin_popcountll (unsigned long long)
8767 Generates the @code{popcntq} machine instruction.
8770 The following built-in functions are available when @option{-mavx} is
8771 used. All of them generate the machine instruction that is part of the
8775 v4df __builtin_ia32_addpd256 (v4df,v4df)
8776 v8sf __builtin_ia32_addps256 (v8sf,v8sf)
8777 v4df __builtin_ia32_addsubpd256 (v4df,v4df)
8778 v8sf __builtin_ia32_addsubps256 (v8sf,v8sf)
8779 v4df __builtin_ia32_andnpd256 (v4df,v4df)
8780 v8sf __builtin_ia32_andnps256 (v8sf,v8sf)
8781 v4df __builtin_ia32_andpd256 (v4df,v4df)
8782 v8sf __builtin_ia32_andps256 (v8sf,v8sf)
8783 v4df __builtin_ia32_blendpd256 (v4df,v4df,int)
8784 v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int)
8785 v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df)
8786 v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf)
8787 v2df __builtin_ia32_cmppd (v2df,v2df,int)
8788 v4df __builtin_ia32_cmppd256 (v4df,v4df,int)
8789 v4sf __builtin_ia32_cmpps (v4sf,v4sf,int)
8790 v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int)
8791 v2df __builtin_ia32_cmpsd (v2df,v2df,int)
8792 v4sf __builtin_ia32_cmpss (v4sf,v4sf,int)
8793 v4df __builtin_ia32_cvtdq2pd256 (v4si)
8794 v8sf __builtin_ia32_cvtdq2ps256 (v8si)
8795 v4si __builtin_ia32_cvtpd2dq256 (v4df)
8796 v4sf __builtin_ia32_cvtpd2ps256 (v4df)
8797 v8si __builtin_ia32_cvtps2dq256 (v8sf)
8798 v4df __builtin_ia32_cvtps2pd256 (v4sf)
8799 v4si __builtin_ia32_cvttpd2dq256 (v4df)
8800 v8si __builtin_ia32_cvttps2dq256 (v8sf)
8801 v4df __builtin_ia32_divpd256 (v4df,v4df)
8802 v8sf __builtin_ia32_divps256 (v8sf,v8sf)
8803 v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int)
8804 v4df __builtin_ia32_haddpd256 (v4df,v4df)
8805 v8sf __builtin_ia32_haddps256 (v8sf,v8sf)
8806 v4df __builtin_ia32_hsubpd256 (v4df,v4df)
8807 v8sf __builtin_ia32_hsubps256 (v8sf,v8sf)
8808 v32qi __builtin_ia32_lddqu256 (pcchar)
8809 v32qi __builtin_ia32_loaddqu256 (pcchar)
8810 v4df __builtin_ia32_loadupd256 (pcdouble)
8811 v8sf __builtin_ia32_loadups256 (pcfloat)
8812 v2df __builtin_ia32_maskloadpd (pcv2df,v2df)
8813 v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df)
8814 v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf)
8815 v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf)
8816 void __builtin_ia32_maskstorepd (pv2df,v2df,v2df)
8817 void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df)
8818 void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf)
8819 void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf)
8820 v4df __builtin_ia32_maxpd256 (v4df,v4df)
8821 v8sf __builtin_ia32_maxps256 (v8sf,v8sf)
8822 v4df __builtin_ia32_minpd256 (v4df,v4df)
8823 v8sf __builtin_ia32_minps256 (v8sf,v8sf)
8824 v4df __builtin_ia32_movddup256 (v4df)
8825 int __builtin_ia32_movmskpd256 (v4df)
8826 int __builtin_ia32_movmskps256 (v8sf)
8827 v8sf __builtin_ia32_movshdup256 (v8sf)
8828 v8sf __builtin_ia32_movsldup256 (v8sf)
8829 v4df __builtin_ia32_mulpd256 (v4df,v4df)
8830 v8sf __builtin_ia32_mulps256 (v8sf,v8sf)
8831 v4df __builtin_ia32_orpd256 (v4df,v4df)
8832 v8sf __builtin_ia32_orps256 (v8sf,v8sf)
8833 v2df __builtin_ia32_pd_pd256 (v4df)
8834 v4df __builtin_ia32_pd256_pd (v2df)
8835 v4sf __builtin_ia32_ps_ps256 (v8sf)
8836 v8sf __builtin_ia32_ps256_ps (v4sf)
8837 int __builtin_ia32_ptestc256 (v4di,v4di,ptest)
8838 int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest)
8839 int __builtin_ia32_ptestz256 (v4di,v4di,ptest)
8840 v8sf __builtin_ia32_rcpps256 (v8sf)
8841 v4df __builtin_ia32_roundpd256 (v4df,int)
8842 v8sf __builtin_ia32_roundps256 (v8sf,int)
8843 v8sf __builtin_ia32_rsqrtps_nr256 (v8sf)
8844 v8sf __builtin_ia32_rsqrtps256 (v8sf)
8845 v4df __builtin_ia32_shufpd256 (v4df,v4df,int)
8846 v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int)
8847 v4si __builtin_ia32_si_si256 (v8si)
8848 v8si __builtin_ia32_si256_si (v4si)
8849 v4df __builtin_ia32_sqrtpd256 (v4df)
8850 v8sf __builtin_ia32_sqrtps_nr256 (v8sf)
8851 v8sf __builtin_ia32_sqrtps256 (v8sf)
8852 void __builtin_ia32_storedqu256 (pchar,v32qi)
8853 void __builtin_ia32_storeupd256 (pdouble,v4df)
8854 void __builtin_ia32_storeups256 (pfloat,v8sf)
8855 v4df __builtin_ia32_subpd256 (v4df,v4df)
8856 v8sf __builtin_ia32_subps256 (v8sf,v8sf)
8857 v4df __builtin_ia32_unpckhpd256 (v4df,v4df)
8858 v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf)
8859 v4df __builtin_ia32_unpcklpd256 (v4df,v4df)
8860 v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf)
8861 v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df)
8862 v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf)
8863 v4df __builtin_ia32_vbroadcastsd256 (pcdouble)
8864 v4sf __builtin_ia32_vbroadcastss (pcfloat)
8865 v8sf __builtin_ia32_vbroadcastss256 (pcfloat)
8866 v2df __builtin_ia32_vextractf128_pd256 (v4df,int)
8867 v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int)
8868 v4si __builtin_ia32_vextractf128_si256 (v8si,int)
8869 v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int)
8870 v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int)
8871 v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int)
8872 v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int)
8873 v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int)
8874 v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int)
8875 v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int)
8876 v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int)
8877 v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int)
8878 v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int)
8879 v2df __builtin_ia32_vpermilpd (v2df,int)
8880 v4df __builtin_ia32_vpermilpd256 (v4df,int)
8881 v4sf __builtin_ia32_vpermilps (v4sf,int)
8882 v8sf __builtin_ia32_vpermilps256 (v8sf,int)
8883 v2df __builtin_ia32_vpermilvarpd (v2df,v2di)
8884 v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di)
8885 v4sf __builtin_ia32_vpermilvarps (v4sf,v4si)
8886 v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si)
8887 int __builtin_ia32_vtestcpd (v2df,v2df,ptest)
8888 int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest)
8889 int __builtin_ia32_vtestcps (v4sf,v4sf,ptest)
8890 int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest)
8891 int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest)
8892 int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest)
8893 int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest)
8894 int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest)
8895 int __builtin_ia32_vtestzpd (v2df,v2df,ptest)
8896 int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest)
8897 int __builtin_ia32_vtestzps (v4sf,v4sf,ptest)
8898 int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest)
8899 void __builtin_ia32_vzeroall (void)
8900 void __builtin_ia32_vzeroupper (void)
8901 v4df __builtin_ia32_xorpd256 (v4df,v4df)
8902 v8sf __builtin_ia32_xorps256 (v8sf,v8sf)
8905 The following built-in functions are available when @option{-maes} is
8906 used. All of them generate the machine instruction that is part of the
8910 v2di __builtin_ia32_aesenc128 (v2di, v2di)
8911 v2di __builtin_ia32_aesenclast128 (v2di, v2di)
8912 v2di __builtin_ia32_aesdec128 (v2di, v2di)
8913 v2di __builtin_ia32_aesdeclast128 (v2di, v2di)
8914 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int)
8915 v2di __builtin_ia32_aesimc128 (v2di)
8918 The following built-in function is available when @option{-mpclmul} is
8922 @item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)
8923 Generates the @code{pclmulqdq} machine instruction.
8926 The following built-in functions are available when @option{-msse4a} is used.
8927 All of them generate the machine instruction that is part of the name.
8930 void __builtin_ia32_movntsd (double *, v2df)
8931 void __builtin_ia32_movntss (float *, v4sf)
8932 v2di __builtin_ia32_extrq (v2di, v16qi)
8933 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
8934 v2di __builtin_ia32_insertq (v2di, v2di)
8935 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
8938 The following built-in functions are available when @option{-mfma4} is used.
8939 All of them generate the machine instruction that is part of the name
8943 v2df __builtin_ia32_fmaddpd (v2df, v2df, v2df)
8944 v4sf __builtin_ia32_fmaddps (v4sf, v4sf, v4sf)
8945 v2df __builtin_ia32_fmaddsd (v2df, v2df, v2df)
8946 v4sf __builtin_ia32_fmaddss (v4sf, v4sf, v4sf)
8947 v2df __builtin_ia32_fmsubpd (v2df, v2df, v2df)
8948 v4sf __builtin_ia32_fmsubps (v4sf, v4sf, v4sf)
8949 v2df __builtin_ia32_fmsubsd (v2df, v2df, v2df)
8950 v4sf __builtin_ia32_fmsubss (v4sf, v4sf, v4sf)
8951 v2df __builtin_ia32_fnmaddpd (v2df, v2df, v2df)
8952 v4sf __builtin_ia32_fnmaddps (v4sf, v4sf, v4sf)
8953 v2df __builtin_ia32_fnmaddsd (v2df, v2df, v2df)
8954 v4sf __builtin_ia32_fnmaddss (v4sf, v4sf, v4sf)
8955 v2df __builtin_ia32_fnmsubpd (v2df, v2df, v2df)
8956 v4sf __builtin_ia32_fnmsubps (v4sf, v4sf, v4sf)
8957 v2df __builtin_ia32_fnmsubsd (v2df, v2df, v2df)
8958 v4sf __builtin_ia32_fnmsubss (v4sf, v4sf, v4sf)
8959 v2df __builtin_ia32_fmaddsubpd (v2df, v2df, v2df)
8960 v4sf __builtin_ia32_fmaddsubps (v4sf, v4sf, v4sf)
8961 v2df __builtin_ia32_fmsubaddpd (v2df, v2df, v2df)
8962 v4sf __builtin_ia32_fmsubaddps (v4sf, v4sf, v4sf)
8963 v4df __builtin_ia32_fmaddpd256 (v4df, v4df, v4df)
8964 v8sf __builtin_ia32_fmaddps256 (v8sf, v8sf, v8sf)
8965 v4df __builtin_ia32_fmsubpd256 (v4df, v4df, v4df)
8966 v8sf __builtin_ia32_fmsubps256 (v8sf, v8sf, v8sf)
8967 v4df __builtin_ia32_fnmaddpd256 (v4df, v4df, v4df)
8968 v8sf __builtin_ia32_fnmaddps256 (v8sf, v8sf, v8sf)
8969 v4df __builtin_ia32_fnmsubpd256 (v4df, v4df, v4df)
8970 v8sf __builtin_ia32_fnmsubps256 (v8sf, v8sf, v8sf)
8971 v4df __builtin_ia32_fmaddsubpd256 (v4df, v4df, v4df)
8972 v8sf __builtin_ia32_fmaddsubps256 (v8sf, v8sf, v8sf)
8973 v4df __builtin_ia32_fmsubaddpd256 (v4df, v4df, v4df)
8974 v8sf __builtin_ia32_fmsubaddps256 (v8sf, v8sf, v8sf)
8978 The following built-in functions are available when @option{-m3dnow} is used.
8979 All of them generate the machine instruction that is part of the name.
8982 void __builtin_ia32_femms (void)
8983 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
8984 v2si __builtin_ia32_pf2id (v2sf)
8985 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
8986 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
8987 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
8988 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
8989 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
8990 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
8991 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
8992 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
8993 v2sf __builtin_ia32_pfrcp (v2sf)
8994 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
8995 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
8996 v2sf __builtin_ia32_pfrsqrt (v2sf)
8997 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
8998 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
8999 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
9000 v2sf __builtin_ia32_pi2fd (v2si)
9001 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
9004 The following built-in functions are available when both @option{-m3dnow}
9005 and @option{-march=athlon} are used. All of them generate the machine
9006 instruction that is part of the name.
9009 v2si __builtin_ia32_pf2iw (v2sf)
9010 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
9011 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
9012 v2sf __builtin_ia32_pi2fw (v2si)
9013 v2sf __builtin_ia32_pswapdsf (v2sf)
9014 v2si __builtin_ia32_pswapdsi (v2si)
9017 @node MIPS DSP Built-in Functions
9018 @subsection MIPS DSP Built-in Functions
9020 The MIPS DSP Application-Specific Extension (ASE) includes new
9021 instructions that are designed to improve the performance of DSP and
9022 media applications. It provides instructions that operate on packed
9023 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
9025 GCC supports MIPS DSP operations using both the generic
9026 vector extensions (@pxref{Vector Extensions}) and a collection of
9027 MIPS-specific built-in functions. Both kinds of support are
9028 enabled by the @option{-mdsp} command-line option.
9030 Revision 2 of the ASE was introduced in the second half of 2006.
9031 This revision adds extra instructions to the original ASE, but is
9032 otherwise backwards-compatible with it. You can select revision 2
9033 using the command-line option @option{-mdspr2}; this option implies
9036 The SCOUNT and POS bits of the DSP control register are global. The
9037 WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and
9038 POS bits. During optimization, the compiler will not delete these
9039 instructions and it will not delete calls to functions containing
9042 At present, GCC only provides support for operations on 32-bit
9043 vectors. The vector type associated with 8-bit integer data is
9044 usually called @code{v4i8}, the vector type associated with Q7
9045 is usually called @code{v4q7}, the vector type associated with 16-bit
9046 integer data is usually called @code{v2i16}, and the vector type
9047 associated with Q15 is usually called @code{v2q15}. They can be
9048 defined in C as follows:
9051 typedef signed char v4i8 __attribute__ ((vector_size(4)));
9052 typedef signed char v4q7 __attribute__ ((vector_size(4)));
9053 typedef short v2i16 __attribute__ ((vector_size(4)));
9054 typedef short v2q15 __attribute__ ((vector_size(4)));
9057 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
9058 initialized in the same way as aggregates. For example:
9061 v4i8 a = @{1, 2, 3, 4@};
9063 b = (v4i8) @{5, 6, 7, 8@};
9065 v2q15 c = @{0x0fcb, 0x3a75@};
9067 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
9070 @emph{Note:} The CPU's endianness determines the order in which values
9071 are packed. On little-endian targets, the first value is the least
9072 significant and the last value is the most significant. The opposite
9073 order applies to big-endian targets. For example, the code above will
9074 set the lowest byte of @code{a} to @code{1} on little-endian targets
9075 and @code{4} on big-endian targets.
9077 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
9078 representation. As shown in this example, the integer representation
9079 of a Q7 value can be obtained by multiplying the fractional value by
9080 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
9081 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
9084 The table below lists the @code{v4i8} and @code{v2q15} operations for which
9085 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
9086 and @code{c} and @code{d} are @code{v2q15} values.
9088 @multitable @columnfractions .50 .50
9089 @item C code @tab MIPS instruction
9090 @item @code{a + b} @tab @code{addu.qb}
9091 @item @code{c + d} @tab @code{addq.ph}
9092 @item @code{a - b} @tab @code{subu.qb}
9093 @item @code{c - d} @tab @code{subq.ph}
9096 The table below lists the @code{v2i16} operation for which
9097 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
9098 @code{v2i16} values.
9100 @multitable @columnfractions .50 .50
9101 @item C code @tab MIPS instruction
9102 @item @code{e * f} @tab @code{mul.ph}
9105 It is easier to describe the DSP built-in functions if we first define
9106 the following types:
9111 typedef unsigned int ui32;
9112 typedef long long a64;
9115 @code{q31} and @code{i32} are actually the same as @code{int}, but we
9116 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
9117 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
9118 @code{long long}, but we use @code{a64} to indicate values that will
9119 be placed in one of the four DSP accumulators (@code{$ac0},
9120 @code{$ac1}, @code{$ac2} or @code{$ac3}).
9122 Also, some built-in functions prefer or require immediate numbers as
9123 parameters, because the corresponding DSP instructions accept both immediate
9124 numbers and register operands, or accept immediate numbers only. The
9125 immediate parameters are listed as follows.
9134 imm_n32_31: -32 to 31.
9135 imm_n512_511: -512 to 511.
9138 The following built-in functions map directly to a particular MIPS DSP
9139 instruction. Please refer to the architecture specification
9140 for details on what each instruction does.
9143 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
9144 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
9145 q31 __builtin_mips_addq_s_w (q31, q31)
9146 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
9147 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
9148 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
9149 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
9150 q31 __builtin_mips_subq_s_w (q31, q31)
9151 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
9152 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
9153 i32 __builtin_mips_addsc (i32, i32)
9154 i32 __builtin_mips_addwc (i32, i32)
9155 i32 __builtin_mips_modsub (i32, i32)
9156 i32 __builtin_mips_raddu_w_qb (v4i8)
9157 v2q15 __builtin_mips_absq_s_ph (v2q15)
9158 q31 __builtin_mips_absq_s_w (q31)
9159 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
9160 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
9161 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
9162 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
9163 q31 __builtin_mips_preceq_w_phl (v2q15)
9164 q31 __builtin_mips_preceq_w_phr (v2q15)
9165 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
9166 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
9167 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
9168 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
9169 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
9170 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
9171 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
9172 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
9173 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
9174 v4i8 __builtin_mips_shll_qb (v4i8, i32)
9175 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
9176 v2q15 __builtin_mips_shll_ph (v2q15, i32)
9177 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
9178 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
9179 q31 __builtin_mips_shll_s_w (q31, imm0_31)
9180 q31 __builtin_mips_shll_s_w (q31, i32)
9181 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
9182 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
9183 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
9184 v2q15 __builtin_mips_shra_ph (v2q15, i32)
9185 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
9186 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
9187 q31 __builtin_mips_shra_r_w (q31, imm0_31)
9188 q31 __builtin_mips_shra_r_w (q31, i32)
9189 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
9190 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
9191 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
9192 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
9193 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
9194 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
9195 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
9196 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
9197 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
9198 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
9199 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
9200 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
9201 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
9202 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
9203 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
9204 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
9205 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
9206 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
9207 i32 __builtin_mips_bitrev (i32)
9208 i32 __builtin_mips_insv (i32, i32)
9209 v4i8 __builtin_mips_repl_qb (imm0_255)
9210 v4i8 __builtin_mips_repl_qb (i32)
9211 v2q15 __builtin_mips_repl_ph (imm_n512_511)
9212 v2q15 __builtin_mips_repl_ph (i32)
9213 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
9214 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
9215 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
9216 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
9217 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
9218 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
9219 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
9220 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
9221 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
9222 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
9223 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
9224 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
9225 i32 __builtin_mips_extr_w (a64, imm0_31)
9226 i32 __builtin_mips_extr_w (a64, i32)
9227 i32 __builtin_mips_extr_r_w (a64, imm0_31)
9228 i32 __builtin_mips_extr_s_h (a64, i32)
9229 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
9230 i32 __builtin_mips_extr_rs_w (a64, i32)
9231 i32 __builtin_mips_extr_s_h (a64, imm0_31)
9232 i32 __builtin_mips_extr_r_w (a64, i32)
9233 i32 __builtin_mips_extp (a64, imm0_31)
9234 i32 __builtin_mips_extp (a64, i32)
9235 i32 __builtin_mips_extpdp (a64, imm0_31)
9236 i32 __builtin_mips_extpdp (a64, i32)
9237 a64 __builtin_mips_shilo (a64, imm_n32_31)
9238 a64 __builtin_mips_shilo (a64, i32)
9239 a64 __builtin_mips_mthlip (a64, i32)
9240 void __builtin_mips_wrdsp (i32, imm0_63)
9241 i32 __builtin_mips_rddsp (imm0_63)
9242 i32 __builtin_mips_lbux (void *, i32)
9243 i32 __builtin_mips_lhx (void *, i32)
9244 i32 __builtin_mips_lwx (void *, i32)
9245 i32 __builtin_mips_bposge32 (void)
9248 The following built-in functions map directly to a particular MIPS DSP REV 2
9249 instruction. Please refer to the architecture specification
9250 for details on what each instruction does.
9253 v4q7 __builtin_mips_absq_s_qb (v4q7);
9254 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
9255 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
9256 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
9257 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
9258 i32 __builtin_mips_append (i32, i32, imm0_31);
9259 i32 __builtin_mips_balign (i32, i32, imm0_3);
9260 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
9261 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
9262 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
9263 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
9264 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
9265 a64 __builtin_mips_madd (a64, i32, i32);
9266 a64 __builtin_mips_maddu (a64, ui32, ui32);
9267 a64 __builtin_mips_msub (a64, i32, i32);
9268 a64 __builtin_mips_msubu (a64, ui32, ui32);
9269 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
9270 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
9271 q31 __builtin_mips_mulq_rs_w (q31, q31);
9272 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
9273 q31 __builtin_mips_mulq_s_w (q31, q31);
9274 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
9275 a64 __builtin_mips_mult (i32, i32);
9276 a64 __builtin_mips_multu (ui32, ui32);
9277 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
9278 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
9279 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
9280 i32 __builtin_mips_prepend (i32, i32, imm0_31);
9281 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
9282 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
9283 v4i8 __builtin_mips_shra_qb (v4i8, i32);
9284 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
9285 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
9286 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
9287 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
9288 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
9289 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
9290 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
9291 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
9292 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
9293 q31 __builtin_mips_addqh_w (q31, q31);
9294 q31 __builtin_mips_addqh_r_w (q31, q31);
9295 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
9296 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
9297 q31 __builtin_mips_subqh_w (q31, q31);
9298 q31 __builtin_mips_subqh_r_w (q31, q31);
9299 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
9300 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
9301 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
9302 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
9303 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
9304 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
9308 @node MIPS Paired-Single Support
9309 @subsection MIPS Paired-Single Support
9311 The MIPS64 architecture includes a number of instructions that
9312 operate on pairs of single-precision floating-point values.
9313 Each pair is packed into a 64-bit floating-point register,
9314 with one element being designated the ``upper half'' and
9315 the other being designated the ``lower half''.
9317 GCC supports paired-single operations using both the generic
9318 vector extensions (@pxref{Vector Extensions}) and a collection of
9319 MIPS-specific built-in functions. Both kinds of support are
9320 enabled by the @option{-mpaired-single} command-line option.
9322 The vector type associated with paired-single values is usually
9323 called @code{v2sf}. It can be defined in C as follows:
9326 typedef float v2sf __attribute__ ((vector_size (8)));
9329 @code{v2sf} values are initialized in the same way as aggregates.
9333 v2sf a = @{1.5, 9.1@};
9336 b = (v2sf) @{e, f@};
9339 @emph{Note:} The CPU's endianness determines which value is stored in
9340 the upper half of a register and which value is stored in the lower half.
9341 On little-endian targets, the first value is the lower one and the second
9342 value is the upper one. The opposite order applies to big-endian targets.
9343 For example, the code above will set the lower half of @code{a} to
9344 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
9346 @node MIPS Loongson Built-in Functions
9347 @subsection MIPS Loongson Built-in Functions
9349 GCC provides intrinsics to access the SIMD instructions provided by the
9350 ST Microelectronics Loongson-2E and -2F processors. These intrinsics,
9351 available after inclusion of the @code{loongson.h} header file,
9352 operate on the following 64-bit vector types:
9355 @item @code{uint8x8_t}, a vector of eight unsigned 8-bit integers;
9356 @item @code{uint16x4_t}, a vector of four unsigned 16-bit integers;
9357 @item @code{uint32x2_t}, a vector of two unsigned 32-bit integers;
9358 @item @code{int8x8_t}, a vector of eight signed 8-bit integers;
9359 @item @code{int16x4_t}, a vector of four signed 16-bit integers;
9360 @item @code{int32x2_t}, a vector of two signed 32-bit integers.
9363 The intrinsics provided are listed below; each is named after the
9364 machine instruction to which it corresponds, with suffixes added as
9365 appropriate to distinguish intrinsics that expand to the same machine
9366 instruction yet have different argument types. Refer to the architecture
9367 documentation for a description of the functionality of each
9371 int16x4_t packsswh (int32x2_t s, int32x2_t t);
9372 int8x8_t packsshb (int16x4_t s, int16x4_t t);
9373 uint8x8_t packushb (uint16x4_t s, uint16x4_t t);
9374 uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t);
9375 uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t);
9376 uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t);
9377 int32x2_t paddw_s (int32x2_t s, int32x2_t t);
9378 int16x4_t paddh_s (int16x4_t s, int16x4_t t);
9379 int8x8_t paddb_s (int8x8_t s, int8x8_t t);
9380 uint64_t paddd_u (uint64_t s, uint64_t t);
9381 int64_t paddd_s (int64_t s, int64_t t);
9382 int16x4_t paddsh (int16x4_t s, int16x4_t t);
9383 int8x8_t paddsb (int8x8_t s, int8x8_t t);
9384 uint16x4_t paddush (uint16x4_t s, uint16x4_t t);
9385 uint8x8_t paddusb (uint8x8_t s, uint8x8_t t);
9386 uint64_t pandn_ud (uint64_t s, uint64_t t);
9387 uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t);
9388 uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t);
9389 uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t);
9390 int64_t pandn_sd (int64_t s, int64_t t);
9391 int32x2_t pandn_sw (int32x2_t s, int32x2_t t);
9392 int16x4_t pandn_sh (int16x4_t s, int16x4_t t);
9393 int8x8_t pandn_sb (int8x8_t s, int8x8_t t);
9394 uint16x4_t pavgh (uint16x4_t s, uint16x4_t t);
9395 uint8x8_t pavgb (uint8x8_t s, uint8x8_t t);
9396 uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t);
9397 uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t);
9398 uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t);
9399 int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t);
9400 int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t);
9401 int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t);
9402 uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t);
9403 uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t);
9404 uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t);
9405 int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t);
9406 int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t);
9407 int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t);
9408 uint16x4_t pextrh_u (uint16x4_t s, int field);
9409 int16x4_t pextrh_s (int16x4_t s, int field);
9410 uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t);
9411 uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t);
9412 uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t);
9413 uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t);
9414 int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t);
9415 int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t);
9416 int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t);
9417 int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t);
9418 int32x2_t pmaddhw (int16x4_t s, int16x4_t t);
9419 int16x4_t pmaxsh (int16x4_t s, int16x4_t t);
9420 uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t);
9421 int16x4_t pminsh (int16x4_t s, int16x4_t t);
9422 uint8x8_t pminub (uint8x8_t s, uint8x8_t t);
9423 uint8x8_t pmovmskb_u (uint8x8_t s);
9424 int8x8_t pmovmskb_s (int8x8_t s);
9425 uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t);
9426 int16x4_t pmulhh (int16x4_t s, int16x4_t t);
9427 int16x4_t pmullh (int16x4_t s, int16x4_t t);
9428 int64_t pmuluw (uint32x2_t s, uint32x2_t t);
9429 uint8x8_t pasubub (uint8x8_t s, uint8x8_t t);
9430 uint16x4_t biadd (uint8x8_t s);
9431 uint16x4_t psadbh (uint8x8_t s, uint8x8_t t);
9432 uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order);
9433 int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order);
9434 uint16x4_t psllh_u (uint16x4_t s, uint8_t amount);
9435 int16x4_t psllh_s (int16x4_t s, uint8_t amount);
9436 uint32x2_t psllw_u (uint32x2_t s, uint8_t amount);
9437 int32x2_t psllw_s (int32x2_t s, uint8_t amount);
9438 uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount);
9439 int16x4_t psrlh_s (int16x4_t s, uint8_t amount);
9440 uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount);
9441 int32x2_t psrlw_s (int32x2_t s, uint8_t amount);
9442 uint16x4_t psrah_u (uint16x4_t s, uint8_t amount);
9443 int16x4_t psrah_s (int16x4_t s, uint8_t amount);
9444 uint32x2_t psraw_u (uint32x2_t s, uint8_t amount);
9445 int32x2_t psraw_s (int32x2_t s, uint8_t amount);
9446 uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t);
9447 uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t);
9448 uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t);
9449 int32x2_t psubw_s (int32x2_t s, int32x2_t t);
9450 int16x4_t psubh_s (int16x4_t s, int16x4_t t);
9451 int8x8_t psubb_s (int8x8_t s, int8x8_t t);
9452 uint64_t psubd_u (uint64_t s, uint64_t t);
9453 int64_t psubd_s (int64_t s, int64_t t);
9454 int16x4_t psubsh (int16x4_t s, int16x4_t t);
9455 int8x8_t psubsb (int8x8_t s, int8x8_t t);
9456 uint16x4_t psubush (uint16x4_t s, uint16x4_t t);
9457 uint8x8_t psubusb (uint8x8_t s, uint8x8_t t);
9458 uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t);
9459 uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t);
9460 uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t);
9461 int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t);
9462 int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t);
9463 int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t);
9464 uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t);
9465 uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t);
9466 uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t);
9467 int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t);
9468 int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t);
9469 int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t);
9473 * Paired-Single Arithmetic::
9474 * Paired-Single Built-in Functions::
9475 * MIPS-3D Built-in Functions::
9478 @node Paired-Single Arithmetic
9479 @subsubsection Paired-Single Arithmetic
9481 The table below lists the @code{v2sf} operations for which hardware
9482 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
9483 values and @code{x} is an integral value.
9485 @multitable @columnfractions .50 .50
9486 @item C code @tab MIPS instruction
9487 @item @code{a + b} @tab @code{add.ps}
9488 @item @code{a - b} @tab @code{sub.ps}
9489 @item @code{-a} @tab @code{neg.ps}
9490 @item @code{a * b} @tab @code{mul.ps}
9491 @item @code{a * b + c} @tab @code{madd.ps}
9492 @item @code{a * b - c} @tab @code{msub.ps}
9493 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
9494 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
9495 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
9498 Note that the multiply-accumulate instructions can be disabled
9499 using the command-line option @code{-mno-fused-madd}.
9501 @node Paired-Single Built-in Functions
9502 @subsubsection Paired-Single Built-in Functions
9504 The following paired-single functions map directly to a particular
9505 MIPS instruction. Please refer to the architecture specification
9506 for details on what each instruction does.
9509 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
9510 Pair lower lower (@code{pll.ps}).
9512 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
9513 Pair upper lower (@code{pul.ps}).
9515 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
9516 Pair lower upper (@code{plu.ps}).
9518 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
9519 Pair upper upper (@code{puu.ps}).
9521 @item v2sf __builtin_mips_cvt_ps_s (float, float)
9522 Convert pair to paired single (@code{cvt.ps.s}).
9524 @item float __builtin_mips_cvt_s_pl (v2sf)
9525 Convert pair lower to single (@code{cvt.s.pl}).
9527 @item float __builtin_mips_cvt_s_pu (v2sf)
9528 Convert pair upper to single (@code{cvt.s.pu}).
9530 @item v2sf __builtin_mips_abs_ps (v2sf)
9531 Absolute value (@code{abs.ps}).
9533 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
9534 Align variable (@code{alnv.ps}).
9536 @emph{Note:} The value of the third parameter must be 0 or 4
9537 modulo 8, otherwise the result will be unpredictable. Please read the
9538 instruction description for details.
9541 The following multi-instruction functions are also available.
9542 In each case, @var{cond} can be any of the 16 floating-point conditions:
9543 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
9544 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
9545 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
9548 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9549 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9550 Conditional move based on floating point comparison (@code{c.@var{cond}.ps},
9551 @code{movt.ps}/@code{movf.ps}).
9553 The @code{movt} functions return the value @var{x} computed by:
9556 c.@var{cond}.ps @var{cc},@var{a},@var{b}
9557 mov.ps @var{x},@var{c}
9558 movt.ps @var{x},@var{d},@var{cc}
9561 The @code{movf} functions are similar but use @code{movf.ps} instead
9564 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9565 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9566 Comparison of two paired-single values (@code{c.@var{cond}.ps},
9567 @code{bc1t}/@code{bc1f}).
9569 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
9570 and return either the upper or lower half of the result. For example:
9574 if (__builtin_mips_upper_c_eq_ps (a, b))
9575 upper_halves_are_equal ();
9577 upper_halves_are_unequal ();
9579 if (__builtin_mips_lower_c_eq_ps (a, b))
9580 lower_halves_are_equal ();
9582 lower_halves_are_unequal ();
9586 @node MIPS-3D Built-in Functions
9587 @subsubsection MIPS-3D Built-in Functions
9589 The MIPS-3D Application-Specific Extension (ASE) includes additional
9590 paired-single instructions that are designed to improve the performance
9591 of 3D graphics operations. Support for these instructions is controlled
9592 by the @option{-mips3d} command-line option.
9594 The functions listed below map directly to a particular MIPS-3D
9595 instruction. Please refer to the architecture specification for
9596 more details on what each instruction does.
9599 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
9600 Reduction add (@code{addr.ps}).
9602 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
9603 Reduction multiply (@code{mulr.ps}).
9605 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
9606 Convert paired single to paired word (@code{cvt.pw.ps}).
9608 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
9609 Convert paired word to paired single (@code{cvt.ps.pw}).
9611 @item float __builtin_mips_recip1_s (float)
9612 @itemx double __builtin_mips_recip1_d (double)
9613 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
9614 Reduced precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
9616 @item float __builtin_mips_recip2_s (float, float)
9617 @itemx double __builtin_mips_recip2_d (double, double)
9618 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
9619 Reduced precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
9621 @item float __builtin_mips_rsqrt1_s (float)
9622 @itemx double __builtin_mips_rsqrt1_d (double)
9623 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
9624 Reduced precision reciprocal square root (sequence step 1)
9625 (@code{rsqrt1.@var{fmt}}).
9627 @item float __builtin_mips_rsqrt2_s (float, float)
9628 @itemx double __builtin_mips_rsqrt2_d (double, double)
9629 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
9630 Reduced precision reciprocal square root (sequence step 2)
9631 (@code{rsqrt2.@var{fmt}}).
9634 The following multi-instruction functions are also available.
9635 In each case, @var{cond} can be any of the 16 floating-point conditions:
9636 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
9637 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
9638 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
9641 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
9642 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
9643 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
9644 @code{bc1t}/@code{bc1f}).
9646 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
9647 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
9652 if (__builtin_mips_cabs_eq_s (a, b))
9658 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9659 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9660 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
9661 @code{bc1t}/@code{bc1f}).
9663 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
9664 and return either the upper or lower half of the result. For example:
9668 if (__builtin_mips_upper_cabs_eq_ps (a, b))
9669 upper_halves_are_equal ();
9671 upper_halves_are_unequal ();
9673 if (__builtin_mips_lower_cabs_eq_ps (a, b))
9674 lower_halves_are_equal ();
9676 lower_halves_are_unequal ();
9679 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9680 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9681 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
9682 @code{movt.ps}/@code{movf.ps}).
9684 The @code{movt} functions return the value @var{x} computed by:
9687 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
9688 mov.ps @var{x},@var{c}
9689 movt.ps @var{x},@var{d},@var{cc}
9692 The @code{movf} functions are similar but use @code{movf.ps} instead
9695 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9696 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9697 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9698 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
9699 Comparison of two paired-single values
9700 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
9701 @code{bc1any2t}/@code{bc1any2f}).
9703 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
9704 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
9705 result is true and the @code{all} forms return true if both results are true.
9710 if (__builtin_mips_any_c_eq_ps (a, b))
9715 if (__builtin_mips_all_c_eq_ps (a, b))
9721 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9722 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9723 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9724 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
9725 Comparison of four paired-single values
9726 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
9727 @code{bc1any4t}/@code{bc1any4f}).
9729 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
9730 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
9731 The @code{any} forms return true if any of the four results are true
9732 and the @code{all} forms return true if all four results are true.
9737 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
9742 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
9749 @node picoChip Built-in Functions
9750 @subsection picoChip Built-in Functions
9752 GCC provides an interface to selected machine instructions from the
9753 picoChip instruction set.
9756 @item int __builtin_sbc (int @var{value})
9757 Sign bit count. Return the number of consecutive bits in @var{value}
9758 which have the same value as the sign-bit. The result is the number of
9759 leading sign bits minus one, giving the number of redundant sign bits in
9762 @item int __builtin_byteswap (int @var{value})
9763 Byte swap. Return the result of swapping the upper and lower bytes of
9766 @item int __builtin_brev (int @var{value})
9767 Bit reversal. Return the result of reversing the bits in
9768 @var{value}. Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1,
9771 @item int __builtin_adds (int @var{x}, int @var{y})
9772 Saturating addition. Return the result of adding @var{x} and @var{y},
9773 storing the value 32767 if the result overflows.
9775 @item int __builtin_subs (int @var{x}, int @var{y})
9776 Saturating subtraction. Return the result of subtracting @var{y} from
9777 @var{x}, storing the value @minus{}32768 if the result overflows.
9779 @item void __builtin_halt (void)
9780 Halt. The processor will stop execution. This built-in is useful for
9781 implementing assertions.
9785 @node Other MIPS Built-in Functions
9786 @subsection Other MIPS Built-in Functions
9788 GCC provides other MIPS-specific built-in functions:
9791 @item void __builtin_mips_cache (int @var{op}, const volatile void *@var{addr})
9792 Insert a @samp{cache} instruction with operands @var{op} and @var{addr}.
9793 GCC defines the preprocessor macro @code{___GCC_HAVE_BUILTIN_MIPS_CACHE}
9794 when this function is available.
9797 @node PowerPC AltiVec/VSX Built-in Functions
9798 @subsection PowerPC AltiVec Built-in Functions
9800 GCC provides an interface for the PowerPC family of processors to access
9801 the AltiVec operations described in Motorola's AltiVec Programming
9802 Interface Manual. The interface is made available by including
9803 @code{<altivec.h>} and using @option{-maltivec} and
9804 @option{-mabi=altivec}. The interface supports the following vector
9808 vector unsigned char
9812 vector unsigned short
9823 If @option{-mvsx} is used the following additional vector types are
9827 vector unsigned long
9832 The long types are only implemented for 64-bit code generation, and
9833 the long type is only used in the floating point/integer conversion
9836 GCC's implementation of the high-level language interface available from
9837 C and C++ code differs from Motorola's documentation in several ways.
9842 A vector constant is a list of constant expressions within curly braces.
9845 A vector initializer requires no cast if the vector constant is of the
9846 same type as the variable it is initializing.
9849 If @code{signed} or @code{unsigned} is omitted, the signedness of the
9850 vector type is the default signedness of the base type. The default
9851 varies depending on the operating system, so a portable program should
9852 always specify the signedness.
9855 Compiling with @option{-maltivec} adds keywords @code{__vector},
9856 @code{vector}, @code{__pixel}, @code{pixel}, @code{__bool} and
9857 @code{bool}. When compiling ISO C, the context-sensitive substitution
9858 of the keywords @code{vector}, @code{pixel} and @code{bool} is
9859 disabled. To use them, you must include @code{<altivec.h>} instead.
9862 GCC allows using a @code{typedef} name as the type specifier for a
9866 For C, overloaded functions are implemented with macros so the following
9870 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
9873 Since @code{vec_add} is a macro, the vector constant in the example
9874 is treated as four separate arguments. Wrap the entire argument in
9875 parentheses for this to work.
9878 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
9879 Internally, GCC uses built-in functions to achieve the functionality in
9880 the aforementioned header file, but they are not supported and are
9881 subject to change without notice.
9883 The following interfaces are supported for the generic and specific
9884 AltiVec operations and the AltiVec predicates. In cases where there
9885 is a direct mapping between generic and specific operations, only the
9886 generic names are shown here, although the specific operations can also
9889 Arguments that are documented as @code{const int} require literal
9890 integral values within the range required for that operation.
9893 vector signed char vec_abs (vector signed char);
9894 vector signed short vec_abs (vector signed short);
9895 vector signed int vec_abs (vector signed int);
9896 vector float vec_abs (vector float);
9898 vector signed char vec_abss (vector signed char);
9899 vector signed short vec_abss (vector signed short);
9900 vector signed int vec_abss (vector signed int);
9902 vector signed char vec_add (vector bool char, vector signed char);
9903 vector signed char vec_add (vector signed char, vector bool char);
9904 vector signed char vec_add (vector signed char, vector signed char);
9905 vector unsigned char vec_add (vector bool char, vector unsigned char);
9906 vector unsigned char vec_add (vector unsigned char, vector bool char);
9907 vector unsigned char vec_add (vector unsigned char,
9908 vector unsigned char);
9909 vector signed short vec_add (vector bool short, vector signed short);
9910 vector signed short vec_add (vector signed short, vector bool short);
9911 vector signed short vec_add (vector signed short, vector signed short);
9912 vector unsigned short vec_add (vector bool short,
9913 vector unsigned short);
9914 vector unsigned short vec_add (vector unsigned short,
9916 vector unsigned short vec_add (vector unsigned short,
9917 vector unsigned short);
9918 vector signed int vec_add (vector bool int, vector signed int);
9919 vector signed int vec_add (vector signed int, vector bool int);
9920 vector signed int vec_add (vector signed int, vector signed int);
9921 vector unsigned int vec_add (vector bool int, vector unsigned int);
9922 vector unsigned int vec_add (vector unsigned int, vector bool int);
9923 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
9924 vector float vec_add (vector float, vector float);
9926 vector float vec_vaddfp (vector float, vector float);
9928 vector signed int vec_vadduwm (vector bool int, vector signed int);
9929 vector signed int vec_vadduwm (vector signed int, vector bool int);
9930 vector signed int vec_vadduwm (vector signed int, vector signed int);
9931 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
9932 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
9933 vector unsigned int vec_vadduwm (vector unsigned int,
9934 vector unsigned int);
9936 vector signed short vec_vadduhm (vector bool short,
9937 vector signed short);
9938 vector signed short vec_vadduhm (vector signed short,
9940 vector signed short vec_vadduhm (vector signed short,
9941 vector signed short);
9942 vector unsigned short vec_vadduhm (vector bool short,
9943 vector unsigned short);
9944 vector unsigned short vec_vadduhm (vector unsigned short,
9946 vector unsigned short vec_vadduhm (vector unsigned short,
9947 vector unsigned short);
9949 vector signed char vec_vaddubm (vector bool char, vector signed char);
9950 vector signed char vec_vaddubm (vector signed char, vector bool char);
9951 vector signed char vec_vaddubm (vector signed char, vector signed char);
9952 vector unsigned char vec_vaddubm (vector bool char,
9953 vector unsigned char);
9954 vector unsigned char vec_vaddubm (vector unsigned char,
9956 vector unsigned char vec_vaddubm (vector unsigned char,
9957 vector unsigned char);
9959 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
9961 vector unsigned char vec_adds (vector bool char, vector unsigned char);
9962 vector unsigned char vec_adds (vector unsigned char, vector bool char);
9963 vector unsigned char vec_adds (vector unsigned char,
9964 vector unsigned char);
9965 vector signed char vec_adds (vector bool char, vector signed char);
9966 vector signed char vec_adds (vector signed char, vector bool char);
9967 vector signed char vec_adds (vector signed char, vector signed char);
9968 vector unsigned short vec_adds (vector bool short,
9969 vector unsigned short);
9970 vector unsigned short vec_adds (vector unsigned short,
9972 vector unsigned short vec_adds (vector unsigned short,
9973 vector unsigned short);
9974 vector signed short vec_adds (vector bool short, vector signed short);
9975 vector signed short vec_adds (vector signed short, vector bool short);
9976 vector signed short vec_adds (vector signed short, vector signed short);
9977 vector unsigned int vec_adds (vector bool int, vector unsigned int);
9978 vector unsigned int vec_adds (vector unsigned int, vector bool int);
9979 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
9980 vector signed int vec_adds (vector bool int, vector signed int);
9981 vector signed int vec_adds (vector signed int, vector bool int);
9982 vector signed int vec_adds (vector signed int, vector signed int);
9984 vector signed int vec_vaddsws (vector bool int, vector signed int);
9985 vector signed int vec_vaddsws (vector signed int, vector bool int);
9986 vector signed int vec_vaddsws (vector signed int, vector signed int);
9988 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
9989 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
9990 vector unsigned int vec_vadduws (vector unsigned int,
9991 vector unsigned int);
9993 vector signed short vec_vaddshs (vector bool short,
9994 vector signed short);
9995 vector signed short vec_vaddshs (vector signed short,
9997 vector signed short vec_vaddshs (vector signed short,
9998 vector signed short);
10000 vector unsigned short vec_vadduhs (vector bool short,
10001 vector unsigned short);
10002 vector unsigned short vec_vadduhs (vector unsigned short,
10003 vector bool short);
10004 vector unsigned short vec_vadduhs (vector unsigned short,
10005 vector unsigned short);
10007 vector signed char vec_vaddsbs (vector bool char, vector signed char);
10008 vector signed char vec_vaddsbs (vector signed char, vector bool char);
10009 vector signed char vec_vaddsbs (vector signed char, vector signed char);
10011 vector unsigned char vec_vaddubs (vector bool char,
10012 vector unsigned char);
10013 vector unsigned char vec_vaddubs (vector unsigned char,
10015 vector unsigned char vec_vaddubs (vector unsigned char,
10016 vector unsigned char);
10018 vector float vec_and (vector float, vector float);
10019 vector float vec_and (vector float, vector bool int);
10020 vector float vec_and (vector bool int, vector float);
10021 vector bool int vec_and (vector bool int, vector bool int);
10022 vector signed int vec_and (vector bool int, vector signed int);
10023 vector signed int vec_and (vector signed int, vector bool int);
10024 vector signed int vec_and (vector signed int, vector signed int);
10025 vector unsigned int vec_and (vector bool int, vector unsigned int);
10026 vector unsigned int vec_and (vector unsigned int, vector bool int);
10027 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
10028 vector bool short vec_and (vector bool short, vector bool short);
10029 vector signed short vec_and (vector bool short, vector signed short);
10030 vector signed short vec_and (vector signed short, vector bool short);
10031 vector signed short vec_and (vector signed short, vector signed short);
10032 vector unsigned short vec_and (vector bool short,
10033 vector unsigned short);
10034 vector unsigned short vec_and (vector unsigned short,
10035 vector bool short);
10036 vector unsigned short vec_and (vector unsigned short,
10037 vector unsigned short);
10038 vector signed char vec_and (vector bool char, vector signed char);
10039 vector bool char vec_and (vector bool char, vector bool char);
10040 vector signed char vec_and (vector signed char, vector bool char);
10041 vector signed char vec_and (vector signed char, vector signed char);
10042 vector unsigned char vec_and (vector bool char, vector unsigned char);
10043 vector unsigned char vec_and (vector unsigned char, vector bool char);
10044 vector unsigned char vec_and (vector unsigned char,
10045 vector unsigned char);
10047 vector float vec_andc (vector float, vector float);
10048 vector float vec_andc (vector float, vector bool int);
10049 vector float vec_andc (vector bool int, vector float);
10050 vector bool int vec_andc (vector bool int, vector bool int);
10051 vector signed int vec_andc (vector bool int, vector signed int);
10052 vector signed int vec_andc (vector signed int, vector bool int);
10053 vector signed int vec_andc (vector signed int, vector signed int);
10054 vector unsigned int vec_andc (vector bool int, vector unsigned int);
10055 vector unsigned int vec_andc (vector unsigned int, vector bool int);
10056 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
10057 vector bool short vec_andc (vector bool short, vector bool short);
10058 vector signed short vec_andc (vector bool short, vector signed short);
10059 vector signed short vec_andc (vector signed short, vector bool short);
10060 vector signed short vec_andc (vector signed short, vector signed short);
10061 vector unsigned short vec_andc (vector bool short,
10062 vector unsigned short);
10063 vector unsigned short vec_andc (vector unsigned short,
10064 vector bool short);
10065 vector unsigned short vec_andc (vector unsigned short,
10066 vector unsigned short);
10067 vector signed char vec_andc (vector bool char, vector signed char);
10068 vector bool char vec_andc (vector bool char, vector bool char);
10069 vector signed char vec_andc (vector signed char, vector bool char);
10070 vector signed char vec_andc (vector signed char, vector signed char);
10071 vector unsigned char vec_andc (vector bool char, vector unsigned char);
10072 vector unsigned char vec_andc (vector unsigned char, vector bool char);
10073 vector unsigned char vec_andc (vector unsigned char,
10074 vector unsigned char);
10076 vector unsigned char vec_avg (vector unsigned char,
10077 vector unsigned char);
10078 vector signed char vec_avg (vector signed char, vector signed char);
10079 vector unsigned short vec_avg (vector unsigned short,
10080 vector unsigned short);
10081 vector signed short vec_avg (vector signed short, vector signed short);
10082 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
10083 vector signed int vec_avg (vector signed int, vector signed int);
10085 vector signed int vec_vavgsw (vector signed int, vector signed int);
10087 vector unsigned int vec_vavguw (vector unsigned int,
10088 vector unsigned int);
10090 vector signed short vec_vavgsh (vector signed short,
10091 vector signed short);
10093 vector unsigned short vec_vavguh (vector unsigned short,
10094 vector unsigned short);
10096 vector signed char vec_vavgsb (vector signed char, vector signed char);
10098 vector unsigned char vec_vavgub (vector unsigned char,
10099 vector unsigned char);
10101 vector float vec_copysign (vector float);
10103 vector float vec_ceil (vector float);
10105 vector signed int vec_cmpb (vector float, vector float);
10107 vector bool char vec_cmpeq (vector signed char, vector signed char);
10108 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
10109 vector bool short vec_cmpeq (vector signed short, vector signed short);
10110 vector bool short vec_cmpeq (vector unsigned short,
10111 vector unsigned short);
10112 vector bool int vec_cmpeq (vector signed int, vector signed int);
10113 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
10114 vector bool int vec_cmpeq (vector float, vector float);
10116 vector bool int vec_vcmpeqfp (vector float, vector float);
10118 vector bool int vec_vcmpequw (vector signed int, vector signed int);
10119 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
10121 vector bool short vec_vcmpequh (vector signed short,
10122 vector signed short);
10123 vector bool short vec_vcmpequh (vector unsigned short,
10124 vector unsigned short);
10126 vector bool char vec_vcmpequb (vector signed char, vector signed char);
10127 vector bool char vec_vcmpequb (vector unsigned char,
10128 vector unsigned char);
10130 vector bool int vec_cmpge (vector float, vector float);
10132 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
10133 vector bool char vec_cmpgt (vector signed char, vector signed char);
10134 vector bool short vec_cmpgt (vector unsigned short,
10135 vector unsigned short);
10136 vector bool short vec_cmpgt (vector signed short, vector signed short);
10137 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
10138 vector bool int vec_cmpgt (vector signed int, vector signed int);
10139 vector bool int vec_cmpgt (vector float, vector float);
10141 vector bool int vec_vcmpgtfp (vector float, vector float);
10143 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
10145 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
10147 vector bool short vec_vcmpgtsh (vector signed short,
10148 vector signed short);
10150 vector bool short vec_vcmpgtuh (vector unsigned short,
10151 vector unsigned short);
10153 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
10155 vector bool char vec_vcmpgtub (vector unsigned char,
10156 vector unsigned char);
10158 vector bool int vec_cmple (vector float, vector float);
10160 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
10161 vector bool char vec_cmplt (vector signed char, vector signed char);
10162 vector bool short vec_cmplt (vector unsigned short,
10163 vector unsigned short);
10164 vector bool short vec_cmplt (vector signed short, vector signed short);
10165 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
10166 vector bool int vec_cmplt (vector signed int, vector signed int);
10167 vector bool int vec_cmplt (vector float, vector float);
10169 vector float vec_ctf (vector unsigned int, const int);
10170 vector float vec_ctf (vector signed int, const int);
10172 vector float vec_vcfsx (vector signed int, const int);
10174 vector float vec_vcfux (vector unsigned int, const int);
10176 vector signed int vec_cts (vector float, const int);
10178 vector unsigned int vec_ctu (vector float, const int);
10180 void vec_dss (const int);
10182 void vec_dssall (void);
10184 void vec_dst (const vector unsigned char *, int, const int);
10185 void vec_dst (const vector signed char *, int, const int);
10186 void vec_dst (const vector bool char *, int, const int);
10187 void vec_dst (const vector unsigned short *, int, const int);
10188 void vec_dst (const vector signed short *, int, const int);
10189 void vec_dst (const vector bool short *, int, const int);
10190 void vec_dst (const vector pixel *, int, const int);
10191 void vec_dst (const vector unsigned int *, int, const int);
10192 void vec_dst (const vector signed int *, int, const int);
10193 void vec_dst (const vector bool int *, int, const int);
10194 void vec_dst (const vector float *, int, const int);
10195 void vec_dst (const unsigned char *, int, const int);
10196 void vec_dst (const signed char *, int, const int);
10197 void vec_dst (const unsigned short *, int, const int);
10198 void vec_dst (const short *, int, const int);
10199 void vec_dst (const unsigned int *, int, const int);
10200 void vec_dst (const int *, int, const int);
10201 void vec_dst (const unsigned long *, int, const int);
10202 void vec_dst (const long *, int, const int);
10203 void vec_dst (const float *, int, const int);
10205 void vec_dstst (const vector unsigned char *, int, const int);
10206 void vec_dstst (const vector signed char *, int, const int);
10207 void vec_dstst (const vector bool char *, int, const int);
10208 void vec_dstst (const vector unsigned short *, int, const int);
10209 void vec_dstst (const vector signed short *, int, const int);
10210 void vec_dstst (const vector bool short *, int, const int);
10211 void vec_dstst (const vector pixel *, int, const int);
10212 void vec_dstst (const vector unsigned int *, int, const int);
10213 void vec_dstst (const vector signed int *, int, const int);
10214 void vec_dstst (const vector bool int *, int, const int);
10215 void vec_dstst (const vector float *, int, const int);
10216 void vec_dstst (const unsigned char *, int, const int);
10217 void vec_dstst (const signed char *, int, const int);
10218 void vec_dstst (const unsigned short *, int, const int);
10219 void vec_dstst (const short *, int, const int);
10220 void vec_dstst (const unsigned int *, int, const int);
10221 void vec_dstst (const int *, int, const int);
10222 void vec_dstst (const unsigned long *, int, const int);
10223 void vec_dstst (const long *, int, const int);
10224 void vec_dstst (const float *, int, const int);
10226 void vec_dststt (const vector unsigned char *, int, const int);
10227 void vec_dststt (const vector signed char *, int, const int);
10228 void vec_dststt (const vector bool char *, int, const int);
10229 void vec_dststt (const vector unsigned short *, int, const int);
10230 void vec_dststt (const vector signed short *, int, const int);
10231 void vec_dststt (const vector bool short *, int, const int);
10232 void vec_dststt (const vector pixel *, int, const int);
10233 void vec_dststt (const vector unsigned int *, int, const int);
10234 void vec_dststt (const vector signed int *, int, const int);
10235 void vec_dststt (const vector bool int *, int, const int);
10236 void vec_dststt (const vector float *, int, const int);
10237 void vec_dststt (const unsigned char *, int, const int);
10238 void vec_dststt (const signed char *, int, const int);
10239 void vec_dststt (const unsigned short *, int, const int);
10240 void vec_dststt (const short *, int, const int);
10241 void vec_dststt (const unsigned int *, int, const int);
10242 void vec_dststt (const int *, int, const int);
10243 void vec_dststt (const unsigned long *, int, const int);
10244 void vec_dststt (const long *, int, const int);
10245 void vec_dststt (const float *, int, const int);
10247 void vec_dstt (const vector unsigned char *, int, const int);
10248 void vec_dstt (const vector signed char *, int, const int);
10249 void vec_dstt (const vector bool char *, int, const int);
10250 void vec_dstt (const vector unsigned short *, int, const int);
10251 void vec_dstt (const vector signed short *, int, const int);
10252 void vec_dstt (const vector bool short *, int, const int);
10253 void vec_dstt (const vector pixel *, int, const int);
10254 void vec_dstt (const vector unsigned int *, int, const int);
10255 void vec_dstt (const vector signed int *, int, const int);
10256 void vec_dstt (const vector bool int *, int, const int);
10257 void vec_dstt (const vector float *, int, const int);
10258 void vec_dstt (const unsigned char *, int, const int);
10259 void vec_dstt (const signed char *, int, const int);
10260 void vec_dstt (const unsigned short *, int, const int);
10261 void vec_dstt (const short *, int, const int);
10262 void vec_dstt (const unsigned int *, int, const int);
10263 void vec_dstt (const int *, int, const int);
10264 void vec_dstt (const unsigned long *, int, const int);
10265 void vec_dstt (const long *, int, const int);
10266 void vec_dstt (const float *, int, const int);
10268 vector float vec_expte (vector float);
10270 vector float vec_floor (vector float);
10272 vector float vec_ld (int, const vector float *);
10273 vector float vec_ld (int, const float *);
10274 vector bool int vec_ld (int, const vector bool int *);
10275 vector signed int vec_ld (int, const vector signed int *);
10276 vector signed int vec_ld (int, const int *);
10277 vector signed int vec_ld (int, const long *);
10278 vector unsigned int vec_ld (int, const vector unsigned int *);
10279 vector unsigned int vec_ld (int, const unsigned int *);
10280 vector unsigned int vec_ld (int, const unsigned long *);
10281 vector bool short vec_ld (int, const vector bool short *);
10282 vector pixel vec_ld (int, const vector pixel *);
10283 vector signed short vec_ld (int, const vector signed short *);
10284 vector signed short vec_ld (int, const short *);
10285 vector unsigned short vec_ld (int, const vector unsigned short *);
10286 vector unsigned short vec_ld (int, const unsigned short *);
10287 vector bool char vec_ld (int, const vector bool char *);
10288 vector signed char vec_ld (int, const vector signed char *);
10289 vector signed char vec_ld (int, const signed char *);
10290 vector unsigned char vec_ld (int, const vector unsigned char *);
10291 vector unsigned char vec_ld (int, const unsigned char *);
10293 vector signed char vec_lde (int, const signed char *);
10294 vector unsigned char vec_lde (int, const unsigned char *);
10295 vector signed short vec_lde (int, const short *);
10296 vector unsigned short vec_lde (int, const unsigned short *);
10297 vector float vec_lde (int, const float *);
10298 vector signed int vec_lde (int, const int *);
10299 vector unsigned int vec_lde (int, const unsigned int *);
10300 vector signed int vec_lde (int, const long *);
10301 vector unsigned int vec_lde (int, const unsigned long *);
10303 vector float vec_lvewx (int, float *);
10304 vector signed int vec_lvewx (int, int *);
10305 vector unsigned int vec_lvewx (int, unsigned int *);
10306 vector signed int vec_lvewx (int, long *);
10307 vector unsigned int vec_lvewx (int, unsigned long *);
10309 vector signed short vec_lvehx (int, short *);
10310 vector unsigned short vec_lvehx (int, unsigned short *);
10312 vector signed char vec_lvebx (int, char *);
10313 vector unsigned char vec_lvebx (int, unsigned char *);
10315 vector float vec_ldl (int, const vector float *);
10316 vector float vec_ldl (int, const float *);
10317 vector bool int vec_ldl (int, const vector bool int *);
10318 vector signed int vec_ldl (int, const vector signed int *);
10319 vector signed int vec_ldl (int, const int *);
10320 vector signed int vec_ldl (int, const long *);
10321 vector unsigned int vec_ldl (int, const vector unsigned int *);
10322 vector unsigned int vec_ldl (int, const unsigned int *);
10323 vector unsigned int vec_ldl (int, const unsigned long *);
10324 vector bool short vec_ldl (int, const vector bool short *);
10325 vector pixel vec_ldl (int, const vector pixel *);
10326 vector signed short vec_ldl (int, const vector signed short *);
10327 vector signed short vec_ldl (int, const short *);
10328 vector unsigned short vec_ldl (int, const vector unsigned short *);
10329 vector unsigned short vec_ldl (int, const unsigned short *);
10330 vector bool char vec_ldl (int, const vector bool char *);
10331 vector signed char vec_ldl (int, const vector signed char *);
10332 vector signed char vec_ldl (int, const signed char *);
10333 vector unsigned char vec_ldl (int, const vector unsigned char *);
10334 vector unsigned char vec_ldl (int, const unsigned char *);
10336 vector float vec_loge (vector float);
10338 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
10339 vector unsigned char vec_lvsl (int, const volatile signed char *);
10340 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
10341 vector unsigned char vec_lvsl (int, const volatile short *);
10342 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
10343 vector unsigned char vec_lvsl (int, const volatile int *);
10344 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
10345 vector unsigned char vec_lvsl (int, const volatile long *);
10346 vector unsigned char vec_lvsl (int, const volatile float *);
10348 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
10349 vector unsigned char vec_lvsr (int, const volatile signed char *);
10350 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
10351 vector unsigned char vec_lvsr (int, const volatile short *);
10352 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
10353 vector unsigned char vec_lvsr (int, const volatile int *);
10354 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
10355 vector unsigned char vec_lvsr (int, const volatile long *);
10356 vector unsigned char vec_lvsr (int, const volatile float *);
10358 vector float vec_madd (vector float, vector float, vector float);
10360 vector signed short vec_madds (vector signed short,
10361 vector signed short,
10362 vector signed short);
10364 vector unsigned char vec_max (vector bool char, vector unsigned char);
10365 vector unsigned char vec_max (vector unsigned char, vector bool char);
10366 vector unsigned char vec_max (vector unsigned char,
10367 vector unsigned char);
10368 vector signed char vec_max (vector bool char, vector signed char);
10369 vector signed char vec_max (vector signed char, vector bool char);
10370 vector signed char vec_max (vector signed char, vector signed char);
10371 vector unsigned short vec_max (vector bool short,
10372 vector unsigned short);
10373 vector unsigned short vec_max (vector unsigned short,
10374 vector bool short);
10375 vector unsigned short vec_max (vector unsigned short,
10376 vector unsigned short);
10377 vector signed short vec_max (vector bool short, vector signed short);
10378 vector signed short vec_max (vector signed short, vector bool short);
10379 vector signed short vec_max (vector signed short, vector signed short);
10380 vector unsigned int vec_max (vector bool int, vector unsigned int);
10381 vector unsigned int vec_max (vector unsigned int, vector bool int);
10382 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
10383 vector signed int vec_max (vector bool int, vector signed int);
10384 vector signed int vec_max (vector signed int, vector bool int);
10385 vector signed int vec_max (vector signed int, vector signed int);
10386 vector float vec_max (vector float, vector float);
10388 vector float vec_vmaxfp (vector float, vector float);
10390 vector signed int vec_vmaxsw (vector bool int, vector signed int);
10391 vector signed int vec_vmaxsw (vector signed int, vector bool int);
10392 vector signed int vec_vmaxsw (vector signed int, vector signed int);
10394 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
10395 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
10396 vector unsigned int vec_vmaxuw (vector unsigned int,
10397 vector unsigned int);
10399 vector signed short vec_vmaxsh (vector bool short, vector signed short);
10400 vector signed short vec_vmaxsh (vector signed short, vector bool short);
10401 vector signed short vec_vmaxsh (vector signed short,
10402 vector signed short);
10404 vector unsigned short vec_vmaxuh (vector bool short,
10405 vector unsigned short);
10406 vector unsigned short vec_vmaxuh (vector unsigned short,
10407 vector bool short);
10408 vector unsigned short vec_vmaxuh (vector unsigned short,
10409 vector unsigned short);
10411 vector signed char vec_vmaxsb (vector bool char, vector signed char);
10412 vector signed char vec_vmaxsb (vector signed char, vector bool char);
10413 vector signed char vec_vmaxsb (vector signed char, vector signed char);
10415 vector unsigned char vec_vmaxub (vector bool char,
10416 vector unsigned char);
10417 vector unsigned char vec_vmaxub (vector unsigned char,
10419 vector unsigned char vec_vmaxub (vector unsigned char,
10420 vector unsigned char);
10422 vector bool char vec_mergeh (vector bool char, vector bool char);
10423 vector signed char vec_mergeh (vector signed char, vector signed char);
10424 vector unsigned char vec_mergeh (vector unsigned char,
10425 vector unsigned char);
10426 vector bool short vec_mergeh (vector bool short, vector bool short);
10427 vector pixel vec_mergeh (vector pixel, vector pixel);
10428 vector signed short vec_mergeh (vector signed short,
10429 vector signed short);
10430 vector unsigned short vec_mergeh (vector unsigned short,
10431 vector unsigned short);
10432 vector float vec_mergeh (vector float, vector float);
10433 vector bool int vec_mergeh (vector bool int, vector bool int);
10434 vector signed int vec_mergeh (vector signed int, vector signed int);
10435 vector unsigned int vec_mergeh (vector unsigned int,
10436 vector unsigned int);
10438 vector float vec_vmrghw (vector float, vector float);
10439 vector bool int vec_vmrghw (vector bool int, vector bool int);
10440 vector signed int vec_vmrghw (vector signed int, vector signed int);
10441 vector unsigned int vec_vmrghw (vector unsigned int,
10442 vector unsigned int);
10444 vector bool short vec_vmrghh (vector bool short, vector bool short);
10445 vector signed short vec_vmrghh (vector signed short,
10446 vector signed short);
10447 vector unsigned short vec_vmrghh (vector unsigned short,
10448 vector unsigned short);
10449 vector pixel vec_vmrghh (vector pixel, vector pixel);
10451 vector bool char vec_vmrghb (vector bool char, vector bool char);
10452 vector signed char vec_vmrghb (vector signed char, vector signed char);
10453 vector unsigned char vec_vmrghb (vector unsigned char,
10454 vector unsigned char);
10456 vector bool char vec_mergel (vector bool char, vector bool char);
10457 vector signed char vec_mergel (vector signed char, vector signed char);
10458 vector unsigned char vec_mergel (vector unsigned char,
10459 vector unsigned char);
10460 vector bool short vec_mergel (vector bool short, vector bool short);
10461 vector pixel vec_mergel (vector pixel, vector pixel);
10462 vector signed short vec_mergel (vector signed short,
10463 vector signed short);
10464 vector unsigned short vec_mergel (vector unsigned short,
10465 vector unsigned short);
10466 vector float vec_mergel (vector float, vector float);
10467 vector bool int vec_mergel (vector bool int, vector bool int);
10468 vector signed int vec_mergel (vector signed int, vector signed int);
10469 vector unsigned int vec_mergel (vector unsigned int,
10470 vector unsigned int);
10472 vector float vec_vmrglw (vector float, vector float);
10473 vector signed int vec_vmrglw (vector signed int, vector signed int);
10474 vector unsigned int vec_vmrglw (vector unsigned int,
10475 vector unsigned int);
10476 vector bool int vec_vmrglw (vector bool int, vector bool int);
10478 vector bool short vec_vmrglh (vector bool short, vector bool short);
10479 vector signed short vec_vmrglh (vector signed short,
10480 vector signed short);
10481 vector unsigned short vec_vmrglh (vector unsigned short,
10482 vector unsigned short);
10483 vector pixel vec_vmrglh (vector pixel, vector pixel);
10485 vector bool char vec_vmrglb (vector bool char, vector bool char);
10486 vector signed char vec_vmrglb (vector signed char, vector signed char);
10487 vector unsigned char vec_vmrglb (vector unsigned char,
10488 vector unsigned char);
10490 vector unsigned short vec_mfvscr (void);
10492 vector unsigned char vec_min (vector bool char, vector unsigned char);
10493 vector unsigned char vec_min (vector unsigned char, vector bool char);
10494 vector unsigned char vec_min (vector unsigned char,
10495 vector unsigned char);
10496 vector signed char vec_min (vector bool char, vector signed char);
10497 vector signed char vec_min (vector signed char, vector bool char);
10498 vector signed char vec_min (vector signed char, vector signed char);
10499 vector unsigned short vec_min (vector bool short,
10500 vector unsigned short);
10501 vector unsigned short vec_min (vector unsigned short,
10502 vector bool short);
10503 vector unsigned short vec_min (vector unsigned short,
10504 vector unsigned short);
10505 vector signed short vec_min (vector bool short, vector signed short);
10506 vector signed short vec_min (vector signed short, vector bool short);
10507 vector signed short vec_min (vector signed short, vector signed short);
10508 vector unsigned int vec_min (vector bool int, vector unsigned int);
10509 vector unsigned int vec_min (vector unsigned int, vector bool int);
10510 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
10511 vector signed int vec_min (vector bool int, vector signed int);
10512 vector signed int vec_min (vector signed int, vector bool int);
10513 vector signed int vec_min (vector signed int, vector signed int);
10514 vector float vec_min (vector float, vector float);
10516 vector float vec_vminfp (vector float, vector float);
10518 vector signed int vec_vminsw (vector bool int, vector signed int);
10519 vector signed int vec_vminsw (vector signed int, vector bool int);
10520 vector signed int vec_vminsw (vector signed int, vector signed int);
10522 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
10523 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
10524 vector unsigned int vec_vminuw (vector unsigned int,
10525 vector unsigned int);
10527 vector signed short vec_vminsh (vector bool short, vector signed short);
10528 vector signed short vec_vminsh (vector signed short, vector bool short);
10529 vector signed short vec_vminsh (vector signed short,
10530 vector signed short);
10532 vector unsigned short vec_vminuh (vector bool short,
10533 vector unsigned short);
10534 vector unsigned short vec_vminuh (vector unsigned short,
10535 vector bool short);
10536 vector unsigned short vec_vminuh (vector unsigned short,
10537 vector unsigned short);
10539 vector signed char vec_vminsb (vector bool char, vector signed char);
10540 vector signed char vec_vminsb (vector signed char, vector bool char);
10541 vector signed char vec_vminsb (vector signed char, vector signed char);
10543 vector unsigned char vec_vminub (vector bool char,
10544 vector unsigned char);
10545 vector unsigned char vec_vminub (vector unsigned char,
10547 vector unsigned char vec_vminub (vector unsigned char,
10548 vector unsigned char);
10550 vector signed short vec_mladd (vector signed short,
10551 vector signed short,
10552 vector signed short);
10553 vector signed short vec_mladd (vector signed short,
10554 vector unsigned short,
10555 vector unsigned short);
10556 vector signed short vec_mladd (vector unsigned short,
10557 vector signed short,
10558 vector signed short);
10559 vector unsigned short vec_mladd (vector unsigned short,
10560 vector unsigned short,
10561 vector unsigned short);
10563 vector signed short vec_mradds (vector signed short,
10564 vector signed short,
10565 vector signed short);
10567 vector unsigned int vec_msum (vector unsigned char,
10568 vector unsigned char,
10569 vector unsigned int);
10570 vector signed int vec_msum (vector signed char,
10571 vector unsigned char,
10572 vector signed int);
10573 vector unsigned int vec_msum (vector unsigned short,
10574 vector unsigned short,
10575 vector unsigned int);
10576 vector signed int vec_msum (vector signed short,
10577 vector signed short,
10578 vector signed int);
10580 vector signed int vec_vmsumshm (vector signed short,
10581 vector signed short,
10582 vector signed int);
10584 vector unsigned int vec_vmsumuhm (vector unsigned short,
10585 vector unsigned short,
10586 vector unsigned int);
10588 vector signed int vec_vmsummbm (vector signed char,
10589 vector unsigned char,
10590 vector signed int);
10592 vector unsigned int vec_vmsumubm (vector unsigned char,
10593 vector unsigned char,
10594 vector unsigned int);
10596 vector unsigned int vec_msums (vector unsigned short,
10597 vector unsigned short,
10598 vector unsigned int);
10599 vector signed int vec_msums (vector signed short,
10600 vector signed short,
10601 vector signed int);
10603 vector signed int vec_vmsumshs (vector signed short,
10604 vector signed short,
10605 vector signed int);
10607 vector unsigned int vec_vmsumuhs (vector unsigned short,
10608 vector unsigned short,
10609 vector unsigned int);
10611 void vec_mtvscr (vector signed int);
10612 void vec_mtvscr (vector unsigned int);
10613 void vec_mtvscr (vector bool int);
10614 void vec_mtvscr (vector signed short);
10615 void vec_mtvscr (vector unsigned short);
10616 void vec_mtvscr (vector bool short);
10617 void vec_mtvscr (vector pixel);
10618 void vec_mtvscr (vector signed char);
10619 void vec_mtvscr (vector unsigned char);
10620 void vec_mtvscr (vector bool char);
10622 vector unsigned short vec_mule (vector unsigned char,
10623 vector unsigned char);
10624 vector signed short vec_mule (vector signed char,
10625 vector signed char);
10626 vector unsigned int vec_mule (vector unsigned short,
10627 vector unsigned short);
10628 vector signed int vec_mule (vector signed short, vector signed short);
10630 vector signed int vec_vmulesh (vector signed short,
10631 vector signed short);
10633 vector unsigned int vec_vmuleuh (vector unsigned short,
10634 vector unsigned short);
10636 vector signed short vec_vmulesb (vector signed char,
10637 vector signed char);
10639 vector unsigned short vec_vmuleub (vector unsigned char,
10640 vector unsigned char);
10642 vector unsigned short vec_mulo (vector unsigned char,
10643 vector unsigned char);
10644 vector signed short vec_mulo (vector signed char, vector signed char);
10645 vector unsigned int vec_mulo (vector unsigned short,
10646 vector unsigned short);
10647 vector signed int vec_mulo (vector signed short, vector signed short);
10649 vector signed int vec_vmulosh (vector signed short,
10650 vector signed short);
10652 vector unsigned int vec_vmulouh (vector unsigned short,
10653 vector unsigned short);
10655 vector signed short vec_vmulosb (vector signed char,
10656 vector signed char);
10658 vector unsigned short vec_vmuloub (vector unsigned char,
10659 vector unsigned char);
10661 vector float vec_nmsub (vector float, vector float, vector float);
10663 vector float vec_nor (vector float, vector float);
10664 vector signed int vec_nor (vector signed int, vector signed int);
10665 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
10666 vector bool int vec_nor (vector bool int, vector bool int);
10667 vector signed short vec_nor (vector signed short, vector signed short);
10668 vector unsigned short vec_nor (vector unsigned short,
10669 vector unsigned short);
10670 vector bool short vec_nor (vector bool short, vector bool short);
10671 vector signed char vec_nor (vector signed char, vector signed char);
10672 vector unsigned char vec_nor (vector unsigned char,
10673 vector unsigned char);
10674 vector bool char vec_nor (vector bool char, vector bool char);
10676 vector float vec_or (vector float, vector float);
10677 vector float vec_or (vector float, vector bool int);
10678 vector float vec_or (vector bool int, vector float);
10679 vector bool int vec_or (vector bool int, vector bool int);
10680 vector signed int vec_or (vector bool int, vector signed int);
10681 vector signed int vec_or (vector signed int, vector bool int);
10682 vector signed int vec_or (vector signed int, vector signed int);
10683 vector unsigned int vec_or (vector bool int, vector unsigned int);
10684 vector unsigned int vec_or (vector unsigned int, vector bool int);
10685 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
10686 vector bool short vec_or (vector bool short, vector bool short);
10687 vector signed short vec_or (vector bool short, vector signed short);
10688 vector signed short vec_or (vector signed short, vector bool short);
10689 vector signed short vec_or (vector signed short, vector signed short);
10690 vector unsigned short vec_or (vector bool short, vector unsigned short);
10691 vector unsigned short vec_or (vector unsigned short, vector bool short);
10692 vector unsigned short vec_or (vector unsigned short,
10693 vector unsigned short);
10694 vector signed char vec_or (vector bool char, vector signed char);
10695 vector bool char vec_or (vector bool char, vector bool char);
10696 vector signed char vec_or (vector signed char, vector bool char);
10697 vector signed char vec_or (vector signed char, vector signed char);
10698 vector unsigned char vec_or (vector bool char, vector unsigned char);
10699 vector unsigned char vec_or (vector unsigned char, vector bool char);
10700 vector unsigned char vec_or (vector unsigned char,
10701 vector unsigned char);
10703 vector signed char vec_pack (vector signed short, vector signed short);
10704 vector unsigned char vec_pack (vector unsigned short,
10705 vector unsigned short);
10706 vector bool char vec_pack (vector bool short, vector bool short);
10707 vector signed short vec_pack (vector signed int, vector signed int);
10708 vector unsigned short vec_pack (vector unsigned int,
10709 vector unsigned int);
10710 vector bool short vec_pack (vector bool int, vector bool int);
10712 vector bool short vec_vpkuwum (vector bool int, vector bool int);
10713 vector signed short vec_vpkuwum (vector signed int, vector signed int);
10714 vector unsigned short vec_vpkuwum (vector unsigned int,
10715 vector unsigned int);
10717 vector bool char vec_vpkuhum (vector bool short, vector bool short);
10718 vector signed char vec_vpkuhum (vector signed short,
10719 vector signed short);
10720 vector unsigned char vec_vpkuhum (vector unsigned short,
10721 vector unsigned short);
10723 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
10725 vector unsigned char vec_packs (vector unsigned short,
10726 vector unsigned short);
10727 vector signed char vec_packs (vector signed short, vector signed short);
10728 vector unsigned short vec_packs (vector unsigned int,
10729 vector unsigned int);
10730 vector signed short vec_packs (vector signed int, vector signed int);
10732 vector signed short vec_vpkswss (vector signed int, vector signed int);
10734 vector unsigned short vec_vpkuwus (vector unsigned int,
10735 vector unsigned int);
10737 vector signed char vec_vpkshss (vector signed short,
10738 vector signed short);
10740 vector unsigned char vec_vpkuhus (vector unsigned short,
10741 vector unsigned short);
10743 vector unsigned char vec_packsu (vector unsigned short,
10744 vector unsigned short);
10745 vector unsigned char vec_packsu (vector signed short,
10746 vector signed short);
10747 vector unsigned short vec_packsu (vector unsigned int,
10748 vector unsigned int);
10749 vector unsigned short vec_packsu (vector signed int, vector signed int);
10751 vector unsigned short vec_vpkswus (vector signed int,
10752 vector signed int);
10754 vector unsigned char vec_vpkshus (vector signed short,
10755 vector signed short);
10757 vector float vec_perm (vector float,
10759 vector unsigned char);
10760 vector signed int vec_perm (vector signed int,
10762 vector unsigned char);
10763 vector unsigned int vec_perm (vector unsigned int,
10764 vector unsigned int,
10765 vector unsigned char);
10766 vector bool int vec_perm (vector bool int,
10768 vector unsigned char);
10769 vector signed short vec_perm (vector signed short,
10770 vector signed short,
10771 vector unsigned char);
10772 vector unsigned short vec_perm (vector unsigned short,
10773 vector unsigned short,
10774 vector unsigned char);
10775 vector bool short vec_perm (vector bool short,
10777 vector unsigned char);
10778 vector pixel vec_perm (vector pixel,
10780 vector unsigned char);
10781 vector signed char vec_perm (vector signed char,
10782 vector signed char,
10783 vector unsigned char);
10784 vector unsigned char vec_perm (vector unsigned char,
10785 vector unsigned char,
10786 vector unsigned char);
10787 vector bool char vec_perm (vector bool char,
10789 vector unsigned char);
10791 vector float vec_re (vector float);
10793 vector signed char vec_rl (vector signed char,
10794 vector unsigned char);
10795 vector unsigned char vec_rl (vector unsigned char,
10796 vector unsigned char);
10797 vector signed short vec_rl (vector signed short, vector unsigned short);
10798 vector unsigned short vec_rl (vector unsigned short,
10799 vector unsigned short);
10800 vector signed int vec_rl (vector signed int, vector unsigned int);
10801 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
10803 vector signed int vec_vrlw (vector signed int, vector unsigned int);
10804 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
10806 vector signed short vec_vrlh (vector signed short,
10807 vector unsigned short);
10808 vector unsigned short vec_vrlh (vector unsigned short,
10809 vector unsigned short);
10811 vector signed char vec_vrlb (vector signed char, vector unsigned char);
10812 vector unsigned char vec_vrlb (vector unsigned char,
10813 vector unsigned char);
10815 vector float vec_round (vector float);
10817 vector float vec_rsqrte (vector float);
10819 vector float vec_sel (vector float, vector float, vector bool int);
10820 vector float vec_sel (vector float, vector float, vector unsigned int);
10821 vector signed int vec_sel (vector signed int,
10824 vector signed int vec_sel (vector signed int,
10826 vector unsigned int);
10827 vector unsigned int vec_sel (vector unsigned int,
10828 vector unsigned int,
10830 vector unsigned int vec_sel (vector unsigned int,
10831 vector unsigned int,
10832 vector unsigned int);
10833 vector bool int vec_sel (vector bool int,
10836 vector bool int vec_sel (vector bool int,
10838 vector unsigned int);
10839 vector signed short vec_sel (vector signed short,
10840 vector signed short,
10841 vector bool short);
10842 vector signed short vec_sel (vector signed short,
10843 vector signed short,
10844 vector unsigned short);
10845 vector unsigned short vec_sel (vector unsigned short,
10846 vector unsigned short,
10847 vector bool short);
10848 vector unsigned short vec_sel (vector unsigned short,
10849 vector unsigned short,
10850 vector unsigned short);
10851 vector bool short vec_sel (vector bool short,
10853 vector bool short);
10854 vector bool short vec_sel (vector bool short,
10856 vector unsigned short);
10857 vector signed char vec_sel (vector signed char,
10858 vector signed char,
10860 vector signed char vec_sel (vector signed char,
10861 vector signed char,
10862 vector unsigned char);
10863 vector unsigned char vec_sel (vector unsigned char,
10864 vector unsigned char,
10866 vector unsigned char vec_sel (vector unsigned char,
10867 vector unsigned char,
10868 vector unsigned char);
10869 vector bool char vec_sel (vector bool char,
10872 vector bool char vec_sel (vector bool char,
10874 vector unsigned char);
10876 vector signed char vec_sl (vector signed char,
10877 vector unsigned char);
10878 vector unsigned char vec_sl (vector unsigned char,
10879 vector unsigned char);
10880 vector signed short vec_sl (vector signed short, vector unsigned short);
10881 vector unsigned short vec_sl (vector unsigned short,
10882 vector unsigned short);
10883 vector signed int vec_sl (vector signed int, vector unsigned int);
10884 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
10886 vector signed int vec_vslw (vector signed int, vector unsigned int);
10887 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
10889 vector signed short vec_vslh (vector signed short,
10890 vector unsigned short);
10891 vector unsigned short vec_vslh (vector unsigned short,
10892 vector unsigned short);
10894 vector signed char vec_vslb (vector signed char, vector unsigned char);
10895 vector unsigned char vec_vslb (vector unsigned char,
10896 vector unsigned char);
10898 vector float vec_sld (vector float, vector float, const int);
10899 vector signed int vec_sld (vector signed int,
10902 vector unsigned int vec_sld (vector unsigned int,
10903 vector unsigned int,
10905 vector bool int vec_sld (vector bool int,
10908 vector signed short vec_sld (vector signed short,
10909 vector signed short,
10911 vector unsigned short vec_sld (vector unsigned short,
10912 vector unsigned short,
10914 vector bool short vec_sld (vector bool short,
10917 vector pixel vec_sld (vector pixel,
10920 vector signed char vec_sld (vector signed char,
10921 vector signed char,
10923 vector unsigned char vec_sld (vector unsigned char,
10924 vector unsigned char,
10926 vector bool char vec_sld (vector bool char,
10930 vector signed int vec_sll (vector signed int,
10931 vector unsigned int);
10932 vector signed int vec_sll (vector signed int,
10933 vector unsigned short);
10934 vector signed int vec_sll (vector signed int,
10935 vector unsigned char);
10936 vector unsigned int vec_sll (vector unsigned int,
10937 vector unsigned int);
10938 vector unsigned int vec_sll (vector unsigned int,
10939 vector unsigned short);
10940 vector unsigned int vec_sll (vector unsigned int,
10941 vector unsigned char);
10942 vector bool int vec_sll (vector bool int,
10943 vector unsigned int);
10944 vector bool int vec_sll (vector bool int,
10945 vector unsigned short);
10946 vector bool int vec_sll (vector bool int,
10947 vector unsigned char);
10948 vector signed short vec_sll (vector signed short,
10949 vector unsigned int);
10950 vector signed short vec_sll (vector signed short,
10951 vector unsigned short);
10952 vector signed short vec_sll (vector signed short,
10953 vector unsigned char);
10954 vector unsigned short vec_sll (vector unsigned short,
10955 vector unsigned int);
10956 vector unsigned short vec_sll (vector unsigned short,
10957 vector unsigned short);
10958 vector unsigned short vec_sll (vector unsigned short,
10959 vector unsigned char);
10960 vector bool short vec_sll (vector bool short, vector unsigned int);
10961 vector bool short vec_sll (vector bool short, vector unsigned short);
10962 vector bool short vec_sll (vector bool short, vector unsigned char);
10963 vector pixel vec_sll (vector pixel, vector unsigned int);
10964 vector pixel vec_sll (vector pixel, vector unsigned short);
10965 vector pixel vec_sll (vector pixel, vector unsigned char);
10966 vector signed char vec_sll (vector signed char, vector unsigned int);
10967 vector signed char vec_sll (vector signed char, vector unsigned short);
10968 vector signed char vec_sll (vector signed char, vector unsigned char);
10969 vector unsigned char vec_sll (vector unsigned char,
10970 vector unsigned int);
10971 vector unsigned char vec_sll (vector unsigned char,
10972 vector unsigned short);
10973 vector unsigned char vec_sll (vector unsigned char,
10974 vector unsigned char);
10975 vector bool char vec_sll (vector bool char, vector unsigned int);
10976 vector bool char vec_sll (vector bool char, vector unsigned short);
10977 vector bool char vec_sll (vector bool char, vector unsigned char);
10979 vector float vec_slo (vector float, vector signed char);
10980 vector float vec_slo (vector float, vector unsigned char);
10981 vector signed int vec_slo (vector signed int, vector signed char);
10982 vector signed int vec_slo (vector signed int, vector unsigned char);
10983 vector unsigned int vec_slo (vector unsigned int, vector signed char);
10984 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
10985 vector signed short vec_slo (vector signed short, vector signed char);
10986 vector signed short vec_slo (vector signed short, vector unsigned char);
10987 vector unsigned short vec_slo (vector unsigned short,
10988 vector signed char);
10989 vector unsigned short vec_slo (vector unsigned short,
10990 vector unsigned char);
10991 vector pixel vec_slo (vector pixel, vector signed char);
10992 vector pixel vec_slo (vector pixel, vector unsigned char);
10993 vector signed char vec_slo (vector signed char, vector signed char);
10994 vector signed char vec_slo (vector signed char, vector unsigned char);
10995 vector unsigned char vec_slo (vector unsigned char, vector signed char);
10996 vector unsigned char vec_slo (vector unsigned char,
10997 vector unsigned char);
10999 vector signed char vec_splat (vector signed char, const int);
11000 vector unsigned char vec_splat (vector unsigned char, const int);
11001 vector bool char vec_splat (vector bool char, const int);
11002 vector signed short vec_splat (vector signed short, const int);
11003 vector unsigned short vec_splat (vector unsigned short, const int);
11004 vector bool short vec_splat (vector bool short, const int);
11005 vector pixel vec_splat (vector pixel, const int);
11006 vector float vec_splat (vector float, const int);
11007 vector signed int vec_splat (vector signed int, const int);
11008 vector unsigned int vec_splat (vector unsigned int, const int);
11009 vector bool int vec_splat (vector bool int, const int);
11011 vector float vec_vspltw (vector float, const int);
11012 vector signed int vec_vspltw (vector signed int, const int);
11013 vector unsigned int vec_vspltw (vector unsigned int, const int);
11014 vector bool int vec_vspltw (vector bool int, const int);
11016 vector bool short vec_vsplth (vector bool short, const int);
11017 vector signed short vec_vsplth (vector signed short, const int);
11018 vector unsigned short vec_vsplth (vector unsigned short, const int);
11019 vector pixel vec_vsplth (vector pixel, const int);
11021 vector signed char vec_vspltb (vector signed char, const int);
11022 vector unsigned char vec_vspltb (vector unsigned char, const int);
11023 vector bool char vec_vspltb (vector bool char, const int);
11025 vector signed char vec_splat_s8 (const int);
11027 vector signed short vec_splat_s16 (const int);
11029 vector signed int vec_splat_s32 (const int);
11031 vector unsigned char vec_splat_u8 (const int);
11033 vector unsigned short vec_splat_u16 (const int);
11035 vector unsigned int vec_splat_u32 (const int);
11037 vector signed char vec_sr (vector signed char, vector unsigned char);
11038 vector unsigned char vec_sr (vector unsigned char,
11039 vector unsigned char);
11040 vector signed short vec_sr (vector signed short,
11041 vector unsigned short);
11042 vector unsigned short vec_sr (vector unsigned short,
11043 vector unsigned short);
11044 vector signed int vec_sr (vector signed int, vector unsigned int);
11045 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
11047 vector signed int vec_vsrw (vector signed int, vector unsigned int);
11048 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
11050 vector signed short vec_vsrh (vector signed short,
11051 vector unsigned short);
11052 vector unsigned short vec_vsrh (vector unsigned short,
11053 vector unsigned short);
11055 vector signed char vec_vsrb (vector signed char, vector unsigned char);
11056 vector unsigned char vec_vsrb (vector unsigned char,
11057 vector unsigned char);
11059 vector signed char vec_sra (vector signed char, vector unsigned char);
11060 vector unsigned char vec_sra (vector unsigned char,
11061 vector unsigned char);
11062 vector signed short vec_sra (vector signed short,
11063 vector unsigned short);
11064 vector unsigned short vec_sra (vector unsigned short,
11065 vector unsigned short);
11066 vector signed int vec_sra (vector signed int, vector unsigned int);
11067 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
11069 vector signed int vec_vsraw (vector signed int, vector unsigned int);
11070 vector unsigned int vec_vsraw (vector unsigned int,
11071 vector unsigned int);
11073 vector signed short vec_vsrah (vector signed short,
11074 vector unsigned short);
11075 vector unsigned short vec_vsrah (vector unsigned short,
11076 vector unsigned short);
11078 vector signed char vec_vsrab (vector signed char, vector unsigned char);
11079 vector unsigned char vec_vsrab (vector unsigned char,
11080 vector unsigned char);
11082 vector signed int vec_srl (vector signed int, vector unsigned int);
11083 vector signed int vec_srl (vector signed int, vector unsigned short);
11084 vector signed int vec_srl (vector signed int, vector unsigned char);
11085 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
11086 vector unsigned int vec_srl (vector unsigned int,
11087 vector unsigned short);
11088 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
11089 vector bool int vec_srl (vector bool int, vector unsigned int);
11090 vector bool int vec_srl (vector bool int, vector unsigned short);
11091 vector bool int vec_srl (vector bool int, vector unsigned char);
11092 vector signed short vec_srl (vector signed short, vector unsigned int);
11093 vector signed short vec_srl (vector signed short,
11094 vector unsigned short);
11095 vector signed short vec_srl (vector signed short, vector unsigned char);
11096 vector unsigned short vec_srl (vector unsigned short,
11097 vector unsigned int);
11098 vector unsigned short vec_srl (vector unsigned short,
11099 vector unsigned short);
11100 vector unsigned short vec_srl (vector unsigned short,
11101 vector unsigned char);
11102 vector bool short vec_srl (vector bool short, vector unsigned int);
11103 vector bool short vec_srl (vector bool short, vector unsigned short);
11104 vector bool short vec_srl (vector bool short, vector unsigned char);
11105 vector pixel vec_srl (vector pixel, vector unsigned int);
11106 vector pixel vec_srl (vector pixel, vector unsigned short);
11107 vector pixel vec_srl (vector pixel, vector unsigned char);
11108 vector signed char vec_srl (vector signed char, vector unsigned int);
11109 vector signed char vec_srl (vector signed char, vector unsigned short);
11110 vector signed char vec_srl (vector signed char, vector unsigned char);
11111 vector unsigned char vec_srl (vector unsigned char,
11112 vector unsigned int);
11113 vector unsigned char vec_srl (vector unsigned char,
11114 vector unsigned short);
11115 vector unsigned char vec_srl (vector unsigned char,
11116 vector unsigned char);
11117 vector bool char vec_srl (vector bool char, vector unsigned int);
11118 vector bool char vec_srl (vector bool char, vector unsigned short);
11119 vector bool char vec_srl (vector bool char, vector unsigned char);
11121 vector float vec_sro (vector float, vector signed char);
11122 vector float vec_sro (vector float, vector unsigned char);
11123 vector signed int vec_sro (vector signed int, vector signed char);
11124 vector signed int vec_sro (vector signed int, vector unsigned char);
11125 vector unsigned int vec_sro (vector unsigned int, vector signed char);
11126 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
11127 vector signed short vec_sro (vector signed short, vector signed char);
11128 vector signed short vec_sro (vector signed short, vector unsigned char);
11129 vector unsigned short vec_sro (vector unsigned short,
11130 vector signed char);
11131 vector unsigned short vec_sro (vector unsigned short,
11132 vector unsigned char);
11133 vector pixel vec_sro (vector pixel, vector signed char);
11134 vector pixel vec_sro (vector pixel, vector unsigned char);
11135 vector signed char vec_sro (vector signed char, vector signed char);
11136 vector signed char vec_sro (vector signed char, vector unsigned char);
11137 vector unsigned char vec_sro (vector unsigned char, vector signed char);
11138 vector unsigned char vec_sro (vector unsigned char,
11139 vector unsigned char);
11141 void vec_st (vector float, int, vector float *);
11142 void vec_st (vector float, int, float *);
11143 void vec_st (vector signed int, int, vector signed int *);
11144 void vec_st (vector signed int, int, int *);
11145 void vec_st (vector unsigned int, int, vector unsigned int *);
11146 void vec_st (vector unsigned int, int, unsigned int *);
11147 void vec_st (vector bool int, int, vector bool int *);
11148 void vec_st (vector bool int, int, unsigned int *);
11149 void vec_st (vector bool int, int, int *);
11150 void vec_st (vector signed short, int, vector signed short *);
11151 void vec_st (vector signed short, int, short *);
11152 void vec_st (vector unsigned short, int, vector unsigned short *);
11153 void vec_st (vector unsigned short, int, unsigned short *);
11154 void vec_st (vector bool short, int, vector bool short *);
11155 void vec_st (vector bool short, int, unsigned short *);
11156 void vec_st (vector pixel, int, vector pixel *);
11157 void vec_st (vector pixel, int, unsigned short *);
11158 void vec_st (vector pixel, int, short *);
11159 void vec_st (vector bool short, int, short *);
11160 void vec_st (vector signed char, int, vector signed char *);
11161 void vec_st (vector signed char, int, signed char *);
11162 void vec_st (vector unsigned char, int, vector unsigned char *);
11163 void vec_st (vector unsigned char, int, unsigned char *);
11164 void vec_st (vector bool char, int, vector bool char *);
11165 void vec_st (vector bool char, int, unsigned char *);
11166 void vec_st (vector bool char, int, signed char *);
11168 void vec_ste (vector signed char, int, signed char *);
11169 void vec_ste (vector unsigned char, int, unsigned char *);
11170 void vec_ste (vector bool char, int, signed char *);
11171 void vec_ste (vector bool char, int, unsigned char *);
11172 void vec_ste (vector signed short, int, short *);
11173 void vec_ste (vector unsigned short, int, unsigned short *);
11174 void vec_ste (vector bool short, int, short *);
11175 void vec_ste (vector bool short, int, unsigned short *);
11176 void vec_ste (vector pixel, int, short *);
11177 void vec_ste (vector pixel, int, unsigned short *);
11178 void vec_ste (vector float, int, float *);
11179 void vec_ste (vector signed int, int, int *);
11180 void vec_ste (vector unsigned int, int, unsigned int *);
11181 void vec_ste (vector bool int, int, int *);
11182 void vec_ste (vector bool int, int, unsigned int *);
11184 void vec_stvewx (vector float, int, float *);
11185 void vec_stvewx (vector signed int, int, int *);
11186 void vec_stvewx (vector unsigned int, int, unsigned int *);
11187 void vec_stvewx (vector bool int, int, int *);
11188 void vec_stvewx (vector bool int, int, unsigned int *);
11190 void vec_stvehx (vector signed short, int, short *);
11191 void vec_stvehx (vector unsigned short, int, unsigned short *);
11192 void vec_stvehx (vector bool short, int, short *);
11193 void vec_stvehx (vector bool short, int, unsigned short *);
11194 void vec_stvehx (vector pixel, int, short *);
11195 void vec_stvehx (vector pixel, int, unsigned short *);
11197 void vec_stvebx (vector signed char, int, signed char *);
11198 void vec_stvebx (vector unsigned char, int, unsigned char *);
11199 void vec_stvebx (vector bool char, int, signed char *);
11200 void vec_stvebx (vector bool char, int, unsigned char *);
11202 void vec_stl (vector float, int, vector float *);
11203 void vec_stl (vector float, int, float *);
11204 void vec_stl (vector signed int, int, vector signed int *);
11205 void vec_stl (vector signed int, int, int *);
11206 void vec_stl (vector unsigned int, int, vector unsigned int *);
11207 void vec_stl (vector unsigned int, int, unsigned int *);
11208 void vec_stl (vector bool int, int, vector bool int *);
11209 void vec_stl (vector bool int, int, unsigned int *);
11210 void vec_stl (vector bool int, int, int *);
11211 void vec_stl (vector signed short, int, vector signed short *);
11212 void vec_stl (vector signed short, int, short *);
11213 void vec_stl (vector unsigned short, int, vector unsigned short *);
11214 void vec_stl (vector unsigned short, int, unsigned short *);
11215 void vec_stl (vector bool short, int, vector bool short *);
11216 void vec_stl (vector bool short, int, unsigned short *);
11217 void vec_stl (vector bool short, int, short *);
11218 void vec_stl (vector pixel, int, vector pixel *);
11219 void vec_stl (vector pixel, int, unsigned short *);
11220 void vec_stl (vector pixel, int, short *);
11221 void vec_stl (vector signed char, int, vector signed char *);
11222 void vec_stl (vector signed char, int, signed char *);
11223 void vec_stl (vector unsigned char, int, vector unsigned char *);
11224 void vec_stl (vector unsigned char, int, unsigned char *);
11225 void vec_stl (vector bool char, int, vector bool char *);
11226 void vec_stl (vector bool char, int, unsigned char *);
11227 void vec_stl (vector bool char, int, signed char *);
11229 vector signed char vec_sub (vector bool char, vector signed char);
11230 vector signed char vec_sub (vector signed char, vector bool char);
11231 vector signed char vec_sub (vector signed char, vector signed char);
11232 vector unsigned char vec_sub (vector bool char, vector unsigned char);
11233 vector unsigned char vec_sub (vector unsigned char, vector bool char);
11234 vector unsigned char vec_sub (vector unsigned char,
11235 vector unsigned char);
11236 vector signed short vec_sub (vector bool short, vector signed short);
11237 vector signed short vec_sub (vector signed short, vector bool short);
11238 vector signed short vec_sub (vector signed short, vector signed short);
11239 vector unsigned short vec_sub (vector bool short,
11240 vector unsigned short);
11241 vector unsigned short vec_sub (vector unsigned short,
11242 vector bool short);
11243 vector unsigned short vec_sub (vector unsigned short,
11244 vector unsigned short);
11245 vector signed int vec_sub (vector bool int, vector signed int);
11246 vector signed int vec_sub (vector signed int, vector bool int);
11247 vector signed int vec_sub (vector signed int, vector signed int);
11248 vector unsigned int vec_sub (vector bool int, vector unsigned int);
11249 vector unsigned int vec_sub (vector unsigned int, vector bool int);
11250 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
11251 vector float vec_sub (vector float, vector float);
11253 vector float vec_vsubfp (vector float, vector float);
11255 vector signed int vec_vsubuwm (vector bool int, vector signed int);
11256 vector signed int vec_vsubuwm (vector signed int, vector bool int);
11257 vector signed int vec_vsubuwm (vector signed int, vector signed int);
11258 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
11259 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
11260 vector unsigned int vec_vsubuwm (vector unsigned int,
11261 vector unsigned int);
11263 vector signed short vec_vsubuhm (vector bool short,
11264 vector signed short);
11265 vector signed short vec_vsubuhm (vector signed short,
11266 vector bool short);
11267 vector signed short vec_vsubuhm (vector signed short,
11268 vector signed short);
11269 vector unsigned short vec_vsubuhm (vector bool short,
11270 vector unsigned short);
11271 vector unsigned short vec_vsubuhm (vector unsigned short,
11272 vector bool short);
11273 vector unsigned short vec_vsubuhm (vector unsigned short,
11274 vector unsigned short);
11276 vector signed char vec_vsububm (vector bool char, vector signed char);
11277 vector signed char vec_vsububm (vector signed char, vector bool char);
11278 vector signed char vec_vsububm (vector signed char, vector signed char);
11279 vector unsigned char vec_vsububm (vector bool char,
11280 vector unsigned char);
11281 vector unsigned char vec_vsububm (vector unsigned char,
11283 vector unsigned char vec_vsububm (vector unsigned char,
11284 vector unsigned char);
11286 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
11288 vector unsigned char vec_subs (vector bool char, vector unsigned char);
11289 vector unsigned char vec_subs (vector unsigned char, vector bool char);
11290 vector unsigned char vec_subs (vector unsigned char,
11291 vector unsigned char);
11292 vector signed char vec_subs (vector bool char, vector signed char);
11293 vector signed char vec_subs (vector signed char, vector bool char);
11294 vector signed char vec_subs (vector signed char, vector signed char);
11295 vector unsigned short vec_subs (vector bool short,
11296 vector unsigned short);
11297 vector unsigned short vec_subs (vector unsigned short,
11298 vector bool short);
11299 vector unsigned short vec_subs (vector unsigned short,
11300 vector unsigned short);
11301 vector signed short vec_subs (vector bool short, vector signed short);
11302 vector signed short vec_subs (vector signed short, vector bool short);
11303 vector signed short vec_subs (vector signed short, vector signed short);
11304 vector unsigned int vec_subs (vector bool int, vector unsigned int);
11305 vector unsigned int vec_subs (vector unsigned int, vector bool int);
11306 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
11307 vector signed int vec_subs (vector bool int, vector signed int);
11308 vector signed int vec_subs (vector signed int, vector bool int);
11309 vector signed int vec_subs (vector signed int, vector signed int);
11311 vector signed int vec_vsubsws (vector bool int, vector signed int);
11312 vector signed int vec_vsubsws (vector signed int, vector bool int);
11313 vector signed int vec_vsubsws (vector signed int, vector signed int);
11315 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
11316 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
11317 vector unsigned int vec_vsubuws (vector unsigned int,
11318 vector unsigned int);
11320 vector signed short vec_vsubshs (vector bool short,
11321 vector signed short);
11322 vector signed short vec_vsubshs (vector signed short,
11323 vector bool short);
11324 vector signed short vec_vsubshs (vector signed short,
11325 vector signed short);
11327 vector unsigned short vec_vsubuhs (vector bool short,
11328 vector unsigned short);
11329 vector unsigned short vec_vsubuhs (vector unsigned short,
11330 vector bool short);
11331 vector unsigned short vec_vsubuhs (vector unsigned short,
11332 vector unsigned short);
11334 vector signed char vec_vsubsbs (vector bool char, vector signed char);
11335 vector signed char vec_vsubsbs (vector signed char, vector bool char);
11336 vector signed char vec_vsubsbs (vector signed char, vector signed char);
11338 vector unsigned char vec_vsububs (vector bool char,
11339 vector unsigned char);
11340 vector unsigned char vec_vsububs (vector unsigned char,
11342 vector unsigned char vec_vsububs (vector unsigned char,
11343 vector unsigned char);
11345 vector unsigned int vec_sum4s (vector unsigned char,
11346 vector unsigned int);
11347 vector signed int vec_sum4s (vector signed char, vector signed int);
11348 vector signed int vec_sum4s (vector signed short, vector signed int);
11350 vector signed int vec_vsum4shs (vector signed short, vector signed int);
11352 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
11354 vector unsigned int vec_vsum4ubs (vector unsigned char,
11355 vector unsigned int);
11357 vector signed int vec_sum2s (vector signed int, vector signed int);
11359 vector signed int vec_sums (vector signed int, vector signed int);
11361 vector float vec_trunc (vector float);
11363 vector signed short vec_unpackh (vector signed char);
11364 vector bool short vec_unpackh (vector bool char);
11365 vector signed int vec_unpackh (vector signed short);
11366 vector bool int vec_unpackh (vector bool short);
11367 vector unsigned int vec_unpackh (vector pixel);
11369 vector bool int vec_vupkhsh (vector bool short);
11370 vector signed int vec_vupkhsh (vector signed short);
11372 vector unsigned int vec_vupkhpx (vector pixel);
11374 vector bool short vec_vupkhsb (vector bool char);
11375 vector signed short vec_vupkhsb (vector signed char);
11377 vector signed short vec_unpackl (vector signed char);
11378 vector bool short vec_unpackl (vector bool char);
11379 vector unsigned int vec_unpackl (vector pixel);
11380 vector signed int vec_unpackl (vector signed short);
11381 vector bool int vec_unpackl (vector bool short);
11383 vector unsigned int vec_vupklpx (vector pixel);
11385 vector bool int vec_vupklsh (vector bool short);
11386 vector signed int vec_vupklsh (vector signed short);
11388 vector bool short vec_vupklsb (vector bool char);
11389 vector signed short vec_vupklsb (vector signed char);
11391 vector float vec_xor (vector float, vector float);
11392 vector float vec_xor (vector float, vector bool int);
11393 vector float vec_xor (vector bool int, vector float);
11394 vector bool int vec_xor (vector bool int, vector bool int);
11395 vector signed int vec_xor (vector bool int, vector signed int);
11396 vector signed int vec_xor (vector signed int, vector bool int);
11397 vector signed int vec_xor (vector signed int, vector signed int);
11398 vector unsigned int vec_xor (vector bool int, vector unsigned int);
11399 vector unsigned int vec_xor (vector unsigned int, vector bool int);
11400 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
11401 vector bool short vec_xor (vector bool short, vector bool short);
11402 vector signed short vec_xor (vector bool short, vector signed short);
11403 vector signed short vec_xor (vector signed short, vector bool short);
11404 vector signed short vec_xor (vector signed short, vector signed short);
11405 vector unsigned short vec_xor (vector bool short,
11406 vector unsigned short);
11407 vector unsigned short vec_xor (vector unsigned short,
11408 vector bool short);
11409 vector unsigned short vec_xor (vector unsigned short,
11410 vector unsigned short);
11411 vector signed char vec_xor (vector bool char, vector signed char);
11412 vector bool char vec_xor (vector bool char, vector bool char);
11413 vector signed char vec_xor (vector signed char, vector bool char);
11414 vector signed char vec_xor (vector signed char, vector signed char);
11415 vector unsigned char vec_xor (vector bool char, vector unsigned char);
11416 vector unsigned char vec_xor (vector unsigned char, vector bool char);
11417 vector unsigned char vec_xor (vector unsigned char,
11418 vector unsigned char);
11420 int vec_all_eq (vector signed char, vector bool char);
11421 int vec_all_eq (vector signed char, vector signed char);
11422 int vec_all_eq (vector unsigned char, vector bool char);
11423 int vec_all_eq (vector unsigned char, vector unsigned char);
11424 int vec_all_eq (vector bool char, vector bool char);
11425 int vec_all_eq (vector bool char, vector unsigned char);
11426 int vec_all_eq (vector bool char, vector signed char);
11427 int vec_all_eq (vector signed short, vector bool short);
11428 int vec_all_eq (vector signed short, vector signed short);
11429 int vec_all_eq (vector unsigned short, vector bool short);
11430 int vec_all_eq (vector unsigned short, vector unsigned short);
11431 int vec_all_eq (vector bool short, vector bool short);
11432 int vec_all_eq (vector bool short, vector unsigned short);
11433 int vec_all_eq (vector bool short, vector signed short);
11434 int vec_all_eq (vector pixel, vector pixel);
11435 int vec_all_eq (vector signed int, vector bool int);
11436 int vec_all_eq (vector signed int, vector signed int);
11437 int vec_all_eq (vector unsigned int, vector bool int);
11438 int vec_all_eq (vector unsigned int, vector unsigned int);
11439 int vec_all_eq (vector bool int, vector bool int);
11440 int vec_all_eq (vector bool int, vector unsigned int);
11441 int vec_all_eq (vector bool int, vector signed int);
11442 int vec_all_eq (vector float, vector float);
11444 int vec_all_ge (vector bool char, vector unsigned char);
11445 int vec_all_ge (vector unsigned char, vector bool char);
11446 int vec_all_ge (vector unsigned char, vector unsigned char);
11447 int vec_all_ge (vector bool char, vector signed char);
11448 int vec_all_ge (vector signed char, vector bool char);
11449 int vec_all_ge (vector signed char, vector signed char);
11450 int vec_all_ge (vector bool short, vector unsigned short);
11451 int vec_all_ge (vector unsigned short, vector bool short);
11452 int vec_all_ge (vector unsigned short, vector unsigned short);
11453 int vec_all_ge (vector signed short, vector signed short);
11454 int vec_all_ge (vector bool short, vector signed short);
11455 int vec_all_ge (vector signed short, vector bool short);
11456 int vec_all_ge (vector bool int, vector unsigned int);
11457 int vec_all_ge (vector unsigned int, vector bool int);
11458 int vec_all_ge (vector unsigned int, vector unsigned int);
11459 int vec_all_ge (vector bool int, vector signed int);
11460 int vec_all_ge (vector signed int, vector bool int);
11461 int vec_all_ge (vector signed int, vector signed int);
11462 int vec_all_ge (vector float, vector float);
11464 int vec_all_gt (vector bool char, vector unsigned char);
11465 int vec_all_gt (vector unsigned char, vector bool char);
11466 int vec_all_gt (vector unsigned char, vector unsigned char);
11467 int vec_all_gt (vector bool char, vector signed char);
11468 int vec_all_gt (vector signed char, vector bool char);
11469 int vec_all_gt (vector signed char, vector signed char);
11470 int vec_all_gt (vector bool short, vector unsigned short);
11471 int vec_all_gt (vector unsigned short, vector bool short);
11472 int vec_all_gt (vector unsigned short, vector unsigned short);
11473 int vec_all_gt (vector bool short, vector signed short);
11474 int vec_all_gt (vector signed short, vector bool short);
11475 int vec_all_gt (vector signed short, vector signed short);
11476 int vec_all_gt (vector bool int, vector unsigned int);
11477 int vec_all_gt (vector unsigned int, vector bool int);
11478 int vec_all_gt (vector unsigned int, vector unsigned int);
11479 int vec_all_gt (vector bool int, vector signed int);
11480 int vec_all_gt (vector signed int, vector bool int);
11481 int vec_all_gt (vector signed int, vector signed int);
11482 int vec_all_gt (vector float, vector float);
11484 int vec_all_in (vector float, vector float);
11486 int vec_all_le (vector bool char, vector unsigned char);
11487 int vec_all_le (vector unsigned char, vector bool char);
11488 int vec_all_le (vector unsigned char, vector unsigned char);
11489 int vec_all_le (vector bool char, vector signed char);
11490 int vec_all_le (vector signed char, vector bool char);
11491 int vec_all_le (vector signed char, vector signed char);
11492 int vec_all_le (vector bool short, vector unsigned short);
11493 int vec_all_le (vector unsigned short, vector bool short);
11494 int vec_all_le (vector unsigned short, vector unsigned short);
11495 int vec_all_le (vector bool short, vector signed short);
11496 int vec_all_le (vector signed short, vector bool short);
11497 int vec_all_le (vector signed short, vector signed short);
11498 int vec_all_le (vector bool int, vector unsigned int);
11499 int vec_all_le (vector unsigned int, vector bool int);
11500 int vec_all_le (vector unsigned int, vector unsigned int);
11501 int vec_all_le (vector bool int, vector signed int);
11502 int vec_all_le (vector signed int, vector bool int);
11503 int vec_all_le (vector signed int, vector signed int);
11504 int vec_all_le (vector float, vector float);
11506 int vec_all_lt (vector bool char, vector unsigned char);
11507 int vec_all_lt (vector unsigned char, vector bool char);
11508 int vec_all_lt (vector unsigned char, vector unsigned char);
11509 int vec_all_lt (vector bool char, vector signed char);
11510 int vec_all_lt (vector signed char, vector bool char);
11511 int vec_all_lt (vector signed char, vector signed char);
11512 int vec_all_lt (vector bool short, vector unsigned short);
11513 int vec_all_lt (vector unsigned short, vector bool short);
11514 int vec_all_lt (vector unsigned short, vector unsigned short);
11515 int vec_all_lt (vector bool short, vector signed short);
11516 int vec_all_lt (vector signed short, vector bool short);
11517 int vec_all_lt (vector signed short, vector signed short);
11518 int vec_all_lt (vector bool int, vector unsigned int);
11519 int vec_all_lt (vector unsigned int, vector bool int);
11520 int vec_all_lt (vector unsigned int, vector unsigned int);
11521 int vec_all_lt (vector bool int, vector signed int);
11522 int vec_all_lt (vector signed int, vector bool int);
11523 int vec_all_lt (vector signed int, vector signed int);
11524 int vec_all_lt (vector float, vector float);
11526 int vec_all_nan (vector float);
11528 int vec_all_ne (vector signed char, vector bool char);
11529 int vec_all_ne (vector signed char, vector signed char);
11530 int vec_all_ne (vector unsigned char, vector bool char);
11531 int vec_all_ne (vector unsigned char, vector unsigned char);
11532 int vec_all_ne (vector bool char, vector bool char);
11533 int vec_all_ne (vector bool char, vector unsigned char);
11534 int vec_all_ne (vector bool char, vector signed char);
11535 int vec_all_ne (vector signed short, vector bool short);
11536 int vec_all_ne (vector signed short, vector signed short);
11537 int vec_all_ne (vector unsigned short, vector bool short);
11538 int vec_all_ne (vector unsigned short, vector unsigned short);
11539 int vec_all_ne (vector bool short, vector bool short);
11540 int vec_all_ne (vector bool short, vector unsigned short);
11541 int vec_all_ne (vector bool short, vector signed short);
11542 int vec_all_ne (vector pixel, vector pixel);
11543 int vec_all_ne (vector signed int, vector bool int);
11544 int vec_all_ne (vector signed int, vector signed int);
11545 int vec_all_ne (vector unsigned int, vector bool int);
11546 int vec_all_ne (vector unsigned int, vector unsigned int);
11547 int vec_all_ne (vector bool int, vector bool int);
11548 int vec_all_ne (vector bool int, vector unsigned int);
11549 int vec_all_ne (vector bool int, vector signed int);
11550 int vec_all_ne (vector float, vector float);
11552 int vec_all_nge (vector float, vector float);
11554 int vec_all_ngt (vector float, vector float);
11556 int vec_all_nle (vector float, vector float);
11558 int vec_all_nlt (vector float, vector float);
11560 int vec_all_numeric (vector float);
11562 int vec_any_eq (vector signed char, vector bool char);
11563 int vec_any_eq (vector signed char, vector signed char);
11564 int vec_any_eq (vector unsigned char, vector bool char);
11565 int vec_any_eq (vector unsigned char, vector unsigned char);
11566 int vec_any_eq (vector bool char, vector bool char);
11567 int vec_any_eq (vector bool char, vector unsigned char);
11568 int vec_any_eq (vector bool char, vector signed char);
11569 int vec_any_eq (vector signed short, vector bool short);
11570 int vec_any_eq (vector signed short, vector signed short);
11571 int vec_any_eq (vector unsigned short, vector bool short);
11572 int vec_any_eq (vector unsigned short, vector unsigned short);
11573 int vec_any_eq (vector bool short, vector bool short);
11574 int vec_any_eq (vector bool short, vector unsigned short);
11575 int vec_any_eq (vector bool short, vector signed short);
11576 int vec_any_eq (vector pixel, vector pixel);
11577 int vec_any_eq (vector signed int, vector bool int);
11578 int vec_any_eq (vector signed int, vector signed int);
11579 int vec_any_eq (vector unsigned int, vector bool int);
11580 int vec_any_eq (vector unsigned int, vector unsigned int);
11581 int vec_any_eq (vector bool int, vector bool int);
11582 int vec_any_eq (vector bool int, vector unsigned int);
11583 int vec_any_eq (vector bool int, vector signed int);
11584 int vec_any_eq (vector float, vector float);
11586 int vec_any_ge (vector signed char, vector bool char);
11587 int vec_any_ge (vector unsigned char, vector bool char);
11588 int vec_any_ge (vector unsigned char, vector unsigned char);
11589 int vec_any_ge (vector signed char, vector signed char);
11590 int vec_any_ge (vector bool char, vector unsigned char);
11591 int vec_any_ge (vector bool char, vector signed char);
11592 int vec_any_ge (vector unsigned short, vector bool short);
11593 int vec_any_ge (vector unsigned short, vector unsigned short);
11594 int vec_any_ge (vector signed short, vector signed short);
11595 int vec_any_ge (vector signed short, vector bool short);
11596 int vec_any_ge (vector bool short, vector unsigned short);
11597 int vec_any_ge (vector bool short, vector signed short);
11598 int vec_any_ge (vector signed int, vector bool int);
11599 int vec_any_ge (vector unsigned int, vector bool int);
11600 int vec_any_ge (vector unsigned int, vector unsigned int);
11601 int vec_any_ge (vector signed int, vector signed int);
11602 int vec_any_ge (vector bool int, vector unsigned int);
11603 int vec_any_ge (vector bool int, vector signed int);
11604 int vec_any_ge (vector float, vector float);
11606 int vec_any_gt (vector bool char, vector unsigned char);
11607 int vec_any_gt (vector unsigned char, vector bool char);
11608 int vec_any_gt (vector unsigned char, vector unsigned char);
11609 int vec_any_gt (vector bool char, vector signed char);
11610 int vec_any_gt (vector signed char, vector bool char);
11611 int vec_any_gt (vector signed char, vector signed char);
11612 int vec_any_gt (vector bool short, vector unsigned short);
11613 int vec_any_gt (vector unsigned short, vector bool short);
11614 int vec_any_gt (vector unsigned short, vector unsigned short);
11615 int vec_any_gt (vector bool short, vector signed short);
11616 int vec_any_gt (vector signed short, vector bool short);
11617 int vec_any_gt (vector signed short, vector signed short);
11618 int vec_any_gt (vector bool int, vector unsigned int);
11619 int vec_any_gt (vector unsigned int, vector bool int);
11620 int vec_any_gt (vector unsigned int, vector unsigned int);
11621 int vec_any_gt (vector bool int, vector signed int);
11622 int vec_any_gt (vector signed int, vector bool int);
11623 int vec_any_gt (vector signed int, vector signed int);
11624 int vec_any_gt (vector float, vector float);
11626 int vec_any_le (vector bool char, vector unsigned char);
11627 int vec_any_le (vector unsigned char, vector bool char);
11628 int vec_any_le (vector unsigned char, vector unsigned char);
11629 int vec_any_le (vector bool char, vector signed char);
11630 int vec_any_le (vector signed char, vector bool char);
11631 int vec_any_le (vector signed char, vector signed char);
11632 int vec_any_le (vector bool short, vector unsigned short);
11633 int vec_any_le (vector unsigned short, vector bool short);
11634 int vec_any_le (vector unsigned short, vector unsigned short);
11635 int vec_any_le (vector bool short, vector signed short);
11636 int vec_any_le (vector signed short, vector bool short);
11637 int vec_any_le (vector signed short, vector signed short);
11638 int vec_any_le (vector bool int, vector unsigned int);
11639 int vec_any_le (vector unsigned int, vector bool int);
11640 int vec_any_le (vector unsigned int, vector unsigned int);
11641 int vec_any_le (vector bool int, vector signed int);
11642 int vec_any_le (vector signed int, vector bool int);
11643 int vec_any_le (vector signed int, vector signed int);
11644 int vec_any_le (vector float, vector float);
11646 int vec_any_lt (vector bool char, vector unsigned char);
11647 int vec_any_lt (vector unsigned char, vector bool char);
11648 int vec_any_lt (vector unsigned char, vector unsigned char);
11649 int vec_any_lt (vector bool char, vector signed char);
11650 int vec_any_lt (vector signed char, vector bool char);
11651 int vec_any_lt (vector signed char, vector signed char);
11652 int vec_any_lt (vector bool short, vector unsigned short);
11653 int vec_any_lt (vector unsigned short, vector bool short);
11654 int vec_any_lt (vector unsigned short, vector unsigned short);
11655 int vec_any_lt (vector bool short, vector signed short);
11656 int vec_any_lt (vector signed short, vector bool short);
11657 int vec_any_lt (vector signed short, vector signed short);
11658 int vec_any_lt (vector bool int, vector unsigned int);
11659 int vec_any_lt (vector unsigned int, vector bool int);
11660 int vec_any_lt (vector unsigned int, vector unsigned int);
11661 int vec_any_lt (vector bool int, vector signed int);
11662 int vec_any_lt (vector signed int, vector bool int);
11663 int vec_any_lt (vector signed int, vector signed int);
11664 int vec_any_lt (vector float, vector float);
11666 int vec_any_nan (vector float);
11668 int vec_any_ne (vector signed char, vector bool char);
11669 int vec_any_ne (vector signed char, vector signed char);
11670 int vec_any_ne (vector unsigned char, vector bool char);
11671 int vec_any_ne (vector unsigned char, vector unsigned char);
11672 int vec_any_ne (vector bool char, vector bool char);
11673 int vec_any_ne (vector bool char, vector unsigned char);
11674 int vec_any_ne (vector bool char, vector signed char);
11675 int vec_any_ne (vector signed short, vector bool short);
11676 int vec_any_ne (vector signed short, vector signed short);
11677 int vec_any_ne (vector unsigned short, vector bool short);
11678 int vec_any_ne (vector unsigned short, vector unsigned short);
11679 int vec_any_ne (vector bool short, vector bool short);
11680 int vec_any_ne (vector bool short, vector unsigned short);
11681 int vec_any_ne (vector bool short, vector signed short);
11682 int vec_any_ne (vector pixel, vector pixel);
11683 int vec_any_ne (vector signed int, vector bool int);
11684 int vec_any_ne (vector signed int, vector signed int);
11685 int vec_any_ne (vector unsigned int, vector bool int);
11686 int vec_any_ne (vector unsigned int, vector unsigned int);
11687 int vec_any_ne (vector bool int, vector bool int);
11688 int vec_any_ne (vector bool int, vector unsigned int);
11689 int vec_any_ne (vector bool int, vector signed int);
11690 int vec_any_ne (vector float, vector float);
11692 int vec_any_nge (vector float, vector float);
11694 int vec_any_ngt (vector float, vector float);
11696 int vec_any_nle (vector float, vector float);
11698 int vec_any_nlt (vector float, vector float);
11700 int vec_any_numeric (vector float);
11702 int vec_any_out (vector float, vector float);
11705 If the vector/scalar (VSX) instruction set is available, the following
11706 additional functions are available:
11709 vector double vec_abs (vector double);
11710 vector double vec_add (vector double, vector double);
11711 vector double vec_and (vector double, vector double);
11712 vector double vec_and (vector double, vector bool long);
11713 vector double vec_and (vector bool long, vector double);
11714 vector double vec_andc (vector double, vector double);
11715 vector double vec_andc (vector double, vector bool long);
11716 vector double vec_andc (vector bool long, vector double);
11717 vector double vec_ceil (vector double);
11718 vector bool long vec_cmpeq (vector double, vector double);
11719 vector bool long vec_cmpge (vector double, vector double);
11720 vector bool long vec_cmpgt (vector double, vector double);
11721 vector bool long vec_cmple (vector double, vector double);
11722 vector bool long vec_cmplt (vector double, vector double);
11723 vector float vec_div (vector float, vector float);
11724 vector double vec_div (vector double, vector double);
11725 vector double vec_floor (vector double);
11726 vector double vec_madd (vector double, vector double, vector double);
11727 vector double vec_max (vector double, vector double);
11728 vector double vec_min (vector double, vector double);
11729 vector float vec_msub (vector float, vector float, vector float);
11730 vector double vec_msub (vector double, vector double, vector double);
11731 vector float vec_mul (vector float, vector float);
11732 vector double vec_mul (vector double, vector double);
11733 vector float vec_nearbyint (vector float);
11734 vector double vec_nearbyint (vector double);
11735 vector float vec_nmadd (vector float, vector float, vector float);
11736 vector double vec_nmadd (vector double, vector double, vector double);
11737 vector double vec_nmsub (vector double, vector double, vector double);
11738 vector double vec_nor (vector double, vector double);
11739 vector double vec_or (vector double, vector double);
11740 vector double vec_or (vector double, vector bool long);
11741 vector double vec_or (vector bool long, vector double);
11742 vector double vec_perm (vector double,
11744 vector unsigned char);
11745 vector float vec_rint (vector float);
11746 vector double vec_rint (vector double);
11747 vector double vec_sel (vector double, vector double, vector bool long);
11748 vector double vec_sel (vector double, vector double, vector unsigned long);
11749 vector double vec_sub (vector double, vector double);
11750 vector float vec_sqrt (vector float);
11751 vector double vec_sqrt (vector double);
11752 vector double vec_trunc (vector double);
11753 vector double vec_xor (vector double, vector double);
11754 vector double vec_xor (vector double, vector bool long);
11755 vector double vec_xor (vector bool long, vector double);
11756 int vec_all_eq (vector double, vector double);
11757 int vec_all_ge (vector double, vector double);
11758 int vec_all_gt (vector double, vector double);
11759 int vec_all_le (vector double, vector double);
11760 int vec_all_lt (vector double, vector double);
11761 int vec_all_nan (vector double);
11762 int vec_all_ne (vector double, vector double);
11763 int vec_all_nge (vector double, vector double);
11764 int vec_all_ngt (vector double, vector double);
11765 int vec_all_nle (vector double, vector double);
11766 int vec_all_nlt (vector double, vector double);
11767 int vec_all_numeric (vector double);
11768 int vec_any_eq (vector double, vector double);
11769 int vec_any_ge (vector double, vector double);
11770 int vec_any_gt (vector double, vector double);
11771 int vec_any_le (vector double, vector double);
11772 int vec_any_lt (vector double, vector double);
11773 int vec_any_nan (vector double);
11774 int vec_any_ne (vector double, vector double);
11775 int vec_any_nge (vector double, vector double);
11776 int vec_any_ngt (vector double, vector double);
11777 int vec_any_nle (vector double, vector double);
11778 int vec_any_nlt (vector double, vector double);
11779 int vec_any_numeric (vector double);
11782 GCC provides a few other builtins on Powerpc to access certain instructions:
11784 float __builtin_recipdivf (float, float);
11785 float __builtin_rsqrtf (float);
11786 double __builtin_recipdiv (double, double);
11787 long __builtin_bpermd (long, long);
11788 int __builtin_bswap16 (int);
11791 @node RX Built-in Functions
11792 @subsection RX Built-in Functions
11793 GCC supports some of the RX instructions which cannot be expressed in
11794 the C programming language via the use of built-in functions. The
11795 following functions are supported:
11797 @deftypefn {Built-in Function} void __builtin_rx_brk (void)
11798 Generates the @code{brk} machine instruction.
11801 @deftypefn {Built-in Function} void __builtin_rx_clrpsw (int)
11802 Generates the @code{clrpsw} machine instruction to clear the specified
11803 bit in the processor status word.
11806 @deftypefn {Built-in Function} void __builtin_rx_int (int)
11807 Generates the @code{int} machine instruction to generate an interrupt
11808 with the specified value.
11811 @deftypefn {Built-in Function} void __builtin_rx_machi (int, int)
11812 Generates the @code{machi} machine instruction to add the result of
11813 multiplying the top 16-bits of the two arguments into the
11817 @deftypefn {Built-in Function} void __builtin_rx_maclo (int, int)
11818 Generates the @code{maclo} machine instruction to add the result of
11819 multiplying the bottom 16-bits of the two arguments into the
11823 @deftypefn {Built-in Function} void __builtin_rx_mulhi (int, int)
11824 Generates the @code{mulhi} machine instruction to place the result of
11825 multiplying the top 16-bits of the two arguments into the
11829 @deftypefn {Built-in Function} void __builtin_rx_mullo (int, int)
11830 Generates the @code{mullo} machine instruction to place the result of
11831 multiplying the bottom 16-bits of the two arguments into the
11835 @deftypefn {Built-in Function} int __builtin_rx_mvfachi (void)
11836 Generates the @code{mvfachi} machine instruction to read the top
11837 32-bits of the accumulator.
11840 @deftypefn {Built-in Function} int __builtin_rx_mvfacmi (void)
11841 Generates the @code{mvfacmi} machine instruction to read the middle
11842 32-bits of the accumulator.
11845 @deftypefn {Built-in Function} int __builtin_rx_mvfc (int)
11846 Generates the @code{mvfc} machine instruction which reads the control
11847 register specified in its argument and returns its value.
11850 @deftypefn {Built-in Function} void __builtin_rx_mvtachi (int)
11851 Generates the @code{mvtachi} machine instruction to set the top
11852 32-bits of the accumulator.
11855 @deftypefn {Built-in Function} void __builtin_rx_mvtaclo (int)
11856 Generates the @code{mvtaclo} machine instruction to set the bottom
11857 32-bits of the accumulator.
11860 @deftypefn {Built-in Function} void __builtin_rx_mvtc (int reg, int val)
11861 Generates the @code{mvtc} machine instruction which sets control
11862 register number @code{reg} to @code{val}.
11865 @deftypefn {Built-in Function} void __builtin_rx_mvtipl (int)
11866 Generates the @code{mvtipl} machine instruction set the interrupt
11870 @deftypefn {Built-in Function} void __builtin_rx_racw (int)
11871 Generates the @code{racw} machine instruction to round the accumulator
11872 according to the specified mode.
11875 @deftypefn {Built-in Function} int __builtin_rx_revw (int)
11876 Generates the @code{revw} machine instruction which swaps the bytes in
11877 the argument so that bits 0--7 now occupy bits 8--15 and vice versa,
11878 and also bits 16--23 occupy bits 24--31 and vice versa.
11881 @deftypefn {Built-in Function} void __builtin_rx_rmpa (void)
11882 Generates the @code{rmpa} machine instruction which initiates a
11883 repeated multiply and accumulate sequence.
11886 @deftypefn {Built-in Function} void __builtin_rx_round (float)
11887 Generates the @code{round} machine instruction which returns the
11888 floating point argument rounded according to the current rounding mode
11889 set in the floating point status word register.
11892 @deftypefn {Built-in Function} int __builtin_rx_sat (int)
11893 Generates the @code{sat} machine instruction which returns the
11894 saturated value of the argument.
11897 @deftypefn {Built-in Function} void __builtin_rx_setpsw (int)
11898 Generates the @code{setpsw} machine instruction to set the specified
11899 bit in the processor status word.
11902 @deftypefn {Built-in Function} void __builtin_rx_wait (void)
11903 Generates the @code{wait} machine instruction.
11906 @node SPARC VIS Built-in Functions
11907 @subsection SPARC VIS Built-in Functions
11909 GCC supports SIMD operations on the SPARC using both the generic vector
11910 extensions (@pxref{Vector Extensions}) as well as built-in functions for
11911 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
11912 switch, the VIS extension is exposed as the following built-in functions:
11915 typedef int v2si __attribute__ ((vector_size (8)));
11916 typedef short v4hi __attribute__ ((vector_size (8)));
11917 typedef short v2hi __attribute__ ((vector_size (4)));
11918 typedef char v8qi __attribute__ ((vector_size (8)));
11919 typedef char v4qi __attribute__ ((vector_size (4)));
11921 void * __builtin_vis_alignaddr (void *, long);
11922 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
11923 v2si __builtin_vis_faligndatav2si (v2si, v2si);
11924 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
11925 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
11927 v4hi __builtin_vis_fexpand (v4qi);
11929 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
11930 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
11931 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
11932 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
11933 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
11934 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
11935 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
11937 v4qi __builtin_vis_fpack16 (v4hi);
11938 v8qi __builtin_vis_fpack32 (v2si, v2si);
11939 v2hi __builtin_vis_fpackfix (v2si);
11940 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
11942 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
11945 @node SPU Built-in Functions
11946 @subsection SPU Built-in Functions
11948 GCC provides extensions for the SPU processor as described in the
11949 Sony/Toshiba/IBM SPU Language Extensions Specification, which can be
11950 found at @uref{http://cell.scei.co.jp/} or
11951 @uref{http://www.ibm.com/developerworks/power/cell/}. GCC's
11952 implementation differs in several ways.
11957 The optional extension of specifying vector constants in parentheses is
11961 A vector initializer requires no cast if the vector constant is of the
11962 same type as the variable it is initializing.
11965 If @code{signed} or @code{unsigned} is omitted, the signedness of the
11966 vector type is the default signedness of the base type. The default
11967 varies depending on the operating system, so a portable program should
11968 always specify the signedness.
11971 By default, the keyword @code{__vector} is added. The macro
11972 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
11976 GCC allows using a @code{typedef} name as the type specifier for a
11980 For C, overloaded functions are implemented with macros so the following
11984 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
11987 Since @code{spu_add} is a macro, the vector constant in the example
11988 is treated as four separate arguments. Wrap the entire argument in
11989 parentheses for this to work.
11992 The extended version of @code{__builtin_expect} is not supported.
11996 @emph{Note:} Only the interface described in the aforementioned
11997 specification is supported. Internally, GCC uses built-in functions to
11998 implement the required functionality, but these are not supported and
11999 are subject to change without notice.
12001 @node Target Format Checks
12002 @section Format Checks Specific to Particular Target Machines
12004 For some target machines, GCC supports additional options to the
12006 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
12009 * Solaris Format Checks::
12012 @node Solaris Format Checks
12013 @subsection Solaris Format Checks
12015 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
12016 check. @code{cmn_err} accepts a subset of the standard @code{printf}
12017 conversions, and the two-argument @code{%b} conversion for displaying
12018 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
12021 @section Pragmas Accepted by GCC
12025 GCC supports several types of pragmas, primarily in order to compile
12026 code originally written for other compilers. Note that in general
12027 we do not recommend the use of pragmas; @xref{Function Attributes},
12028 for further explanation.
12034 * RS/6000 and PowerPC Pragmas::
12036 * Solaris Pragmas::
12037 * Symbol-Renaming Pragmas::
12038 * Structure-Packing Pragmas::
12040 * Diagnostic Pragmas::
12041 * Visibility Pragmas::
12042 * Push/Pop Macro Pragmas::
12043 * Function Specific Option Pragmas::
12047 @subsection ARM Pragmas
12049 The ARM target defines pragmas for controlling the default addition of
12050 @code{long_call} and @code{short_call} attributes to functions.
12051 @xref{Function Attributes}, for information about the effects of these
12056 @cindex pragma, long_calls
12057 Set all subsequent functions to have the @code{long_call} attribute.
12059 @item no_long_calls
12060 @cindex pragma, no_long_calls
12061 Set all subsequent functions to have the @code{short_call} attribute.
12063 @item long_calls_off
12064 @cindex pragma, long_calls_off
12065 Do not affect the @code{long_call} or @code{short_call} attributes of
12066 subsequent functions.
12070 @subsection M32C Pragmas
12073 @item memregs @var{number}
12074 @cindex pragma, memregs
12075 Overrides the command line option @code{-memregs=} for the current
12076 file. Use with care! This pragma must be before any function in the
12077 file, and mixing different memregs values in different objects may
12078 make them incompatible. This pragma is useful when a
12079 performance-critical function uses a memreg for temporary values,
12080 as it may allow you to reduce the number of memregs used.
12085 @subsection MeP Pragmas
12089 @item custom io_volatile (on|off)
12090 @cindex pragma, custom io_volatile
12091 Overrides the command line option @code{-mio-volatile} for the current
12092 file. Note that for compatibility with future GCC releases, this
12093 option should only be used once before any @code{io} variables in each
12096 @item GCC coprocessor available @var{registers}
12097 @cindex pragma, coprocessor available
12098 Specifies which coprocessor registers are available to the register
12099 allocator. @var{registers} may be a single register, register range
12100 separated by ellipses, or comma-separated list of those. Example:
12103 #pragma GCC coprocessor available $c0...$c10, $c28
12106 @item GCC coprocessor call_saved @var{registers}
12107 @cindex pragma, coprocessor call_saved
12108 Specifies which coprocessor registers are to be saved and restored by
12109 any function using them. @var{registers} may be a single register,
12110 register range separated by ellipses, or comma-separated list of
12114 #pragma GCC coprocessor call_saved $c4...$c6, $c31
12117 @item GCC coprocessor subclass '(A|B|C|D)' = @var{registers}
12118 @cindex pragma, coprocessor subclass
12119 Creates and defines a register class. These register classes can be
12120 used by inline @code{asm} constructs. @var{registers} may be a single
12121 register, register range separated by ellipses, or comma-separated
12122 list of those. Example:
12125 #pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6
12127 asm ("cpfoo %0" : "=B" (x));
12130 @item GCC disinterrupt @var{name} , @var{name} @dots{}
12131 @cindex pragma, disinterrupt
12132 For the named functions, the compiler adds code to disable interrupts
12133 for the duration of those functions. Any functions so named, which
12134 are not encountered in the source, cause a warning that the pragma was
12135 not used. Examples:
12138 #pragma disinterrupt foo
12139 #pragma disinterrupt bar, grill
12140 int foo () @{ @dots{} @}
12143 @item GCC call @var{name} , @var{name} @dots{}
12144 @cindex pragma, call
12145 For the named functions, the compiler always uses a register-indirect
12146 call model when calling the named functions. Examples:
12155 @node RS/6000 and PowerPC Pragmas
12156 @subsection RS/6000 and PowerPC Pragmas
12158 The RS/6000 and PowerPC targets define one pragma for controlling
12159 whether or not the @code{longcall} attribute is added to function
12160 declarations by default. This pragma overrides the @option{-mlongcall}
12161 option, but not the @code{longcall} and @code{shortcall} attributes.
12162 @xref{RS/6000 and PowerPC Options}, for more information about when long
12163 calls are and are not necessary.
12167 @cindex pragma, longcall
12168 Apply the @code{longcall} attribute to all subsequent function
12172 Do not apply the @code{longcall} attribute to subsequent function
12176 @c Describe h8300 pragmas here.
12177 @c Describe sh pragmas here.
12178 @c Describe v850 pragmas here.
12180 @node Darwin Pragmas
12181 @subsection Darwin Pragmas
12183 The following pragmas are available for all architectures running the
12184 Darwin operating system. These are useful for compatibility with other
12188 @item mark @var{tokens}@dots{}
12189 @cindex pragma, mark
12190 This pragma is accepted, but has no effect.
12192 @item options align=@var{alignment}
12193 @cindex pragma, options align
12194 This pragma sets the alignment of fields in structures. The values of
12195 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
12196 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
12197 properly; to restore the previous setting, use @code{reset} for the
12200 @item segment @var{tokens}@dots{}
12201 @cindex pragma, segment
12202 This pragma is accepted, but has no effect.
12204 @item unused (@var{var} [, @var{var}]@dots{})
12205 @cindex pragma, unused
12206 This pragma declares variables to be possibly unused. GCC will not
12207 produce warnings for the listed variables. The effect is similar to
12208 that of the @code{unused} attribute, except that this pragma may appear
12209 anywhere within the variables' scopes.
12212 @node Solaris Pragmas
12213 @subsection Solaris Pragmas
12215 The Solaris target supports @code{#pragma redefine_extname}
12216 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
12217 @code{#pragma} directives for compatibility with the system compiler.
12220 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
12221 @cindex pragma, align
12223 Increase the minimum alignment of each @var{variable} to @var{alignment}.
12224 This is the same as GCC's @code{aligned} attribute @pxref{Variable
12225 Attributes}). Macro expansion occurs on the arguments to this pragma
12226 when compiling C and Objective-C@. It does not currently occur when
12227 compiling C++, but this is a bug which may be fixed in a future
12230 @item fini (@var{function} [, @var{function}]...)
12231 @cindex pragma, fini
12233 This pragma causes each listed @var{function} to be called after
12234 main, or during shared module unloading, by adding a call to the
12235 @code{.fini} section.
12237 @item init (@var{function} [, @var{function}]...)
12238 @cindex pragma, init
12240 This pragma causes each listed @var{function} to be called during
12241 initialization (before @code{main}) or during shared module loading, by
12242 adding a call to the @code{.init} section.
12246 @node Symbol-Renaming Pragmas
12247 @subsection Symbol-Renaming Pragmas
12249 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
12250 supports two @code{#pragma} directives which change the name used in
12251 assembly for a given declaration. @code{#pragma_extern_prefix} is only
12252 available on platforms whose system headers need it. To get this effect
12253 on all platforms supported by GCC, use the asm labels extension (@pxref{Asm
12257 @item redefine_extname @var{oldname} @var{newname}
12258 @cindex pragma, redefine_extname
12260 This pragma gives the C function @var{oldname} the assembly symbol
12261 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
12262 will be defined if this pragma is available (currently on all platforms).
12264 @item extern_prefix @var{string}
12265 @cindex pragma, extern_prefix
12267 This pragma causes all subsequent external function and variable
12268 declarations to have @var{string} prepended to their assembly symbols.
12269 This effect may be terminated with another @code{extern_prefix} pragma
12270 whose argument is an empty string. The preprocessor macro
12271 @code{__PRAGMA_EXTERN_PREFIX} will be defined if this pragma is
12272 available (currently only on Tru64 UNIX)@.
12275 These pragmas and the asm labels extension interact in a complicated
12276 manner. Here are some corner cases you may want to be aware of.
12279 @item Both pragmas silently apply only to declarations with external
12280 linkage. Asm labels do not have this restriction.
12282 @item In C++, both pragmas silently apply only to declarations with
12283 ``C'' linkage. Again, asm labels do not have this restriction.
12285 @item If any of the three ways of changing the assembly name of a
12286 declaration is applied to a declaration whose assembly name has
12287 already been determined (either by a previous use of one of these
12288 features, or because the compiler needed the assembly name in order to
12289 generate code), and the new name is different, a warning issues and
12290 the name does not change.
12292 @item The @var{oldname} used by @code{#pragma redefine_extname} is
12293 always the C-language name.
12295 @item If @code{#pragma extern_prefix} is in effect, and a declaration
12296 occurs with an asm label attached, the prefix is silently ignored for
12299 @item If @code{#pragma extern_prefix} and @code{#pragma redefine_extname}
12300 apply to the same declaration, whichever triggered first wins, and a
12301 warning issues if they contradict each other. (We would like to have
12302 @code{#pragma redefine_extname} always win, for consistency with asm
12303 labels, but if @code{#pragma extern_prefix} triggers first we have no
12304 way of knowing that that happened.)
12307 @node Structure-Packing Pragmas
12308 @subsection Structure-Packing Pragmas
12310 For compatibility with Microsoft Windows compilers, GCC supports a
12311 set of @code{#pragma} directives which change the maximum alignment of
12312 members of structures (other than zero-width bitfields), unions, and
12313 classes subsequently defined. The @var{n} value below always is required
12314 to be a small power of two and specifies the new alignment in bytes.
12317 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
12318 @item @code{#pragma pack()} sets the alignment to the one that was in
12319 effect when compilation started (see also command line option
12320 @option{-fpack-struct[=<n>]} @pxref{Code Gen Options}).
12321 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
12322 setting on an internal stack and then optionally sets the new alignment.
12323 @item @code{#pragma pack(pop)} restores the alignment setting to the one
12324 saved at the top of the internal stack (and removes that stack entry).
12325 Note that @code{#pragma pack([@var{n}])} does not influence this internal
12326 stack; thus it is possible to have @code{#pragma pack(push)} followed by
12327 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
12328 @code{#pragma pack(pop)}.
12331 Some targets, e.g.@: i386 and powerpc, support the @code{ms_struct}
12332 @code{#pragma} which lays out a structure as the documented
12333 @code{__attribute__ ((ms_struct))}.
12335 @item @code{#pragma ms_struct on} turns on the layout for structures
12337 @item @code{#pragma ms_struct off} turns off the layout for structures
12339 @item @code{#pragma ms_struct reset} goes back to the default layout.
12343 @subsection Weak Pragmas
12345 For compatibility with SVR4, GCC supports a set of @code{#pragma}
12346 directives for declaring symbols to be weak, and defining weak
12350 @item #pragma weak @var{symbol}
12351 @cindex pragma, weak
12352 This pragma declares @var{symbol} to be weak, as if the declaration
12353 had the attribute of the same name. The pragma may appear before
12354 or after the declaration of @var{symbol}, but must appear before
12355 either its first use or its definition. It is not an error for
12356 @var{symbol} to never be defined at all.
12358 @item #pragma weak @var{symbol1} = @var{symbol2}
12359 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
12360 It is an error if @var{symbol2} is not defined in the current
12364 @node Diagnostic Pragmas
12365 @subsection Diagnostic Pragmas
12367 GCC allows the user to selectively enable or disable certain types of
12368 diagnostics, and change the kind of the diagnostic. For example, a
12369 project's policy might require that all sources compile with
12370 @option{-Werror} but certain files might have exceptions allowing
12371 specific types of warnings. Or, a project might selectively enable
12372 diagnostics and treat them as errors depending on which preprocessor
12373 macros are defined.
12376 @item #pragma GCC diagnostic @var{kind} @var{option}
12377 @cindex pragma, diagnostic
12379 Modifies the disposition of a diagnostic. Note that not all
12380 diagnostics are modifiable; at the moment only warnings (normally
12381 controlled by @samp{-W@dots{}}) can be controlled, and not all of them.
12382 Use @option{-fdiagnostics-show-option} to determine which diagnostics
12383 are controllable and which option controls them.
12385 @var{kind} is @samp{error} to treat this diagnostic as an error,
12386 @samp{warning} to treat it like a warning (even if @option{-Werror} is
12387 in effect), or @samp{ignored} if the diagnostic is to be ignored.
12388 @var{option} is a double quoted string which matches the command line
12392 #pragma GCC diagnostic warning "-Wformat"
12393 #pragma GCC diagnostic error "-Wformat"
12394 #pragma GCC diagnostic ignored "-Wformat"
12397 Note that these pragmas override any command line options. Also,
12398 while it is syntactically valid to put these pragmas anywhere in your
12399 sources, the only supported location for them is before any data or
12400 functions are defined. Doing otherwise may result in unpredictable
12401 results depending on how the optimizer manages your sources. If the
12402 same option is listed multiple times, the last one specified is the
12403 one that is in effect. This pragma is not intended to be a general
12404 purpose replacement for command line options, but for implementing
12405 strict control over project policies.
12409 GCC also offers a simple mechanism for printing messages during
12413 @item #pragma message @var{string}
12414 @cindex pragma, diagnostic
12416 Prints @var{string} as a compiler message on compilation. The message
12417 is informational only, and is neither a compilation warning nor an error.
12420 #pragma message "Compiling " __FILE__ "..."
12423 @var{string} may be parenthesized, and is printed with location
12424 information. For example,
12427 #define DO_PRAGMA(x) _Pragma (#x)
12428 #define TODO(x) DO_PRAGMA(message ("TODO - " #x))
12430 TODO(Remember to fix this)
12433 prints @samp{/tmp/file.c:4: note: #pragma message:
12434 TODO - Remember to fix this}.
12438 @node Visibility Pragmas
12439 @subsection Visibility Pragmas
12442 @item #pragma GCC visibility push(@var{visibility})
12443 @itemx #pragma GCC visibility pop
12444 @cindex pragma, visibility
12446 This pragma allows the user to set the visibility for multiple
12447 declarations without having to give each a visibility attribute
12448 @xref{Function Attributes}, for more information about visibility and
12449 the attribute syntax.
12451 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
12452 declarations. Class members and template specializations are not
12453 affected; if you want to override the visibility for a particular
12454 member or instantiation, you must use an attribute.
12459 @node Push/Pop Macro Pragmas
12460 @subsection Push/Pop Macro Pragmas
12462 For compatibility with Microsoft Windows compilers, GCC supports
12463 @samp{#pragma push_macro(@var{"macro_name"})}
12464 and @samp{#pragma pop_macro(@var{"macro_name"})}.
12467 @item #pragma push_macro(@var{"macro_name"})
12468 @cindex pragma, push_macro
12469 This pragma saves the value of the macro named as @var{macro_name} to
12470 the top of the stack for this macro.
12472 @item #pragma pop_macro(@var{"macro_name"})
12473 @cindex pragma, pop_macro
12474 This pragma sets the value of the macro named as @var{macro_name} to
12475 the value on top of the stack for this macro. If the stack for
12476 @var{macro_name} is empty, the value of the macro remains unchanged.
12483 #pragma push_macro("X")
12486 #pragma pop_macro("X")
12490 In this example, the definition of X as 1 is saved by @code{#pragma
12491 push_macro} and restored by @code{#pragma pop_macro}.
12493 @node Function Specific Option Pragmas
12494 @subsection Function Specific Option Pragmas
12497 @item #pragma GCC target (@var{"string"}...)
12498 @cindex pragma GCC target
12500 This pragma allows you to set target specific options for functions
12501 defined later in the source file. One or more strings can be
12502 specified. Each function that is defined after this point will be as
12503 if @code{attribute((target("STRING")))} was specified for that
12504 function. The parenthesis around the options is optional.
12505 @xref{Function Attributes}, for more information about the
12506 @code{target} attribute and the attribute syntax.
12508 The @samp{#pragma GCC target} pragma is not implemented in GCC
12509 versions earlier than 4.4, and is currently only implemented for the
12510 386 and x86_64 backends.
12514 @item #pragma GCC optimize (@var{"string"}...)
12515 @cindex pragma GCC optimize
12517 This pragma allows you to set global optimization options for functions
12518 defined later in the source file. One or more strings can be
12519 specified. Each function that is defined after this point will be as
12520 if @code{attribute((optimize("STRING")))} was specified for that
12521 function. The parenthesis around the options is optional.
12522 @xref{Function Attributes}, for more information about the
12523 @code{optimize} attribute and the attribute syntax.
12525 The @samp{#pragma GCC optimize} pragma is not implemented in GCC
12526 versions earlier than 4.4.
12530 @item #pragma GCC push_options
12531 @itemx #pragma GCC pop_options
12532 @cindex pragma GCC push_options
12533 @cindex pragma GCC pop_options
12535 These pragmas maintain a stack of the current target and optimization
12536 options. It is intended for include files where you temporarily want
12537 to switch to using a different @samp{#pragma GCC target} or
12538 @samp{#pragma GCC optimize} and then to pop back to the previous
12541 The @samp{#pragma GCC push_options} and @samp{#pragma GCC pop_options}
12542 pragmas are not implemented in GCC versions earlier than 4.4.
12546 @item #pragma GCC reset_options
12547 @cindex pragma GCC reset_options
12549 This pragma clears the current @code{#pragma GCC target} and
12550 @code{#pragma GCC optimize} to use the default switches as specified
12551 on the command line.
12553 The @samp{#pragma GCC reset_options} pragma is not implemented in GCC
12554 versions earlier than 4.4.
12557 @node Unnamed Fields
12558 @section Unnamed struct/union fields within structs/unions
12562 For compatibility with other compilers, GCC allows you to define
12563 a structure or union that contains, as fields, structures and unions
12564 without names. For example:
12577 In this example, the user would be able to access members of the unnamed
12578 union with code like @samp{foo.b}. Note that only unnamed structs and
12579 unions are allowed, you may not have, for example, an unnamed
12582 You must never create such structures that cause ambiguous field definitions.
12583 For example, this structure:
12594 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
12595 Such constructs are not supported and must be avoided. In the future,
12596 such constructs may be detected and treated as compilation errors.
12598 @opindex fms-extensions
12599 Unless @option{-fms-extensions} is used, the unnamed field must be a
12600 structure or union definition without a tag (for example, @samp{struct
12601 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
12602 also be a definition with a tag such as @samp{struct foo @{ int a;
12603 @};}, a reference to a previously defined structure or union such as
12604 @samp{struct foo;}, or a reference to a @code{typedef} name for a
12605 previously defined structure or union type.
12608 @section Thread-Local Storage
12609 @cindex Thread-Local Storage
12610 @cindex @acronym{TLS}
12613 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
12614 are allocated such that there is one instance of the variable per extant
12615 thread. The run-time model GCC uses to implement this originates
12616 in the IA-64 processor-specific ABI, but has since been migrated
12617 to other processors as well. It requires significant support from
12618 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
12619 system libraries (@file{libc.so} and @file{libpthread.so}), so it
12620 is not available everywhere.
12622 At the user level, the extension is visible with a new storage
12623 class keyword: @code{__thread}. For example:
12627 extern __thread struct state s;
12628 static __thread char *p;
12631 The @code{__thread} specifier may be used alone, with the @code{extern}
12632 or @code{static} specifiers, but with no other storage class specifier.
12633 When used with @code{extern} or @code{static}, @code{__thread} must appear
12634 immediately after the other storage class specifier.
12636 The @code{__thread} specifier may be applied to any global, file-scoped
12637 static, function-scoped static, or static data member of a class. It may
12638 not be applied to block-scoped automatic or non-static data member.
12640 When the address-of operator is applied to a thread-local variable, it is
12641 evaluated at run-time and returns the address of the current thread's
12642 instance of that variable. An address so obtained may be used by any
12643 thread. When a thread terminates, any pointers to thread-local variables
12644 in that thread become invalid.
12646 No static initialization may refer to the address of a thread-local variable.
12648 In C++, if an initializer is present for a thread-local variable, it must
12649 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
12652 See @uref{http://people.redhat.com/drepper/tls.pdf,
12653 ELF Handling For Thread-Local Storage} for a detailed explanation of
12654 the four thread-local storage addressing models, and how the run-time
12655 is expected to function.
12658 * C99 Thread-Local Edits::
12659 * C++98 Thread-Local Edits::
12662 @node C99 Thread-Local Edits
12663 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
12665 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
12666 that document the exact semantics of the language extension.
12670 @cite{5.1.2 Execution environments}
12672 Add new text after paragraph 1
12675 Within either execution environment, a @dfn{thread} is a flow of
12676 control within a program. It is implementation defined whether
12677 or not there may be more than one thread associated with a program.
12678 It is implementation defined how threads beyond the first are
12679 created, the name and type of the function called at thread
12680 startup, and how threads may be terminated. However, objects
12681 with thread storage duration shall be initialized before thread
12686 @cite{6.2.4 Storage durations of objects}
12688 Add new text before paragraph 3
12691 An object whose identifier is declared with the storage-class
12692 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
12693 Its lifetime is the entire execution of the thread, and its
12694 stored value is initialized only once, prior to thread startup.
12698 @cite{6.4.1 Keywords}
12700 Add @code{__thread}.
12703 @cite{6.7.1 Storage-class specifiers}
12705 Add @code{__thread} to the list of storage class specifiers in
12708 Change paragraph 2 to
12711 With the exception of @code{__thread}, at most one storage-class
12712 specifier may be given [@dots{}]. The @code{__thread} specifier may
12713 be used alone, or immediately following @code{extern} or
12717 Add new text after paragraph 6
12720 The declaration of an identifier for a variable that has
12721 block scope that specifies @code{__thread} shall also
12722 specify either @code{extern} or @code{static}.
12724 The @code{__thread} specifier shall be used only with
12729 @node C++98 Thread-Local Edits
12730 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
12732 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
12733 that document the exact semantics of the language extension.
12737 @b{[intro.execution]}
12739 New text after paragraph 4
12742 A @dfn{thread} is a flow of control within the abstract machine.
12743 It is implementation defined whether or not there may be more than
12747 New text after paragraph 7
12750 It is unspecified whether additional action must be taken to
12751 ensure when and whether side effects are visible to other threads.
12757 Add @code{__thread}.
12760 @b{[basic.start.main]}
12762 Add after paragraph 5
12765 The thread that begins execution at the @code{main} function is called
12766 the @dfn{main thread}. It is implementation defined how functions
12767 beginning threads other than the main thread are designated or typed.
12768 A function so designated, as well as the @code{main} function, is called
12769 a @dfn{thread startup function}. It is implementation defined what
12770 happens if a thread startup function returns. It is implementation
12771 defined what happens to other threads when any thread calls @code{exit}.
12775 @b{[basic.start.init]}
12777 Add after paragraph 4
12780 The storage for an object of thread storage duration shall be
12781 statically initialized before the first statement of the thread startup
12782 function. An object of thread storage duration shall not require
12783 dynamic initialization.
12787 @b{[basic.start.term]}
12789 Add after paragraph 3
12792 The type of an object with thread storage duration shall not have a
12793 non-trivial destructor, nor shall it be an array type whose elements
12794 (directly or indirectly) have non-trivial destructors.
12800 Add ``thread storage duration'' to the list in paragraph 1.
12805 Thread, static, and automatic storage durations are associated with
12806 objects introduced by declarations [@dots{}].
12809 Add @code{__thread} to the list of specifiers in paragraph 3.
12812 @b{[basic.stc.thread]}
12814 New section before @b{[basic.stc.static]}
12817 The keyword @code{__thread} applied to a non-local object gives the
12818 object thread storage duration.
12820 A local variable or class data member declared both @code{static}
12821 and @code{__thread} gives the variable or member thread storage
12826 @b{[basic.stc.static]}
12831 All objects which have neither thread storage duration, dynamic
12832 storage duration nor are local [@dots{}].
12838 Add @code{__thread} to the list in paragraph 1.
12843 With the exception of @code{__thread}, at most one
12844 @var{storage-class-specifier} shall appear in a given
12845 @var{decl-specifier-seq}. The @code{__thread} specifier may
12846 be used alone, or immediately following the @code{extern} or
12847 @code{static} specifiers. [@dots{}]
12850 Add after paragraph 5
12853 The @code{__thread} specifier can be applied only to the names of objects
12854 and to anonymous unions.
12860 Add after paragraph 6
12863 Non-@code{static} members shall not be @code{__thread}.
12867 @node Binary constants
12868 @section Binary constants using the @samp{0b} prefix
12869 @cindex Binary constants using the @samp{0b} prefix
12871 Integer constants can be written as binary constants, consisting of a
12872 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
12873 @samp{0B}. This is particularly useful in environments that operate a
12874 lot on the bit-level (like microcontrollers).
12876 The following statements are identical:
12885 The type of these constants follows the same rules as for octal or
12886 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
12889 @node C++ Extensions
12890 @chapter Extensions to the C++ Language
12891 @cindex extensions, C++ language
12892 @cindex C++ language extensions
12894 The GNU compiler provides these extensions to the C++ language (and you
12895 can also use most of the C language extensions in your C++ programs). If you
12896 want to write code that checks whether these features are available, you can
12897 test for the GNU compiler the same way as for C programs: check for a
12898 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
12899 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
12900 Predefined Macros,cpp,The GNU C Preprocessor}).
12903 * Volatiles:: What constitutes an access to a volatile object.
12904 * Restricted Pointers:: C99 restricted pointers and references.
12905 * Vague Linkage:: Where G++ puts inlines, vtables and such.
12906 * C++ Interface:: You can use a single C++ header file for both
12907 declarations and definitions.
12908 * Template Instantiation:: Methods for ensuring that exactly one copy of
12909 each needed template instantiation is emitted.
12910 * Bound member functions:: You can extract a function pointer to the
12911 method denoted by a @samp{->*} or @samp{.*} expression.
12912 * C++ Attributes:: Variable, function, and type attributes for C++ only.
12913 * Namespace Association:: Strong using-directives for namespace association.
12914 * Type Traits:: Compiler support for type traits
12915 * Java Exceptions:: Tweaking exception handling to work with Java.
12916 * Deprecated Features:: Things will disappear from g++.
12917 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
12921 @section When is a Volatile Object Accessed?
12922 @cindex accessing volatiles
12923 @cindex volatile read
12924 @cindex volatile write
12925 @cindex volatile access
12927 Both the C and C++ standard have the concept of volatile objects. These
12928 are normally accessed by pointers and used for accessing hardware. The
12929 standards encourage compilers to refrain from optimizations concerning
12930 accesses to volatile objects. The C standard leaves it implementation
12931 defined as to what constitutes a volatile access. The C++ standard omits
12932 to specify this, except to say that C++ should behave in a similar manner
12933 to C with respect to volatiles, where possible. The minimum either
12934 standard specifies is that at a sequence point all previous accesses to
12935 volatile objects have stabilized and no subsequent accesses have
12936 occurred. Thus an implementation is free to reorder and combine
12937 volatile accesses which occur between sequence points, but cannot do so
12938 for accesses across a sequence point. The use of volatiles does not
12939 allow you to violate the restriction on updating objects multiple times
12940 within a sequence point.
12942 @xref{Qualifiers implementation, , Volatile qualifier and the C compiler}.
12944 The behavior differs slightly between C and C++ in the non-obvious cases:
12947 volatile int *src = @var{somevalue};
12951 With C, such expressions are rvalues, and GCC interprets this either as a
12952 read of the volatile object being pointed to or only as request to evaluate
12953 the side-effects. The C++ standard specifies that such expressions do not
12954 undergo lvalue to rvalue conversion, and that the type of the dereferenced
12955 object may be incomplete. The C++ standard does not specify explicitly
12956 that it is this lvalue to rvalue conversion which may be responsible for
12957 causing an access. However, there is reason to believe that it is,
12958 because otherwise certain simple expressions become undefined. However,
12959 because it would surprise most programmers, G++ treats dereferencing a
12960 pointer to volatile object of complete type when the value is unused as
12961 GCC would do for an equivalent type in C@. When the object has incomplete
12962 type, G++ issues a warning; if you wish to force an error, you must
12963 force a conversion to rvalue with, for instance, a static cast.
12965 When using a reference to volatile, G++ does not treat equivalent
12966 expressions as accesses to volatiles, but instead issues a warning that
12967 no volatile is accessed. The rationale for this is that otherwise it
12968 becomes difficult to determine where volatile access occur, and not
12969 possible to ignore the return value from functions returning volatile
12970 references. Again, if you wish to force a read, cast the reference to
12973 @node Restricted Pointers
12974 @section Restricting Pointer Aliasing
12975 @cindex restricted pointers
12976 @cindex restricted references
12977 @cindex restricted this pointer
12979 As with the C front end, G++ understands the C99 feature of restricted pointers,
12980 specified with the @code{__restrict__}, or @code{__restrict} type
12981 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
12982 language flag, @code{restrict} is not a keyword in C++.
12984 In addition to allowing restricted pointers, you can specify restricted
12985 references, which indicate that the reference is not aliased in the local
12989 void fn (int *__restrict__ rptr, int &__restrict__ rref)
12996 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
12997 @var{rref} refers to a (different) unaliased integer.
12999 You may also specify whether a member function's @var{this} pointer is
13000 unaliased by using @code{__restrict__} as a member function qualifier.
13003 void T::fn () __restrict__
13010 Within the body of @code{T::fn}, @var{this} will have the effective
13011 definition @code{T *__restrict__ const this}. Notice that the
13012 interpretation of a @code{__restrict__} member function qualifier is
13013 different to that of @code{const} or @code{volatile} qualifier, in that it
13014 is applied to the pointer rather than the object. This is consistent with
13015 other compilers which implement restricted pointers.
13017 As with all outermost parameter qualifiers, @code{__restrict__} is
13018 ignored in function definition matching. This means you only need to
13019 specify @code{__restrict__} in a function definition, rather than
13020 in a function prototype as well.
13022 @node Vague Linkage
13023 @section Vague Linkage
13024 @cindex vague linkage
13026 There are several constructs in C++ which require space in the object
13027 file but are not clearly tied to a single translation unit. We say that
13028 these constructs have ``vague linkage''. Typically such constructs are
13029 emitted wherever they are needed, though sometimes we can be more
13033 @item Inline Functions
13034 Inline functions are typically defined in a header file which can be
13035 included in many different compilations. Hopefully they can usually be
13036 inlined, but sometimes an out-of-line copy is necessary, if the address
13037 of the function is taken or if inlining fails. In general, we emit an
13038 out-of-line copy in all translation units where one is needed. As an
13039 exception, we only emit inline virtual functions with the vtable, since
13040 it will always require a copy.
13042 Local static variables and string constants used in an inline function
13043 are also considered to have vague linkage, since they must be shared
13044 between all inlined and out-of-line instances of the function.
13048 C++ virtual functions are implemented in most compilers using a lookup
13049 table, known as a vtable. The vtable contains pointers to the virtual
13050 functions provided by a class, and each object of the class contains a
13051 pointer to its vtable (or vtables, in some multiple-inheritance
13052 situations). If the class declares any non-inline, non-pure virtual
13053 functions, the first one is chosen as the ``key method'' for the class,
13054 and the vtable is only emitted in the translation unit where the key
13057 @emph{Note:} If the chosen key method is later defined as inline, the
13058 vtable will still be emitted in every translation unit which defines it.
13059 Make sure that any inline virtuals are declared inline in the class
13060 body, even if they are not defined there.
13062 @item type_info objects
13065 C++ requires information about types to be written out in order to
13066 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
13067 For polymorphic classes (classes with virtual functions), the type_info
13068 object is written out along with the vtable so that @samp{dynamic_cast}
13069 can determine the dynamic type of a class object at runtime. For all
13070 other types, we write out the type_info object when it is used: when
13071 applying @samp{typeid} to an expression, throwing an object, or
13072 referring to a type in a catch clause or exception specification.
13074 @item Template Instantiations
13075 Most everything in this section also applies to template instantiations,
13076 but there are other options as well.
13077 @xref{Template Instantiation,,Where's the Template?}.
13081 When used with GNU ld version 2.8 or later on an ELF system such as
13082 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
13083 these constructs will be discarded at link time. This is known as
13086 On targets that don't support COMDAT, but do support weak symbols, GCC
13087 will use them. This way one copy will override all the others, but
13088 the unused copies will still take up space in the executable.
13090 For targets which do not support either COMDAT or weak symbols,
13091 most entities with vague linkage will be emitted as local symbols to
13092 avoid duplicate definition errors from the linker. This will not happen
13093 for local statics in inlines, however, as having multiple copies will
13094 almost certainly break things.
13096 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
13097 another way to control placement of these constructs.
13099 @node C++ Interface
13100 @section #pragma interface and implementation
13102 @cindex interface and implementation headers, C++
13103 @cindex C++ interface and implementation headers
13104 @cindex pragmas, interface and implementation
13106 @code{#pragma interface} and @code{#pragma implementation} provide the
13107 user with a way of explicitly directing the compiler to emit entities
13108 with vague linkage (and debugging information) in a particular
13111 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
13112 most cases, because of COMDAT support and the ``key method'' heuristic
13113 mentioned in @ref{Vague Linkage}. Using them can actually cause your
13114 program to grow due to unnecessary out-of-line copies of inline
13115 functions. Currently (3.4) the only benefit of these
13116 @code{#pragma}s is reduced duplication of debugging information, and
13117 that should be addressed soon on DWARF 2 targets with the use of
13121 @item #pragma interface
13122 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
13123 @kindex #pragma interface
13124 Use this directive in @emph{header files} that define object classes, to save
13125 space in most of the object files that use those classes. Normally,
13126 local copies of certain information (backup copies of inline member
13127 functions, debugging information, and the internal tables that implement
13128 virtual functions) must be kept in each object file that includes class
13129 definitions. You can use this pragma to avoid such duplication. When a
13130 header file containing @samp{#pragma interface} is included in a
13131 compilation, this auxiliary information will not be generated (unless
13132 the main input source file itself uses @samp{#pragma implementation}).
13133 Instead, the object files will contain references to be resolved at link
13136 The second form of this directive is useful for the case where you have
13137 multiple headers with the same name in different directories. If you
13138 use this form, you must specify the same string to @samp{#pragma
13141 @item #pragma implementation
13142 @itemx #pragma implementation "@var{objects}.h"
13143 @kindex #pragma implementation
13144 Use this pragma in a @emph{main input file}, when you want full output from
13145 included header files to be generated (and made globally visible). The
13146 included header file, in turn, should use @samp{#pragma interface}.
13147 Backup copies of inline member functions, debugging information, and the
13148 internal tables used to implement virtual functions are all generated in
13149 implementation files.
13151 @cindex implied @code{#pragma implementation}
13152 @cindex @code{#pragma implementation}, implied
13153 @cindex naming convention, implementation headers
13154 If you use @samp{#pragma implementation} with no argument, it applies to
13155 an include file with the same basename@footnote{A file's @dfn{basename}
13156 was the name stripped of all leading path information and of trailing
13157 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
13158 file. For example, in @file{allclass.cc}, giving just
13159 @samp{#pragma implementation}
13160 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
13162 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
13163 an implementation file whenever you would include it from
13164 @file{allclass.cc} even if you never specified @samp{#pragma
13165 implementation}. This was deemed to be more trouble than it was worth,
13166 however, and disabled.
13168 Use the string argument if you want a single implementation file to
13169 include code from multiple header files. (You must also use
13170 @samp{#include} to include the header file; @samp{#pragma
13171 implementation} only specifies how to use the file---it doesn't actually
13174 There is no way to split up the contents of a single header file into
13175 multiple implementation files.
13178 @cindex inlining and C++ pragmas
13179 @cindex C++ pragmas, effect on inlining
13180 @cindex pragmas in C++, effect on inlining
13181 @samp{#pragma implementation} and @samp{#pragma interface} also have an
13182 effect on function inlining.
13184 If you define a class in a header file marked with @samp{#pragma
13185 interface}, the effect on an inline function defined in that class is
13186 similar to an explicit @code{extern} declaration---the compiler emits
13187 no code at all to define an independent version of the function. Its
13188 definition is used only for inlining with its callers.
13190 @opindex fno-implement-inlines
13191 Conversely, when you include the same header file in a main source file
13192 that declares it as @samp{#pragma implementation}, the compiler emits
13193 code for the function itself; this defines a version of the function
13194 that can be found via pointers (or by callers compiled without
13195 inlining). If all calls to the function can be inlined, you can avoid
13196 emitting the function by compiling with @option{-fno-implement-inlines}.
13197 If any calls were not inlined, you will get linker errors.
13199 @node Template Instantiation
13200 @section Where's the Template?
13201 @cindex template instantiation
13203 C++ templates are the first language feature to require more
13204 intelligence from the environment than one usually finds on a UNIX
13205 system. Somehow the compiler and linker have to make sure that each
13206 template instance occurs exactly once in the executable if it is needed,
13207 and not at all otherwise. There are two basic approaches to this
13208 problem, which are referred to as the Borland model and the Cfront model.
13211 @item Borland model
13212 Borland C++ solved the template instantiation problem by adding the code
13213 equivalent of common blocks to their linker; the compiler emits template
13214 instances in each translation unit that uses them, and the linker
13215 collapses them together. The advantage of this model is that the linker
13216 only has to consider the object files themselves; there is no external
13217 complexity to worry about. This disadvantage is that compilation time
13218 is increased because the template code is being compiled repeatedly.
13219 Code written for this model tends to include definitions of all
13220 templates in the header file, since they must be seen to be
13224 The AT&T C++ translator, Cfront, solved the template instantiation
13225 problem by creating the notion of a template repository, an
13226 automatically maintained place where template instances are stored. A
13227 more modern version of the repository works as follows: As individual
13228 object files are built, the compiler places any template definitions and
13229 instantiations encountered in the repository. At link time, the link
13230 wrapper adds in the objects in the repository and compiles any needed
13231 instances that were not previously emitted. The advantages of this
13232 model are more optimal compilation speed and the ability to use the
13233 system linker; to implement the Borland model a compiler vendor also
13234 needs to replace the linker. The disadvantages are vastly increased
13235 complexity, and thus potential for error; for some code this can be
13236 just as transparent, but in practice it can been very difficult to build
13237 multiple programs in one directory and one program in multiple
13238 directories. Code written for this model tends to separate definitions
13239 of non-inline member templates into a separate file, which should be
13240 compiled separately.
13243 When used with GNU ld version 2.8 or later on an ELF system such as
13244 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
13245 Borland model. On other systems, G++ implements neither automatic
13248 A future version of G++ will support a hybrid model whereby the compiler
13249 will emit any instantiations for which the template definition is
13250 included in the compile, and store template definitions and
13251 instantiation context information into the object file for the rest.
13252 The link wrapper will extract that information as necessary and invoke
13253 the compiler to produce the remaining instantiations. The linker will
13254 then combine duplicate instantiations.
13256 In the mean time, you have the following options for dealing with
13257 template instantiations:
13262 Compile your template-using code with @option{-frepo}. The compiler will
13263 generate files with the extension @samp{.rpo} listing all of the
13264 template instantiations used in the corresponding object files which
13265 could be instantiated there; the link wrapper, @samp{collect2}, will
13266 then update the @samp{.rpo} files to tell the compiler where to place
13267 those instantiations and rebuild any affected object files. The
13268 link-time overhead is negligible after the first pass, as the compiler
13269 will continue to place the instantiations in the same files.
13271 This is your best option for application code written for the Borland
13272 model, as it will just work. Code written for the Cfront model will
13273 need to be modified so that the template definitions are available at
13274 one or more points of instantiation; usually this is as simple as adding
13275 @code{#include <tmethods.cc>} to the end of each template header.
13277 For library code, if you want the library to provide all of the template
13278 instantiations it needs, just try to link all of its object files
13279 together; the link will fail, but cause the instantiations to be
13280 generated as a side effect. Be warned, however, that this may cause
13281 conflicts if multiple libraries try to provide the same instantiations.
13282 For greater control, use explicit instantiation as described in the next
13286 @opindex fno-implicit-templates
13287 Compile your code with @option{-fno-implicit-templates} to disable the
13288 implicit generation of template instances, and explicitly instantiate
13289 all the ones you use. This approach requires more knowledge of exactly
13290 which instances you need than do the others, but it's less
13291 mysterious and allows greater control. You can scatter the explicit
13292 instantiations throughout your program, perhaps putting them in the
13293 translation units where the instances are used or the translation units
13294 that define the templates themselves; you can put all of the explicit
13295 instantiations you need into one big file; or you can create small files
13302 template class Foo<int>;
13303 template ostream& operator <<
13304 (ostream&, const Foo<int>&);
13307 for each of the instances you need, and create a template instantiation
13308 library from those.
13310 If you are using Cfront-model code, you can probably get away with not
13311 using @option{-fno-implicit-templates} when compiling files that don't
13312 @samp{#include} the member template definitions.
13314 If you use one big file to do the instantiations, you may want to
13315 compile it without @option{-fno-implicit-templates} so you get all of the
13316 instances required by your explicit instantiations (but not by any
13317 other files) without having to specify them as well.
13319 G++ has extended the template instantiation syntax given in the ISO
13320 standard to allow forward declaration of explicit instantiations
13321 (with @code{extern}), instantiation of the compiler support data for a
13322 template class (i.e.@: the vtable) without instantiating any of its
13323 members (with @code{inline}), and instantiation of only the static data
13324 members of a template class, without the support data or member
13325 functions (with (@code{static}):
13328 extern template int max (int, int);
13329 inline template class Foo<int>;
13330 static template class Foo<int>;
13334 Do nothing. Pretend G++ does implement automatic instantiation
13335 management. Code written for the Borland model will work fine, but
13336 each translation unit will contain instances of each of the templates it
13337 uses. In a large program, this can lead to an unacceptable amount of code
13341 @node Bound member functions
13342 @section Extracting the function pointer from a bound pointer to member function
13344 @cindex pointer to member function
13345 @cindex bound pointer to member function
13347 In C++, pointer to member functions (PMFs) are implemented using a wide
13348 pointer of sorts to handle all the possible call mechanisms; the PMF
13349 needs to store information about how to adjust the @samp{this} pointer,
13350 and if the function pointed to is virtual, where to find the vtable, and
13351 where in the vtable to look for the member function. If you are using
13352 PMFs in an inner loop, you should really reconsider that decision. If
13353 that is not an option, you can extract the pointer to the function that
13354 would be called for a given object/PMF pair and call it directly inside
13355 the inner loop, to save a bit of time.
13357 Note that you will still be paying the penalty for the call through a
13358 function pointer; on most modern architectures, such a call defeats the
13359 branch prediction features of the CPU@. This is also true of normal
13360 virtual function calls.
13362 The syntax for this extension is
13366 extern int (A::*fp)();
13367 typedef int (*fptr)(A *);
13369 fptr p = (fptr)(a.*fp);
13372 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
13373 no object is needed to obtain the address of the function. They can be
13374 converted to function pointers directly:
13377 fptr p1 = (fptr)(&A::foo);
13380 @opindex Wno-pmf-conversions
13381 You must specify @option{-Wno-pmf-conversions} to use this extension.
13383 @node C++ Attributes
13384 @section C++-Specific Variable, Function, and Type Attributes
13386 Some attributes only make sense for C++ programs.
13389 @item init_priority (@var{priority})
13390 @cindex init_priority attribute
13393 In Standard C++, objects defined at namespace scope are guaranteed to be
13394 initialized in an order in strict accordance with that of their definitions
13395 @emph{in a given translation unit}. No guarantee is made for initializations
13396 across translation units. However, GNU C++ allows users to control the
13397 order of initialization of objects defined at namespace scope with the
13398 @code{init_priority} attribute by specifying a relative @var{priority},
13399 a constant integral expression currently bounded between 101 and 65535
13400 inclusive. Lower numbers indicate a higher priority.
13402 In the following example, @code{A} would normally be created before
13403 @code{B}, but the @code{init_priority} attribute has reversed that order:
13406 Some_Class A __attribute__ ((init_priority (2000)));
13407 Some_Class B __attribute__ ((init_priority (543)));
13411 Note that the particular values of @var{priority} do not matter; only their
13414 @item java_interface
13415 @cindex java_interface attribute
13417 This type attribute informs C++ that the class is a Java interface. It may
13418 only be applied to classes declared within an @code{extern "Java"} block.
13419 Calls to methods declared in this interface will be dispatched using GCJ's
13420 interface table mechanism, instead of regular virtual table dispatch.
13424 See also @ref{Namespace Association}.
13426 @node Namespace Association
13427 @section Namespace Association
13429 @strong{Caution:} The semantics of this extension are not fully
13430 defined. Users should refrain from using this extension as its
13431 semantics may change subtly over time. It is possible that this
13432 extension will be removed in future versions of G++.
13434 A using-directive with @code{__attribute ((strong))} is stronger
13435 than a normal using-directive in two ways:
13439 Templates from the used namespace can be specialized and explicitly
13440 instantiated as though they were members of the using namespace.
13443 The using namespace is considered an associated namespace of all
13444 templates in the used namespace for purposes of argument-dependent
13448 The used namespace must be nested within the using namespace so that
13449 normal unqualified lookup works properly.
13451 This is useful for composing a namespace transparently from
13452 implementation namespaces. For example:
13457 template <class T> struct A @{ @};
13459 using namespace debug __attribute ((__strong__));
13460 template <> struct A<int> @{ @}; // @r{ok to specialize}
13462 template <class T> void f (A<T>);
13467 f (std::A<float>()); // @r{lookup finds} std::f
13473 @section Type Traits
13475 The C++ front-end implements syntactic extensions that allow to
13476 determine at compile time various characteristics of a type (or of a
13480 @item __has_nothrow_assign (type)
13481 If @code{type} is const qualified or is a reference type then the trait is
13482 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
13483 is true, else if @code{type} is a cv class or union type with copy assignment
13484 operators that are known not to throw an exception then the trait is true,
13485 else it is false. Requires: @code{type} shall be a complete type, an array
13486 type of unknown bound, or is a @code{void} type.
13488 @item __has_nothrow_copy (type)
13489 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
13490 @code{type} is a cv class or union type with copy constructors that
13491 are known not to throw an exception then the trait is true, else it is false.
13492 Requires: @code{type} shall be a complete type, an array type of
13493 unknown bound, or is a @code{void} type.
13495 @item __has_nothrow_constructor (type)
13496 If @code{__has_trivial_constructor (type)} is true then the trait is
13497 true, else if @code{type} is a cv class or union type (or array
13498 thereof) with a default constructor that is known not to throw an
13499 exception then the trait is true, else it is false. Requires:
13500 @code{type} shall be a complete type, an array type of unknown bound,
13501 or is a @code{void} type.
13503 @item __has_trivial_assign (type)
13504 If @code{type} is const qualified or is a reference type then the trait is
13505 false. Otherwise if @code{__is_pod (type)} is true then the trait is
13506 true, else if @code{type} is a cv class or union type with a trivial
13507 copy assignment ([class.copy]) then the trait is true, else it is
13508 false. Requires: @code{type} shall be a complete type, an array type
13509 of unknown bound, or is a @code{void} type.
13511 @item __has_trivial_copy (type)
13512 If @code{__is_pod (type)} is true or @code{type} is a reference type
13513 then the trait is true, else if @code{type} is a cv class or union type
13514 with a trivial copy constructor ([class.copy]) then the trait
13515 is true, else it is false. Requires: @code{type} shall be a complete
13516 type, an array type of unknown bound, or is a @code{void} type.
13518 @item __has_trivial_constructor (type)
13519 If @code{__is_pod (type)} is true then the trait is true, else if
13520 @code{type} is a cv class or union type (or array thereof) with a
13521 trivial default constructor ([class.ctor]) then the trait is true,
13522 else it is false. Requires: @code{type} shall be a complete type, an
13523 array type of unknown bound, or is a @code{void} type.
13525 @item __has_trivial_destructor (type)
13526 If @code{__is_pod (type)} is true or @code{type} is a reference type then
13527 the trait is true, else if @code{type} is a cv class or union type (or
13528 array thereof) with a trivial destructor ([class.dtor]) then the trait
13529 is true, else it is false. Requires: @code{type} shall be a complete
13530 type, an array type of unknown bound, or is a @code{void} type.
13532 @item __has_virtual_destructor (type)
13533 If @code{type} is a class type with a virtual destructor
13534 ([class.dtor]) then the trait is true, else it is false. Requires:
13535 @code{type} shall be a complete type, an array type of unknown bound,
13536 or is a @code{void} type.
13538 @item __is_abstract (type)
13539 If @code{type} is an abstract class ([class.abstract]) then the trait
13540 is true, else it is false. Requires: @code{type} shall be a complete
13541 type, an array type of unknown bound, or is a @code{void} type.
13543 @item __is_base_of (base_type, derived_type)
13544 If @code{base_type} is a base class of @code{derived_type}
13545 ([class.derived]) then the trait is true, otherwise it is false.
13546 Top-level cv qualifications of @code{base_type} and
13547 @code{derived_type} are ignored. For the purposes of this trait, a
13548 class type is considered is own base. Requires: if @code{__is_class
13549 (base_type)} and @code{__is_class (derived_type)} are true and
13550 @code{base_type} and @code{derived_type} are not the same type
13551 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
13552 type. Diagnostic is produced if this requirement is not met.
13554 @item __is_class (type)
13555 If @code{type} is a cv class type, and not a union type
13556 ([basic.compound]) the trait is true, else it is false.
13558 @item __is_empty (type)
13559 If @code{__is_class (type)} is false then the trait is false.
13560 Otherwise @code{type} is considered empty if and only if: @code{type}
13561 has no non-static data members, or all non-static data members, if
13562 any, are bit-fields of length 0, and @code{type} has no virtual
13563 members, and @code{type} has no virtual base classes, and @code{type}
13564 has no base classes @code{base_type} for which
13565 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
13566 be a complete type, an array type of unknown bound, or is a
13569 @item __is_enum (type)
13570 If @code{type} is a cv enumeration type ([basic.compound]) the trait is
13571 true, else it is false.
13573 @item __is_pod (type)
13574 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
13575 else it is false. Requires: @code{type} shall be a complete type,
13576 an array type of unknown bound, or is a @code{void} type.
13578 @item __is_polymorphic (type)
13579 If @code{type} is a polymorphic class ([class.virtual]) then the trait
13580 is true, else it is false. Requires: @code{type} shall be a complete
13581 type, an array type of unknown bound, or is a @code{void} type.
13583 @item __is_union (type)
13584 If @code{type} is a cv union type ([basic.compound]) the trait is
13585 true, else it is false.
13589 @node Java Exceptions
13590 @section Java Exceptions
13592 The Java language uses a slightly different exception handling model
13593 from C++. Normally, GNU C++ will automatically detect when you are
13594 writing C++ code that uses Java exceptions, and handle them
13595 appropriately. However, if C++ code only needs to execute destructors
13596 when Java exceptions are thrown through it, GCC will guess incorrectly.
13597 Sample problematic code is:
13600 struct S @{ ~S(); @};
13601 extern void bar(); // @r{is written in Java, and may throw exceptions}
13610 The usual effect of an incorrect guess is a link failure, complaining of
13611 a missing routine called @samp{__gxx_personality_v0}.
13613 You can inform the compiler that Java exceptions are to be used in a
13614 translation unit, irrespective of what it might think, by writing
13615 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
13616 @samp{#pragma} must appear before any functions that throw or catch
13617 exceptions, or run destructors when exceptions are thrown through them.
13619 You cannot mix Java and C++ exceptions in the same translation unit. It
13620 is believed to be safe to throw a C++ exception from one file through
13621 another file compiled for the Java exception model, or vice versa, but
13622 there may be bugs in this area.
13624 @node Deprecated Features
13625 @section Deprecated Features
13627 In the past, the GNU C++ compiler was extended to experiment with new
13628 features, at a time when the C++ language was still evolving. Now that
13629 the C++ standard is complete, some of those features are superseded by
13630 superior alternatives. Using the old features might cause a warning in
13631 some cases that the feature will be dropped in the future. In other
13632 cases, the feature might be gone already.
13634 While the list below is not exhaustive, it documents some of the options
13635 that are now deprecated:
13638 @item -fexternal-templates
13639 @itemx -falt-external-templates
13640 These are two of the many ways for G++ to implement template
13641 instantiation. @xref{Template Instantiation}. The C++ standard clearly
13642 defines how template definitions have to be organized across
13643 implementation units. G++ has an implicit instantiation mechanism that
13644 should work just fine for standard-conforming code.
13646 @item -fstrict-prototype
13647 @itemx -fno-strict-prototype
13648 Previously it was possible to use an empty prototype parameter list to
13649 indicate an unspecified number of parameters (like C), rather than no
13650 parameters, as C++ demands. This feature has been removed, except where
13651 it is required for backwards compatibility. @xref{Backwards Compatibility}.
13654 G++ allows a virtual function returning @samp{void *} to be overridden
13655 by one returning a different pointer type. This extension to the
13656 covariant return type rules is now deprecated and will be removed from a
13659 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
13660 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
13661 and are now removed from G++. Code using these operators should be
13662 modified to use @code{std::min} and @code{std::max} instead.
13664 The named return value extension has been deprecated, and is now
13667 The use of initializer lists with new expressions has been deprecated,
13668 and is now removed from G++.
13670 Floating and complex non-type template parameters have been deprecated,
13671 and are now removed from G++.
13673 The implicit typename extension has been deprecated and is now
13676 The use of default arguments in function pointers, function typedefs
13677 and other places where they are not permitted by the standard is
13678 deprecated and will be removed from a future version of G++.
13680 G++ allows floating-point literals to appear in integral constant expressions,
13681 e.g. @samp{ enum E @{ e = int(2.2 * 3.7) @} }
13682 This extension is deprecated and will be removed from a future version.
13684 G++ allows static data members of const floating-point type to be declared
13685 with an initializer in a class definition. The standard only allows
13686 initializers for static members of const integral types and const
13687 enumeration types so this extension has been deprecated and will be removed
13688 from a future version.
13690 @node Backwards Compatibility
13691 @section Backwards Compatibility
13692 @cindex Backwards Compatibility
13693 @cindex ARM [Annotated C++ Reference Manual]
13695 Now that there is a definitive ISO standard C++, G++ has a specification
13696 to adhere to. The C++ language evolved over time, and features that
13697 used to be acceptable in previous drafts of the standard, such as the ARM
13698 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
13699 compilation of C++ written to such drafts, G++ contains some backwards
13700 compatibilities. @emph{All such backwards compatibility features are
13701 liable to disappear in future versions of G++.} They should be considered
13702 deprecated. @xref{Deprecated Features}.
13706 If a variable is declared at for scope, it used to remain in scope until
13707 the end of the scope which contained the for statement (rather than just
13708 within the for scope). G++ retains this, but issues a warning, if such a
13709 variable is accessed outside the for scope.
13711 @item Implicit C language
13712 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
13713 scope to set the language. On such systems, all header files are
13714 implicitly scoped inside a C language scope. Also, an empty prototype
13715 @code{()} will be treated as an unspecified number of arguments, rather
13716 than no arguments, as C++ demands.