1 @c Copyright (C) 1988-2013 Free Software Foundation, Inc.
3 @c This is part of the GCC manual.
4 @c For copying conditions, see the file gcc.texi.
7 @chapter Extensions to the C Language Family
8 @cindex extensions, C language
9 @cindex C language extensions
12 GNU C provides several language features not found in ISO standard C@.
13 (The @option{-pedantic} option directs GCC to print a warning message if
14 any of these features is used.) To test for the availability of these
15 features in conditional compilation, check for a predefined macro
16 @code{__GNUC__}, which is always defined under GCC@.
18 These extensions are available in C and Objective-C@. Most of them are
19 also available in C++. @xref{C++ Extensions,,Extensions to the
20 C++ Language}, for extensions that apply @emph{only} to C++.
22 Some features that are in ISO C99 but not C90 or C++ are also, as
23 extensions, accepted by GCC in C90 mode and in C++.
26 * Statement Exprs:: Putting statements and declarations inside expressions.
27 * Local Labels:: Labels local to a block.
28 * Labels as Values:: Getting pointers to labels, and computed gotos.
29 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
30 * Constructing Calls:: Dispatching a call to another function.
31 * Typeof:: @code{typeof}: referring to the type of an expression.
32 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
33 * Long Long:: Double-word integers---@code{long long int}.
34 * __int128:: 128-bit integers---@code{__int128}.
35 * Complex:: Data types for complex numbers.
36 * Floating Types:: Additional Floating Types.
37 * Half-Precision:: Half-Precision Floating Point.
38 * Decimal Float:: Decimal Floating Types.
39 * Hex Floats:: Hexadecimal floating-point constants.
40 * Fixed-Point:: Fixed-Point Types.
41 * Named Address Spaces::Named address spaces.
42 * Zero Length:: Zero-length arrays.
43 * Variable Length:: Arrays whose length is computed at run time.
44 * Empty Structures:: Structures with no members.
45 * Variadic Macros:: Macros with a variable number of arguments.
46 * Escaped Newlines:: Slightly looser rules for escaped newlines.
47 * Subscripting:: Any array can be subscripted, even if not an lvalue.
48 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
49 * Initializers:: Non-constant initializers.
50 * Compound Literals:: Compound literals give structures, unions
52 * Designated Inits:: Labeling elements of initializers.
53 * Cast to Union:: Casting to union type from any member of the union.
54 * Case Ranges:: `case 1 ... 9' and such.
55 * Mixed Declarations:: Mixing declarations and code.
56 * Function Attributes:: Declaring that functions have no side effects,
57 or that they can never return.
58 * Attribute Syntax:: Formal syntax for attributes.
59 * Function Prototypes:: Prototype declarations and old-style definitions.
60 * C++ Comments:: C++ comments are recognized.
61 * Dollar Signs:: Dollar sign is allowed in identifiers.
62 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
63 * Variable Attributes:: Specifying attributes of variables.
64 * Type Attributes:: Specifying attributes of types.
65 * Alignment:: Inquiring about the alignment of a type or variable.
66 * Inline:: Defining inline functions (as fast as macros).
67 * Volatiles:: What constitutes an access to a volatile object.
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 * __sync Builtins:: Legacy built-in functions for atomic memory access.
81 * __atomic Builtins:: Atomic built-in functions with memory model.
82 * Object Size Checking:: Built-in functions for limited buffer overflow
84 * Other Builtins:: Other built-in functions.
85 * Target Builtins:: Built-in functions specific to particular targets.
86 * Target Format Checks:: Format checks specific to particular targets.
87 * Pragmas:: Pragmas accepted by GCC.
88 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
89 * Thread-Local:: Per-thread variables.
90 * Binary constants:: Binary constants using the @samp{0b} prefix.
94 @section Statements and Declarations in Expressions
95 @cindex statements inside expressions
96 @cindex declarations inside expressions
97 @cindex expressions containing statements
98 @cindex macros, statements in expressions
100 @c the above section title wrapped and causes an underfull hbox.. i
101 @c changed it from "within" to "in". --mew 4feb93
102 A compound statement enclosed in parentheses may appear as an expression
103 in GNU C@. This allows you to use loops, switches, and local variables
104 within an expression.
106 Recall that a compound statement is a sequence of statements surrounded
107 by braces; in this construct, parentheses go around the braces. For
111 (@{ int y = foo (); int z;
118 is a valid (though slightly more complex than necessary) expression
119 for the absolute value of @code{foo ()}.
121 The last thing in the compound statement should be an expression
122 followed by a semicolon; the value of this subexpression serves as the
123 value of the entire construct. (If you use some other kind of statement
124 last within the braces, the construct has type @code{void}, and thus
125 effectively no value.)
127 This feature is especially useful in making macro definitions ``safe'' (so
128 that they evaluate each operand exactly once). For example, the
129 ``maximum'' function is commonly defined as a macro in standard C as
133 #define max(a,b) ((a) > (b) ? (a) : (b))
137 @cindex side effects, macro argument
138 But this definition computes either @var{a} or @var{b} twice, with bad
139 results if the operand has side effects. In GNU C, if you know the
140 type of the operands (here taken as @code{int}), you can define
141 the macro safely as follows:
144 #define maxint(a,b) \
145 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
148 Embedded statements are not allowed in constant expressions, such as
149 the value of an enumeration constant, the width of a bit-field, or
150 the initial value of a static variable.
152 If you don't know the type of the operand, you can still do this, but you
153 must use @code{typeof} (@pxref{Typeof}).
155 In G++, the result value of a statement expression undergoes array and
156 function pointer decay, and is returned by value to the enclosing
157 expression. For instance, if @code{A} is a class, then
166 constructs a temporary @code{A} object to hold the result of the
167 statement expression, and that is used to invoke @code{Foo}.
168 Therefore the @code{this} pointer observed by @code{Foo} is not the
171 In a statement expression, any temporaries created within a statement
172 are destroyed at that statement's end. This makes statement
173 expressions inside macros slightly different from function calls. In
174 the latter case temporaries introduced during argument evaluation are
175 destroyed at the end of the statement that includes the function
176 call. In the statement expression case they are destroyed during
177 the statement expression. For instance,
180 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
181 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
191 has different places where temporaries are destroyed. For the
192 @code{macro} case, the temporary @code{X} is destroyed just after
193 the initialization of @code{b}. In the @code{function} case that
194 temporary is destroyed when the function returns.
196 These considerations mean that it is probably a bad idea to use
197 statement expressions of this form in header files that are designed to
198 work with C++. (Note that some versions of the GNU C Library contained
199 header files using statement expressions that lead to precisely this
202 Jumping into a statement expression with @code{goto} or using a
203 @code{switch} statement outside the statement expression with a
204 @code{case} or @code{default} label inside the statement expression is
205 not permitted. Jumping into a statement expression with a computed
206 @code{goto} (@pxref{Labels as Values}) has undefined behavior.
207 Jumping out of a statement expression is permitted, but if the
208 statement expression is part of a larger expression then it is
209 unspecified which other subexpressions of that expression have been
210 evaluated except where the language definition requires certain
211 subexpressions to be evaluated before or after the statement
212 expression. In any case, as with a function call, the evaluation of a
213 statement expression is not interleaved with the evaluation of other
214 parts of the containing expression. For example,
217 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
221 calls @code{foo} and @code{bar1} and does not call @code{baz} but
222 may or may not call @code{bar2}. If @code{bar2} is called, it is
223 called after @code{foo} and before @code{bar1}.
226 @section Locally Declared Labels
228 @cindex macros, local labels
230 GCC allows you to declare @dfn{local labels} in any nested block
231 scope. A local label is just like an ordinary label, but you can
232 only reference it (with a @code{goto} statement, or by taking its
233 address) within the block in which it is declared.
235 A local label declaration looks like this:
238 __label__ @var{label};
245 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
248 Local label declarations must come at the beginning of the block,
249 before any ordinary declarations or statements.
251 The label declaration defines the label @emph{name}, but does not define
252 the label itself. You must do this in the usual way, with
253 @code{@var{label}:}, within the statements of the statement expression.
255 The local label feature is useful for complex macros. If a macro
256 contains nested loops, a @code{goto} can be useful for breaking out of
257 them. However, an ordinary label whose scope is the whole function
258 cannot be used: if the macro can be expanded several times in one
259 function, the label is multiply defined in that function. A
260 local label avoids this problem. For example:
263 #define SEARCH(value, array, target) \
266 typeof (target) _SEARCH_target = (target); \
267 typeof (*(array)) *_SEARCH_array = (array); \
270 for (i = 0; i < max; i++) \
271 for (j = 0; j < max; j++) \
272 if (_SEARCH_array[i][j] == _SEARCH_target) \
273 @{ (value) = i; goto found; @} \
279 This could also be written using a statement expression:
282 #define SEARCH(array, target) \
285 typeof (target) _SEARCH_target = (target); \
286 typeof (*(array)) *_SEARCH_array = (array); \
289 for (i = 0; i < max; i++) \
290 for (j = 0; j < max; j++) \
291 if (_SEARCH_array[i][j] == _SEARCH_target) \
292 @{ value = i; goto found; @} \
299 Local label declarations also make the labels they declare visible to
300 nested functions, if there are any. @xref{Nested Functions}, for details.
302 @node Labels as Values
303 @section Labels as Values
304 @cindex labels as values
305 @cindex computed gotos
306 @cindex goto with computed label
307 @cindex address of a label
309 You can get the address of a label defined in the current function
310 (or a containing function) with the unary operator @samp{&&}. The
311 value has type @code{void *}. This value is a constant and can be used
312 wherever a constant of that type is valid. For example:
320 To use these values, you need to be able to jump to one. This is done
321 with the computed goto statement@footnote{The analogous feature in
322 Fortran is called an assigned goto, but that name seems inappropriate in
323 C, where one can do more than simply store label addresses in label
324 variables.}, @code{goto *@var{exp};}. For example,
331 Any expression of type @code{void *} is allowed.
333 One way of using these constants is in initializing a static array that
334 serves as a jump table:
337 static void *array[] = @{ &&foo, &&bar, &&hack @};
341 Then you can select a label with indexing, like this:
348 Note that this does not check whether the subscript is in bounds---array
349 indexing in C never does that.
351 Such an array of label values serves a purpose much like that of the
352 @code{switch} statement. The @code{switch} statement is cleaner, so
353 use that rather than an array unless the problem does not fit a
354 @code{switch} statement very well.
356 Another use of label values is in an interpreter for threaded code.
357 The labels within the interpreter function can be stored in the
358 threaded code for super-fast dispatching.
360 You may not use this mechanism to jump to code in a different function.
361 If you do that, totally unpredictable things happen. The best way to
362 avoid this is to store the label address only in automatic variables and
363 never pass it as an argument.
365 An alternate way to write the above example is
368 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
370 goto *(&&foo + array[i]);
374 This is more friendly to code living in shared libraries, as it reduces
375 the number of dynamic relocations that are needed, and by consequence,
376 allows the data to be read-only.
378 The @code{&&foo} expressions for the same label might have different
379 values if the containing function is inlined or cloned. If a program
380 relies on them being always the same,
381 @code{__attribute__((__noinline__,__noclone__))} should be used to
382 prevent inlining and cloning. If @code{&&foo} is used in a static
383 variable initializer, inlining and cloning is forbidden.
385 @node Nested Functions
386 @section Nested Functions
387 @cindex nested functions
388 @cindex downward funargs
391 A @dfn{nested function} is a function defined inside another function.
392 Nested functions are supported as an extension in GNU C, but are not
393 supported by GNU C++.
395 The nested function's name is local to the block where it is defined.
396 For example, here we define a nested function named @code{square}, and
401 foo (double a, double b)
403 double square (double z) @{ return z * z; @}
405 return square (a) + square (b);
410 The nested function can access all the variables of the containing
411 function that are visible at the point of its definition. This is
412 called @dfn{lexical scoping}. For example, here we show a nested
413 function which uses an inherited variable named @code{offset}:
417 bar (int *array, int offset, int size)
419 int access (int *array, int index)
420 @{ return array[index + offset]; @}
423 for (i = 0; i < size; i++)
424 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
429 Nested function definitions are permitted within functions in the places
430 where variable definitions are allowed; that is, in any block, mixed
431 with the other declarations and statements in the block.
433 It is possible to call the nested function from outside the scope of its
434 name by storing its address or passing the address to another function:
437 hack (int *array, int size)
439 void store (int index, int value)
440 @{ array[index] = value; @}
442 intermediate (store, size);
446 Here, the function @code{intermediate} receives the address of
447 @code{store} as an argument. If @code{intermediate} calls @code{store},
448 the arguments given to @code{store} are used to store into @code{array}.
449 But this technique works only so long as the containing function
450 (@code{hack}, in this example) does not exit.
452 If you try to call the nested function through its address after the
453 containing function exits, all hell breaks loose. If you try
454 to call it after a containing scope level exits, and if it refers
455 to some of the variables that are no longer in scope, you may be lucky,
456 but it's not wise to take the risk. If, however, the nested function
457 does not refer to anything that has gone out of scope, you should be
460 GCC implements taking the address of a nested function using a technique
461 called @dfn{trampolines}. This technique was described in
462 @cite{Lexical Closures for C++} (Thomas M. Breuel, USENIX
463 C++ Conference Proceedings, October 17-21, 1988).
465 A nested function can jump to a label inherited from a containing
466 function, provided the label is explicitly declared in the containing
467 function (@pxref{Local Labels}). Such a jump returns instantly to the
468 containing function, exiting the nested function that did the
469 @code{goto} and any intermediate functions as well. Here is an example:
473 bar (int *array, int offset, int size)
476 int access (int *array, int index)
480 return array[index + offset];
484 for (i = 0; i < size; i++)
485 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
489 /* @r{Control comes here from @code{access}
490 if it detects an error.} */
497 A nested function always has no linkage. Declaring one with
498 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
499 before its definition, use @code{auto} (which is otherwise meaningless
500 for function declarations).
503 bar (int *array, int offset, int size)
506 auto int access (int *, int);
508 int access (int *array, int index)
512 return array[index + offset];
518 @node Constructing Calls
519 @section Constructing Function Calls
520 @cindex constructing calls
521 @cindex forwarding calls
523 Using the built-in functions described below, you can record
524 the arguments a function received, and call another function
525 with the same arguments, without knowing the number or types
528 You can also record the return value of that function call,
529 and later return that value, without knowing what data type
530 the function tried to return (as long as your caller expects
533 However, these built-in functions may interact badly with some
534 sophisticated features or other extensions of the language. It
535 is, therefore, not recommended to use them outside very simple
536 functions acting as mere forwarders for their arguments.
538 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
539 This built-in function returns a pointer to data
540 describing how to perform a call with the same arguments as are passed
541 to the current function.
543 The function saves the arg pointer register, structure value address,
544 and all registers that might be used to pass arguments to a function
545 into a block of memory allocated on the stack. Then it returns the
546 address of that block.
549 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
550 This built-in function invokes @var{function}
551 with a copy of the parameters described by @var{arguments}
554 The value of @var{arguments} should be the value returned by
555 @code{__builtin_apply_args}. The argument @var{size} specifies the size
556 of the stack argument data, in bytes.
558 This function returns a pointer to data describing
559 how to return whatever value is returned by @var{function}. The data
560 is saved in a block of memory allocated on the stack.
562 It is not always simple to compute the proper value for @var{size}. The
563 value is used by @code{__builtin_apply} to compute the amount of data
564 that should be pushed on the stack and copied from the incoming argument
568 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
569 This built-in function returns the value described by @var{result} from
570 the containing function. You should specify, for @var{result}, a value
571 returned by @code{__builtin_apply}.
574 @deftypefn {Built-in Function} {} __builtin_va_arg_pack ()
575 This built-in function represents all anonymous arguments of an inline
576 function. It can be used only in inline functions that are always
577 inlined, never compiled as a separate function, such as those using
578 @code{__attribute__ ((__always_inline__))} or
579 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
580 It must be only passed as last argument to some other function
581 with variable arguments. This is useful for writing small wrapper
582 inlines for variable argument functions, when using preprocessor
583 macros is undesirable. For example:
585 extern int myprintf (FILE *f, const char *format, ...);
586 extern inline __attribute__ ((__gnu_inline__)) int
587 myprintf (FILE *f, const char *format, ...)
589 int r = fprintf (f, "myprintf: ");
592 int s = fprintf (f, format, __builtin_va_arg_pack ());
600 @deftypefn {Built-in Function} {size_t} __builtin_va_arg_pack_len ()
601 This built-in function returns the number of anonymous arguments of
602 an inline function. It can be used only in inline functions that
603 are always inlined, never compiled as a separate function, such
604 as those using @code{__attribute__ ((__always_inline__))} or
605 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
606 For example following does link- or run-time checking of open
607 arguments for optimized code:
610 extern inline __attribute__((__gnu_inline__)) int
611 myopen (const char *path, int oflag, ...)
613 if (__builtin_va_arg_pack_len () > 1)
614 warn_open_too_many_arguments ();
616 if (__builtin_constant_p (oflag))
618 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1)
620 warn_open_missing_mode ();
621 return __open_2 (path, oflag);
623 return open (path, oflag, __builtin_va_arg_pack ());
626 if (__builtin_va_arg_pack_len () < 1)
627 return __open_2 (path, oflag);
629 return open (path, oflag, __builtin_va_arg_pack ());
636 @section Referring to a Type with @code{typeof}
639 @cindex macros, types of arguments
641 Another way to refer to the type of an expression is with @code{typeof}.
642 The syntax of using of this keyword looks like @code{sizeof}, but the
643 construct acts semantically like a type name defined with @code{typedef}.
645 There are two ways of writing the argument to @code{typeof}: with an
646 expression or with a type. Here is an example with an expression:
653 This assumes that @code{x} is an array of pointers to functions;
654 the type described is that of the values of the functions.
656 Here is an example with a typename as the argument:
663 Here the type described is that of pointers to @code{int}.
665 If you are writing a header file that must work when included in ISO C
666 programs, write @code{__typeof__} instead of @code{typeof}.
667 @xref{Alternate Keywords}.
669 A @code{typeof} construct can be used anywhere a typedef name can be
670 used. For example, you can use it in a declaration, in a cast, or inside
671 of @code{sizeof} or @code{typeof}.
673 The operand of @code{typeof} is evaluated for its side effects if and
674 only if it is an expression of variably modified type or the name of
677 @code{typeof} is often useful in conjunction with
678 statement expressions (@pxref{Statement Exprs}).
679 Here is how the two together can
680 be used to define a safe ``maximum'' macro which operates on any
681 arithmetic type and evaluates each of its arguments exactly once:
685 (@{ typeof (a) _a = (a); \
686 typeof (b) _b = (b); \
687 _a > _b ? _a : _b; @})
690 @cindex underscores in variables in macros
691 @cindex @samp{_} in variables in macros
692 @cindex local variables in macros
693 @cindex variables, local, in macros
694 @cindex macros, local variables in
696 The reason for using names that start with underscores for the local
697 variables is to avoid conflicts with variable names that occur within the
698 expressions that are substituted for @code{a} and @code{b}. Eventually we
699 hope to design a new form of declaration syntax that allows you to declare
700 variables whose scopes start only after their initializers; this will be a
701 more reliable way to prevent such conflicts.
704 Some more examples of the use of @code{typeof}:
708 This declares @code{y} with the type of what @code{x} points to.
715 This declares @code{y} as an array of such values.
722 This declares @code{y} as an array of pointers to characters:
725 typeof (typeof (char *)[4]) y;
729 It is equivalent to the following traditional C declaration:
735 To see the meaning of the declaration using @code{typeof}, and why it
736 might be a useful way to write, rewrite it with these macros:
739 #define pointer(T) typeof(T *)
740 #define array(T, N) typeof(T [N])
744 Now the declaration can be rewritten this way:
747 array (pointer (char), 4) y;
751 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
752 pointers to @code{char}.
755 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
756 a more limited extension that permitted one to write
759 typedef @var{T} = @var{expr};
763 with the effect of declaring @var{T} to have the type of the expression
764 @var{expr}. This extension does not work with GCC 3 (versions between
765 3.0 and 3.2 crash; 3.2.1 and later give an error). Code that
766 relies on it should be rewritten to use @code{typeof}:
769 typedef typeof(@var{expr}) @var{T};
773 This works with all versions of GCC@.
776 @section Conditionals with Omitted Operands
777 @cindex conditional expressions, extensions
778 @cindex omitted middle-operands
779 @cindex middle-operands, omitted
780 @cindex extensions, @code{?:}
781 @cindex @code{?:} extensions
783 The middle operand in a conditional expression may be omitted. Then
784 if the first operand is nonzero, its value is the value of the conditional
787 Therefore, the expression
794 has the value of @code{x} if that is nonzero; otherwise, the value of
797 This example is perfectly equivalent to
803 @cindex side effect in @code{?:}
804 @cindex @code{?:} side effect
806 In this simple case, the ability to omit the middle operand is not
807 especially useful. When it becomes useful is when the first operand does,
808 or may (if it is a macro argument), contain a side effect. Then repeating
809 the operand in the middle would perform the side effect twice. Omitting
810 the middle operand uses the value already computed without the undesirable
811 effects of recomputing it.
814 @section 128-bit integers
815 @cindex @code{__int128} data types
817 As an extension the integer scalar type @code{__int128} is supported for
818 targets which have an integer mode wide enough to hold 128 bits.
819 Simply write @code{__int128} for a signed 128-bit integer, or
820 @code{unsigned __int128} for an unsigned 128-bit integer. There is no
821 support in GCC for expressing an integer constant of type @code{__int128}
822 for targets with @code{long long} integer less than 128 bits wide.
825 @section Double-Word Integers
826 @cindex @code{long long} data types
827 @cindex double-word arithmetic
828 @cindex multiprecision arithmetic
829 @cindex @code{LL} integer suffix
830 @cindex @code{ULL} integer suffix
832 ISO C99 supports data types for integers that are at least 64 bits wide,
833 and as an extension GCC supports them in C90 mode and in C++.
834 Simply write @code{long long int} for a signed integer, or
835 @code{unsigned long long int} for an unsigned integer. To make an
836 integer constant of type @code{long long int}, add the suffix @samp{LL}
837 to the integer. To make an integer constant of type @code{unsigned long
838 long int}, add the suffix @samp{ULL} to the integer.
840 You can use these types in arithmetic like any other integer types.
841 Addition, subtraction, and bitwise boolean operations on these types
842 are open-coded on all types of machines. Multiplication is open-coded
843 if the machine supports a fullword-to-doubleword widening multiply
844 instruction. Division and shifts are open-coded only on machines that
845 provide special support. The operations that are not open-coded use
846 special library routines that come with GCC@.
848 There may be pitfalls when you use @code{long long} types for function
849 arguments without function prototypes. If a function
850 expects type @code{int} for its argument, and you pass a value of type
851 @code{long long int}, confusion results because the caller and the
852 subroutine disagree about the number of bytes for the argument.
853 Likewise, if the function expects @code{long long int} and you pass
854 @code{int}. The best way to avoid such problems is to use prototypes.
857 @section Complex Numbers
858 @cindex complex numbers
859 @cindex @code{_Complex} keyword
860 @cindex @code{__complex__} keyword
862 ISO C99 supports complex floating data types, and as an extension GCC
863 supports them in C90 mode and in C++. GCC also supports complex integer data
864 types which are not part of ISO C99. You can declare complex types
865 using the keyword @code{_Complex}. As an extension, the older GNU
866 keyword @code{__complex__} is also supported.
868 For example, @samp{_Complex double x;} declares @code{x} as a
869 variable whose real part and imaginary part are both of type
870 @code{double}. @samp{_Complex short int y;} declares @code{y} to
871 have real and imaginary parts of type @code{short int}; this is not
872 likely to be useful, but it shows that the set of complex types is
875 To write a constant with a complex data type, use the suffix @samp{i} or
876 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
877 has type @code{_Complex float} and @code{3i} has type
878 @code{_Complex int}. Such a constant always has a pure imaginary
879 value, but you can form any complex value you like by adding one to a
880 real constant. This is a GNU extension; if you have an ISO C99
881 conforming C library (such as the GNU C Library), and want to construct complex
882 constants of floating type, you should include @code{<complex.h>} and
883 use the macros @code{I} or @code{_Complex_I} instead.
885 @cindex @code{__real__} keyword
886 @cindex @code{__imag__} keyword
887 To extract the real part of a complex-valued expression @var{exp}, write
888 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
889 extract the imaginary part. This is a GNU extension; for values of
890 floating type, you should use the ISO C99 functions @code{crealf},
891 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
892 @code{cimagl}, declared in @code{<complex.h>} and also provided as
893 built-in functions by GCC@.
895 @cindex complex conjugation
896 The operator @samp{~} performs complex conjugation when used on a value
897 with a complex type. This is a GNU extension; for values of
898 floating type, you should use the ISO C99 functions @code{conjf},
899 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
900 provided as built-in functions by GCC@.
902 GCC can allocate complex automatic variables in a noncontiguous
903 fashion; it's even possible for the real part to be in a register while
904 the imaginary part is on the stack (or vice versa). Only the DWARF 2
905 debug info format can represent this, so use of DWARF 2 is recommended.
906 If you are using the stabs debug info format, GCC describes a noncontiguous
907 complex variable as if it were two separate variables of noncomplex type.
908 If the variable's actual name is @code{foo}, the two fictitious
909 variables are named @code{foo$real} and @code{foo$imag}. You can
910 examine and set these two fictitious variables with your debugger.
913 @section Additional Floating Types
914 @cindex additional floating types
915 @cindex @code{__float80} data type
916 @cindex @code{__float128} data type
917 @cindex @code{w} floating point suffix
918 @cindex @code{q} floating point suffix
919 @cindex @code{W} floating point suffix
920 @cindex @code{Q} floating point suffix
922 As an extension, GNU C supports additional floating
923 types, @code{__float80} and @code{__float128} to support 80-bit
924 (@code{XFmode}) and 128-bit (@code{TFmode}) floating types.
925 Support for additional types includes the arithmetic operators:
926 add, subtract, multiply, divide; unary arithmetic operators;
927 relational operators; equality operators; and conversions to and from
928 integer and other floating types. Use a suffix @samp{w} or @samp{W}
929 in a literal constant of type @code{__float80} and @samp{q} or @samp{Q}
930 for @code{_float128}. You can declare complex types using the
931 corresponding internal complex type, @code{XCmode} for @code{__float80}
932 type and @code{TCmode} for @code{__float128} type:
935 typedef _Complex float __attribute__((mode(TC))) _Complex128;
936 typedef _Complex float __attribute__((mode(XC))) _Complex80;
939 Not all targets support additional floating-point types. @code{__float80}
940 and @code{__float128} types are supported on i386, x86_64 and IA-64 targets.
941 The @code{__float128} type is supported on hppa HP-UX targets.
944 @section Half-Precision Floating Point
945 @cindex half-precision floating point
946 @cindex @code{__fp16} data type
948 On ARM targets, GCC supports half-precision (16-bit) floating point via
949 the @code{__fp16} type. You must enable this type explicitly
950 with the @option{-mfp16-format} command-line option in order to use it.
952 ARM supports two incompatible representations for half-precision
953 floating-point values. You must choose one of the representations and
954 use it consistently in your program.
956 Specifying @option{-mfp16-format=ieee} selects the IEEE 754-2008 format.
957 This format can represent normalized values in the range of @math{2^{-14}} to 65504.
958 There are 11 bits of significand precision, approximately 3
961 Specifying @option{-mfp16-format=alternative} selects the ARM
962 alternative format. This representation is similar to the IEEE
963 format, but does not support infinities or NaNs. Instead, the range
964 of exponents is extended, so that this format can represent normalized
965 values in the range of @math{2^{-14}} to 131008.
967 The @code{__fp16} type is a storage format only. For purposes
968 of arithmetic and other operations, @code{__fp16} values in C or C++
969 expressions are automatically promoted to @code{float}. In addition,
970 you cannot declare a function with a return value or parameters
971 of type @code{__fp16}.
973 Note that conversions from @code{double} to @code{__fp16}
974 involve an intermediate conversion to @code{float}. Because
975 of rounding, this can sometimes produce a different result than a
978 ARM provides hardware support for conversions between
979 @code{__fp16} and @code{float} values
980 as an extension to VFP and NEON (Advanced SIMD). GCC generates
981 code using these hardware instructions if you compile with
982 options to select an FPU that provides them;
983 for example, @option{-mfpu=neon-fp16 -mfloat-abi=softfp},
984 in addition to the @option{-mfp16-format} option to select
985 a half-precision format.
987 Language-level support for the @code{__fp16} data type is
988 independent of whether GCC generates code using hardware floating-point
989 instructions. In cases where hardware support is not specified, GCC
990 implements conversions between @code{__fp16} and @code{float} values
994 @section Decimal Floating Types
995 @cindex decimal floating types
996 @cindex @code{_Decimal32} data type
997 @cindex @code{_Decimal64} data type
998 @cindex @code{_Decimal128} data type
999 @cindex @code{df} integer suffix
1000 @cindex @code{dd} integer suffix
1001 @cindex @code{dl} integer suffix
1002 @cindex @code{DF} integer suffix
1003 @cindex @code{DD} integer suffix
1004 @cindex @code{DL} integer suffix
1006 As an extension, GNU C supports decimal floating types as
1007 defined in the N1312 draft of ISO/IEC WDTR24732. Support for decimal
1008 floating types in GCC will evolve as the draft technical report changes.
1009 Calling conventions for any target might also change. Not all targets
1010 support decimal floating types.
1012 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
1013 @code{_Decimal128}. They use a radix of ten, unlike the floating types
1014 @code{float}, @code{double}, and @code{long double} whose radix is not
1015 specified by the C standard but is usually two.
1017 Support for decimal floating types includes the arithmetic operators
1018 add, subtract, multiply, divide; unary arithmetic operators;
1019 relational operators; equality operators; and conversions to and from
1020 integer and other floating types. Use a suffix @samp{df} or
1021 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
1022 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
1025 GCC support of decimal float as specified by the draft technical report
1030 When the value of a decimal floating type cannot be represented in the
1031 integer type to which it is being converted, the result is undefined
1032 rather than the result value specified by the draft technical report.
1035 GCC does not provide the C library functionality associated with
1036 @file{math.h}, @file{fenv.h}, @file{stdio.h}, @file{stdlib.h}, and
1037 @file{wchar.h}, which must come from a separate C library implementation.
1038 Because of this the GNU C compiler does not define macro
1039 @code{__STDC_DEC_FP__} to indicate that the implementation conforms to
1040 the technical report.
1043 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
1044 are supported by the DWARF 2 debug information format.
1050 ISO C99 supports floating-point numbers written not only in the usual
1051 decimal notation, such as @code{1.55e1}, but also numbers such as
1052 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1053 supports this in C90 mode (except in some cases when strictly
1054 conforming) and in C++. In that format the
1055 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1056 mandatory. The exponent is a decimal number that indicates the power of
1057 2 by which the significant part is multiplied. Thus @samp{0x1.f} is
1064 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1065 is the same as @code{1.55e1}.
1067 Unlike for floating-point numbers in the decimal notation the exponent
1068 is always required in the hexadecimal notation. Otherwise the compiler
1069 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1070 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1071 extension for floating-point constants of type @code{float}.
1074 @section Fixed-Point Types
1075 @cindex fixed-point types
1076 @cindex @code{_Fract} data type
1077 @cindex @code{_Accum} data type
1078 @cindex @code{_Sat} data type
1079 @cindex @code{hr} fixed-suffix
1080 @cindex @code{r} fixed-suffix
1081 @cindex @code{lr} fixed-suffix
1082 @cindex @code{llr} fixed-suffix
1083 @cindex @code{uhr} fixed-suffix
1084 @cindex @code{ur} fixed-suffix
1085 @cindex @code{ulr} fixed-suffix
1086 @cindex @code{ullr} fixed-suffix
1087 @cindex @code{hk} fixed-suffix
1088 @cindex @code{k} fixed-suffix
1089 @cindex @code{lk} fixed-suffix
1090 @cindex @code{llk} fixed-suffix
1091 @cindex @code{uhk} fixed-suffix
1092 @cindex @code{uk} fixed-suffix
1093 @cindex @code{ulk} fixed-suffix
1094 @cindex @code{ullk} fixed-suffix
1095 @cindex @code{HR} fixed-suffix
1096 @cindex @code{R} fixed-suffix
1097 @cindex @code{LR} fixed-suffix
1098 @cindex @code{LLR} fixed-suffix
1099 @cindex @code{UHR} fixed-suffix
1100 @cindex @code{UR} fixed-suffix
1101 @cindex @code{ULR} fixed-suffix
1102 @cindex @code{ULLR} fixed-suffix
1103 @cindex @code{HK} fixed-suffix
1104 @cindex @code{K} fixed-suffix
1105 @cindex @code{LK} fixed-suffix
1106 @cindex @code{LLK} fixed-suffix
1107 @cindex @code{UHK} fixed-suffix
1108 @cindex @code{UK} fixed-suffix
1109 @cindex @code{ULK} fixed-suffix
1110 @cindex @code{ULLK} fixed-suffix
1112 As an extension, GNU C supports fixed-point types as
1113 defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point
1114 types in GCC will evolve as the draft technical report changes.
1115 Calling conventions for any target might also change. Not all targets
1116 support fixed-point types.
1118 The fixed-point types are
1119 @code{short _Fract},
1122 @code{long long _Fract},
1123 @code{unsigned short _Fract},
1124 @code{unsigned _Fract},
1125 @code{unsigned long _Fract},
1126 @code{unsigned long long _Fract},
1127 @code{_Sat short _Fract},
1129 @code{_Sat long _Fract},
1130 @code{_Sat long long _Fract},
1131 @code{_Sat unsigned short _Fract},
1132 @code{_Sat unsigned _Fract},
1133 @code{_Sat unsigned long _Fract},
1134 @code{_Sat unsigned long long _Fract},
1135 @code{short _Accum},
1138 @code{long long _Accum},
1139 @code{unsigned short _Accum},
1140 @code{unsigned _Accum},
1141 @code{unsigned long _Accum},
1142 @code{unsigned long long _Accum},
1143 @code{_Sat short _Accum},
1145 @code{_Sat long _Accum},
1146 @code{_Sat long long _Accum},
1147 @code{_Sat unsigned short _Accum},
1148 @code{_Sat unsigned _Accum},
1149 @code{_Sat unsigned long _Accum},
1150 @code{_Sat unsigned long long _Accum}.
1152 Fixed-point data values contain fractional and optional integral parts.
1153 The format of fixed-point data varies and depends on the target machine.
1155 Support for fixed-point types includes:
1158 prefix and postfix increment and decrement operators (@code{++}, @code{--})
1160 unary arithmetic operators (@code{+}, @code{-}, @code{!})
1162 binary arithmetic operators (@code{+}, @code{-}, @code{*}, @code{/})
1164 binary shift operators (@code{<<}, @code{>>})
1166 relational operators (@code{<}, @code{<=}, @code{>=}, @code{>})
1168 equality operators (@code{==}, @code{!=})
1170 assignment operators (@code{+=}, @code{-=}, @code{*=}, @code{/=},
1171 @code{<<=}, @code{>>=})
1173 conversions to and from integer, floating-point, or fixed-point types
1176 Use a suffix in a fixed-point literal constant:
1178 @item @samp{hr} or @samp{HR} for @code{short _Fract} and
1179 @code{_Sat short _Fract}
1180 @item @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract}
1181 @item @samp{lr} or @samp{LR} for @code{long _Fract} and
1182 @code{_Sat long _Fract}
1183 @item @samp{llr} or @samp{LLR} for @code{long long _Fract} and
1184 @code{_Sat long long _Fract}
1185 @item @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and
1186 @code{_Sat unsigned short _Fract}
1187 @item @samp{ur} or @samp{UR} for @code{unsigned _Fract} and
1188 @code{_Sat unsigned _Fract}
1189 @item @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and
1190 @code{_Sat unsigned long _Fract}
1191 @item @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract}
1192 and @code{_Sat unsigned long long _Fract}
1193 @item @samp{hk} or @samp{HK} for @code{short _Accum} and
1194 @code{_Sat short _Accum}
1195 @item @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum}
1196 @item @samp{lk} or @samp{LK} for @code{long _Accum} and
1197 @code{_Sat long _Accum}
1198 @item @samp{llk} or @samp{LLK} for @code{long long _Accum} and
1199 @code{_Sat long long _Accum}
1200 @item @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and
1201 @code{_Sat unsigned short _Accum}
1202 @item @samp{uk} or @samp{UK} for @code{unsigned _Accum} and
1203 @code{_Sat unsigned _Accum}
1204 @item @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and
1205 @code{_Sat unsigned long _Accum}
1206 @item @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum}
1207 and @code{_Sat unsigned long long _Accum}
1210 GCC support of fixed-point types as specified by the draft technical report
1215 Pragmas to control overflow and rounding behaviors are not implemented.
1218 Fixed-point types are supported by the DWARF 2 debug information format.
1220 @node Named Address Spaces
1221 @section Named Address Spaces
1222 @cindex Named Address Spaces
1224 As an extension, GNU C supports named address spaces as
1225 defined in the N1275 draft of ISO/IEC DTR 18037. Support for named
1226 address spaces in GCC will evolve as the draft technical report
1227 changes. Calling conventions for any target might also change. At
1228 present, only the AVR, SPU, M32C, and RL78 targets support address
1229 spaces other than the generic address space.
1231 Address space identifiers may be used exactly like any other C type
1232 qualifier (e.g., @code{const} or @code{volatile}). See the N1275
1233 document for more details.
1235 @anchor{AVR Named Address Spaces}
1236 @subsection AVR Named Address Spaces
1238 On the AVR target, there are several address spaces that can be used
1239 in order to put read-only data into the flash memory and access that
1240 data by means of the special instructions @code{LPM} or @code{ELPM}
1241 needed to read from flash.
1243 Per default, any data including read-only data is located in RAM
1244 (the generic address space) so that non-generic address spaces are
1245 needed to locate read-only data in flash memory
1246 @emph{and} to generate the right instructions to access this data
1247 without using (inline) assembler code.
1251 @cindex @code{__flash} AVR Named Address Spaces
1252 The @code{__flash} qualifier locates data in the
1253 @code{.progmem.data} section. Data is read using the @code{LPM}
1254 instruction. Pointers to this address space are 16 bits wide.
1261 @cindex @code{__flash1} AVR Named Address Spaces
1262 @cindex @code{__flash2} AVR Named Address Spaces
1263 @cindex @code{__flash3} AVR Named Address Spaces
1264 @cindex @code{__flash4} AVR Named Address Spaces
1265 @cindex @code{__flash5} AVR Named Address Spaces
1266 These are 16-bit address spaces locating data in section
1267 @code{.progmem@var{N}.data} where @var{N} refers to
1268 address space @code{__flash@var{N}}.
1269 The compiler sets the @code{RAMPZ} segment register appropriately
1270 before reading data by means of the @code{ELPM} instruction.
1273 @cindex @code{__memx} AVR Named Address Spaces
1274 This is a 24-bit address space that linearizes flash and RAM:
1275 If the high bit of the address is set, data is read from
1276 RAM using the lower two bytes as RAM address.
1277 If the high bit of the address is clear, data is read from flash
1278 with @code{RAMPZ} set according to the high byte of the address.
1279 @xref{AVR Built-in Functions,,@code{__builtin_avr_flash_segment}}.
1281 Objects in this address space are located in @code{.progmemx.data}.
1287 char my_read (const __flash char ** p)
1289 /* p is a pointer to RAM that points to a pointer to flash.
1290 The first indirection of p reads that flash pointer
1291 from RAM and the second indirection reads a char from this
1297 /* Locate array[] in flash memory */
1298 const __flash int array[] = @{ 3, 5, 7, 11, 13, 17, 19 @};
1304 /* Return 17 by reading from flash memory */
1305 return array[array[i]];
1310 For each named address space supported by avr-gcc there is an equally
1311 named but uppercase built-in macro defined.
1312 The purpose is to facilitate testing if respective address space
1313 support is available or not:
1317 const __flash int var = 1;
1324 #include <avr/pgmspace.h> /* From AVR-LibC */
1326 const int var PROGMEM = 1;
1330 return (int) pgm_read_word (&var);
1332 #endif /* __FLASH */
1336 Notice that attribute @ref{AVR Variable Attributes,,@code{progmem}}
1337 locates data in flash but
1338 accesses to these data read from generic address space, i.e.@:
1340 so that you need special accessors like @code{pgm_read_byte}
1341 from @w{@uref{http://nongnu.org/avr-libc/user-manual,AVR-LibC}}
1342 together with attribute @code{progmem}.
1345 @b{Limitations and caveats}
1349 Reading across the 64@tie{}KiB section boundary of
1350 the @code{__flash} or @code{__flash@var{N}} address spaces
1351 shows undefined behavior. The only address space that
1352 supports reading across the 64@tie{}KiB flash segment boundaries is
1356 If you use one of the @code{__flash@var{N}} address spaces
1357 you must arrange your linker script to locate the
1358 @code{.progmem@var{N}.data} sections according to your needs.
1361 Any data or pointers to the non-generic address spaces must
1362 be qualified as @code{const}, i.e.@: as read-only data.
1363 This still applies if the data in one of these address
1364 spaces like software version number or calibration lookup table are intended to
1365 be changed after load time by, say, a boot loader. In this case
1366 the right qualification is @code{const} @code{volatile} so that the compiler
1367 must not optimize away known values or insert them
1368 as immediates into operands of instructions.
1371 The following code initializes a variable @code{pfoo}
1372 located in static storage with a 24-bit address:
1374 extern const __memx char foo;
1375 const __memx void *pfoo = &foo;
1379 Such code requires at least binutils 2.23, see
1380 @w{@uref{http://sourceware.org/PR13503,PR13503}}.
1384 @subsection M32C Named Address Spaces
1385 @cindex @code{__far} M32C Named Address Spaces
1387 On the M32C target, with the R8C and M16C CPU variants, variables
1388 qualified with @code{__far} are accessed using 32-bit addresses in
1389 order to access memory beyond the first 64@tie{}Ki bytes. If
1390 @code{__far} is used with the M32CM or M32C CPU variants, it has no
1393 @subsection RL78 Named Address Spaces
1394 @cindex @code{__far} RL78 Named Address Spaces
1396 On the RL78 target, variables qualified with @code{__far} are accessed
1397 with 32-bit pointers (20-bit addresses) rather than the default 16-bit
1398 addresses. Non-far variables are assumed to appear in the topmost
1399 64@tie{}KiB of the address space.
1401 @subsection SPU Named Address Spaces
1402 @cindex @code{__ea} SPU Named Address Spaces
1404 On the SPU target variables may be declared as
1405 belonging to another address space by qualifying the type with the
1406 @code{__ea} address space identifier:
1413 The compiler generates special code to access the variable @code{i}.
1414 It may use runtime library
1415 support, or generate special machine instructions to access that address
1419 @section Arrays of Length Zero
1420 @cindex arrays of length zero
1421 @cindex zero-length arrays
1422 @cindex length-zero arrays
1423 @cindex flexible array members
1425 Zero-length arrays are allowed in GNU C@. They are very useful as the
1426 last element of a structure that is really a header for a variable-length
1435 struct line *thisline = (struct line *)
1436 malloc (sizeof (struct line) + this_length);
1437 thisline->length = this_length;
1440 In ISO C90, you would have to give @code{contents} a length of 1, which
1441 means either you waste space or complicate the argument to @code{malloc}.
1443 In ISO C99, you would use a @dfn{flexible array member}, which is
1444 slightly different in syntax and semantics:
1448 Flexible array members are written as @code{contents[]} without
1452 Flexible array members have incomplete type, and so the @code{sizeof}
1453 operator may not be applied. As a quirk of the original implementation
1454 of zero-length arrays, @code{sizeof} evaluates to zero.
1457 Flexible array members may only appear as the last member of a
1458 @code{struct} that is otherwise non-empty.
1461 A structure containing a flexible array member, or a union containing
1462 such a structure (possibly recursively), may not be a member of a
1463 structure or an element of an array. (However, these uses are
1464 permitted by GCC as extensions.)
1467 GCC versions before 3.0 allowed zero-length arrays to be statically
1468 initialized, as if they were flexible arrays. In addition to those
1469 cases that were useful, it also allowed initializations in situations
1470 that would corrupt later data. Non-empty initialization of zero-length
1471 arrays is now treated like any case where there are more initializer
1472 elements than the array holds, in that a suitable warning about ``excess
1473 elements in array'' is given, and the excess elements (all of them, in
1474 this case) are ignored.
1476 Instead GCC allows static initialization of flexible array members.
1477 This is equivalent to defining a new structure containing the original
1478 structure followed by an array of sufficient size to contain the data.
1479 E.g.@: in the following, @code{f1} is constructed as if it were declared
1485 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1488 struct f1 f1; int data[3];
1489 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1493 The convenience of this extension is that @code{f1} has the desired
1494 type, eliminating the need to consistently refer to @code{f2.f1}.
1496 This has symmetry with normal static arrays, in that an array of
1497 unknown size is also written with @code{[]}.
1499 Of course, this extension only makes sense if the extra data comes at
1500 the end of a top-level object, as otherwise we would be overwriting
1501 data at subsequent offsets. To avoid undue complication and confusion
1502 with initialization of deeply nested arrays, we simply disallow any
1503 non-empty initialization except when the structure is the top-level
1504 object. For example:
1507 struct foo @{ int x; int y[]; @};
1508 struct bar @{ struct foo z; @};
1510 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1511 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1512 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1513 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1516 @node Empty Structures
1517 @section Structures With No Members
1518 @cindex empty structures
1519 @cindex zero-size structures
1521 GCC permits a C structure to have no members:
1528 The structure has size zero. In C++, empty structures are part
1529 of the language. G++ treats empty structures as if they had a single
1530 member of type @code{char}.
1532 @node Variable Length
1533 @section Arrays of Variable Length
1534 @cindex variable-length arrays
1535 @cindex arrays of variable length
1538 Variable-length automatic arrays are allowed in ISO C99, and as an
1539 extension GCC accepts them in C90 mode and in C++. These arrays are
1540 declared like any other automatic arrays, but with a length that is not
1541 a constant expression. The storage is allocated at the point of
1542 declaration and deallocated when the block scope containing the declaration
1548 concat_fopen (char *s1, char *s2, char *mode)
1550 char str[strlen (s1) + strlen (s2) + 1];
1553 return fopen (str, mode);
1557 @cindex scope of a variable length array
1558 @cindex variable-length array scope
1559 @cindex deallocating variable length arrays
1560 Jumping or breaking out of the scope of the array name deallocates the
1561 storage. Jumping into the scope is not allowed; you get an error
1564 @cindex @code{alloca} vs variable-length arrays
1565 You can use the function @code{alloca} to get an effect much like
1566 variable-length arrays. The function @code{alloca} is available in
1567 many other C implementations (but not in all). On the other hand,
1568 variable-length arrays are more elegant.
1570 There are other differences between these two methods. Space allocated
1571 with @code{alloca} exists until the containing @emph{function} returns.
1572 The space for a variable-length array is deallocated as soon as the array
1573 name's scope ends. (If you use both variable-length arrays and
1574 @code{alloca} in the same function, deallocation of a variable-length array
1575 also deallocates anything more recently allocated with @code{alloca}.)
1577 You can also use variable-length arrays as arguments to functions:
1581 tester (int len, char data[len][len])
1587 The length of an array is computed once when the storage is allocated
1588 and is remembered for the scope of the array in case you access it with
1591 If you want to pass the array first and the length afterward, you can
1592 use a forward declaration in the parameter list---another GNU extension.
1596 tester (int len; char data[len][len], int len)
1602 @cindex parameter forward declaration
1603 The @samp{int len} before the semicolon is a @dfn{parameter forward
1604 declaration}, and it serves the purpose of making the name @code{len}
1605 known when the declaration of @code{data} is parsed.
1607 You can write any number of such parameter forward declarations in the
1608 parameter list. They can be separated by commas or semicolons, but the
1609 last one must end with a semicolon, which is followed by the ``real''
1610 parameter declarations. Each forward declaration must match a ``real''
1611 declaration in parameter name and data type. ISO C99 does not support
1612 parameter forward declarations.
1614 @node Variadic Macros
1615 @section Macros with a Variable Number of Arguments.
1616 @cindex variable number of arguments
1617 @cindex macro with variable arguments
1618 @cindex rest argument (in macro)
1619 @cindex variadic macros
1621 In the ISO C standard of 1999, a macro can be declared to accept a
1622 variable number of arguments much as a function can. The syntax for
1623 defining the macro is similar to that of a function. Here is an
1627 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1631 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1632 such a macro, it represents the zero or more tokens until the closing
1633 parenthesis that ends the invocation, including any commas. This set of
1634 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1635 wherever it appears. See the CPP manual for more information.
1637 GCC has long supported variadic macros, and used a different syntax that
1638 allowed you to give a name to the variable arguments just like any other
1639 argument. Here is an example:
1642 #define debug(format, args...) fprintf (stderr, format, args)
1646 This is in all ways equivalent to the ISO C example above, but arguably
1647 more readable and descriptive.
1649 GNU CPP has two further variadic macro extensions, and permits them to
1650 be used with either of the above forms of macro definition.
1652 In standard C, you are not allowed to leave the variable argument out
1653 entirely; but you are allowed to pass an empty argument. For example,
1654 this invocation is invalid in ISO C, because there is no comma after
1661 GNU CPP permits you to completely omit the variable arguments in this
1662 way. In the above examples, the compiler would complain, though since
1663 the expansion of the macro still has the extra comma after the format
1666 To help solve this problem, CPP behaves specially for variable arguments
1667 used with the token paste operator, @samp{##}. If instead you write
1670 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1674 and if the variable arguments are omitted or empty, the @samp{##}
1675 operator causes the preprocessor to remove the comma before it. If you
1676 do provide some variable arguments in your macro invocation, GNU CPP
1677 does not complain about the paste operation and instead places the
1678 variable arguments after the comma. Just like any other pasted macro
1679 argument, these arguments are not macro expanded.
1681 @node Escaped Newlines
1682 @section Slightly Looser Rules for Escaped Newlines
1683 @cindex escaped newlines
1684 @cindex newlines (escaped)
1686 Recently, the preprocessor has relaxed its treatment of escaped
1687 newlines. Previously, the newline had to immediately follow a
1688 backslash. The current implementation allows whitespace in the form
1689 of spaces, horizontal and vertical tabs, and form feeds between the
1690 backslash and the subsequent newline. The preprocessor issues a
1691 warning, but treats it as a valid escaped newline and combines the two
1692 lines to form a single logical line. This works within comments and
1693 tokens, as well as between tokens. Comments are @emph{not} treated as
1694 whitespace for the purposes of this relaxation, since they have not
1695 yet been replaced with spaces.
1698 @section Non-Lvalue Arrays May Have Subscripts
1699 @cindex subscripting
1700 @cindex arrays, non-lvalue
1702 @cindex subscripting and function values
1703 In ISO C99, arrays that are not lvalues still decay to pointers, and
1704 may be subscripted, although they may not be modified or used after
1705 the next sequence point and the unary @samp{&} operator may not be
1706 applied to them. As an extension, GNU C allows such arrays to be
1707 subscripted in C90 mode, though otherwise they do not decay to
1708 pointers outside C99 mode. For example,
1709 this is valid in GNU C though not valid in C90:
1713 struct foo @{int a[4];@};
1719 return f().a[index];
1725 @section Arithmetic on @code{void}- and Function-Pointers
1726 @cindex void pointers, arithmetic
1727 @cindex void, size of pointer to
1728 @cindex function pointers, arithmetic
1729 @cindex function, size of pointer to
1731 In GNU C, addition and subtraction operations are supported on pointers to
1732 @code{void} and on pointers to functions. This is done by treating the
1733 size of a @code{void} or of a function as 1.
1735 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1736 and on function types, and returns 1.
1738 @opindex Wpointer-arith
1739 The option @option{-Wpointer-arith} requests a warning if these extensions
1743 @section Non-Constant Initializers
1744 @cindex initializers, non-constant
1745 @cindex non-constant initializers
1747 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1748 automatic variable are not required to be constant expressions in GNU C@.
1749 Here is an example of an initializer with run-time varying elements:
1752 foo (float f, float g)
1754 float beat_freqs[2] = @{ f-g, f+g @};
1759 @node Compound Literals
1760 @section Compound Literals
1761 @cindex constructor expressions
1762 @cindex initializations in expressions
1763 @cindex structures, constructor expression
1764 @cindex expressions, constructor
1765 @cindex compound literals
1766 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1768 ISO C99 supports compound literals. A compound literal looks like
1769 a cast containing an initializer. Its value is an object of the
1770 type specified in the cast, containing the elements specified in
1771 the initializer; it is an lvalue. As an extension, GCC supports
1772 compound literals in C90 mode and in C++, though the semantics are
1773 somewhat different in C++.
1775 Usually, the specified type is a structure. Assume that
1776 @code{struct foo} and @code{structure} are declared as shown:
1779 struct foo @{int a; char b[2];@} structure;
1783 Here is an example of constructing a @code{struct foo} with a compound literal:
1786 structure = ((struct foo) @{x + y, 'a', 0@});
1790 This is equivalent to writing the following:
1794 struct foo temp = @{x + y, 'a', 0@};
1799 You can also construct an array, though this is dangerous in C++, as
1800 explained below. If all the elements of the compound literal are
1801 (made up of) simple constant expressions, suitable for use in
1802 initializers of objects of static storage duration, then the compound
1803 literal can be coerced to a pointer to its first element and used in
1804 such an initializer, as shown here:
1807 char **foo = (char *[]) @{ "x", "y", "z" @};
1810 Compound literals for scalar types and union types are
1811 also allowed, but then the compound literal is equivalent
1814 As a GNU extension, GCC allows initialization of objects with static storage
1815 duration by compound literals (which is not possible in ISO C99, because
1816 the initializer is not a constant).
1817 It is handled as if the object is initialized only with the bracket
1818 enclosed list if the types of the compound literal and the object match.
1819 The initializer list of the compound literal must be constant.
1820 If the object being initialized has array type of unknown size, the size is
1821 determined by compound literal size.
1824 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1825 static int y[] = (int []) @{1, 2, 3@};
1826 static int z[] = (int [3]) @{1@};
1830 The above lines are equivalent to the following:
1832 static struct foo x = @{1, 'a', 'b'@};
1833 static int y[] = @{1, 2, 3@};
1834 static int z[] = @{1, 0, 0@};
1837 In C, a compound literal designates an unnamed object with static or
1838 automatic storage duration. In C++, a compound literal designates a
1839 temporary object, which only lives until the end of its
1840 full-expression. As a result, well-defined C code that takes the
1841 address of a subobject of a compound literal can be undefined in C++.
1842 For instance, if the array compound literal example above appeared
1843 inside a function, any subsequent use of @samp{foo} in C++ has
1844 undefined behavior because the lifetime of the array ends after the
1845 declaration of @samp{foo}. As a result, the C++ compiler now rejects
1846 the conversion of a temporary array to a pointer.
1848 As an optimization, the C++ compiler sometimes gives array compound
1849 literals longer lifetimes: when the array either appears outside a
1850 function or has const-qualified type. If @samp{foo} and its
1851 initializer had elements of @samp{char *const} type rather than
1852 @samp{char *}, or if @samp{foo} were a global variable, the array
1853 would have static storage duration. But it is probably safest just to
1854 avoid the use of array compound literals in code compiled as C++.
1856 @node Designated Inits
1857 @section Designated Initializers
1858 @cindex initializers with labeled elements
1859 @cindex labeled elements in initializers
1860 @cindex case labels in initializers
1861 @cindex designated initializers
1863 Standard C90 requires the elements of an initializer to appear in a fixed
1864 order, the same as the order of the elements in the array or structure
1867 In ISO C99 you can give the elements in any order, specifying the array
1868 indices or structure field names they apply to, and GNU C allows this as
1869 an extension in C90 mode as well. This extension is not
1870 implemented in GNU C++.
1872 To specify an array index, write
1873 @samp{[@var{index}] =} before the element value. For example,
1876 int a[6] = @{ [4] = 29, [2] = 15 @};
1883 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1887 The index values must be constant expressions, even if the array being
1888 initialized is automatic.
1890 An alternative syntax for this that has been obsolete since GCC 2.5 but
1891 GCC still accepts is to write @samp{[@var{index}]} before the element
1892 value, with no @samp{=}.
1894 To initialize a range of elements to the same value, write
1895 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1896 extension. For example,
1899 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1903 If the value in it has side-effects, the side-effects happen only once,
1904 not for each initialized field by the range initializer.
1907 Note that the length of the array is the highest value specified
1910 In a structure initializer, specify the name of a field to initialize
1911 with @samp{.@var{fieldname} =} before the element value. For example,
1912 given the following structure,
1915 struct point @{ int x, y; @};
1919 the following initialization
1922 struct point p = @{ .y = yvalue, .x = xvalue @};
1929 struct point p = @{ xvalue, yvalue @};
1932 Another syntax that has the same meaning, obsolete since GCC 2.5, is
1933 @samp{@var{fieldname}:}, as shown here:
1936 struct point p = @{ y: yvalue, x: xvalue @};
1940 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1941 @dfn{designator}. You can also use a designator (or the obsolete colon
1942 syntax) when initializing a union, to specify which element of the union
1943 should be used. For example,
1946 union foo @{ int i; double d; @};
1948 union foo f = @{ .d = 4 @};
1952 converts 4 to a @code{double} to store it in the union using
1953 the second element. By contrast, casting 4 to type @code{union foo}
1954 stores it into the union as the integer @code{i}, since it is
1955 an integer. (@xref{Cast to Union}.)
1957 You can combine this technique of naming elements with ordinary C
1958 initialization of successive elements. Each initializer element that
1959 does not have a designator applies to the next consecutive element of the
1960 array or structure. For example,
1963 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1970 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1973 Labeling the elements of an array initializer is especially useful
1974 when the indices are characters or belong to an @code{enum} type.
1979 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1980 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1983 @cindex designator lists
1984 You can also write a series of @samp{.@var{fieldname}} and
1985 @samp{[@var{index}]} designators before an @samp{=} to specify a
1986 nested subobject to initialize; the list is taken relative to the
1987 subobject corresponding to the closest surrounding brace pair. For
1988 example, with the @samp{struct point} declaration above:
1991 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1995 If the same field is initialized multiple times, it has the value from
1996 the last initialization. If any such overridden initialization has
1997 side-effect, it is unspecified whether the side-effect happens or not.
1998 Currently, GCC discards them and issues a warning.
2001 @section Case Ranges
2003 @cindex ranges in case statements
2005 You can specify a range of consecutive values in a single @code{case} label,
2009 case @var{low} ... @var{high}:
2013 This has the same effect as the proper number of individual @code{case}
2014 labels, one for each integer value from @var{low} to @var{high}, inclusive.
2016 This feature is especially useful for ranges of ASCII character codes:
2022 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
2023 it may be parsed wrong when you use it with integer values. For example,
2038 @section Cast to a Union Type
2039 @cindex cast to a union
2040 @cindex union, casting to a
2042 A cast to union type is similar to other casts, except that the type
2043 specified is a union type. You can specify the type either with
2044 @code{union @var{tag}} or with a typedef name. A cast to union is actually
2045 a constructor, not a cast, and hence does not yield an lvalue like
2046 normal casts. (@xref{Compound Literals}.)
2048 The types that may be cast to the union type are those of the members
2049 of the union. Thus, given the following union and variables:
2052 union foo @{ int i; double d; @};
2058 both @code{x} and @code{y} can be cast to type @code{union foo}.
2060 Using the cast as the right-hand side of an assignment to a variable of
2061 union type is equivalent to storing in a member of the union:
2066 u = (union foo) x @equiv{} u.i = x
2067 u = (union foo) y @equiv{} u.d = y
2070 You can also use the union cast as a function argument:
2073 void hack (union foo);
2075 hack ((union foo) x);
2078 @node Mixed Declarations
2079 @section Mixed Declarations and Code
2080 @cindex mixed declarations and code
2081 @cindex declarations, mixed with code
2082 @cindex code, mixed with declarations
2084 ISO C99 and ISO C++ allow declarations and code to be freely mixed
2085 within compound statements. As an extension, GNU C also allows this in
2086 C90 mode. For example, you could do:
2095 Each identifier is visible from where it is declared until the end of
2096 the enclosing block.
2098 @node Function Attributes
2099 @section Declaring Attributes of Functions
2100 @cindex function attributes
2101 @cindex declaring attributes of functions
2102 @cindex functions that never return
2103 @cindex functions that return more than once
2104 @cindex functions that have no side effects
2105 @cindex functions in arbitrary sections
2106 @cindex functions that behave like malloc
2107 @cindex @code{volatile} applied to function
2108 @cindex @code{const} applied to function
2109 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
2110 @cindex functions with non-null pointer arguments
2111 @cindex functions that are passed arguments in registers on the 386
2112 @cindex functions that pop the argument stack on the 386
2113 @cindex functions that do not pop the argument stack on the 386
2114 @cindex functions that have different compilation options on the 386
2115 @cindex functions that have different optimization options
2116 @cindex functions that are dynamically resolved
2118 In GNU C, you declare certain things about functions called in your program
2119 which help the compiler optimize function calls and check your code more
2122 The keyword @code{__attribute__} allows you to specify special
2123 attributes when making a declaration. This keyword is followed by an
2124 attribute specification inside double parentheses. The following
2125 attributes are currently defined for functions on all targets:
2126 @code{aligned}, @code{alloc_size}, @code{noreturn},
2127 @code{returns_twice}, @code{noinline}, @code{noclone},
2128 @code{always_inline}, @code{flatten}, @code{pure}, @code{const},
2129 @code{nothrow}, @code{sentinel}, @code{format}, @code{format_arg},
2130 @code{no_instrument_function}, @code{no_split_stack},
2131 @code{section}, @code{constructor},
2132 @code{destructor}, @code{used}, @code{unused}, @code{deprecated},
2133 @code{weak}, @code{malloc}, @code{alias}, @code{ifunc},
2134 @code{warn_unused_result}, @code{nonnull}, @code{gnu_inline},
2135 @code{externally_visible}, @code{hot}, @code{cold}, @code{artificial},
2136 @code{no_address_safety_analysis}, @code{error} and @code{warning}.
2137 Several other attributes are defined for functions on particular
2138 target systems. Other attributes, including @code{section} are
2139 supported for variables declarations (@pxref{Variable Attributes})
2140 and for types (@pxref{Type Attributes}).
2142 GCC plugins may provide their own attributes.
2144 You may also specify attributes with @samp{__} preceding and following
2145 each keyword. This allows you to use them in header files without
2146 being concerned about a possible macro of the same name. For example,
2147 you may use @code{__noreturn__} instead of @code{noreturn}.
2149 @xref{Attribute Syntax}, for details of the exact syntax for using
2153 @c Keep this table alphabetized by attribute name. Treat _ as space.
2155 @item alias ("@var{target}")
2156 @cindex @code{alias} attribute
2157 The @code{alias} attribute causes the declaration to be emitted as an
2158 alias for another symbol, which must be specified. For instance,
2161 void __f () @{ /* @r{Do something.} */; @}
2162 void f () __attribute__ ((weak, alias ("__f")));
2166 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
2167 mangled name for the target must be used. It is an error if @samp{__f}
2168 is not defined in the same translation unit.
2170 Not all target machines support this attribute.
2172 @item aligned (@var{alignment})
2173 @cindex @code{aligned} attribute
2174 This attribute specifies a minimum alignment for the function,
2177 You cannot use this attribute to decrease the alignment of a function,
2178 only to increase it. However, when you explicitly specify a function
2179 alignment this overrides the effect of the
2180 @option{-falign-functions} (@pxref{Optimize Options}) option for this
2183 Note that the effectiveness of @code{aligned} attributes may be
2184 limited by inherent limitations in your linker. On many systems, the
2185 linker is only able to arrange for functions to be aligned up to a
2186 certain maximum alignment. (For some linkers, the maximum supported
2187 alignment may be very very small.) See your linker documentation for
2188 further information.
2190 The @code{aligned} attribute can also be used for variables and fields
2191 (@pxref{Variable Attributes}.)
2194 @cindex @code{alloc_size} attribute
2195 The @code{alloc_size} attribute is used to tell the compiler that the
2196 function return value points to memory, where the size is given by
2197 one or two of the functions parameters. GCC uses this
2198 information to improve the correctness of @code{__builtin_object_size}.
2200 The function parameter(s) denoting the allocated size are specified by
2201 one or two integer arguments supplied to the attribute. The allocated size
2202 is either the value of the single function argument specified or the product
2203 of the two function arguments specified. Argument numbering starts at
2209 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
2210 void my_realloc(void*, size_t) __attribute__((alloc_size(2)))
2214 declares that @code{my_calloc} returns memory of the size given by
2215 the product of parameter 1 and 2 and that @code{my_realloc} returns memory
2216 of the size given by parameter 2.
2219 @cindex @code{always_inline} function attribute
2220 Generally, functions are not inlined unless optimization is specified.
2221 For functions declared inline, this attribute inlines the function even
2222 if no optimization level is specified.
2225 @cindex @code{gnu_inline} function attribute
2226 This attribute should be used with a function that is also declared
2227 with the @code{inline} keyword. It directs GCC to treat the function
2228 as if it were defined in gnu90 mode even when compiling in C99 or
2231 If the function is declared @code{extern}, then this definition of the
2232 function is used only for inlining. In no case is the function
2233 compiled as a standalone function, not even if you take its address
2234 explicitly. Such an address becomes an external reference, as if you
2235 had only declared the function, and had not defined it. This has
2236 almost the effect of a macro. The way to use this is to put a
2237 function definition in a header file with this attribute, and put
2238 another copy of the function, without @code{extern}, in a library
2239 file. The definition in the header file causes most calls to the
2240 function to be inlined. If any uses of the function remain, they
2241 refer to the single copy in the library. Note that the two
2242 definitions of the functions need not be precisely the same, although
2243 if they do not have the same effect your program may behave oddly.
2245 In C, if the function is neither @code{extern} nor @code{static}, then
2246 the function is compiled as a standalone function, as well as being
2247 inlined where possible.
2249 This is how GCC traditionally handled functions declared
2250 @code{inline}. Since ISO C99 specifies a different semantics for
2251 @code{inline}, this function attribute is provided as a transition
2252 measure and as a useful feature in its own right. This attribute is
2253 available in GCC 4.1.3 and later. It is available if either of the
2254 preprocessor macros @code{__GNUC_GNU_INLINE__} or
2255 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
2256 Function is As Fast As a Macro}.
2258 In C++, this attribute does not depend on @code{extern} in any way,
2259 but it still requires the @code{inline} keyword to enable its special
2263 @cindex @code{artificial} function attribute
2264 This attribute is useful for small inline wrappers that if possible
2265 should appear during debugging as a unit. Depending on the debug
2266 info format it either means marking the function as artificial
2267 or using the caller location for all instructions within the inlined
2271 @cindex interrupt handler functions
2272 When added to an interrupt handler with the M32C port, causes the
2273 prologue and epilogue to use bank switching to preserve the registers
2274 rather than saving them on the stack.
2277 @cindex @code{flatten} function attribute
2278 Generally, inlining into a function is limited. For a function marked with
2279 this attribute, every call inside this function is inlined, if possible.
2280 Whether the function itself is considered for inlining depends on its size and
2281 the current inlining parameters.
2283 @item error ("@var{message}")
2284 @cindex @code{error} function attribute
2285 If this attribute is used on a function declaration and a call to such a function
2286 is not eliminated through dead code elimination or other optimizations, an error
2287 that includes @var{message} is diagnosed. This is useful
2288 for compile-time checking, especially together with @code{__builtin_constant_p}
2289 and inline functions where checking the inline function arguments is not
2290 possible through @code{extern char [(condition) ? 1 : -1];} tricks.
2291 While it is possible to leave the function undefined and thus invoke
2292 a link failure, when using this attribute the problem is diagnosed
2293 earlier and with exact location of the call even in presence of inline
2294 functions or when not emitting debugging information.
2296 @item warning ("@var{message}")
2297 @cindex @code{warning} function attribute
2298 If this attribute is used on a function declaration and a call to such a function
2299 is not eliminated through dead code elimination or other optimizations, a warning
2300 that includes @var{message} is diagnosed. This is useful
2301 for compile-time checking, especially together with @code{__builtin_constant_p}
2302 and inline functions. While it is possible to define the function with
2303 a message in @code{.gnu.warning*} section, when using this attribute the problem
2304 is diagnosed earlier and with exact location of the call even in presence
2305 of inline functions or when not emitting debugging information.
2308 @cindex functions that do pop the argument stack on the 386
2310 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2311 assume that the calling function pops off the stack space used to
2312 pass arguments. This is
2313 useful to override the effects of the @option{-mrtd} switch.
2316 @cindex @code{const} function attribute
2317 Many functions do not examine any values except their arguments, and
2318 have no effects except the return value. Basically this is just slightly
2319 more strict class than the @code{pure} attribute below, since function is not
2320 allowed to read global memory.
2322 @cindex pointer arguments
2323 Note that a function that has pointer arguments and examines the data
2324 pointed to must @emph{not} be declared @code{const}. Likewise, a
2325 function that calls a non-@code{const} function usually must not be
2326 @code{const}. It does not make sense for a @code{const} function to
2329 The attribute @code{const} is not implemented in GCC versions earlier
2330 than 2.5. An alternative way to declare that a function has no side
2331 effects, which works in the current version and in some older versions,
2335 typedef int intfn ();
2337 extern const intfn square;
2341 This approach does not work in GNU C++ from 2.6.0 on, since the language
2342 specifies that the @samp{const} must be attached to the return value.
2346 @itemx constructor (@var{priority})
2347 @itemx destructor (@var{priority})
2348 @cindex @code{constructor} function attribute
2349 @cindex @code{destructor} function attribute
2350 The @code{constructor} attribute causes the function to be called
2351 automatically before execution enters @code{main ()}. Similarly, the
2352 @code{destructor} attribute causes the function to be called
2353 automatically after @code{main ()} completes or @code{exit ()} is
2354 called. Functions with these attributes are useful for
2355 initializing data that is used implicitly during the execution of
2358 You may provide an optional integer priority to control the order in
2359 which constructor and destructor functions are run. A constructor
2360 with a smaller priority number runs before a constructor with a larger
2361 priority number; the opposite relationship holds for destructors. So,
2362 if you have a constructor that allocates a resource and a destructor
2363 that deallocates the same resource, both functions typically have the
2364 same priority. The priorities for constructor and destructor
2365 functions are the same as those specified for namespace-scope C++
2366 objects (@pxref{C++ Attributes}).
2368 These attributes are not currently implemented for Objective-C@.
2371 @itemx deprecated (@var{msg})
2372 @cindex @code{deprecated} attribute.
2373 The @code{deprecated} attribute results in a warning if the function
2374 is used anywhere in the source file. This is useful when identifying
2375 functions that are expected to be removed in a future version of a
2376 program. The warning also includes the location of the declaration
2377 of the deprecated function, to enable users to easily find further
2378 information about why the function is deprecated, or what they should
2379 do instead. Note that the warnings only occurs for uses:
2382 int old_fn () __attribute__ ((deprecated));
2384 int (*fn_ptr)() = old_fn;
2388 results in a warning on line 3 but not line 2. The optional @var{msg}
2389 argument, which must be a string, is printed in the warning if
2392 The @code{deprecated} attribute can also be used for variables and
2393 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2396 @cindex @code{disinterrupt} attribute
2397 On Epiphany and MeP targets, this attribute causes the compiler to emit
2398 instructions to disable interrupts for the duration of the given
2402 @cindex @code{__declspec(dllexport)}
2403 On Microsoft Windows targets and Symbian OS targets the
2404 @code{dllexport} attribute causes the compiler to provide a global
2405 pointer to a pointer in a DLL, so that it can be referenced with the
2406 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
2407 name is formed by combining @code{_imp__} and the function or variable
2410 You can use @code{__declspec(dllexport)} as a synonym for
2411 @code{__attribute__ ((dllexport))} for compatibility with other
2414 On systems that support the @code{visibility} attribute, this
2415 attribute also implies ``default'' visibility. It is an error to
2416 explicitly specify any other visibility.
2418 In previous versions of GCC, the @code{dllexport} attribute was ignored
2419 for inlined functions, unless the @option{-fkeep-inline-functions} flag
2420 had been used. The default behavior now is to emit all dllexported
2421 inline functions; however, this can cause object file-size bloat, in
2422 which case the old behavior can be restored by using
2423 @option{-fno-keep-inline-dllexport}.
2425 The attribute is also ignored for undefined symbols.
2427 When applied to C++ classes, the attribute marks defined non-inlined
2428 member functions and static data members as exports. Static consts
2429 initialized in-class are not marked unless they are also defined
2432 For Microsoft Windows targets there are alternative methods for
2433 including the symbol in the DLL's export table such as using a
2434 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
2435 the @option{--export-all} linker flag.
2438 @cindex @code{__declspec(dllimport)}
2439 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
2440 attribute causes the compiler to reference a function or variable via
2441 a global pointer to a pointer that is set up by the DLL exporting the
2442 symbol. The attribute implies @code{extern}. On Microsoft Windows
2443 targets, the pointer name is formed by combining @code{_imp__} and the
2444 function or variable name.
2446 You can use @code{__declspec(dllimport)} as a synonym for
2447 @code{__attribute__ ((dllimport))} for compatibility with other
2450 On systems that support the @code{visibility} attribute, this
2451 attribute also implies ``default'' visibility. It is an error to
2452 explicitly specify any other visibility.
2454 Currently, the attribute is ignored for inlined functions. If the
2455 attribute is applied to a symbol @emph{definition}, an error is reported.
2456 If a symbol previously declared @code{dllimport} is later defined, the
2457 attribute is ignored in subsequent references, and a warning is emitted.
2458 The attribute is also overridden by a subsequent declaration as
2461 When applied to C++ classes, the attribute marks non-inlined
2462 member functions and static data members as imports. However, the
2463 attribute is ignored for virtual methods to allow creation of vtables
2466 On the SH Symbian OS target the @code{dllimport} attribute also has
2467 another affect---it can cause the vtable and run-time type information
2468 for a class to be exported. This happens when the class has a
2469 dllimported constructor or a non-inline, non-pure virtual function
2470 and, for either of those two conditions, the class also has an inline
2471 constructor or destructor and has a key function that is defined in
2472 the current translation unit.
2474 For Microsoft Windows targets the use of the @code{dllimport}
2475 attribute on functions is not necessary, but provides a small
2476 performance benefit by eliminating a thunk in the DLL@. The use of the
2477 @code{dllimport} attribute on imported variables was required on older
2478 versions of the GNU linker, but can now be avoided by passing the
2479 @option{--enable-auto-import} switch to the GNU linker. As with
2480 functions, using the attribute for a variable eliminates a thunk in
2483 One drawback to using this attribute is that a pointer to a
2484 @emph{variable} marked as @code{dllimport} cannot be used as a constant
2485 address. However, a pointer to a @emph{function} with the
2486 @code{dllimport} attribute can be used as a constant initializer; in
2487 this case, the address of a stub function in the import lib is
2488 referenced. On Microsoft Windows targets, the attribute can be disabled
2489 for functions by setting the @option{-mnop-fun-dllimport} flag.
2492 @cindex eight-bit data on the H8/300, H8/300H, and H8S
2493 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2494 variable should be placed into the eight-bit data section.
2495 The compiler generates more efficient code for certain operations
2496 on data in the eight-bit data area. Note the eight-bit data area is limited to
2499 You must use GAS and GLD from GNU binutils version 2.7 or later for
2500 this attribute to work correctly.
2502 @item exception_handler
2503 @cindex exception handler functions on the Blackfin processor
2504 Use this attribute on the Blackfin to indicate that the specified function
2505 is an exception handler. The compiler generates function entry and
2506 exit sequences suitable for use in an exception handler when this
2507 attribute is present.
2509 @item externally_visible
2510 @cindex @code{externally_visible} attribute.
2511 This attribute, attached to a global variable or function, nullifies
2512 the effect of the @option{-fwhole-program} command-line option, so the
2513 object remains visible outside the current compilation unit.
2515 If @option{-fwhole-program} is used together with @option{-flto} and
2516 @command{gold} is used as the linker plugin,
2517 @code{externally_visible} attributes are automatically added to functions
2518 (not variable yet due to a current @command{gold} issue)
2519 that are accessed outside of LTO objects according to resolution file
2520 produced by @command{gold}.
2521 For other linkers that cannot generate resolution file,
2522 explicit @code{externally_visible} attributes are still necessary.
2525 @cindex functions that handle memory bank switching
2526 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
2527 use a calling convention that takes care of switching memory banks when
2528 entering and leaving a function. This calling convention is also the
2529 default when using the @option{-mlong-calls} option.
2531 On 68HC12 the compiler uses the @code{call} and @code{rtc} instructions
2532 to call and return from a function.
2534 On 68HC11 the compiler generates a sequence of instructions
2535 to invoke a board-specific routine to switch the memory bank and call the
2536 real function. The board-specific routine simulates a @code{call}.
2537 At the end of a function, it jumps to a board-specific routine
2538 instead of using @code{rts}. The board-specific return routine simulates
2541 On MeP targets this causes the compiler to use a calling convention
2542 that assumes the called function is too far away for the built-in
2545 @item fast_interrupt
2546 @cindex interrupt handler functions
2547 Use this attribute on the M32C and RX ports to indicate that the specified
2548 function is a fast interrupt handler. This is just like the
2549 @code{interrupt} attribute, except that @code{freit} is used to return
2550 instead of @code{reit}.
2553 @cindex functions that pop the argument stack on the 386
2554 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2555 pass the first argument (if of integral type) in the register ECX and
2556 the second argument (if of integral type) in the register EDX@. Subsequent
2557 and other typed arguments are passed on the stack. The called function
2558 pops the arguments off the stack. If the number of arguments is variable all
2559 arguments are pushed on the stack.
2562 @cindex functions that pop the argument stack on the 386
2563 On the Intel 386, the @code{thiscall} attribute causes the compiler to
2564 pass the first argument (if of integral type) in the register ECX.
2565 Subsequent and other typed arguments are passed on the stack. The called
2566 function pops the arguments off the stack.
2567 If the number of arguments is variable all arguments are pushed on the
2569 The @code{thiscall} attribute is intended for C++ non-static member functions.
2570 As a GCC extension, this calling convention can be used for C functions
2571 and for static member methods.
2573 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2574 @cindex @code{format} function attribute
2576 The @code{format} attribute specifies that a function takes @code{printf},
2577 @code{scanf}, @code{strftime} or @code{strfmon} style arguments that
2578 should be type-checked against a format string. For example, the
2583 my_printf (void *my_object, const char *my_format, ...)
2584 __attribute__ ((format (printf, 2, 3)));
2588 causes the compiler to check the arguments in calls to @code{my_printf}
2589 for consistency with the @code{printf} style format string argument
2592 The parameter @var{archetype} determines how the format string is
2593 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime},
2594 @code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or
2595 @code{strfmon}. (You can also use @code{__printf__},
2596 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On
2597 MinGW targets, @code{ms_printf}, @code{ms_scanf}, and
2598 @code{ms_strftime} are also present.
2599 @var{archetype} values such as @code{printf} refer to the formats accepted
2600 by the system's C runtime library,
2601 while values prefixed with @samp{gnu_} always refer
2602 to the formats accepted by the GNU C Library. On Microsoft Windows
2603 targets, values prefixed with @samp{ms_} refer to the formats accepted by the
2604 @file{msvcrt.dll} library.
2605 The parameter @var{string-index}
2606 specifies which argument is the format string argument (starting
2607 from 1), while @var{first-to-check} is the number of the first
2608 argument to check against the format string. For functions
2609 where the arguments are not available to be checked (such as
2610 @code{vprintf}), specify the third parameter as zero. In this case the
2611 compiler only checks the format string for consistency. For
2612 @code{strftime} formats, the third parameter is required to be zero.
2613 Since non-static C++ methods have an implicit @code{this} argument, the
2614 arguments of such methods should be counted from two, not one, when
2615 giving values for @var{string-index} and @var{first-to-check}.
2617 In the example above, the format string (@code{my_format}) is the second
2618 argument of the function @code{my_print}, and the arguments to check
2619 start with the third argument, so the correct parameters for the format
2620 attribute are 2 and 3.
2622 @opindex ffreestanding
2623 @opindex fno-builtin
2624 The @code{format} attribute allows you to identify your own functions
2625 that take format strings as arguments, so that GCC can check the
2626 calls to these functions for errors. The compiler always (unless
2627 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2628 for the standard library functions @code{printf}, @code{fprintf},
2629 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2630 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2631 warnings are requested (using @option{-Wformat}), so there is no need to
2632 modify the header file @file{stdio.h}. In C99 mode, the functions
2633 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2634 @code{vsscanf} are also checked. Except in strictly conforming C
2635 standard modes, the X/Open function @code{strfmon} is also checked as
2636 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2637 @xref{C Dialect Options,,Options Controlling C Dialect}.
2639 For Objective-C dialects, @code{NSString} (or @code{__NSString__}) is
2640 recognized in the same context. Declarations including these format attributes
2641 are parsed for correct syntax, however the result of checking of such format
2642 strings is not yet defined, and is not carried out by this version of the
2645 The target may also provide additional types of format checks.
2646 @xref{Target Format Checks,,Format Checks Specific to Particular
2649 @item format_arg (@var{string-index})
2650 @cindex @code{format_arg} function attribute
2651 @opindex Wformat-nonliteral
2652 The @code{format_arg} attribute specifies that a function takes a format
2653 string for a @code{printf}, @code{scanf}, @code{strftime} or
2654 @code{strfmon} style function and modifies it (for example, to translate
2655 it into another language), so the result can be passed to a
2656 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2657 function (with the remaining arguments to the format function the same
2658 as they would have been for the unmodified string). For example, the
2663 my_dgettext (char *my_domain, const char *my_format)
2664 __attribute__ ((format_arg (2)));
2668 causes the compiler to check the arguments in calls to a @code{printf},
2669 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2670 format string argument is a call to the @code{my_dgettext} function, for
2671 consistency with the format string argument @code{my_format}. If the
2672 @code{format_arg} attribute had not been specified, all the compiler
2673 could tell in such calls to format functions would be that the format
2674 string argument is not constant; this would generate a warning when
2675 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2676 without the attribute.
2678 The parameter @var{string-index} specifies which argument is the format
2679 string argument (starting from one). Since non-static C++ methods have
2680 an implicit @code{this} argument, the arguments of such methods should
2681 be counted from two.
2683 The @code{format_arg} attribute allows you to identify your own
2684 functions that modify format strings, so that GCC can check the
2685 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2686 type function whose operands are a call to one of your own function.
2687 The compiler always treats @code{gettext}, @code{dgettext}, and
2688 @code{dcgettext} in this manner except when strict ISO C support is
2689 requested by @option{-ansi} or an appropriate @option{-std} option, or
2690 @option{-ffreestanding} or @option{-fno-builtin}
2691 is used. @xref{C Dialect Options,,Options
2692 Controlling C Dialect}.
2694 For Objective-C dialects, the @code{format-arg} attribute may refer to an
2695 @code{NSString} reference for compatibility with the @code{format} attribute
2698 The target may also allow additional types in @code{format-arg} attributes.
2699 @xref{Target Format Checks,,Format Checks Specific to Particular
2702 @item function_vector
2703 @cindex calling functions through the function vector on H8/300, M16C, M32C and SH2A processors
2704 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2705 function should be called through the function vector. Calling a
2706 function through the function vector reduces code size, however;
2707 the function vector has a limited size (maximum 128 entries on the H8/300
2708 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
2710 On SH2A targets, this attribute declares a function to be called using the
2711 TBR relative addressing mode. The argument to this attribute is the entry
2712 number of the same function in a vector table containing all the TBR
2713 relative addressable functions. For correct operation the TBR must be setup
2714 accordingly to point to the start of the vector table before any functions with
2715 this attribute are invoked. Usually a good place to do the initialization is
2716 the startup routine. The TBR relative vector table can have at max 256 function
2717 entries. The jumps to these functions are generated using a SH2A specific,
2718 non delayed branch instruction JSR/N @@(disp8,TBR). You must use GAS and GLD
2719 from GNU binutils version 2.7 or later for this attribute to work correctly.
2721 Please refer the example of M16C target, to see the use of this
2722 attribute while declaring a function,
2724 In an application, for a function being called once, this attribute
2725 saves at least 8 bytes of code; and if other successive calls are being
2726 made to the same function, it saves 2 bytes of code per each of these
2729 On M16C/M32C targets, the @code{function_vector} attribute declares a
2730 special page subroutine call function. Use of this attribute reduces
2731 the code size by 2 bytes for each call generated to the
2732 subroutine. The argument to the attribute is the vector number entry
2733 from the special page vector table which contains the 16 low-order
2734 bits of the subroutine's entry address. Each vector table has special
2735 page number (18 to 255) that is used in @code{jsrs} instructions.
2736 Jump addresses of the routines are generated by adding 0x0F0000 (in
2737 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the
2738 2-byte addresses set in the vector table. Therefore you need to ensure
2739 that all the special page vector routines should get mapped within the
2740 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
2743 In the following example 2 bytes are saved for each call to
2744 function @code{foo}.
2747 void foo (void) __attribute__((function_vector(0x18)));
2758 If functions are defined in one file and are called in another file,
2759 then be sure to write this declaration in both files.
2761 This attribute is ignored for R8C target.
2763 @item ifunc ("@var{resolver}")
2764 @cindex @code{ifunc} attribute
2765 The @code{ifunc} attribute is used to mark a function as an indirect
2766 function using the STT_GNU_IFUNC symbol type extension to the ELF
2767 standard. This allows the resolution of the symbol value to be
2768 determined dynamically at load time, and an optimized version of the
2769 routine can be selected for the particular processor or other system
2770 characteristics determined then. To use this attribute, first define
2771 the implementation functions available, and a resolver function that
2772 returns a pointer to the selected implementation function. The
2773 implementation functions' declarations must match the API of the
2774 function being implemented, the resolver's declaration is be a
2775 function returning pointer to void function returning void:
2778 void *my_memcpy (void *dst, const void *src, size_t len)
2783 static void (*resolve_memcpy (void)) (void)
2785 return my_memcpy; // we'll just always select this routine
2790 The exported header file declaring the function the user calls would
2794 extern void *memcpy (void *, const void *, size_t);
2798 allowing the user to call this as a regular function, unaware of the
2799 implementation. Finally, the indirect function needs to be defined in
2800 the same translation unit as the resolver function:
2803 void *memcpy (void *, const void *, size_t)
2804 __attribute__ ((ifunc ("resolve_memcpy")));
2807 Indirect functions cannot be weak, and require a recent binutils (at
2808 least version 2.20.1), and GNU C library (at least version 2.11.1).
2811 @cindex interrupt handler functions
2812 Use this attribute on the ARM, AVR, CR16, Epiphany, M32C, M32R/D, m68k, MeP, MIPS,
2813 RL78, RX and Xstormy16 ports to indicate that the specified function is an
2814 interrupt handler. The compiler generates function entry and exit
2815 sequences suitable for use in an interrupt handler when this attribute
2816 is present. With Epiphany targets it may also generate a special section with
2817 code to initialize the interrupt vector table.
2819 Note, interrupt handlers for the Blackfin, H8/300, H8/300H, H8S, MicroBlaze,
2820 and SH processors can be specified via the @code{interrupt_handler} attribute.
2822 Note, on the AVR, the hardware globally disables interrupts when an
2823 interrupt is executed. The first instruction of an interrupt handler
2824 declared with this attribute is a @code{SEI} instruction to
2825 re-enable interrupts. See also the @code{signal} function attribute
2826 that does not insert a @code{SEI} instruction. If both @code{signal} and
2827 @code{interrupt} are specified for the same function, @code{signal}
2828 is silently ignored.
2830 Note, for the ARM, you can specify the kind of interrupt to be handled by
2831 adding an optional parameter to the interrupt attribute like this:
2834 void f () __attribute__ ((interrupt ("IRQ")));
2838 Permissible values for this parameter are: @code{IRQ}, @code{FIQ},
2839 @code{SWI}, @code{ABORT} and @code{UNDEF}.
2841 On ARMv7-M the interrupt type is ignored, and the attribute means the function
2842 may be called with a word-aligned stack pointer.
2844 On Epiphany targets one or more optional parameters can be added like this:
2847 void __attribute__ ((interrupt ("dma0, dma1"))) universal_dma_handler ();
2850 Permissible values for these parameters are: @w{@code{reset}},
2851 @w{@code{software_exception}}, @w{@code{page_miss}},
2852 @w{@code{timer0}}, @w{@code{timer1}}, @w{@code{message}},
2853 @w{@code{dma0}}, @w{@code{dma1}}, @w{@code{wand}} and @w{@code{swi}}.
2854 Multiple parameters indicate that multiple entries in the interrupt
2855 vector table should be initialized for this function, i.e.@: for each
2856 parameter @w{@var{name}}, a jump to the function is emitted in
2857 the section @w{ivt_entry_@var{name}}. The parameter(s) may be omitted
2858 entirely, in which case no interrupt vector table entry is provided.
2860 Note, on Epiphany targets, interrupts are enabled inside the function
2861 unless the @code{disinterrupt} attribute is also specified.
2863 On Epiphany targets, you can also use the following attribute to
2864 modify the behavior of an interrupt handler:
2866 @item forwarder_section
2867 @cindex @code{forwarder_section} attribute
2868 The interrupt handler may be in external memory which cannot be
2869 reached by a branch instruction, so generate a local memory trampoline
2870 to transfer control. The single parameter identifies the section where
2871 the trampoline is placed.
2874 The following examples are all valid uses of these attributes on
2877 void __attribute__ ((interrupt)) universal_handler ();
2878 void __attribute__ ((interrupt ("dma1"))) dma1_handler ();
2879 void __attribute__ ((interrupt ("dma0, dma1"))) universal_dma_handler ();
2880 void __attribute__ ((interrupt ("timer0"), disinterrupt))
2881 fast_timer_handler ();
2882 void __attribute__ ((interrupt ("dma0, dma1"), forwarder_section ("tramp")))
2883 external_dma_handler ();
2886 On MIPS targets, you can use the following attributes to modify the behavior
2887 of an interrupt handler:
2889 @item use_shadow_register_set
2890 @cindex @code{use_shadow_register_set} attribute
2891 Assume that the handler uses a shadow register set, instead of
2892 the main general-purpose registers.
2894 @item keep_interrupts_masked
2895 @cindex @code{keep_interrupts_masked} attribute
2896 Keep interrupts masked for the whole function. Without this attribute,
2897 GCC tries to reenable interrupts for as much of the function as it can.
2899 @item use_debug_exception_return
2900 @cindex @code{use_debug_exception_return} attribute
2901 Return using the @code{deret} instruction. Interrupt handlers that don't
2902 have this attribute return using @code{eret} instead.
2905 You can use any combination of these attributes, as shown below:
2907 void __attribute__ ((interrupt)) v0 ();
2908 void __attribute__ ((interrupt, use_shadow_register_set)) v1 ();
2909 void __attribute__ ((interrupt, keep_interrupts_masked)) v2 ();
2910 void __attribute__ ((interrupt, use_debug_exception_return)) v3 ();
2911 void __attribute__ ((interrupt, use_shadow_register_set,
2912 keep_interrupts_masked)) v4 ();
2913 void __attribute__ ((interrupt, use_shadow_register_set,
2914 use_debug_exception_return)) v5 ();
2915 void __attribute__ ((interrupt, keep_interrupts_masked,
2916 use_debug_exception_return)) v6 ();
2917 void __attribute__ ((interrupt, use_shadow_register_set,
2918 keep_interrupts_masked,
2919 use_debug_exception_return)) v7 ();
2922 On RL78, use @code{brk_interrupt} instead of @code{interrupt} for
2923 handlers intended to be used with the @code{BRK} opcode (i.e.@: those
2924 that must end with @code{RETB} instead of @code{RETI}).
2926 @item interrupt_handler
2927 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
2928 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
2929 indicate that the specified function is an interrupt handler. The compiler
2930 generates function entry and exit sequences suitable for use in an
2931 interrupt handler when this attribute is present.
2933 @item interrupt_thread
2934 @cindex interrupt thread functions on fido
2935 Use this attribute on fido, a subarchitecture of the m68k, to indicate
2936 that the specified function is an interrupt handler that is designed
2937 to run as a thread. The compiler omits generate prologue/epilogue
2938 sequences and replaces the return instruction with a @code{sleep}
2939 instruction. This attribute is available only on fido.
2942 @cindex interrupt service routines on ARM
2943 Use this attribute on ARM to write Interrupt Service Routines. This is an
2944 alias to the @code{interrupt} attribute above.
2947 @cindex User stack pointer in interrupts on the Blackfin
2948 When used together with @code{interrupt_handler}, @code{exception_handler}
2949 or @code{nmi_handler}, code is generated to load the stack pointer
2950 from the USP register in the function prologue.
2953 @cindex @code{l1_text} function attribute
2954 This attribute specifies a function to be placed into L1 Instruction
2955 SRAM@. The function is put into a specific section named @code{.l1.text}.
2956 With @option{-mfdpic}, function calls with a such function as the callee
2957 or caller uses inlined PLT.
2960 @cindex @code{l2} function attribute
2961 On the Blackfin, this attribute specifies a function to be placed into L2
2962 SRAM. The function is put into a specific section named
2963 @code{.l1.text}. With @option{-mfdpic}, callers of such functions use
2967 @cindex @code{leaf} function attribute
2968 Calls to external functions with this attribute must return to the current
2969 compilation unit only by return or by exception handling. In particular, leaf
2970 functions are not allowed to call callback function passed to it from the current
2971 compilation unit or directly call functions exported by the unit or longjmp
2972 into the unit. Leaf function might still call functions from other compilation
2973 units and thus they are not necessarily leaf in the sense that they contain no
2974 function calls at all.
2976 The attribute is intended for library functions to improve dataflow analysis.
2977 The compiler takes the hint that any data not escaping the current compilation unit can
2978 not be used or modified by the leaf function. For example, the @code{sin} function
2979 is a leaf function, but @code{qsort} is not.
2981 Note that leaf functions might invoke signals and signal handlers might be
2982 defined in the current compilation unit and use static variables. The only
2983 compliant way to write such a signal handler is to declare such variables
2986 The attribute has no effect on functions defined within the current compilation
2987 unit. This is to allow easy merging of multiple compilation units into one,
2988 for example, by using the link-time optimization. For this reason the
2989 attribute is not allowed on types to annotate indirect calls.
2991 @item long_call/short_call
2992 @cindex indirect calls on ARM
2993 This attribute specifies how a particular function is called on
2994 ARM and Epiphany. Both attributes override the
2995 @option{-mlong-calls} (@pxref{ARM Options})
2996 command-line switch and @code{#pragma long_calls} settings. The
2997 @code{long_call} attribute indicates that the function might be far
2998 away from the call site and require a different (more expensive)
2999 calling sequence. The @code{short_call} attribute always places
3000 the offset to the function from the call site into the @samp{BL}
3001 instruction directly.
3003 @item longcall/shortcall
3004 @cindex functions called via pointer on the RS/6000 and PowerPC
3005 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute
3006 indicates that the function might be far away from the call site and
3007 require a different (more expensive) calling sequence. The
3008 @code{shortcall} attribute indicates that the function is always close
3009 enough for the shorter calling sequence to be used. These attributes
3010 override both the @option{-mlongcall} switch and, on the RS/6000 and
3011 PowerPC, the @code{#pragma longcall} setting.
3013 @xref{RS/6000 and PowerPC Options}, for more information on whether long
3014 calls are necessary.
3016 @item long_call/near/far
3017 @cindex indirect calls on MIPS
3018 These attributes specify how a particular function is called on MIPS@.
3019 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
3020 command-line switch. The @code{long_call} and @code{far} attributes are
3021 synonyms, and cause the compiler to always call
3022 the function by first loading its address into a register, and then using
3023 the contents of that register. The @code{near} attribute has the opposite
3024 effect; it specifies that non-PIC calls should be made using the more
3025 efficient @code{jal} instruction.
3028 @cindex @code{malloc} attribute
3029 The @code{malloc} attribute is used to tell the compiler that a function
3030 may be treated as if any non-@code{NULL} pointer it returns cannot
3031 alias any other pointer valid when the function returns and that the memory
3032 has undefined content.
3033 This often improves optimization.
3034 Standard functions with this property include @code{malloc} and
3035 @code{calloc}. @code{realloc}-like functions do not have this
3036 property as the memory pointed to does not have undefined content.
3038 @item mips16/nomips16
3039 @cindex @code{mips16} attribute
3040 @cindex @code{nomips16} attribute
3042 On MIPS targets, you can use the @code{mips16} and @code{nomips16}
3043 function attributes to locally select or turn off MIPS16 code generation.
3044 A function with the @code{mips16} attribute is emitted as MIPS16 code,
3045 while MIPS16 code generation is disabled for functions with the
3046 @code{nomips16} attribute. These attributes override the
3047 @option{-mips16} and @option{-mno-mips16} options on the command line
3048 (@pxref{MIPS Options}).
3050 When compiling files containing mixed MIPS16 and non-MIPS16 code, the
3051 preprocessor symbol @code{__mips16} reflects the setting on the command line,
3052 not that within individual functions. Mixed MIPS16 and non-MIPS16 code
3053 may interact badly with some GCC extensions such as @code{__builtin_apply}
3054 (@pxref{Constructing Calls}).
3056 @item model (@var{model-name})
3057 @cindex function addressability on the M32R/D
3058 @cindex variable addressability on the IA-64
3060 On the M32R/D, use this attribute to set the addressability of an
3061 object, and of the code generated for a function. The identifier
3062 @var{model-name} is one of @code{small}, @code{medium}, or
3063 @code{large}, representing each of the code models.
3065 Small model objects live in the lower 16MB of memory (so that their
3066 addresses can be loaded with the @code{ld24} instruction), and are
3067 callable with the @code{bl} instruction.
3069 Medium model objects may live anywhere in the 32-bit address space (the
3070 compiler generates @code{seth/add3} instructions to load their addresses),
3071 and are callable with the @code{bl} instruction.
3073 Large model objects may live anywhere in the 32-bit address space (the
3074 compiler generates @code{seth/add3} instructions to load their addresses),
3075 and may not be reachable with the @code{bl} instruction (the compiler
3076 generates the much slower @code{seth/add3/jl} instruction sequence).
3078 On IA-64, use this attribute to set the addressability of an object.
3079 At present, the only supported identifier for @var{model-name} is
3080 @code{small}, indicating addressability via ``small'' (22-bit)
3081 addresses (so that their addresses can be loaded with the @code{addl}
3082 instruction). Caveat: such addressing is by definition not position
3083 independent and hence this attribute must not be used for objects
3084 defined by shared libraries.
3086 @item ms_abi/sysv_abi
3087 @cindex @code{ms_abi} attribute
3088 @cindex @code{sysv_abi} attribute
3090 On 32-bit and 64-bit (i?86|x86_64)-*-* targets, you can use an ABI attribute
3091 to indicate which calling convention should be used for a function. The
3092 @code{ms_abi} attribute tells the compiler to use the Microsoft ABI,
3093 while the @code{sysv_abi} attribute tells the compiler to use the ABI
3094 used on GNU/Linux and other systems. The default is to use the Microsoft ABI
3095 when targeting Windows. On all other systems, the default is the x86/AMD ABI.
3097 Note, the @code{ms_abi} attribute for Microsoft Windows 64-bit targets currently
3098 requires the @option{-maccumulate-outgoing-args} option.
3100 @item callee_pop_aggregate_return (@var{number})
3101 @cindex @code{callee_pop_aggregate_return} attribute
3103 On 32-bit i?86-*-* targets, you can use this attribute to control how
3104 aggregates are returned in memory. If the caller is responsible for
3105 popping the hidden pointer together with the rest of the arguments, specify
3106 @var{number} equal to zero. If callee is responsible for popping the
3107 hidden pointer, specify @var{number} equal to one.
3109 The default i386 ABI assumes that the callee pops the
3110 stack for hidden pointer. However, on 32-bit i386 Microsoft Windows targets,
3111 the compiler assumes that the
3112 caller pops the stack for hidden pointer.
3114 @item ms_hook_prologue
3115 @cindex @code{ms_hook_prologue} attribute
3117 On 32-bit i[34567]86-*-* targets and 64-bit x86_64-*-* targets, you can use
3118 this function attribute to make GCC generate the ``hot-patching'' function
3119 prologue used in Win32 API functions in Microsoft Windows XP Service Pack 2
3123 @cindex function without a prologue/epilogue code
3124 Use this attribute on the ARM, AVR, MCORE, RX and SPU ports to indicate that
3125 the specified function does not need prologue/epilogue sequences generated by
3126 the compiler. It is up to the programmer to provide these sequences. The
3127 only statements that can be safely included in naked functions are
3128 @code{asm} statements that do not have operands. All other statements,
3129 including declarations of local variables, @code{if} statements, and so
3130 forth, should be avoided. Naked functions should be used to implement the
3131 body of an assembly function, while allowing the compiler to construct
3132 the requisite function declaration for the assembler.
3135 @cindex functions that do not handle memory bank switching on 68HC11/68HC12
3136 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
3137 use the normal calling convention based on @code{jsr} and @code{rts}.
3138 This attribute can be used to cancel the effect of the @option{-mlong-calls}
3141 On MeP targets this attribute causes the compiler to assume the called
3142 function is close enough to use the normal calling convention,
3143 overriding the @option{-mtf} command-line option.
3146 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
3147 Use this attribute together with @code{interrupt_handler},
3148 @code{exception_handler} or @code{nmi_handler} to indicate that the function
3149 entry code should enable nested interrupts or exceptions.
3152 @cindex NMI handler functions on the Blackfin processor
3153 Use this attribute on the Blackfin to indicate that the specified function
3154 is an NMI handler. The compiler generates function entry and
3155 exit sequences suitable for use in an NMI handler when this
3156 attribute is present.
3158 @item no_instrument_function
3159 @cindex @code{no_instrument_function} function attribute
3160 @opindex finstrument-functions
3161 If @option{-finstrument-functions} is given, profiling function calls are
3162 generated at entry and exit of most user-compiled functions.
3163 Functions with this attribute are not so instrumented.
3165 @item no_split_stack
3166 @cindex @code{no_split_stack} function attribute
3167 @opindex fsplit-stack
3168 If @option{-fsplit-stack} is given, functions have a small
3169 prologue which decides whether to split the stack. Functions with the
3170 @code{no_split_stack} attribute do not have that prologue, and thus
3171 may run with only a small amount of stack space available.
3174 @cindex @code{noinline} function attribute
3175 This function attribute prevents a function from being considered for
3177 @c Don't enumerate the optimizations by name here; we try to be
3178 @c future-compatible with this mechanism.
3179 If the function does not have side-effects, there are optimizations
3180 other than inlining that cause function calls to be optimized away,
3181 although the function call is live. To keep such calls from being
3188 (@pxref{Extended Asm}) in the called function, to serve as a special
3192 @cindex @code{noclone} function attribute
3193 This function attribute prevents a function from being considered for
3194 cloning---a mechanism that produces specialized copies of functions
3195 and which is (currently) performed by interprocedural constant
3198 @item nonnull (@var{arg-index}, @dots{})
3199 @cindex @code{nonnull} function attribute
3200 The @code{nonnull} attribute specifies that some function parameters should
3201 be non-null pointers. For instance, the declaration:
3205 my_memcpy (void *dest, const void *src, size_t len)
3206 __attribute__((nonnull (1, 2)));
3210 causes the compiler to check that, in calls to @code{my_memcpy},
3211 arguments @var{dest} and @var{src} are non-null. If the compiler
3212 determines that a null pointer is passed in an argument slot marked
3213 as non-null, and the @option{-Wnonnull} option is enabled, a warning
3214 is issued. The compiler may also choose to make optimizations based
3215 on the knowledge that certain function arguments will never be null.
3217 If no argument index list is given to the @code{nonnull} attribute,
3218 all pointer arguments are marked as non-null. To illustrate, the
3219 following declaration is equivalent to the previous example:
3223 my_memcpy (void *dest, const void *src, size_t len)
3224 __attribute__((nonnull));
3228 @cindex @code{noreturn} function attribute
3229 A few standard library functions, such as @code{abort} and @code{exit},
3230 cannot return. GCC knows this automatically. Some programs define
3231 their own functions that never return. You can declare them
3232 @code{noreturn} to tell the compiler this fact. For example,
3236 void fatal () __attribute__ ((noreturn));
3239 fatal (/* @r{@dots{}} */)
3241 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
3247 The @code{noreturn} keyword tells the compiler to assume that
3248 @code{fatal} cannot return. It can then optimize without regard to what
3249 would happen if @code{fatal} ever did return. This makes slightly
3250 better code. More importantly, it helps avoid spurious warnings of
3251 uninitialized variables.
3253 The @code{noreturn} keyword does not affect the exceptional path when that
3254 applies: a @code{noreturn}-marked function may still return to the caller
3255 by throwing an exception or calling @code{longjmp}.
3257 Do not assume that registers saved by the calling function are
3258 restored before calling the @code{noreturn} function.
3260 It does not make sense for a @code{noreturn} function to have a return
3261 type other than @code{void}.
3263 The attribute @code{noreturn} is not implemented in GCC versions
3264 earlier than 2.5. An alternative way to declare that a function does
3265 not return, which works in the current version and in some older
3266 versions, is as follows:
3269 typedef void voidfn ();
3271 volatile voidfn fatal;
3275 This approach does not work in GNU C++.
3278 @cindex @code{nothrow} function attribute
3279 The @code{nothrow} attribute is used to inform the compiler that a
3280 function cannot throw an exception. For example, most functions in
3281 the standard C library can be guaranteed not to throw an exception
3282 with the notable exceptions of @code{qsort} and @code{bsearch} that
3283 take function pointer arguments. The @code{nothrow} attribute is not
3284 implemented in GCC versions earlier than 3.3.
3286 @item nosave_low_regs
3287 @cindex @code{nosave_low_regs} attribute
3288 Use this attribute on SH targets to indicate that an @code{interrupt_handler}
3289 function should not save and restore registers R0..R7. This can be used on SH3*
3290 and SH4* targets that have a second R0..R7 register bank for non-reentrant
3294 @cindex @code{optimize} function attribute
3295 The @code{optimize} attribute is used to specify that a function is to
3296 be compiled with different optimization options than specified on the
3297 command line. Arguments can either be numbers or strings. Numbers
3298 are assumed to be an optimization level. Strings that begin with
3299 @code{O} are assumed to be an optimization option, while other options
3300 are assumed to be used with a @code{-f} prefix. You can also use the
3301 @samp{#pragma GCC optimize} pragma to set the optimization options
3302 that affect more than one function.
3303 @xref{Function Specific Option Pragmas}, for details about the
3304 @samp{#pragma GCC optimize} pragma.
3306 This can be used for instance to have frequently-executed functions
3307 compiled with more aggressive optimization options that produce faster
3308 and larger code, while other functions can be compiled with less
3311 @item OS_main/OS_task
3312 @cindex @code{OS_main} AVR function attribute
3313 @cindex @code{OS_task} AVR function attribute
3314 On AVR, functions with the @code{OS_main} or @code{OS_task} attribute
3315 do not save/restore any call-saved register in their prologue/epilogue.
3317 The @code{OS_main} attribute can be used when there @emph{is
3318 guarantee} that interrupts are disabled at the time when the function
3319 is entered. This saves resources when the stack pointer has to be
3320 changed to set up a frame for local variables.
3322 The @code{OS_task} attribute can be used when there is @emph{no
3323 guarantee} that interrupts are disabled at that time when the function
3324 is entered like for, e@.g@. task functions in a multi-threading operating
3325 system. In that case, changing the stack pointer register is
3326 guarded by save/clear/restore of the global interrupt enable flag.
3328 The differences to the @code{naked} function attribute are:
3330 @item @code{naked} functions do not have a return instruction whereas
3331 @code{OS_main} and @code{OS_task} functions have a @code{RET} or
3332 @code{RETI} return instruction.
3333 @item @code{naked} functions do not set up a frame for local variables
3334 or a frame pointer whereas @code{OS_main} and @code{OS_task} do this
3339 @cindex @code{pcs} function attribute
3341 The @code{pcs} attribute can be used to control the calling convention
3342 used for a function on ARM. The attribute takes an argument that specifies
3343 the calling convention to use.
3345 When compiling using the AAPCS ABI (or a variant of it) then valid
3346 values for the argument are @code{"aapcs"} and @code{"aapcs-vfp"}. In
3347 order to use a variant other than @code{"aapcs"} then the compiler must
3348 be permitted to use the appropriate co-processor registers (i.e., the
3349 VFP registers must be available in order to use @code{"aapcs-vfp"}).
3353 /* Argument passed in r0, and result returned in r0+r1. */
3354 double f2d (float) __attribute__((pcs("aapcs")));
3357 Variadic functions always use the @code{"aapcs"} calling convention and
3358 the compiler rejects attempts to specify an alternative.
3361 @cindex @code{pure} function attribute
3362 Many functions have no effects except the return value and their
3363 return value depends only on the parameters and/or global variables.
3364 Such a function can be subject
3365 to common subexpression elimination and loop optimization just as an
3366 arithmetic operator would be. These functions should be declared
3367 with the attribute @code{pure}. For example,
3370 int square (int) __attribute__ ((pure));
3374 says that the hypothetical function @code{square} is safe to call
3375 fewer times than the program says.
3377 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
3378 Interesting non-pure functions are functions with infinite loops or those
3379 depending on volatile memory or other system resource, that may change between
3380 two consecutive calls (such as @code{feof} in a multithreading environment).
3382 The attribute @code{pure} is not implemented in GCC versions earlier
3386 @cindex @code{hot} function attribute
3387 The @code{hot} attribute on a function is used to inform the compiler that
3388 the function is a hot spot of the compiled program. The function is
3389 optimized more aggressively and on many target it is placed into special
3390 subsection of the text section so all hot functions appears close together
3393 When profile feedback is available, via @option{-fprofile-use}, hot functions
3394 are automatically detected and this attribute is ignored.
3396 The @code{hot} attribute on functions is not implemented in GCC versions
3399 @cindex @code{hot} label attribute
3400 The @code{hot} attribute on a label is used to inform the compiler that
3401 path following the label are more likely than paths that are not so
3402 annotated. This attribute is used in cases where @code{__builtin_expect}
3403 cannot be used, for instance with computed goto or @code{asm goto}.
3405 The @code{hot} attribute on labels is not implemented in GCC versions
3409 @cindex @code{cold} function attribute
3410 The @code{cold} attribute on functions is used to inform the compiler that
3411 the function is unlikely to be executed. The function is optimized for
3412 size rather than speed and on many targets it is placed into special
3413 subsection of the text section so all cold functions appears close together
3414 improving code locality of non-cold parts of program. The paths leading
3415 to call of cold functions within code are marked as unlikely by the branch
3416 prediction mechanism. It is thus useful to mark functions used to handle
3417 unlikely conditions, such as @code{perror}, as cold to improve optimization
3418 of hot functions that do call marked functions in rare occasions.
3420 When profile feedback is available, via @option{-fprofile-use}, cold functions
3421 are automatically detected and this attribute is ignored.
3423 The @code{cold} attribute on functions is not implemented in GCC versions
3426 @cindex @code{cold} label attribute
3427 The @code{cold} attribute on labels is used to inform the compiler that
3428 the path following the label is unlikely to be executed. This attribute
3429 is used in cases where @code{__builtin_expect} cannot be used, for instance
3430 with computed goto or @code{asm goto}.
3432 The @code{cold} attribute on labels is not implemented in GCC versions
3435 @item no_address_safety_analysis
3436 @cindex @code{no_address_safety_analysis} function attribute
3437 The @code{no_address_safety_analysis} attribute on functions is used
3438 to inform the compiler that it should not instrument memory accesses
3439 in the function when compiling with the @option{-fsanitize=address} option.
3441 @item regparm (@var{number})
3442 @cindex @code{regparm} attribute
3443 @cindex functions that are passed arguments in registers on the 386
3444 On the Intel 386, the @code{regparm} attribute causes the compiler to
3445 pass arguments number one to @var{number} if they are of integral type
3446 in registers EAX, EDX, and ECX instead of on the stack. Functions that
3447 take a variable number of arguments continue to be passed all of their
3448 arguments on the stack.
3450 Beware that on some ELF systems this attribute is unsuitable for
3451 global functions in shared libraries with lazy binding (which is the
3452 default). Lazy binding sends the first call via resolving code in
3453 the loader, which might assume EAX, EDX and ECX can be clobbered, as
3454 per the standard calling conventions. Solaris 8 is affected by this.
3455 Systems with the GNU C Library version 2.1 or higher
3456 and FreeBSD are believed to be
3457 safe since the loaders there save EAX, EDX and ECX. (Lazy binding can be
3458 disabled with the linker or the loader if desired, to avoid the
3462 @cindex @code{sseregparm} attribute
3463 On the Intel 386 with SSE support, the @code{sseregparm} attribute
3464 causes the compiler to pass up to 3 floating-point arguments in
3465 SSE registers instead of on the stack. Functions that take a
3466 variable number of arguments continue to pass all of their
3467 floating-point arguments on the stack.
3469 @item force_align_arg_pointer
3470 @cindex @code{force_align_arg_pointer} attribute
3471 On the Intel x86, the @code{force_align_arg_pointer} attribute may be
3472 applied to individual function definitions, generating an alternate
3473 prologue and epilogue that realigns the run-time stack if necessary.
3474 This supports mixing legacy codes that run with a 4-byte aligned stack
3475 with modern codes that keep a 16-byte stack for SSE compatibility.
3478 @cindex @code{renesas} attribute
3479 On SH targets this attribute specifies that the function or struct follows the
3483 @cindex @code{resbank} attribute
3484 On the SH2A target, this attribute enables the high-speed register
3485 saving and restoration using a register bank for @code{interrupt_handler}
3486 routines. Saving to the bank is performed automatically after the CPU
3487 accepts an interrupt that uses a register bank.
3489 The nineteen 32-bit registers comprising general register R0 to R14,
3490 control register GBR, and system registers MACH, MACL, and PR and the
3491 vector table address offset are saved into a register bank. Register
3492 banks are stacked in first-in last-out (FILO) sequence. Restoration
3493 from the bank is executed by issuing a RESBANK instruction.
3496 @cindex @code{returns_twice} attribute
3497 The @code{returns_twice} attribute tells the compiler that a function may
3498 return more than one time. The compiler ensures that all registers
3499 are dead before calling such a function and emits a warning about
3500 the variables that may be clobbered after the second return from the
3501 function. Examples of such functions are @code{setjmp} and @code{vfork}.
3502 The @code{longjmp}-like counterpart of such function, if any, might need
3503 to be marked with the @code{noreturn} attribute.
3506 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
3507 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
3508 all registers except the stack pointer should be saved in the prologue
3509 regardless of whether they are used or not.
3511 @item save_volatiles
3512 @cindex save volatile registers on the MicroBlaze
3513 Use this attribute on the MicroBlaze to indicate that the function is
3514 an interrupt handler. All volatile registers (in addition to non-volatile
3515 registers) are saved in the function prologue. If the function is a leaf
3516 function, only volatiles used by the function are saved. A normal function
3517 return is generated instead of a return from interrupt.
3519 @item section ("@var{section-name}")
3520 @cindex @code{section} function attribute
3521 Normally, the compiler places the code it generates in the @code{text} section.
3522 Sometimes, however, you need additional sections, or you need certain
3523 particular functions to appear in special sections. The @code{section}
3524 attribute specifies that a function lives in a particular section.
3525 For example, the declaration:
3528 extern void foobar (void) __attribute__ ((section ("bar")));
3532 puts the function @code{foobar} in the @code{bar} section.
3534 Some file formats do not support arbitrary sections so the @code{section}
3535 attribute is not available on all platforms.
3536 If you need to map the entire contents of a module to a particular
3537 section, consider using the facilities of the linker instead.
3540 @cindex @code{sentinel} function attribute
3541 This function attribute ensures that a parameter in a function call is
3542 an explicit @code{NULL}. The attribute is only valid on variadic
3543 functions. By default, the sentinel is located at position zero, the
3544 last parameter of the function call. If an optional integer position
3545 argument P is supplied to the attribute, the sentinel must be located at
3546 position P counting backwards from the end of the argument list.
3549 __attribute__ ((sentinel))
3551 __attribute__ ((sentinel(0)))
3554 The attribute is automatically set with a position of 0 for the built-in
3555 functions @code{execl} and @code{execlp}. The built-in function
3556 @code{execle} has the attribute set with a position of 1.
3558 A valid @code{NULL} in this context is defined as zero with any pointer
3559 type. If your system defines the @code{NULL} macro with an integer type
3560 then you need to add an explicit cast. GCC replaces @code{stddef.h}
3561 with a copy that redefines NULL appropriately.
3563 The warnings for missing or incorrect sentinels are enabled with
3567 See @code{long_call/short_call}.
3570 See @code{longcall/shortcall}.
3573 @cindex interrupt handler functions on the AVR processors
3574 Use this attribute on the AVR to indicate that the specified
3575 function is an interrupt handler. The compiler generates function
3576 entry and exit sequences suitable for use in an interrupt handler when this
3577 attribute is present.
3579 See also the @code{interrupt} function attribute.
3581 The AVR hardware globally disables interrupts when an interrupt is executed.
3582 Interrupt handler functions defined with the @code{signal} attribute
3583 do not re-enable interrupts. It is save to enable interrupts in a
3584 @code{signal} handler. This ``save'' only applies to the code
3585 generated by the compiler and not to the IRQ layout of the
3586 application which is responsibility of the application.
3588 If both @code{signal} and @code{interrupt} are specified for the same
3589 function, @code{signal} is silently ignored.
3592 @cindex @code{sp_switch} attribute
3593 Use this attribute on the SH to indicate an @code{interrupt_handler}
3594 function should switch to an alternate stack. It expects a string
3595 argument that names a global variable holding the address of the
3600 void f () __attribute__ ((interrupt_handler,
3601 sp_switch ("alt_stack")));
3605 @cindex functions that pop the argument stack on the 386
3606 On the Intel 386, the @code{stdcall} attribute causes the compiler to
3607 assume that the called function pops off the stack space used to
3608 pass arguments, unless it takes a variable number of arguments.
3610 @item syscall_linkage
3611 @cindex @code{syscall_linkage} attribute
3612 This attribute is used to modify the IA-64 calling convention by marking
3613 all input registers as live at all function exits. This makes it possible
3614 to restart a system call after an interrupt without having to save/restore
3615 the input registers. This also prevents kernel data from leaking into
3619 @cindex @code{target} function attribute
3620 The @code{target} attribute is used to specify that a function is to
3621 be compiled with different target options than specified on the
3622 command line. This can be used for instance to have functions
3623 compiled with a different ISA (instruction set architecture) than the
3624 default. You can also use the @samp{#pragma GCC target} pragma to set
3625 more than one function to be compiled with specific target options.
3626 @xref{Function Specific Option Pragmas}, for details about the
3627 @samp{#pragma GCC target} pragma.
3629 For instance on a 386, you could compile one function with
3630 @code{target("sse4.1,arch=core2")} and another with
3631 @code{target("sse4a,arch=amdfam10")}. This is equivalent to
3632 compiling the first function with @option{-msse4.1} and
3633 @option{-march=core2} options, and the second function with
3634 @option{-msse4a} and @option{-march=amdfam10} options. It is up to the
3635 user to make sure that a function is only invoked on a machine that
3636 supports the particular ISA it is compiled for (for example by using
3637 @code{cpuid} on 386 to determine what feature bits and architecture
3641 int core2_func (void) __attribute__ ((__target__ ("arch=core2")));
3642 int sse3_func (void) __attribute__ ((__target__ ("sse3")));
3645 On the 386, the following options are allowed:
3650 @cindex @code{target("abm")} attribute
3651 Enable/disable the generation of the advanced bit instructions.
3655 @cindex @code{target("aes")} attribute
3656 Enable/disable the generation of the AES instructions.
3660 @cindex @code{target("mmx")} attribute
3661 Enable/disable the generation of the MMX instructions.
3665 @cindex @code{target("pclmul")} attribute
3666 Enable/disable the generation of the PCLMUL instructions.
3670 @cindex @code{target("popcnt")} attribute
3671 Enable/disable the generation of the POPCNT instruction.
3675 @cindex @code{target("sse")} attribute
3676 Enable/disable the generation of the SSE instructions.
3680 @cindex @code{target("sse2")} attribute
3681 Enable/disable the generation of the SSE2 instructions.
3685 @cindex @code{target("sse3")} attribute
3686 Enable/disable the generation of the SSE3 instructions.
3690 @cindex @code{target("sse4")} attribute
3691 Enable/disable the generation of the SSE4 instructions (both SSE4.1
3696 @cindex @code{target("sse4.1")} attribute
3697 Enable/disable the generation of the sse4.1 instructions.
3701 @cindex @code{target("sse4.2")} attribute
3702 Enable/disable the generation of the sse4.2 instructions.
3706 @cindex @code{target("sse4a")} attribute
3707 Enable/disable the generation of the SSE4A instructions.
3711 @cindex @code{target("fma4")} attribute
3712 Enable/disable the generation of the FMA4 instructions.
3716 @cindex @code{target("xop")} attribute
3717 Enable/disable the generation of the XOP instructions.
3721 @cindex @code{target("lwp")} attribute
3722 Enable/disable the generation of the LWP instructions.
3726 @cindex @code{target("ssse3")} attribute
3727 Enable/disable the generation of the SSSE3 instructions.
3731 @cindex @code{target("cld")} attribute
3732 Enable/disable the generation of the CLD before string moves.
3734 @item fancy-math-387
3735 @itemx no-fancy-math-387
3736 @cindex @code{target("fancy-math-387")} attribute
3737 Enable/disable the generation of the @code{sin}, @code{cos}, and
3738 @code{sqrt} instructions on the 387 floating-point unit.
3741 @itemx no-fused-madd
3742 @cindex @code{target("fused-madd")} attribute
3743 Enable/disable the generation of the fused multiply/add instructions.
3747 @cindex @code{target("ieee-fp")} attribute
3748 Enable/disable the generation of floating point that depends on IEEE arithmetic.
3750 @item inline-all-stringops
3751 @itemx no-inline-all-stringops
3752 @cindex @code{target("inline-all-stringops")} attribute
3753 Enable/disable inlining of string operations.
3755 @item inline-stringops-dynamically
3756 @itemx no-inline-stringops-dynamically
3757 @cindex @code{target("inline-stringops-dynamically")} attribute
3758 Enable/disable the generation of the inline code to do small string
3759 operations and calling the library routines for large operations.
3761 @item align-stringops
3762 @itemx no-align-stringops
3763 @cindex @code{target("align-stringops")} attribute
3764 Do/do not align destination of inlined string operations.
3768 @cindex @code{target("recip")} attribute
3769 Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS
3770 instructions followed an additional Newton-Raphson step instead of
3771 doing a floating-point division.
3773 @item arch=@var{ARCH}
3774 @cindex @code{target("arch=@var{ARCH}")} attribute
3775 Specify the architecture to generate code for in compiling the function.
3777 @item tune=@var{TUNE}
3778 @cindex @code{target("tune=@var{TUNE}")} attribute
3779 Specify the architecture to tune for in compiling the function.
3781 @item fpmath=@var{FPMATH}
3782 @cindex @code{target("fpmath=@var{FPMATH}")} attribute
3783 Specify which floating-point unit to use. The
3784 @code{target("fpmath=sse,387")} option must be specified as
3785 @code{target("fpmath=sse+387")} because the comma would separate
3789 On the PowerPC, the following options are allowed:
3794 @cindex @code{target("altivec")} attribute
3795 Generate code that uses (does not use) AltiVec instructions. In
3796 32-bit code, you cannot enable AltiVec instructions unless
3797 @option{-mabi=altivec} is used on the command line.
3801 @cindex @code{target("cmpb")} attribute
3802 Generate code that uses (does not use) the compare bytes instruction
3803 implemented on the POWER6 processor and other processors that support
3804 the PowerPC V2.05 architecture.
3808 @cindex @code{target("dlmzb")} attribute
3809 Generate code that uses (does not use) the string-search @samp{dlmzb}
3810 instruction on the IBM 405, 440, 464 and 476 processors. This instruction is
3811 generated by default when targeting those processors.
3815 @cindex @code{target("fprnd")} attribute
3816 Generate code that uses (does not use) the FP round to integer
3817 instructions implemented on the POWER5+ processor and other processors
3818 that support the PowerPC V2.03 architecture.
3822 @cindex @code{target("hard-dfp")} attribute
3823 Generate code that uses (does not use) the decimal floating-point
3824 instructions implemented on some POWER processors.
3828 @cindex @code{target("isel")} attribute
3829 Generate code that uses (does not use) ISEL instruction.
3833 @cindex @code{target("mfcrf")} attribute
3834 Generate code that uses (does not use) the move from condition
3835 register field instruction implemented on the POWER4 processor and
3836 other processors that support the PowerPC V2.01 architecture.
3840 @cindex @code{target("mfpgpr")} attribute
3841 Generate code that uses (does not use) the FP move to/from general
3842 purpose register instructions implemented on the POWER6X processor and
3843 other processors that support the extended PowerPC V2.05 architecture.
3847 @cindex @code{target("mulhw")} attribute
3848 Generate code that uses (does not use) the half-word multiply and
3849 multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors.
3850 These instructions are generated by default when targeting those
3855 @cindex @code{target("multiple")} attribute
3856 Generate code that uses (does not use) the load multiple word
3857 instructions and the store multiple word instructions.
3861 @cindex @code{target("update")} attribute
3862 Generate code that uses (does not use) the load or store instructions
3863 that update the base register to the address of the calculated memory
3868 @cindex @code{target("popcntb")} attribute
3869 Generate code that uses (does not use) the popcount and double-precision
3870 FP reciprocal estimate instruction implemented on the POWER5
3871 processor and other processors that support the PowerPC V2.02
3876 @cindex @code{target("popcntd")} attribute
3877 Generate code that uses (does not use) the popcount instruction
3878 implemented on the POWER7 processor and other processors that support
3879 the PowerPC V2.06 architecture.
3881 @item powerpc-gfxopt
3882 @itemx no-powerpc-gfxopt
3883 @cindex @code{target("powerpc-gfxopt")} attribute
3884 Generate code that uses (does not use) the optional PowerPC
3885 architecture instructions in the Graphics group, including
3886 floating-point select.
3889 @itemx no-powerpc-gpopt
3890 @cindex @code{target("powerpc-gpopt")} attribute
3891 Generate code that uses (does not use) the optional PowerPC
3892 architecture instructions in the General Purpose group, including
3893 floating-point square root.
3895 @item recip-precision
3896 @itemx no-recip-precision
3897 @cindex @code{target("recip-precision")} attribute
3898 Assume (do not assume) that the reciprocal estimate instructions
3899 provide higher-precision estimates than is mandated by the powerpc
3904 @cindex @code{target("string")} attribute
3905 Generate code that uses (does not use) the load string instructions
3906 and the store string word instructions to save multiple registers and
3907 do small block moves.
3911 @cindex @code{target("vsx")} attribute
3912 Generate code that uses (does not use) vector/scalar (VSX)
3913 instructions, and also enable the use of built-in functions that allow
3914 more direct access to the VSX instruction set. In 32-bit code, you
3915 cannot enable VSX or AltiVec instructions unless
3916 @option{-mabi=altivec} is used on the command line.
3920 @cindex @code{target("friz")} attribute
3921 Generate (do not generate) the @code{friz} instruction when the
3922 @option{-funsafe-math-optimizations} option is used to optimize
3923 rounding a floating-point value to 64-bit integer and back to floating
3924 point. The @code{friz} instruction does not return the same value if
3925 the floating-point number is too large to fit in an integer.
3927 @item avoid-indexed-addresses
3928 @itemx no-avoid-indexed-addresses
3929 @cindex @code{target("avoid-indexed-addresses")} attribute
3930 Generate code that tries to avoid (not avoid) the use of indexed load
3931 or store instructions.
3935 @cindex @code{target("paired")} attribute
3936 Generate code that uses (does not use) the generation of PAIRED simd
3941 @cindex @code{target("longcall")} attribute
3942 Generate code that assumes (does not assume) that all calls are far
3943 away so that a longer more expensive calling sequence is required.
3946 @cindex @code{target("cpu=@var{CPU}")} attribute
3947 Specify the architecture to generate code for when compiling the
3948 function. If you select the @code{target("cpu=power7")} attribute when
3949 generating 32-bit code, VSX and AltiVec instructions are not generated
3950 unless you use the @option{-mabi=altivec} option on the command line.
3952 @item tune=@var{TUNE}
3953 @cindex @code{target("tune=@var{TUNE}")} attribute
3954 Specify the architecture to tune for when compiling the function. If
3955 you do not specify the @code{target("tune=@var{TUNE}")} attribute and
3956 you do specify the @code{target("cpu=@var{CPU}")} attribute,
3957 compilation tunes for the @var{CPU} architecture, and not the
3958 default tuning specified on the command line.
3961 On the 386/x86_64 and PowerPC back ends, you can use either multiple
3962 strings to specify multiple options, or you can separate the option
3963 with a comma (@code{,}).
3965 On the 386/x86_64 and PowerPC back ends, the inliner does not inline a
3966 function that has different target options than the caller, unless the
3967 callee has a subset of the target options of the caller. For example
3968 a function declared with @code{target("sse3")} can inline a function
3969 with @code{target("sse2")}, since @code{-msse3} implies @code{-msse2}.
3971 The @code{target} attribute is not implemented in GCC versions earlier
3972 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC back ends. It is
3973 not currently implemented for other back ends.
3976 @cindex tiny data section on the H8/300H and H8S
3977 Use this attribute on the H8/300H and H8S to indicate that the specified
3978 variable should be placed into the tiny data section.
3979 The compiler generates more efficient code for loads and stores
3980 on data in the tiny data section. Note the tiny data area is limited to
3981 slightly under 32KB of data.
3984 @cindex @code{trap_exit} attribute
3985 Use this attribute on the SH for an @code{interrupt_handler} to return using
3986 @code{trapa} instead of @code{rte}. This attribute expects an integer
3987 argument specifying the trap number to be used.
3990 @cindex @code{trapa_handler} attribute
3991 On SH targets this function attribute is similar to @code{interrupt_handler}
3992 but it does not save and restore all registers.
3995 @cindex @code{unused} attribute.
3996 This attribute, attached to a function, means that the function is meant
3997 to be possibly unused. GCC does not produce a warning for this
4001 @cindex @code{used} attribute.
4002 This attribute, attached to a function, means that code must be emitted
4003 for the function even if it appears that the function is not referenced.
4004 This is useful, for example, when the function is referenced only in
4007 When applied to a member function of a C++ class template, the
4008 attribute also means that the function is instantiated if the
4009 class itself is instantiated.
4012 @cindex @code{version_id} attribute
4013 This IA-64 HP-UX attribute, attached to a global variable or function, renames a
4014 symbol to contain a version string, thus allowing for function level
4015 versioning. HP-UX system header files may use version level functioning
4016 for some system calls.
4019 extern int foo () __attribute__((version_id ("20040821")));
4023 Calls to @var{foo} are mapped to calls to @var{foo@{20040821@}}.
4025 @item visibility ("@var{visibility_type}")
4026 @cindex @code{visibility} attribute
4027 This attribute affects the linkage of the declaration to which it is attached.
4028 There are four supported @var{visibility_type} values: default,
4029 hidden, protected or internal visibility.
4032 void __attribute__ ((visibility ("protected")))
4033 f () @{ /* @r{Do something.} */; @}
4034 int i __attribute__ ((visibility ("hidden")));
4037 The possible values of @var{visibility_type} correspond to the
4038 visibility settings in the ELF gABI.
4041 @c keep this list of visibilities in alphabetical order.
4044 Default visibility is the normal case for the object file format.
4045 This value is available for the visibility attribute to override other
4046 options that may change the assumed visibility of entities.
4048 On ELF, default visibility means that the declaration is visible to other
4049 modules and, in shared libraries, means that the declared entity may be
4052 On Darwin, default visibility means that the declaration is visible to
4055 Default visibility corresponds to ``external linkage'' in the language.
4058 Hidden visibility indicates that the entity declared has a new
4059 form of linkage, which we call ``hidden linkage''. Two
4060 declarations of an object with hidden linkage refer to the same object
4061 if they are in the same shared object.
4064 Internal visibility is like hidden visibility, but with additional
4065 processor specific semantics. Unless otherwise specified by the
4066 psABI, GCC defines internal visibility to mean that a function is
4067 @emph{never} called from another module. Compare this with hidden
4068 functions which, while they cannot be referenced directly by other
4069 modules, can be referenced indirectly via function pointers. By
4070 indicating that a function cannot be called from outside the module,
4071 GCC may for instance omit the load of a PIC register since it is known
4072 that the calling function loaded the correct value.
4075 Protected visibility is like default visibility except that it
4076 indicates that references within the defining module bind to the
4077 definition in that module. That is, the declared entity cannot be
4078 overridden by another module.
4082 All visibilities are supported on many, but not all, ELF targets
4083 (supported when the assembler supports the @samp{.visibility}
4084 pseudo-op). Default visibility is supported everywhere. Hidden
4085 visibility is supported on Darwin targets.
4087 The visibility attribute should be applied only to declarations that
4088 would otherwise have external linkage. The attribute should be applied
4089 consistently, so that the same entity should not be declared with
4090 different settings of the attribute.
4092 In C++, the visibility attribute applies to types as well as functions
4093 and objects, because in C++ types have linkage. A class must not have
4094 greater visibility than its non-static data member types and bases,
4095 and class members default to the visibility of their class. Also, a
4096 declaration without explicit visibility is limited to the visibility
4099 In C++, you can mark member functions and static member variables of a
4100 class with the visibility attribute. This is useful if you know a
4101 particular method or static member variable should only be used from
4102 one shared object; then you can mark it hidden while the rest of the
4103 class has default visibility. Care must be taken to avoid breaking
4104 the One Definition Rule; for example, it is usually not useful to mark
4105 an inline method as hidden without marking the whole class as hidden.
4107 A C++ namespace declaration can also have the visibility attribute.
4108 This attribute applies only to the particular namespace body, not to
4109 other definitions of the same namespace; it is equivalent to using
4110 @samp{#pragma GCC visibility} before and after the namespace
4111 definition (@pxref{Visibility Pragmas}).
4113 In C++, if a template argument has limited visibility, this
4114 restriction is implicitly propagated to the template instantiation.
4115 Otherwise, template instantiations and specializations default to the
4116 visibility of their template.
4118 If both the template and enclosing class have explicit visibility, the
4119 visibility from the template is used.
4122 @cindex @code{vliw} attribute
4123 On MeP, the @code{vliw} attribute tells the compiler to emit
4124 instructions in VLIW mode instead of core mode. Note that this
4125 attribute is not allowed unless a VLIW coprocessor has been configured
4126 and enabled through command-line options.
4128 @item warn_unused_result
4129 @cindex @code{warn_unused_result} attribute
4130 The @code{warn_unused_result} attribute causes a warning to be emitted
4131 if a caller of the function with this attribute does not use its
4132 return value. This is useful for functions where not checking
4133 the result is either a security problem or always a bug, such as
4137 int fn () __attribute__ ((warn_unused_result));
4140 if (fn () < 0) return -1;
4147 results in warning on line 5.
4150 @cindex @code{weak} attribute
4151 The @code{weak} attribute causes the declaration to be emitted as a weak
4152 symbol rather than a global. This is primarily useful in defining
4153 library functions that can be overridden in user code, though it can
4154 also be used with non-function declarations. Weak symbols are supported
4155 for ELF targets, and also for a.out targets when using the GNU assembler
4159 @itemx weakref ("@var{target}")
4160 @cindex @code{weakref} attribute
4161 The @code{weakref} attribute marks a declaration as a weak reference.
4162 Without arguments, it should be accompanied by an @code{alias} attribute
4163 naming the target symbol. Optionally, the @var{target} may be given as
4164 an argument to @code{weakref} itself. In either case, @code{weakref}
4165 implicitly marks the declaration as @code{weak}. Without a
4166 @var{target}, given as an argument to @code{weakref} or to @code{alias},
4167 @code{weakref} is equivalent to @code{weak}.
4170 static int x() __attribute__ ((weakref ("y")));
4171 /* is equivalent to... */
4172 static int x() __attribute__ ((weak, weakref, alias ("y")));
4174 static int x() __attribute__ ((weakref));
4175 static int x() __attribute__ ((alias ("y")));
4178 A weak reference is an alias that does not by itself require a
4179 definition to be given for the target symbol. If the target symbol is
4180 only referenced through weak references, then it becomes a @code{weak}
4181 undefined symbol. If it is directly referenced, however, then such
4182 strong references prevail, and a definition is required for the
4183 symbol, not necessarily in the same translation unit.
4185 The effect is equivalent to moving all references to the alias to a
4186 separate translation unit, renaming the alias to the aliased symbol,
4187 declaring it as weak, compiling the two separate translation units and
4188 performing a reloadable link on them.
4190 At present, a declaration to which @code{weakref} is attached can
4191 only be @code{static}.
4195 You can specify multiple attributes in a declaration by separating them
4196 by commas within the double parentheses or by immediately following an
4197 attribute declaration with another attribute declaration.
4199 @cindex @code{#pragma}, reason for not using
4200 @cindex pragma, reason for not using
4201 Some people object to the @code{__attribute__} feature, suggesting that
4202 ISO C's @code{#pragma} should be used instead. At the time
4203 @code{__attribute__} was designed, there were two reasons for not doing
4208 It is impossible to generate @code{#pragma} commands from a macro.
4211 There is no telling what the same @code{#pragma} might mean in another
4215 These two reasons applied to almost any application that might have been
4216 proposed for @code{#pragma}. It was basically a mistake to use
4217 @code{#pragma} for @emph{anything}.
4219 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
4220 to be generated from macros. In addition, a @code{#pragma GCC}
4221 namespace is now in use for GCC-specific pragmas. However, it has been
4222 found convenient to use @code{__attribute__} to achieve a natural
4223 attachment of attributes to their corresponding declarations, whereas
4224 @code{#pragma GCC} is of use for constructs that do not naturally form
4225 part of the grammar. @xref{Pragmas,,Pragmas Accepted by GCC}.
4227 @node Attribute Syntax
4228 @section Attribute Syntax
4229 @cindex attribute syntax
4231 This section describes the syntax with which @code{__attribute__} may be
4232 used, and the constructs to which attribute specifiers bind, for the C
4233 language. Some details may vary for C++ and Objective-C@. Because of
4234 infelicities in the grammar for attributes, some forms described here
4235 may not be successfully parsed in all cases.
4237 There are some problems with the semantics of attributes in C++. For
4238 example, there are no manglings for attributes, although they may affect
4239 code generation, so problems may arise when attributed types are used in
4240 conjunction with templates or overloading. Similarly, @code{typeid}
4241 does not distinguish between types with different attributes. Support
4242 for attributes in C++ may be restricted in future to attributes on
4243 declarations only, but not on nested declarators.
4245 @xref{Function Attributes}, for details of the semantics of attributes
4246 applying to functions. @xref{Variable Attributes}, for details of the
4247 semantics of attributes applying to variables. @xref{Type Attributes},
4248 for details of the semantics of attributes applying to structure, union
4249 and enumerated types.
4251 An @dfn{attribute specifier} is of the form
4252 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
4253 is a possibly empty comma-separated sequence of @dfn{attributes}, where
4254 each attribute is one of the following:
4258 Empty. Empty attributes are ignored.
4261 A word (which may be an identifier such as @code{unused}, or a reserved
4262 word such as @code{const}).
4265 A word, followed by, in parentheses, parameters for the attribute.
4266 These parameters take one of the following forms:
4270 An identifier. For example, @code{mode} attributes use this form.
4273 An identifier followed by a comma and a non-empty comma-separated list
4274 of expressions. For example, @code{format} attributes use this form.
4277 A possibly empty comma-separated list of expressions. For example,
4278 @code{format_arg} attributes use this form with the list being a single
4279 integer constant expression, and @code{alias} attributes use this form
4280 with the list being a single string constant.
4284 An @dfn{attribute specifier list} is a sequence of one or more attribute
4285 specifiers, not separated by any other tokens.
4287 In GNU C, an attribute specifier list may appear after the colon following a
4288 label, other than a @code{case} or @code{default} label. The only
4289 attribute it makes sense to use after a label is @code{unused}. This
4290 feature is intended for program-generated code that may contain unused labels,
4291 but which is compiled with @option{-Wall}. It is
4292 not normally appropriate to use in it human-written code, though it
4293 could be useful in cases where the code that jumps to the label is
4294 contained within an @code{#ifdef} conditional. GNU C++ only permits
4295 attributes on labels if the attribute specifier is immediately
4296 followed by a semicolon (i.e., the label applies to an empty
4297 statement). If the semicolon is missing, C++ label attributes are
4298 ambiguous, as it is permissible for a declaration, which could begin
4299 with an attribute list, to be labelled in C++. Declarations cannot be
4300 labelled in C90 or C99, so the ambiguity does not arise there.
4302 An attribute specifier list may appear as part of a @code{struct},
4303 @code{union} or @code{enum} specifier. It may go either immediately
4304 after the @code{struct}, @code{union} or @code{enum} keyword, or after
4305 the closing brace. The former syntax is preferred.
4306 Where attribute specifiers follow the closing brace, they are considered
4307 to relate to the structure, union or enumerated type defined, not to any
4308 enclosing declaration the type specifier appears in, and the type
4309 defined is not complete until after the attribute specifiers.
4310 @c Otherwise, there would be the following problems: a shift/reduce
4311 @c conflict between attributes binding the struct/union/enum and
4312 @c binding to the list of specifiers/qualifiers; and "aligned"
4313 @c attributes could use sizeof for the structure, but the size could be
4314 @c changed later by "packed" attributes.
4316 Otherwise, an attribute specifier appears as part of a declaration,
4317 counting declarations of unnamed parameters and type names, and relates
4318 to that declaration (which may be nested in another declaration, for
4319 example in the case of a parameter declaration), or to a particular declarator
4320 within a declaration. Where an
4321 attribute specifier is applied to a parameter declared as a function or
4322 an array, it should apply to the function or array rather than the
4323 pointer to which the parameter is implicitly converted, but this is not
4324 yet correctly implemented.
4326 Any list of specifiers and qualifiers at the start of a declaration may
4327 contain attribute specifiers, whether or not such a list may in that
4328 context contain storage class specifiers. (Some attributes, however,
4329 are essentially in the nature of storage class specifiers, and only make
4330 sense where storage class specifiers may be used; for example,
4331 @code{section}.) There is one necessary limitation to this syntax: the
4332 first old-style parameter declaration in a function definition cannot
4333 begin with an attribute specifier, because such an attribute applies to
4334 the function instead by syntax described below (which, however, is not
4335 yet implemented in this case). In some other cases, attribute
4336 specifiers are permitted by this grammar but not yet supported by the
4337 compiler. All attribute specifiers in this place relate to the
4338 declaration as a whole. In the obsolescent usage where a type of
4339 @code{int} is implied by the absence of type specifiers, such a list of
4340 specifiers and qualifiers may be an attribute specifier list with no
4341 other specifiers or qualifiers.
4343 At present, the first parameter in a function prototype must have some
4344 type specifier that is not an attribute specifier; this resolves an
4345 ambiguity in the interpretation of @code{void f(int
4346 (__attribute__((foo)) x))}, but is subject to change. At present, if
4347 the parentheses of a function declarator contain only attributes then
4348 those attributes are ignored, rather than yielding an error or warning
4349 or implying a single parameter of type int, but this is subject to
4352 An attribute specifier list may appear immediately before a declarator
4353 (other than the first) in a comma-separated list of declarators in a
4354 declaration of more than one identifier using a single list of
4355 specifiers and qualifiers. Such attribute specifiers apply
4356 only to the identifier before whose declarator they appear. For
4360 __attribute__((noreturn)) void d0 (void),
4361 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
4366 the @code{noreturn} attribute applies to all the functions
4367 declared; the @code{format} attribute only applies to @code{d1}.
4369 An attribute specifier list may appear immediately before the comma,
4370 @code{=} or semicolon terminating the declaration of an identifier other
4371 than a function definition. Such attribute specifiers apply
4372 to the declared object or function. Where an
4373 assembler name for an object or function is specified (@pxref{Asm
4374 Labels}), the attribute must follow the @code{asm}
4377 An attribute specifier list may, in future, be permitted to appear after
4378 the declarator in a function definition (before any old-style parameter
4379 declarations or the function body).
4381 Attribute specifiers may be mixed with type qualifiers appearing inside
4382 the @code{[]} of a parameter array declarator, in the C99 construct by
4383 which such qualifiers are applied to the pointer to which the array is
4384 implicitly converted. Such attribute specifiers apply to the pointer,
4385 not to the array, but at present this is not implemented and they are
4388 An attribute specifier list may appear at the start of a nested
4389 declarator. At present, there are some limitations in this usage: the
4390 attributes correctly apply to the declarator, but for most individual
4391 attributes the semantics this implies are not implemented.
4392 When attribute specifiers follow the @code{*} of a pointer
4393 declarator, they may be mixed with any type qualifiers present.
4394 The following describes the formal semantics of this syntax. It makes the
4395 most sense if you are familiar with the formal specification of
4396 declarators in the ISO C standard.
4398 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
4399 D1}, where @code{T} contains declaration specifiers that specify a type
4400 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
4401 contains an identifier @var{ident}. The type specified for @var{ident}
4402 for derived declarators whose type does not include an attribute
4403 specifier is as in the ISO C standard.
4405 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
4406 and the declaration @code{T D} specifies the type
4407 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4408 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4409 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
4411 If @code{D1} has the form @code{*
4412 @var{type-qualifier-and-attribute-specifier-list} D}, and the
4413 declaration @code{T D} specifies the type
4414 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4415 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4416 @var{type-qualifier-and-attribute-specifier-list} pointer to @var{Type}'' for
4422 void (__attribute__((noreturn)) ****f) (void);
4426 specifies the type ``pointer to pointer to pointer to pointer to
4427 non-returning function returning @code{void}''. As another example,
4430 char *__attribute__((aligned(8))) *f;
4434 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
4435 Note again that this does not work with most attributes; for example,
4436 the usage of @samp{aligned} and @samp{noreturn} attributes given above
4437 is not yet supported.
4439 For compatibility with existing code written for compiler versions that
4440 did not implement attributes on nested declarators, some laxity is
4441 allowed in the placing of attributes. If an attribute that only applies
4442 to types is applied to a declaration, it is treated as applying to
4443 the type of that declaration. If an attribute that only applies to
4444 declarations is applied to the type of a declaration, it is treated
4445 as applying to that declaration; and, for compatibility with code
4446 placing the attributes immediately before the identifier declared, such
4447 an attribute applied to a function return type is treated as
4448 applying to the function type, and such an attribute applied to an array
4449 element type is treated as applying to the array type. If an
4450 attribute that only applies to function types is applied to a
4451 pointer-to-function type, it is treated as applying to the pointer
4452 target type; if such an attribute is applied to a function return type
4453 that is not a pointer-to-function type, it is treated as applying
4454 to the function type.
4456 @node Function Prototypes
4457 @section Prototypes and Old-Style Function Definitions
4458 @cindex function prototype declarations
4459 @cindex old-style function definitions
4460 @cindex promotion of formal parameters
4462 GNU C extends ISO C to allow a function prototype to override a later
4463 old-style non-prototype definition. Consider the following example:
4466 /* @r{Use prototypes unless the compiler is old-fashioned.} */
4473 /* @r{Prototype function declaration.} */
4474 int isroot P((uid_t));
4476 /* @r{Old-style function definition.} */
4478 isroot (x) /* @r{??? lossage here ???} */
4485 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
4486 not allow this example, because subword arguments in old-style
4487 non-prototype definitions are promoted. Therefore in this example the
4488 function definition's argument is really an @code{int}, which does not
4489 match the prototype argument type of @code{short}.
4491 This restriction of ISO C makes it hard to write code that is portable
4492 to traditional C compilers, because the programmer does not know
4493 whether the @code{uid_t} type is @code{short}, @code{int}, or
4494 @code{long}. Therefore, in cases like these GNU C allows a prototype
4495 to override a later old-style definition. More precisely, in GNU C, a
4496 function prototype argument type overrides the argument type specified
4497 by a later old-style definition if the former type is the same as the
4498 latter type before promotion. Thus in GNU C the above example is
4499 equivalent to the following:
4512 GNU C++ does not support old-style function definitions, so this
4513 extension is irrelevant.
4516 @section C++ Style Comments
4518 @cindex C++ comments
4519 @cindex comments, C++ style
4521 In GNU C, you may use C++ style comments, which start with @samp{//} and
4522 continue until the end of the line. Many other C implementations allow
4523 such comments, and they are included in the 1999 C standard. However,
4524 C++ style comments are not recognized if you specify an @option{-std}
4525 option specifying a version of ISO C before C99, or @option{-ansi}
4526 (equivalent to @option{-std=c90}).
4529 @section Dollar Signs in Identifier Names
4531 @cindex dollar signs in identifier names
4532 @cindex identifier names, dollar signs in
4534 In GNU C, you may normally use dollar signs in identifier names.
4535 This is because many traditional C implementations allow such identifiers.
4536 However, dollar signs in identifiers are not supported on a few target
4537 machines, typically because the target assembler does not allow them.
4539 @node Character Escapes
4540 @section The Character @key{ESC} in Constants
4542 You can use the sequence @samp{\e} in a string or character constant to
4543 stand for the ASCII character @key{ESC}.
4545 @node Variable Attributes
4546 @section Specifying Attributes of Variables
4547 @cindex attribute of variables
4548 @cindex variable attributes
4550 The keyword @code{__attribute__} allows you to specify special
4551 attributes of variables or structure fields. This keyword is followed
4552 by an attribute specification inside double parentheses. Some
4553 attributes are currently defined generically for variables.
4554 Other attributes are defined for variables on particular target
4555 systems. Other attributes are available for functions
4556 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
4557 Other front ends might define more attributes
4558 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
4560 You may also specify attributes with @samp{__} preceding and following
4561 each keyword. This allows you to use them in header files without
4562 being concerned about a possible macro of the same name. For example,
4563 you may use @code{__aligned__} instead of @code{aligned}.
4565 @xref{Attribute Syntax}, for details of the exact syntax for using
4569 @cindex @code{aligned} attribute
4570 @item aligned (@var{alignment})
4571 This attribute specifies a minimum alignment for the variable or
4572 structure field, measured in bytes. For example, the declaration:
4575 int x __attribute__ ((aligned (16))) = 0;
4579 causes the compiler to allocate the global variable @code{x} on a
4580 16-byte boundary. On a 68040, this could be used in conjunction with
4581 an @code{asm} expression to access the @code{move16} instruction which
4582 requires 16-byte aligned operands.
4584 You can also specify the alignment of structure fields. For example, to
4585 create a double-word aligned @code{int} pair, you could write:
4588 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
4592 This is an alternative to creating a union with a @code{double} member,
4593 which forces the union to be double-word aligned.
4595 As in the preceding examples, you can explicitly specify the alignment
4596 (in bytes) that you wish the compiler to use for a given variable or
4597 structure field. Alternatively, you can leave out the alignment factor
4598 and just ask the compiler to align a variable or field to the
4599 default alignment for the target architecture you are compiling for.
4600 The default alignment is sufficient for all scalar types, but may not be
4601 enough for all vector types on a target that supports vector operations.
4602 The default alignment is fixed for a particular target ABI.
4604 GCC also provides a target specific macro @code{__BIGGEST_ALIGNMENT__},
4605 which is the largest alignment ever used for any data type on the
4606 target machine you are compiling for. For example, you could write:
4609 short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__)));
4612 The compiler automatically sets the alignment for the declared
4613 variable or field to @code{__BIGGEST_ALIGNMENT__}. Doing this can
4614 often make copy operations more efficient, because the compiler can
4615 use whatever instructions copy the biggest chunks of memory when
4616 performing copies to or from the variables or fields that you have
4617 aligned this way. Note that the value of @code{__BIGGEST_ALIGNMENT__}
4618 may change depending on command-line options.
4620 When used on a struct, or struct member, the @code{aligned} attribute can
4621 only increase the alignment; in order to decrease it, the @code{packed}
4622 attribute must be specified as well. When used as part of a typedef, the
4623 @code{aligned} attribute can both increase and decrease alignment, and
4624 specifying the @code{packed} attribute generates a warning.
4626 Note that the effectiveness of @code{aligned} attributes may be limited
4627 by inherent limitations in your linker. On many systems, the linker is
4628 only able to arrange for variables to be aligned up to a certain maximum
4629 alignment. (For some linkers, the maximum supported alignment may
4630 be very very small.) If your linker is only able to align variables
4631 up to a maximum of 8-byte alignment, then specifying @code{aligned(16)}
4632 in an @code{__attribute__} still only provides you with 8-byte
4633 alignment. See your linker documentation for further information.
4635 The @code{aligned} attribute can also be used for functions
4636 (@pxref{Function Attributes}.)
4638 @item cleanup (@var{cleanup_function})
4639 @cindex @code{cleanup} attribute
4640 The @code{cleanup} attribute runs a function when the variable goes
4641 out of scope. This attribute can only be applied to auto function
4642 scope variables; it may not be applied to parameters or variables
4643 with static storage duration. The function must take one parameter,
4644 a pointer to a type compatible with the variable. The return value
4645 of the function (if any) is ignored.
4647 If @option{-fexceptions} is enabled, then @var{cleanup_function}
4648 is run during the stack unwinding that happens during the
4649 processing of the exception. Note that the @code{cleanup} attribute
4650 does not allow the exception to be caught, only to perform an action.
4651 It is undefined what happens if @var{cleanup_function} does not
4656 @cindex @code{common} attribute
4657 @cindex @code{nocommon} attribute
4660 The @code{common} attribute requests GCC to place a variable in
4661 ``common'' storage. The @code{nocommon} attribute requests the
4662 opposite---to allocate space for it directly.
4664 These attributes override the default chosen by the
4665 @option{-fno-common} and @option{-fcommon} flags respectively.
4668 @itemx deprecated (@var{msg})
4669 @cindex @code{deprecated} attribute
4670 The @code{deprecated} attribute results in a warning if the variable
4671 is used anywhere in the source file. This is useful when identifying
4672 variables that are expected to be removed in a future version of a
4673 program. The warning also includes the location of the declaration
4674 of the deprecated variable, to enable users to easily find further
4675 information about why the variable is deprecated, or what they should
4676 do instead. Note that the warning only occurs for uses:
4679 extern int old_var __attribute__ ((deprecated));
4681 int new_fn () @{ return old_var; @}
4685 results in a warning on line 3 but not line 2. The optional @var{msg}
4686 argument, which must be a string, is printed in the warning if
4689 The @code{deprecated} attribute can also be used for functions and
4690 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
4692 @item mode (@var{mode})
4693 @cindex @code{mode} attribute
4694 This attribute specifies the data type for the declaration---whichever
4695 type corresponds to the mode @var{mode}. This in effect lets you
4696 request an integer or floating-point type according to its width.
4698 You may also specify a mode of @code{byte} or @code{__byte__} to
4699 indicate the mode corresponding to a one-byte integer, @code{word} or
4700 @code{__word__} for the mode of a one-word integer, and @code{pointer}
4701 or @code{__pointer__} for the mode used to represent pointers.
4704 @cindex @code{packed} attribute
4705 The @code{packed} attribute specifies that a variable or structure field
4706 should have the smallest possible alignment---one byte for a variable,
4707 and one bit for a field, unless you specify a larger value with the
4708 @code{aligned} attribute.
4710 Here is a structure in which the field @code{x} is packed, so that it
4711 immediately follows @code{a}:
4717 int x[2] __attribute__ ((packed));
4721 @emph{Note:} The 4.1, 4.2 and 4.3 series of GCC ignore the
4722 @code{packed} attribute on bit-fields of type @code{char}. This has
4723 been fixed in GCC 4.4 but the change can lead to differences in the
4724 structure layout. See the documentation of
4725 @option{-Wpacked-bitfield-compat} for more information.
4727 @item section ("@var{section-name}")
4728 @cindex @code{section} variable attribute
4729 Normally, the compiler places the objects it generates in sections like
4730 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
4731 or you need certain particular variables to appear in special sections,
4732 for example to map to special hardware. The @code{section}
4733 attribute specifies that a variable (or function) lives in a particular
4734 section. For example, this small program uses several specific section names:
4737 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
4738 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
4739 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
4740 int init_data __attribute__ ((section ("INITDATA")));
4744 /* @r{Initialize stack pointer} */
4745 init_sp (stack + sizeof (stack));
4747 /* @r{Initialize initialized data} */
4748 memcpy (&init_data, &data, &edata - &data);
4750 /* @r{Turn on the serial ports} */
4757 Use the @code{section} attribute with
4758 @emph{global} variables and not @emph{local} variables,
4759 as shown in the example.
4761 You may use the @code{section} attribute with initialized or
4762 uninitialized global variables but the linker requires
4763 each object be defined once, with the exception that uninitialized
4764 variables tentatively go in the @code{common} (or @code{bss}) section
4765 and can be multiply ``defined''. Using the @code{section} attribute
4766 changes what section the variable goes into and may cause the
4767 linker to issue an error if an uninitialized variable has multiple
4768 definitions. You can force a variable to be initialized with the
4769 @option{-fno-common} flag or the @code{nocommon} attribute.
4771 Some file formats do not support arbitrary sections so the @code{section}
4772 attribute is not available on all platforms.
4773 If you need to map the entire contents of a module to a particular
4774 section, consider using the facilities of the linker instead.
4777 @cindex @code{shared} variable attribute
4778 On Microsoft Windows, in addition to putting variable definitions in a named
4779 section, the section can also be shared among all running copies of an
4780 executable or DLL@. For example, this small program defines shared data
4781 by putting it in a named section @code{shared} and marking the section
4785 int foo __attribute__((section ("shared"), shared)) = 0;
4790 /* @r{Read and write foo. All running
4791 copies see the same value.} */
4797 You may only use the @code{shared} attribute along with @code{section}
4798 attribute with a fully-initialized global definition because of the way
4799 linkers work. See @code{section} attribute for more information.
4801 The @code{shared} attribute is only available on Microsoft Windows@.
4803 @item tls_model ("@var{tls_model}")
4804 @cindex @code{tls_model} attribute
4805 The @code{tls_model} attribute sets thread-local storage model
4806 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
4807 overriding @option{-ftls-model=} command-line switch on a per-variable
4809 The @var{tls_model} argument should be one of @code{global-dynamic},
4810 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
4812 Not all targets support this attribute.
4815 This attribute, attached to a variable, means that the variable is meant
4816 to be possibly unused. GCC does not produce a warning for this
4820 This attribute, attached to a variable, means that the variable must be
4821 emitted even if it appears that the variable is not referenced.
4823 When applied to a static data member of a C++ class template, the
4824 attribute also means that the member is instantiated if the
4825 class itself is instantiated.
4827 @item vector_size (@var{bytes})
4828 This attribute specifies the vector size for the variable, measured in
4829 bytes. For example, the declaration:
4832 int foo __attribute__ ((vector_size (16)));
4836 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
4837 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
4838 4 units of 4 bytes), the corresponding mode of @code{foo} is V4SI@.
4840 This attribute is only applicable to integral and float scalars,
4841 although arrays, pointers, and function return values are allowed in
4842 conjunction with this construct.
4844 Aggregates with this attribute are invalid, even if they are of the same
4845 size as a corresponding scalar. For example, the declaration:
4848 struct S @{ int a; @};
4849 struct S __attribute__ ((vector_size (16))) foo;
4853 is invalid even if the size of the structure is the same as the size of
4857 The @code{selectany} attribute causes an initialized global variable to
4858 have link-once semantics. When multiple definitions of the variable are
4859 encountered by the linker, the first is selected and the remainder are
4860 discarded. Following usage by the Microsoft compiler, the linker is told
4861 @emph{not} to warn about size or content differences of the multiple
4864 Although the primary usage of this attribute is for POD types, the
4865 attribute can also be applied to global C++ objects that are initialized
4866 by a constructor. In this case, the static initialization and destruction
4867 code for the object is emitted in each translation defining the object,
4868 but the calls to the constructor and destructor are protected by a
4869 link-once guard variable.
4871 The @code{selectany} attribute is only available on Microsoft Windows
4872 targets. You can use @code{__declspec (selectany)} as a synonym for
4873 @code{__attribute__ ((selectany))} for compatibility with other
4877 The @code{weak} attribute is described in @ref{Function Attributes}.
4880 The @code{dllimport} attribute is described in @ref{Function Attributes}.
4883 The @code{dllexport} attribute is described in @ref{Function Attributes}.
4887 @anchor{AVR Variable Attributes}
4888 @subsection AVR Variable Attributes
4892 @cindex @code{progmem} AVR variable attribute
4893 The @code{progmem} attribute is used on the AVR to place read-only
4894 data in the non-volatile program memory (flash). The @code{progmem}
4895 attribute accomplishes this by putting respective variables into a
4896 section whose name starts with @code{.progmem}.
4898 This attribute works similar to the @code{section} attribute
4899 but adds additional checking. Notice that just like the
4900 @code{section} attribute, @code{progmem} affects the location
4901 of the data but not how this data is accessed.
4903 In order to read data located with the @code{progmem} attribute
4904 (inline) assembler must be used.
4906 /* Use custom macros from @w{@uref{http://nongnu.org/avr-libc/user-manual,AVR-LibC}} */
4907 #include <avr/pgmspace.h>
4909 /* Locate var in flash memory */
4910 const int var[2] PROGMEM = @{ 1, 2 @};
4912 int read_var (int i)
4914 /* Access var[] by accessor macro from avr/pgmspace.h */
4915 return (int) pgm_read_word (& var[i]);
4919 AVR is a Harvard architecture processor and data and read-only data
4920 normally resides in the data memory (RAM).
4922 See also the @ref{AVR Named Address Spaces} section for
4923 an alternate way to locate and access data in flash memory.
4926 @subsection Blackfin Variable Attributes
4928 Three attributes are currently defined for the Blackfin.
4934 @cindex @code{l1_data} variable attribute
4935 @cindex @code{l1_data_A} variable attribute
4936 @cindex @code{l1_data_B} variable attribute
4937 Use these attributes on the Blackfin to place the variable into L1 Data SRAM.
4938 Variables with @code{l1_data} attribute are put into the specific section
4939 named @code{.l1.data}. Those with @code{l1_data_A} attribute are put into
4940 the specific section named @code{.l1.data.A}. Those with @code{l1_data_B}
4941 attribute are put into the specific section named @code{.l1.data.B}.
4944 @cindex @code{l2} variable attribute
4945 Use this attribute on the Blackfin to place the variable into L2 SRAM.
4946 Variables with @code{l2} attribute are put into the specific section
4947 named @code{.l2.data}.
4950 @subsection M32R/D Variable Attributes
4952 One attribute is currently defined for the M32R/D@.
4955 @item model (@var{model-name})
4956 @cindex variable addressability on the M32R/D
4957 Use this attribute on the M32R/D to set the addressability of an object.
4958 The identifier @var{model-name} is one of @code{small}, @code{medium},
4959 or @code{large}, representing each of the code models.
4961 Small model objects live in the lower 16MB of memory (so that their
4962 addresses can be loaded with the @code{ld24} instruction).
4964 Medium and large model objects may live anywhere in the 32-bit address space
4965 (the compiler generates @code{seth/add3} instructions to load their
4969 @anchor{MeP Variable Attributes}
4970 @subsection MeP Variable Attributes
4972 The MeP target has a number of addressing modes and busses. The
4973 @code{near} space spans the standard memory space's first 16 megabytes
4974 (24 bits). The @code{far} space spans the entire 32-bit memory space.
4975 The @code{based} space is a 128-byte region in the memory space that
4976 is addressed relative to the @code{$tp} register. The @code{tiny}
4977 space is a 65536-byte region relative to the @code{$gp} register. In
4978 addition to these memory regions, the MeP target has a separate 16-bit
4979 control bus which is specified with @code{cb} attributes.
4984 Any variable with the @code{based} attribute is assigned to the
4985 @code{.based} section, and is accessed with relative to the
4986 @code{$tp} register.
4989 Likewise, the @code{tiny} attribute assigned variables to the
4990 @code{.tiny} section, relative to the @code{$gp} register.
4993 Variables with the @code{near} attribute are assumed to have addresses
4994 that fit in a 24-bit addressing mode. This is the default for large
4995 variables (@code{-mtiny=4} is the default) but this attribute can
4996 override @code{-mtiny=} for small variables, or override @code{-ml}.
4999 Variables with the @code{far} attribute are addressed using a full
5000 32-bit address. Since this covers the entire memory space, this
5001 allows modules to make no assumptions about where variables might be
5005 @itemx io (@var{addr})
5006 Variables with the @code{io} attribute are used to address
5007 memory-mapped peripherals. If an address is specified, the variable
5008 is assigned that address, else it is not assigned an address (it is
5009 assumed some other module assigns an address). Example:
5012 int timer_count __attribute__((io(0x123)));
5016 @itemx cb (@var{addr})
5017 Variables with the @code{cb} attribute are used to access the control
5018 bus, using special instructions. @code{addr} indicates the control bus
5022 int cpu_clock __attribute__((cb(0x123)));
5027 @anchor{i386 Variable Attributes}
5028 @subsection i386 Variable Attributes
5030 Two attributes are currently defined for i386 configurations:
5031 @code{ms_struct} and @code{gcc_struct}
5036 @cindex @code{ms_struct} attribute
5037 @cindex @code{gcc_struct} attribute
5039 If @code{packed} is used on a structure, or if bit-fields are used,
5040 it may be that the Microsoft ABI lays out the structure differently
5041 than the way GCC normally does. Particularly when moving packed
5042 data between functions compiled with GCC and the native Microsoft compiler
5043 (either via function call or as data in a file), it may be necessary to access
5046 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
5047 compilers to match the native Microsoft compiler.
5049 The Microsoft structure layout algorithm is fairly simple with the exception
5050 of the bit-field packing.
5051 The padding and alignment of members of structures and whether a bit-field
5052 can straddle a storage-unit boundary are determine by these rules:
5055 @item Structure members are stored sequentially in the order in which they are
5056 declared: the first member has the lowest memory address and the last member
5059 @item Every data object has an alignment requirement. The alignment requirement
5060 for all data except structures, unions, and arrays is either the size of the
5061 object or the current packing size (specified with either the
5062 @code{aligned} attribute or the @code{pack} pragma),
5063 whichever is less. For structures, unions, and arrays,
5064 the alignment requirement is the largest alignment requirement of its members.
5065 Every object is allocated an offset so that:
5068 offset % alignment_requirement == 0
5071 @item Adjacent bit-fields are packed into the same 1-, 2-, or 4-byte allocation
5072 unit if the integral types are the same size and if the next bit-field fits
5073 into the current allocation unit without crossing the boundary imposed by the
5074 common alignment requirements of the bit-fields.
5077 MSVC interprets zero-length bit-fields in the following ways:
5080 @item If a zero-length bit-field is inserted between two bit-fields that
5081 are normally coalesced, the bit-fields are not coalesced.
5088 unsigned long bf_1 : 12;
5090 unsigned long bf_2 : 12;
5095 The size of @code{t1} is 8 bytes with the zero-length bit-field. If the
5096 zero-length bit-field were removed, @code{t1}'s size would be 4 bytes.
5098 @item If a zero-length bit-field is inserted after a bit-field, @code{foo}, and the
5099 alignment of the zero-length bit-field is greater than the member that follows it,
5100 @code{bar}, @code{bar} is aligned as the type of the zero-length bit-field.
5121 For @code{t2}, @code{bar} is placed at offset 2, rather than offset 1.
5122 Accordingly, the size of @code{t2} is 4. For @code{t3}, the zero-length
5123 bit-field does not affect the alignment of @code{bar} or, as a result, the size
5126 Taking this into account, it is important to note the following:
5129 @item If a zero-length bit-field follows a normal bit-field, the type of the
5130 zero-length bit-field may affect the alignment of the structure as whole. For
5131 example, @code{t2} has a size of 4 bytes, since the zero-length bit-field follows a
5132 normal bit-field, and is of type short.
5134 @item Even if a zero-length bit-field is not followed by a normal bit-field, it may
5135 still affect the alignment of the structure:
5146 Here, @code{t4} takes up 4 bytes.
5149 @item Zero-length bit-fields following non-bit-field members are ignored:
5161 Here, @code{t5} takes up 2 bytes.
5165 @subsection PowerPC Variable Attributes
5167 Three attributes currently are defined for PowerPC configurations:
5168 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
5170 For full documentation of the struct attributes please see the
5171 documentation in @ref{i386 Variable Attributes}.
5173 For documentation of @code{altivec} attribute please see the
5174 documentation in @ref{PowerPC Type Attributes}.
5176 @subsection SPU Variable Attributes
5178 The SPU supports the @code{spu_vector} attribute for variables. For
5179 documentation of this attribute please see the documentation in
5180 @ref{SPU Type Attributes}.
5182 @subsection Xstormy16 Variable Attributes
5184 One attribute is currently defined for xstormy16 configurations:
5189 @cindex @code{below100} attribute
5191 If a variable has the @code{below100} attribute (@code{BELOW100} is
5192 allowed also), GCC places the variable in the first 0x100 bytes of
5193 memory and use special opcodes to access it. Such variables are
5194 placed in either the @code{.bss_below100} section or the
5195 @code{.data_below100} section.
5199 @node Type Attributes
5200 @section Specifying Attributes of Types
5201 @cindex attribute of types
5202 @cindex type attributes
5204 The keyword @code{__attribute__} allows you to specify special
5205 attributes of @code{struct} and @code{union} types when you define
5206 such types. This keyword is followed by an attribute specification
5207 inside double parentheses. Seven attributes are currently defined for
5208 types: @code{aligned}, @code{packed}, @code{transparent_union},
5209 @code{unused}, @code{deprecated}, @code{visibility}, and
5210 @code{may_alias}. Other attributes are defined for functions
5211 (@pxref{Function Attributes}) and for variables (@pxref{Variable
5214 You may also specify any one of these attributes with @samp{__}
5215 preceding and following its keyword. This allows you to use these
5216 attributes in header files without being concerned about a possible
5217 macro of the same name. For example, you may use @code{__aligned__}
5218 instead of @code{aligned}.
5220 You may specify type attributes in an enum, struct or union type
5221 declaration or definition, or for other types in a @code{typedef}
5224 For an enum, struct or union type, you may specify attributes either
5225 between the enum, struct or union tag and the name of the type, or
5226 just past the closing curly brace of the @emph{definition}. The
5227 former syntax is preferred.
5229 @xref{Attribute Syntax}, for details of the exact syntax for using
5233 @cindex @code{aligned} attribute
5234 @item aligned (@var{alignment})
5235 This attribute specifies a minimum alignment (in bytes) for variables
5236 of the specified type. For example, the declarations:
5239 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
5240 typedef int more_aligned_int __attribute__ ((aligned (8)));
5244 force the compiler to ensure (as far as it can) that each variable whose
5245 type is @code{struct S} or @code{more_aligned_int} is allocated and
5246 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
5247 variables of type @code{struct S} aligned to 8-byte boundaries allows
5248 the compiler to use the @code{ldd} and @code{std} (doubleword load and
5249 store) instructions when copying one variable of type @code{struct S} to
5250 another, thus improving run-time efficiency.
5252 Note that the alignment of any given @code{struct} or @code{union} type
5253 is required by the ISO C standard to be at least a perfect multiple of
5254 the lowest common multiple of the alignments of all of the members of
5255 the @code{struct} or @code{union} in question. This means that you @emph{can}
5256 effectively adjust the alignment of a @code{struct} or @code{union}
5257 type by attaching an @code{aligned} attribute to any one of the members
5258 of such a type, but the notation illustrated in the example above is a
5259 more obvious, intuitive, and readable way to request the compiler to
5260 adjust the alignment of an entire @code{struct} or @code{union} type.
5262 As in the preceding example, you can explicitly specify the alignment
5263 (in bytes) that you wish the compiler to use for a given @code{struct}
5264 or @code{union} type. Alternatively, you can leave out the alignment factor
5265 and just ask the compiler to align a type to the maximum
5266 useful alignment for the target machine you are compiling for. For
5267 example, you could write:
5270 struct S @{ short f[3]; @} __attribute__ ((aligned));
5273 Whenever you leave out the alignment factor in an @code{aligned}
5274 attribute specification, the compiler automatically sets the alignment
5275 for the type to the largest alignment that is ever used for any data
5276 type on the target machine you are compiling for. Doing this can often
5277 make copy operations more efficient, because the compiler can use
5278 whatever instructions copy the biggest chunks of memory when performing
5279 copies to or from the variables that have types that you have aligned
5282 In the example above, if the size of each @code{short} is 2 bytes, then
5283 the size of the entire @code{struct S} type is 6 bytes. The smallest
5284 power of two that is greater than or equal to that is 8, so the
5285 compiler sets the alignment for the entire @code{struct S} type to 8
5288 Note that although you can ask the compiler to select a time-efficient
5289 alignment for a given type and then declare only individual stand-alone
5290 objects of that type, the compiler's ability to select a time-efficient
5291 alignment is primarily useful only when you plan to create arrays of
5292 variables having the relevant (efficiently aligned) type. If you
5293 declare or use arrays of variables of an efficiently-aligned type, then
5294 it is likely that your program also does pointer arithmetic (or
5295 subscripting, which amounts to the same thing) on pointers to the
5296 relevant type, and the code that the compiler generates for these
5297 pointer arithmetic operations is often more efficient for
5298 efficiently-aligned types than for other types.
5300 The @code{aligned} attribute can only increase the alignment; but you
5301 can decrease it by specifying @code{packed} as well. See below.
5303 Note that the effectiveness of @code{aligned} attributes may be limited
5304 by inherent limitations in your linker. On many systems, the linker is
5305 only able to arrange for variables to be aligned up to a certain maximum
5306 alignment. (For some linkers, the maximum supported alignment may
5307 be very very small.) If your linker is only able to align variables
5308 up to a maximum of 8-byte alignment, then specifying @code{aligned(16)}
5309 in an @code{__attribute__} still only provides you with 8-byte
5310 alignment. See your linker documentation for further information.
5313 This attribute, attached to @code{struct} or @code{union} type
5314 definition, specifies that each member (other than zero-width bit-fields)
5315 of the structure or union is placed to minimize the memory required. When
5316 attached to an @code{enum} definition, it indicates that the smallest
5317 integral type should be used.
5319 @opindex fshort-enums
5320 Specifying this attribute for @code{struct} and @code{union} types is
5321 equivalent to specifying the @code{packed} attribute on each of the
5322 structure or union members. Specifying the @option{-fshort-enums}
5323 flag on the line is equivalent to specifying the @code{packed}
5324 attribute on all @code{enum} definitions.
5326 In the following example @code{struct my_packed_struct}'s members are
5327 packed closely together, but the internal layout of its @code{s} member
5328 is not packed---to do that, @code{struct my_unpacked_struct} needs to
5332 struct my_unpacked_struct
5338 struct __attribute__ ((__packed__)) my_packed_struct
5342 struct my_unpacked_struct s;
5346 You may only specify this attribute on the definition of an @code{enum},
5347 @code{struct} or @code{union}, not on a @code{typedef} that does not
5348 also define the enumerated type, structure or union.
5350 @item transparent_union
5351 This attribute, attached to a @code{union} type definition, indicates
5352 that any function parameter having that union type causes calls to that
5353 function to be treated in a special way.
5355 First, the argument corresponding to a transparent union type can be of
5356 any type in the union; no cast is required. Also, if the union contains
5357 a pointer type, the corresponding argument can be a null pointer
5358 constant or a void pointer expression; and if the union contains a void
5359 pointer type, the corresponding argument can be any pointer expression.
5360 If the union member type is a pointer, qualifiers like @code{const} on
5361 the referenced type must be respected, just as with normal pointer
5364 Second, the argument is passed to the function using the calling
5365 conventions of the first member of the transparent union, not the calling
5366 conventions of the union itself. All members of the union must have the
5367 same machine representation; this is necessary for this argument passing
5370 Transparent unions are designed for library functions that have multiple
5371 interfaces for compatibility reasons. For example, suppose the
5372 @code{wait} function must accept either a value of type @code{int *} to
5373 comply with POSIX, or a value of type @code{union wait *} to comply with
5374 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
5375 @code{wait} would accept both kinds of arguments, but it would also
5376 accept any other pointer type and this would make argument type checking
5377 less useful. Instead, @code{<sys/wait.h>} might define the interface
5381 typedef union __attribute__ ((__transparent_union__))
5385 @} wait_status_ptr_t;
5387 pid_t wait (wait_status_ptr_t);
5391 This interface allows either @code{int *} or @code{union wait *}
5392 arguments to be passed, using the @code{int *} calling convention.
5393 The program can call @code{wait} with arguments of either type:
5396 int w1 () @{ int w; return wait (&w); @}
5397 int w2 () @{ union wait w; return wait (&w); @}
5401 With this interface, @code{wait}'s implementation might look like this:
5404 pid_t wait (wait_status_ptr_t p)
5406 return waitpid (-1, p.__ip, 0);
5411 When attached to a type (including a @code{union} or a @code{struct}),
5412 this attribute means that variables of that type are meant to appear
5413 possibly unused. GCC does not produce a warning for any variables of
5414 that type, even if the variable appears to do nothing. This is often
5415 the case with lock or thread classes, which are usually defined and then
5416 not referenced, but contain constructors and destructors that have
5417 nontrivial bookkeeping functions.
5420 @itemx deprecated (@var{msg})
5421 The @code{deprecated} attribute results in a warning if the type
5422 is used anywhere in the source file. This is useful when identifying
5423 types that are expected to be removed in a future version of a program.
5424 If possible, the warning also includes the location of the declaration
5425 of the deprecated type, to enable users to easily find further
5426 information about why the type is deprecated, or what they should do
5427 instead. Note that the warnings only occur for uses and then only
5428 if the type is being applied to an identifier that itself is not being
5429 declared as deprecated.
5432 typedef int T1 __attribute__ ((deprecated));
5436 typedef T1 T3 __attribute__ ((deprecated));
5437 T3 z __attribute__ ((deprecated));
5441 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
5442 warning is issued for line 4 because T2 is not explicitly
5443 deprecated. Line 5 has no warning because T3 is explicitly
5444 deprecated. Similarly for line 6. The optional @var{msg}
5445 argument, which must be a string, is printed in the warning if
5448 The @code{deprecated} attribute can also be used for functions and
5449 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
5452 Accesses through pointers to types with this attribute are not subject
5453 to type-based alias analysis, but are instead assumed to be able to alias
5454 any other type of objects.
5455 In the context of section 6.5 paragraph 7 of the C99 standard,
5456 an lvalue expression
5457 dereferencing such a pointer is treated like having a character type.
5458 See @option{-fstrict-aliasing} for more information on aliasing issues.
5459 This extension exists to support some vector APIs, in which pointers to
5460 one vector type are permitted to alias pointers to a different vector type.
5462 Note that an object of a type with this attribute does not have any
5468 typedef short __attribute__((__may_alias__)) short_a;
5474 short_a *b = (short_a *) &a;
5478 if (a == 0x12345678)
5486 If you replaced @code{short_a} with @code{short} in the variable
5487 declaration, the above program would abort when compiled with
5488 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
5489 above in recent GCC versions.
5492 In C++, attribute visibility (@pxref{Function Attributes}) can also be
5493 applied to class, struct, union and enum types. Unlike other type
5494 attributes, the attribute must appear between the initial keyword and
5495 the name of the type; it cannot appear after the body of the type.
5497 Note that the type visibility is applied to vague linkage entities
5498 associated with the class (vtable, typeinfo node, etc.). In
5499 particular, if a class is thrown as an exception in one shared object
5500 and caught in another, the class must have default visibility.
5501 Otherwise the two shared objects are unable to use the same
5502 typeinfo node and exception handling will break.
5506 To specify multiple attributes, separate them by commas within the
5507 double parentheses: for example, @samp{__attribute__ ((aligned (16),
5510 @subsection ARM Type Attributes
5512 On those ARM targets that support @code{dllimport} (such as Symbian
5513 OS), you can use the @code{notshared} attribute to indicate that the
5514 virtual table and other similar data for a class should not be
5515 exported from a DLL@. For example:
5518 class __declspec(notshared) C @{
5520 __declspec(dllimport) C();
5524 __declspec(dllexport)
5529 In this code, @code{C::C} is exported from the current DLL, but the
5530 virtual table for @code{C} is not exported. (You can use
5531 @code{__attribute__} instead of @code{__declspec} if you prefer, but
5532 most Symbian OS code uses @code{__declspec}.)
5534 @anchor{MeP Type Attributes}
5535 @subsection MeP Type Attributes
5537 Many of the MeP variable attributes may be applied to types as well.
5538 Specifically, the @code{based}, @code{tiny}, @code{near}, and
5539 @code{far} attributes may be applied to either. The @code{io} and
5540 @code{cb} attributes may not be applied to types.
5542 @anchor{i386 Type Attributes}
5543 @subsection i386 Type Attributes
5545 Two attributes are currently defined for i386 configurations:
5546 @code{ms_struct} and @code{gcc_struct}.
5552 @cindex @code{ms_struct}
5553 @cindex @code{gcc_struct}
5555 If @code{packed} is used on a structure, or if bit-fields are used
5556 it may be that the Microsoft ABI packs them differently
5557 than GCC normally packs them. Particularly when moving packed
5558 data between functions compiled with GCC and the native Microsoft compiler
5559 (either via function call or as data in a file), it may be necessary to access
5562 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
5563 compilers to match the native Microsoft compiler.
5566 @anchor{PowerPC Type Attributes}
5567 @subsection PowerPC Type Attributes
5569 Three attributes currently are defined for PowerPC configurations:
5570 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
5572 For full documentation of the @code{ms_struct} and @code{gcc_struct}
5573 attributes please see the documentation in @ref{i386 Type Attributes}.
5575 The @code{altivec} attribute allows one to declare AltiVec vector data
5576 types supported by the AltiVec Programming Interface Manual. The
5577 attribute requires an argument to specify one of three vector types:
5578 @code{vector__}, @code{pixel__} (always followed by unsigned short),
5579 and @code{bool__} (always followed by unsigned).
5582 __attribute__((altivec(vector__)))
5583 __attribute__((altivec(pixel__))) unsigned short
5584 __attribute__((altivec(bool__))) unsigned
5587 These attributes mainly are intended to support the @code{__vector},
5588 @code{__pixel}, and @code{__bool} AltiVec keywords.
5590 @anchor{SPU Type Attributes}
5591 @subsection SPU Type Attributes
5593 The SPU supports the @code{spu_vector} attribute for types. This attribute
5594 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
5595 Language Extensions Specification. It is intended to support the
5596 @code{__vector} keyword.
5599 @section Inquiring on Alignment of Types or Variables
5601 @cindex type alignment
5602 @cindex variable alignment
5604 The keyword @code{__alignof__} allows you to inquire about how an object
5605 is aligned, or the minimum alignment usually required by a type. Its
5606 syntax is just like @code{sizeof}.
5608 For example, if the target machine requires a @code{double} value to be
5609 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
5610 This is true on many RISC machines. On more traditional machine
5611 designs, @code{__alignof__ (double)} is 4 or even 2.
5613 Some machines never actually require alignment; they allow reference to any
5614 data type even at an odd address. For these machines, @code{__alignof__}
5615 reports the smallest alignment that GCC gives the data type, usually as
5616 mandated by the target ABI.
5618 If the operand of @code{__alignof__} is an lvalue rather than a type,
5619 its value is the required alignment for its type, taking into account
5620 any minimum alignment specified with GCC's @code{__attribute__}
5621 extension (@pxref{Variable Attributes}). For example, after this
5625 struct foo @{ int x; char y; @} foo1;
5629 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
5630 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
5632 It is an error to ask for the alignment of an incomplete type.
5636 @section An Inline Function is As Fast As a Macro
5637 @cindex inline functions
5638 @cindex integrating function code
5640 @cindex macros, inline alternative
5642 By declaring a function inline, you can direct GCC to make
5643 calls to that function faster. One way GCC can achieve this is to
5644 integrate that function's code into the code for its callers. This
5645 makes execution faster by eliminating the function-call overhead; in
5646 addition, if any of the actual argument values are constant, their
5647 known values may permit simplifications at compile time so that not
5648 all of the inline function's code needs to be included. The effect on
5649 code size is less predictable; object code may be larger or smaller
5650 with function inlining, depending on the particular case. You can
5651 also direct GCC to try to integrate all ``simple enough'' functions
5652 into their callers with the option @option{-finline-functions}.
5654 GCC implements three different semantics of declaring a function
5655 inline. One is available with @option{-std=gnu89} or
5656 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
5657 on all inline declarations, another when
5658 @option{-std=c99}, @option{-std=c11},
5659 @option{-std=gnu99} or @option{-std=gnu11}
5660 (without @option{-fgnu89-inline}), and the third
5661 is used when compiling C++.
5663 To declare a function inline, use the @code{inline} keyword in its
5664 declaration, like this:
5674 If you are writing a header file to be included in ISO C90 programs, write
5675 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
5677 The three types of inlining behave similarly in two important cases:
5678 when the @code{inline} keyword is used on a @code{static} function,
5679 like the example above, and when a function is first declared without
5680 using the @code{inline} keyword and then is defined with
5681 @code{inline}, like this:
5684 extern int inc (int *a);
5692 In both of these common cases, the program behaves the same as if you
5693 had not used the @code{inline} keyword, except for its speed.
5695 @cindex inline functions, omission of
5696 @opindex fkeep-inline-functions
5697 When a function is both inline and @code{static}, if all calls to the
5698 function are integrated into the caller, and the function's address is
5699 never used, then the function's own assembler code is never referenced.
5700 In this case, GCC does not actually output assembler code for the
5701 function, unless you specify the option @option{-fkeep-inline-functions}.
5702 Some calls cannot be integrated for various reasons (in particular,
5703 calls that precede the function's definition cannot be integrated, and
5704 neither can recursive calls within the definition). If there is a
5705 nonintegrated call, then the function is compiled to assembler code as
5706 usual. The function must also be compiled as usual if the program
5707 refers to its address, because that can't be inlined.
5710 Note that certain usages in a function definition can make it unsuitable
5711 for inline substitution. Among these usages are: variadic functions, use of
5712 @code{alloca}, use of variable-length data types (@pxref{Variable Length}),
5713 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
5714 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
5715 warns when a function marked @code{inline} could not be substituted,
5716 and gives the reason for the failure.
5718 @cindex automatic @code{inline} for C++ member fns
5719 @cindex @code{inline} automatic for C++ member fns
5720 @cindex member fns, automatically @code{inline}
5721 @cindex C++ member fns, automatically @code{inline}
5722 @opindex fno-default-inline
5723 As required by ISO C++, GCC considers member functions defined within
5724 the body of a class to be marked inline even if they are
5725 not explicitly declared with the @code{inline} keyword. You can
5726 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
5727 Options,,Options Controlling C++ Dialect}.
5729 GCC does not inline any functions when not optimizing unless you specify
5730 the @samp{always_inline} attribute for the function, like this:
5733 /* @r{Prototype.} */
5734 inline void foo (const char) __attribute__((always_inline));
5737 The remainder of this section is specific to GNU C90 inlining.
5739 @cindex non-static inline function
5740 When an inline function is not @code{static}, then the compiler must assume
5741 that there may be calls from other source files; since a global symbol can
5742 be defined only once in any program, the function must not be defined in
5743 the other source files, so the calls therein cannot be integrated.
5744 Therefore, a non-@code{static} inline function is always compiled on its
5745 own in the usual fashion.
5747 If you specify both @code{inline} and @code{extern} in the function
5748 definition, then the definition is used only for inlining. In no case
5749 is the function compiled on its own, not even if you refer to its
5750 address explicitly. Such an address becomes an external reference, as
5751 if you had only declared the function, and had not defined it.
5753 This combination of @code{inline} and @code{extern} has almost the
5754 effect of a macro. The way to use it is to put a function definition in
5755 a header file with these keywords, and put another copy of the
5756 definition (lacking @code{inline} and @code{extern}) in a library file.
5757 The definition in the header file causes most calls to the function
5758 to be inlined. If any uses of the function remain, they refer to
5759 the single copy in the library.
5762 @section When is a Volatile Object Accessed?
5763 @cindex accessing volatiles
5764 @cindex volatile read
5765 @cindex volatile write
5766 @cindex volatile access
5768 C has the concept of volatile objects. These are normally accessed by
5769 pointers and used for accessing hardware or inter-thread
5770 communication. The standard encourages compilers to refrain from
5771 optimizations concerning accesses to volatile objects, but leaves it
5772 implementation defined as to what constitutes a volatile access. The
5773 minimum requirement is that at a sequence point all previous accesses
5774 to volatile objects have stabilized and no subsequent accesses have
5775 occurred. Thus an implementation is free to reorder and combine
5776 volatile accesses that occur between sequence points, but cannot do
5777 so for accesses across a sequence point. The use of volatile does
5778 not allow you to violate the restriction on updating objects multiple
5779 times between two sequence points.
5781 Accesses to non-volatile objects are not ordered with respect to
5782 volatile accesses. You cannot use a volatile object as a memory
5783 barrier to order a sequence of writes to non-volatile memory. For
5787 int *ptr = @var{something};
5789 *ptr = @var{something};
5794 Unless @var{*ptr} and @var{vobj} can be aliased, it is not guaranteed
5795 that the write to @var{*ptr} occurs by the time the update
5796 of @var{vobj} happens. If you need this guarantee, you must use
5797 a stronger memory barrier such as:
5800 int *ptr = @var{something};
5802 *ptr = @var{something};
5803 asm volatile ("" : : : "memory");
5807 A scalar volatile object is read when it is accessed in a void context:
5810 volatile int *src = @var{somevalue};
5814 Such expressions are rvalues, and GCC implements this as a
5815 read of the volatile object being pointed to.
5817 Assignments are also expressions and have an rvalue. However when
5818 assigning to a scalar volatile, the volatile object is not reread,
5819 regardless of whether the assignment expression's rvalue is used or
5820 not. If the assignment's rvalue is used, the value is that assigned
5821 to the volatile object. For instance, there is no read of @var{vobj}
5822 in all the following cases:
5827 vobj = @var{something};
5828 obj = vobj = @var{something};
5829 obj ? vobj = @var{onething} : vobj = @var{anotherthing};
5830 obj = (@var{something}, vobj = @var{anotherthing});
5833 If you need to read the volatile object after an assignment has
5834 occurred, you must use a separate expression with an intervening
5837 As bit-fields are not individually addressable, volatile bit-fields may
5838 be implicitly read when written to, or when adjacent bit-fields are
5839 accessed. Bit-field operations may be optimized such that adjacent
5840 bit-fields are only partially accessed, if they straddle a storage unit
5841 boundary. For these reasons it is unwise to use volatile bit-fields to
5845 @section Assembler Instructions with C Expression Operands
5846 @cindex extended @code{asm}
5847 @cindex @code{asm} expressions
5848 @cindex assembler instructions
5851 In an assembler instruction using @code{asm}, you can specify the
5852 operands of the instruction using C expressions. This means you need not
5853 guess which registers or memory locations contain the data you want
5856 You must specify an assembler instruction template much like what
5857 appears in a machine description, plus an operand constraint string for
5860 For example, here is how to use the 68881's @code{fsinx} instruction:
5863 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
5867 Here @code{angle} is the C expression for the input operand while
5868 @code{result} is that of the output operand. Each has @samp{"f"} as its
5869 operand constraint, saying that a floating-point register is required.
5870 The @samp{=} in @samp{=f} indicates that the operand is an output; all
5871 output operands' constraints must use @samp{=}. The constraints use the
5872 same language used in the machine description (@pxref{Constraints}).
5874 Each operand is described by an operand-constraint string followed by
5875 the C expression in parentheses. A colon separates the assembler
5876 template from the first output operand and another separates the last
5877 output operand from the first input, if any. Commas separate the
5878 operands within each group. The total number of operands is currently
5879 limited to 30; this limitation may be lifted in some future version of
5882 If there are no output operands but there are input operands, you must
5883 place two consecutive colons surrounding the place where the output
5886 As of GCC version 3.1, it is also possible to specify input and output
5887 operands using symbolic names which can be referenced within the
5888 assembler code. These names are specified inside square brackets
5889 preceding the constraint string, and can be referenced inside the
5890 assembler code using @code{%[@var{name}]} instead of a percentage sign
5891 followed by the operand number. Using named operands the above example
5895 asm ("fsinx %[angle],%[output]"
5896 : [output] "=f" (result)
5897 : [angle] "f" (angle));
5901 Note that the symbolic operand names have no relation whatsoever to
5902 other C identifiers. You may use any name you like, even those of
5903 existing C symbols, but you must ensure that no two operands within the same
5904 assembler construct use the same symbolic name.
5906 Output operand expressions must be lvalues; the compiler can check this.
5907 The input operands need not be lvalues. The compiler cannot check
5908 whether the operands have data types that are reasonable for the
5909 instruction being executed. It does not parse the assembler instruction
5910 template and does not know what it means or even whether it is valid
5911 assembler input. The extended @code{asm} feature is most often used for
5912 machine instructions the compiler itself does not know exist. If
5913 the output expression cannot be directly addressed (for example, it is a
5914 bit-field), your constraint must allow a register. In that case, GCC
5915 uses the register as the output of the @code{asm}, and then stores
5916 that register into the output.
5918 The ordinary output operands must be write-only; GCC assumes that
5919 the values in these operands before the instruction are dead and need
5920 not be generated. Extended asm supports input-output or read-write
5921 operands. Use the constraint character @samp{+} to indicate such an
5922 operand and list it with the output operands.
5924 You may, as an alternative, logically split its function into two
5925 separate operands, one input operand and one write-only output
5926 operand. The connection between them is expressed by constraints
5927 that say they need to be in the same location when the instruction
5928 executes. You can use the same C expression for both operands, or
5929 different expressions. For example, here we write the (fictitious)
5930 @samp{combine} instruction with @code{bar} as its read-only source
5931 operand and @code{foo} as its read-write destination:
5934 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
5938 The constraint @samp{"0"} for operand 1 says that it must occupy the
5939 same location as operand 0. A number in constraint is allowed only in
5940 an input operand and it must refer to an output operand.
5942 Only a number in the constraint can guarantee that one operand is in
5943 the same place as another. The mere fact that @code{foo} is the value
5944 of both operands is not enough to guarantee that they are in the
5945 same place in the generated assembler code. The following does not
5949 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
5952 Various optimizations or reloading could cause operands 0 and 1 to be in
5953 different registers; GCC knows no reason not to do so. For example, the
5954 compiler might find a copy of the value of @code{foo} in one register and
5955 use it for operand 1, but generate the output operand 0 in a different
5956 register (copying it afterward to @code{foo}'s own address). Of course,
5957 since the register for operand 1 is not even mentioned in the assembler
5958 code, the result will not work, but GCC can't tell that.
5960 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
5961 the operand number for a matching constraint. For example:
5964 asm ("cmoveq %1,%2,%[result]"
5965 : [result] "=r"(result)
5966 : "r" (test), "r"(new), "[result]"(old));
5969 Sometimes you need to make an @code{asm} operand be a specific register,
5970 but there's no matching constraint letter for that register @emph{by
5971 itself}. To force the operand into that register, use a local variable
5972 for the operand and specify the register in the variable declaration.
5973 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
5974 register constraint letter that matches the register:
5977 register int *p1 asm ("r0") = @dots{};
5978 register int *p2 asm ("r1") = @dots{};
5979 register int *result asm ("r0");
5980 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
5983 @anchor{Example of asm with clobbered asm reg}
5984 In the above example, beware that a register that is call-clobbered by
5985 the target ABI will be overwritten by any function call in the
5986 assignment, including library calls for arithmetic operators.
5987 Also a register may be clobbered when generating some operations,
5988 like variable shift, memory copy or memory move on x86.
5989 Assuming it is a call-clobbered register, this may happen to @code{r0}
5990 above by the assignment to @code{p2}. If you have to use such a
5991 register, use temporary variables for expressions between the register
5996 register int *p1 asm ("r0") = @dots{};
5997 register int *p2 asm ("r1") = t1;
5998 register int *result asm ("r0");
5999 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
6002 Some instructions clobber specific hard registers. To describe this,
6003 write a third colon after the input operands, followed by the names of
6004 the clobbered hard registers (given as strings). Here is a realistic
6005 example for the VAX:
6008 asm volatile ("movc3 %0,%1,%2"
6009 : /* @r{no outputs} */
6010 : "g" (from), "g" (to), "g" (count)
6011 : "r0", "r1", "r2", "r3", "r4", "r5");
6014 You may not write a clobber description in a way that overlaps with an
6015 input or output operand. For example, you may not have an operand
6016 describing a register class with one member if you mention that register
6017 in the clobber list. Variables declared to live in specific registers
6018 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
6019 have no part mentioned in the clobber description.
6020 There is no way for you to specify that an input
6021 operand is modified without also specifying it as an output
6022 operand. Note that if all the output operands you specify are for this
6023 purpose (and hence unused), you then also need to specify
6024 @code{volatile} for the @code{asm} construct, as described below, to
6025 prevent GCC from deleting the @code{asm} statement as unused.
6027 If you refer to a particular hardware register from the assembler code,
6028 you probably have to list the register after the third colon to
6029 tell the compiler the register's value is modified. In some assemblers,
6030 the register names begin with @samp{%}; to produce one @samp{%} in the
6031 assembler code, you must write @samp{%%} in the input.
6033 If your assembler instruction can alter the condition code register, add
6034 @samp{cc} to the list of clobbered registers. GCC on some machines
6035 represents the condition codes as a specific hardware register;
6036 @samp{cc} serves to name this register. On other machines, the
6037 condition code is handled differently, and specifying @samp{cc} has no
6038 effect. But it is valid no matter what the machine.
6040 If your assembler instructions access memory in an unpredictable
6041 fashion, add @samp{memory} to the list of clobbered registers. This
6042 causes GCC to not keep memory values cached in registers across the
6043 assembler instruction and not optimize stores or loads to that memory.
6044 You also should add the @code{volatile} keyword if the memory
6045 affected is not listed in the inputs or outputs of the @code{asm}, as
6046 the @samp{memory} clobber does not count as a side-effect of the
6047 @code{asm}. If you know how large the accessed memory is, you can add
6048 it as input or output but if this is not known, you should add
6049 @samp{memory}. As an example, if you access ten bytes of a string, you
6050 can use a memory input like:
6053 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
6056 Note that in the following example the memory input is necessary,
6057 otherwise GCC might optimize the store to @code{x} away:
6064 asm ("magic stuff accessing an 'int' pointed to by '%1'"
6065 : "=&d" (r) : "a" (y), "m" (*y));
6070 You can put multiple assembler instructions together in a single
6071 @code{asm} template, separated by the characters normally used in assembly
6072 code for the system. A combination that works in most places is a newline
6073 to break the line, plus a tab character to move to the instruction field
6074 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
6075 assembler allows semicolons as a line-breaking character. Note that some
6076 assembler dialects use semicolons to start a comment.
6077 The input operands are guaranteed not to use any of the clobbered
6078 registers, and neither do the output operands' addresses, so you can
6079 read and write the clobbered registers as many times as you like. Here
6080 is an example of multiple instructions in a template; it assumes the
6081 subroutine @code{_foo} accepts arguments in registers 9 and 10:
6084 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
6086 : "g" (from), "g" (to)
6090 Unless an output operand has the @samp{&} constraint modifier, GCC
6091 may allocate it in the same register as an unrelated input operand, on
6092 the assumption the inputs are consumed before the outputs are produced.
6093 This assumption may be false if the assembler code actually consists of
6094 more than one instruction. In such a case, use @samp{&} for each output
6095 operand that may not overlap an input. @xref{Modifiers}.
6097 If you want to test the condition code produced by an assembler
6098 instruction, you must include a branch and a label in the @code{asm}
6099 construct, as follows:
6102 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
6108 This assumes your assembler supports local labels, as the GNU assembler
6109 and most Unix assemblers do.
6111 Speaking of labels, jumps from one @code{asm} to another are not
6112 supported. The compiler's optimizers do not know about these jumps, and
6113 therefore they cannot take account of them when deciding how to
6114 optimize. @xref{Extended asm with goto}.
6116 @cindex macros containing @code{asm}
6117 Usually the most convenient way to use these @code{asm} instructions is to
6118 encapsulate them in macros that look like functions. For example,
6122 (@{ double __value, __arg = (x); \
6123 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
6128 Here the variable @code{__arg} is used to make sure that the instruction
6129 operates on a proper @code{double} value, and to accept only those
6130 arguments @code{x} that can convert automatically to a @code{double}.
6132 Another way to make sure the instruction operates on the correct data
6133 type is to use a cast in the @code{asm}. This is different from using a
6134 variable @code{__arg} in that it converts more different types. For
6135 example, if the desired type is @code{int}, casting the argument to
6136 @code{int} accepts a pointer with no complaint, while assigning the
6137 argument to an @code{int} variable named @code{__arg} warns about
6138 using a pointer unless the caller explicitly casts it.
6140 If an @code{asm} has output operands, GCC assumes for optimization
6141 purposes the instruction has no side effects except to change the output
6142 operands. This does not mean instructions with a side effect cannot be
6143 used, but you must be careful, because the compiler may eliminate them
6144 if the output operands aren't used, or move them out of loops, or
6145 replace two with one if they constitute a common subexpression. Also,
6146 if your instruction does have a side effect on a variable that otherwise
6147 appears not to change, the old value of the variable may be reused later
6148 if it happens to be found in a register.
6150 You can prevent an @code{asm} instruction from being deleted
6151 by writing the keyword @code{volatile} after
6152 the @code{asm}. For example:
6155 #define get_and_set_priority(new) \
6157 asm volatile ("get_and_set_priority %0, %1" \
6158 : "=g" (__old) : "g" (new)); \
6163 The @code{volatile} keyword indicates that the instruction has
6164 important side-effects. GCC does not delete a volatile @code{asm} if
6165 it is reachable. (The instruction can still be deleted if GCC can
6166 prove that control flow never reaches the location of the
6167 instruction.) Note that even a volatile @code{asm} instruction
6168 can be moved relative to other code, including across jump
6169 instructions. For example, on many targets there is a system
6170 register that can be set to control the rounding mode of
6171 floating-point operations. You might try
6172 setting it with a volatile @code{asm}, like this PowerPC example:
6175 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
6180 This does not work reliably, as the compiler may move the addition back
6181 before the volatile @code{asm}. To make it work you need to add an
6182 artificial dependency to the @code{asm} referencing a variable in the code
6183 you don't want moved, for example:
6186 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
6190 Similarly, you can't expect a
6191 sequence of volatile @code{asm} instructions to remain perfectly
6192 consecutive. If you want consecutive output, use a single @code{asm}.
6193 Also, GCC performs some optimizations across a volatile @code{asm}
6194 instruction; GCC does not ``forget everything'' when it encounters
6195 a volatile @code{asm} instruction the way some other compilers do.
6197 An @code{asm} instruction without any output operands is treated
6198 identically to a volatile @code{asm} instruction.
6200 It is a natural idea to look for a way to give access to the condition
6201 code left by the assembler instruction. However, when we attempted to
6202 implement this, we found no way to make it work reliably. The problem
6203 is that output operands might need reloading, which result in
6204 additional following ``store'' instructions. On most machines, these
6205 instructions alter the condition code before there is time to
6206 test it. This problem doesn't arise for ordinary ``test'' and
6207 ``compare'' instructions because they don't have any output operands.
6209 For reasons similar to those described above, it is not possible to give
6210 an assembler instruction access to the condition code left by previous
6213 @anchor{Extended asm with goto}
6214 As of GCC version 4.5, @code{asm goto} may be used to have the assembly
6215 jump to one or more C labels. In this form, a fifth section after the
6216 clobber list contains a list of all C labels to which the assembly may jump.
6217 Each label operand is implicitly self-named. The @code{asm} is also assumed
6218 to fall through to the next statement.
6220 This form of @code{asm} is restricted to not have outputs. This is due
6221 to a internal restriction in the compiler that control transfer instructions
6222 cannot have outputs. This restriction on @code{asm goto} may be lifted
6223 in some future version of the compiler. In the meantime, @code{asm goto}
6224 may include a memory clobber, and so leave outputs in memory.
6230 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
6231 : : "r"(x), "r"(&y) : "r5", "memory" : error);
6239 In this (inefficient) example, the @code{frob} instruction sets the
6240 carry bit to indicate an error. The @code{jc} instruction detects
6241 this and branches to the @code{error} label. Finally, the output
6242 of the @code{frob} instruction (@code{%r5}) is stored into the memory
6243 for variable @code{y}, which is later read by the @code{return} statement.
6249 asm goto ("mfsr %%r1, 123; jmp %%r1;"
6250 ".pushsection doit_table;"
6251 ".long %l0, %l1, %l2, %l3;"
6253 : : : "r1" : label1, label2, label3, label4);
6254 __builtin_unreachable ();
6270 In this (also inefficient) example, the @code{mfsr} instruction reads
6271 an address from some out-of-band machine register, and the following
6272 @code{jmp} instruction branches to that address. The address read by
6273 the @code{mfsr} instruction is assumed to have been previously set via
6274 some application-specific mechanism to be one of the four values stored
6275 in the @code{doit_table} section. Finally, the @code{asm} is followed
6276 by a call to @code{__builtin_unreachable} to indicate that the @code{asm}
6277 does not in fact fall through.
6280 #define TRACE1(NUM) \
6282 asm goto ("0: nop;" \
6283 ".pushsection trace_table;" \
6286 : : : : trace#NUM); \
6287 if (0) @{ trace#NUM: trace(); @} \
6289 #define TRACE TRACE1(__COUNTER__)
6293 In this example (which in fact inspired the @code{asm goto} feature)
6294 we want on rare occasions to call the @code{trace} function; on other
6295 occasions we'd like to keep the overhead to the absolute minimum.
6296 The normal code path consists of a single @code{nop} instruction.
6297 However, we record the address of this @code{nop} together with the
6298 address of a label that calls the @code{trace} function. This allows
6299 the @code{nop} instruction to be patched at run time to be an
6300 unconditional branch to the stored label. It is assumed that an
6301 optimizing compiler moves the labeled block out of line, to
6302 optimize the fall through path from the @code{asm}.
6304 If you are writing a header file that should be includable in ISO C
6305 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
6308 @subsection Size of an @code{asm}
6310 Some targets require that GCC track the size of each instruction used in
6311 order to generate correct code. Because the final length of an
6312 @code{asm} is only known by the assembler, GCC must make an estimate as
6313 to how big it will be. The estimate is formed by counting the number of
6314 statements in the pattern of the @code{asm} and multiplying that by the
6315 length of the longest instruction on that processor. Statements in the
6316 @code{asm} are identified by newline characters and whatever statement
6317 separator characters are supported by the assembler; on most processors
6318 this is the @samp{;} character.
6320 Normally, GCC's estimate is perfectly adequate to ensure that correct
6321 code is generated, but it is possible to confuse the compiler if you use
6322 pseudo instructions or assembler macros that expand into multiple real
6323 instructions or if you use assembler directives that expand to more
6324 space in the object file than is needed for a single instruction.
6325 If this happens then the assembler produces a diagnostic saying that
6326 a label is unreachable.
6328 @subsection i386 floating-point asm operands
6330 On i386 targets, there are several rules on the usage of stack-like registers
6331 in the operands of an @code{asm}. These rules apply only to the operands
6332 that are stack-like registers:
6336 Given a set of input registers that die in an @code{asm}, it is
6337 necessary to know which are implicitly popped by the @code{asm}, and
6338 which must be explicitly popped by GCC@.
6340 An input register that is implicitly popped by the @code{asm} must be
6341 explicitly clobbered, unless it is constrained to match an
6345 For any input register that is implicitly popped by an @code{asm}, it is
6346 necessary to know how to adjust the stack to compensate for the pop.
6347 If any non-popped input is closer to the top of the reg-stack than
6348 the implicitly popped register, it would not be possible to know what the
6349 stack looked like---it's not clear how the rest of the stack ``slides
6352 All implicitly popped input registers must be closer to the top of
6353 the reg-stack than any input that is not implicitly popped.
6355 It is possible that if an input dies in an @code{asm}, the compiler might
6356 use the input register for an output reload. Consider this example:
6359 asm ("foo" : "=t" (a) : "f" (b));
6363 This code says that input @code{b} is not popped by the @code{asm}, and that
6364 the @code{asm} pushes a result onto the reg-stack, i.e., the stack is one
6365 deeper after the @code{asm} than it was before. But, it is possible that
6366 reload may think that it can use the same register for both the input and
6369 To prevent this from happening,
6370 if any input operand uses the @code{f} constraint, all output register
6371 constraints must use the @code{&} early-clobber modifier.
6373 The example above would be correctly written as:
6376 asm ("foo" : "=&t" (a) : "f" (b));
6380 Some operands need to be in particular places on the stack. All
6381 output operands fall in this category---GCC has no other way to
6382 know which registers the outputs appear in unless you indicate
6383 this in the constraints.
6385 Output operands must specifically indicate which register an output
6386 appears in after an @code{asm}. @code{=f} is not allowed: the operand
6387 constraints must select a class with a single register.
6390 Output operands may not be ``inserted'' between existing stack registers.
6391 Since no 387 opcode uses a read/write operand, all output operands
6392 are dead before the @code{asm}, and are pushed by the @code{asm}.
6393 It makes no sense to push anywhere but the top of the reg-stack.
6395 Output operands must start at the top of the reg-stack: output
6396 operands may not ``skip'' a register.
6399 Some @code{asm} statements may need extra stack space for internal
6400 calculations. This can be guaranteed by clobbering stack registers
6401 unrelated to the inputs and outputs.
6405 Here are a couple of reasonable @code{asm}s to want to write. This
6407 takes one input, which is internally popped, and produces two outputs.
6410 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
6414 This @code{asm} takes two inputs, which are popped by the @code{fyl2xp1} opcode,
6415 and replaces them with one output. The @code{st(1)} clobber is necessary
6416 for the compiler to know that @code{fyl2xp1} pops both inputs.
6419 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
6425 @section Controlling Names Used in Assembler Code
6426 @cindex assembler names for identifiers
6427 @cindex names used in assembler code
6428 @cindex identifiers, names in assembler code
6430 You can specify the name to be used in the assembler code for a C
6431 function or variable by writing the @code{asm} (or @code{__asm__})
6432 keyword after the declarator as follows:
6435 int foo asm ("myfoo") = 2;
6439 This specifies that the name to be used for the variable @code{foo} in
6440 the assembler code should be @samp{myfoo} rather than the usual
6443 On systems where an underscore is normally prepended to the name of a C
6444 function or variable, this feature allows you to define names for the
6445 linker that do not start with an underscore.
6447 It does not make sense to use this feature with a non-static local
6448 variable since such variables do not have assembler names. If you are
6449 trying to put the variable in a particular register, see @ref{Explicit
6450 Reg Vars}. GCC presently accepts such code with a warning, but will
6451 probably be changed to issue an error, rather than a warning, in the
6454 You cannot use @code{asm} in this way in a function @emph{definition}; but
6455 you can get the same effect by writing a declaration for the function
6456 before its definition and putting @code{asm} there, like this:
6459 extern func () asm ("FUNC");
6466 It is up to you to make sure that the assembler names you choose do not
6467 conflict with any other assembler symbols. Also, you must not use a
6468 register name; that would produce completely invalid assembler code. GCC
6469 does not as yet have the ability to store static variables in registers.
6470 Perhaps that will be added.
6472 @node Explicit Reg Vars
6473 @section Variables in Specified Registers
6474 @cindex explicit register variables
6475 @cindex variables in specified registers
6476 @cindex specified registers
6477 @cindex registers, global allocation
6479 GNU C allows you to put a few global variables into specified hardware
6480 registers. You can also specify the register in which an ordinary
6481 register variable should be allocated.
6485 Global register variables reserve registers throughout the program.
6486 This may be useful in programs such as programming language
6487 interpreters that have a couple of global variables that are accessed
6491 Local register variables in specific registers do not reserve the
6492 registers, except at the point where they are used as input or output
6493 operands in an @code{asm} statement and the @code{asm} statement itself is
6494 not deleted. The compiler's data flow analysis is capable of determining
6495 where the specified registers contain live values, and where they are
6496 available for other uses. Stores into local register variables may be deleted
6497 when they appear to be dead according to dataflow analysis. References
6498 to local register variables may be deleted or moved or simplified.
6500 These local variables are sometimes convenient for use with the extended
6501 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
6502 output of the assembler instruction directly into a particular register.
6503 (This works provided the register you specify fits the constraints
6504 specified for that operand in the @code{asm}.)
6512 @node Global Reg Vars
6513 @subsection Defining Global Register Variables
6514 @cindex global register variables
6515 @cindex registers, global variables in
6517 You can define a global register variable in GNU C like this:
6520 register int *foo asm ("a5");
6524 Here @code{a5} is the name of the register that should be used. Choose a
6525 register that is normally saved and restored by function calls on your
6526 machine, so that library routines will not clobber it.
6528 Naturally the register name is cpu-dependent, so you need to
6529 conditionalize your program according to cpu type. The register
6530 @code{a5} is a good choice on a 68000 for a variable of pointer
6531 type. On machines with register windows, be sure to choose a ``global''
6532 register that is not affected magically by the function call mechanism.
6534 In addition, different operating systems on the same CPU may differ in how they
6535 name the registers; then you need additional conditionals. For
6536 example, some 68000 operating systems call this register @code{%a5}.
6538 Eventually there may be a way of asking the compiler to choose a register
6539 automatically, but first we need to figure out how it should choose and
6540 how to enable you to guide the choice. No solution is evident.
6542 Defining a global register variable in a certain register reserves that
6543 register entirely for this use, at least within the current compilation.
6544 The register is not allocated for any other purpose in the functions
6545 in the current compilation, and is not saved and restored by
6546 these functions. Stores into this register are never deleted even if they
6547 appear to be dead, but references may be deleted or moved or
6550 It is not safe to access the global register variables from signal
6551 handlers, or from more than one thread of control, because the system
6552 library routines may temporarily use the register for other things (unless
6553 you recompile them specially for the task at hand).
6555 @cindex @code{qsort}, and global register variables
6556 It is not safe for one function that uses a global register variable to
6557 call another such function @code{foo} by way of a third function
6558 @code{lose} that is compiled without knowledge of this variable (i.e.@: in a
6559 different source file in which the variable isn't declared). This is
6560 because @code{lose} might save the register and put some other value there.
6561 For example, you can't expect a global register variable to be available in
6562 the comparison-function that you pass to @code{qsort}, since @code{qsort}
6563 might have put something else in that register. (If you are prepared to
6564 recompile @code{qsort} with the same global register variable, you can
6565 solve this problem.)
6567 If you want to recompile @code{qsort} or other source files that do not
6568 actually use your global register variable, so that they do not use that
6569 register for any other purpose, then it suffices to specify the compiler
6570 option @option{-ffixed-@var{reg}}. You need not actually add a global
6571 register declaration to their source code.
6573 A function that can alter the value of a global register variable cannot
6574 safely be called from a function compiled without this variable, because it
6575 could clobber the value the caller expects to find there on return.
6576 Therefore, the function that is the entry point into the part of the
6577 program that uses the global register variable must explicitly save and
6578 restore the value that belongs to its caller.
6580 @cindex register variable after @code{longjmp}
6581 @cindex global register after @code{longjmp}
6582 @cindex value after @code{longjmp}
6585 On most machines, @code{longjmp} restores to each global register
6586 variable the value it had at the time of the @code{setjmp}. On some
6587 machines, however, @code{longjmp} does not change the value of global
6588 register variables. To be portable, the function that called @code{setjmp}
6589 should make other arrangements to save the values of the global register
6590 variables, and to restore them in a @code{longjmp}. This way, the same
6591 thing happens regardless of what @code{longjmp} does.
6593 All global register variable declarations must precede all function
6594 definitions. If such a declaration could appear after function
6595 definitions, the declaration would be too late to prevent the register from
6596 being used for other purposes in the preceding functions.
6598 Global register variables may not have initial values, because an
6599 executable file has no means to supply initial contents for a register.
6601 On the SPARC, there are reports that g3 @dots{} g7 are suitable
6602 registers, but certain library functions, such as @code{getwd}, as well
6603 as the subroutines for division and remainder, modify g3 and g4. g1 and
6604 g2 are local temporaries.
6606 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
6607 Of course, it does not do to use more than a few of those.
6609 @node Local Reg Vars
6610 @subsection Specifying Registers for Local Variables
6611 @cindex local variables, specifying registers
6612 @cindex specifying registers for local variables
6613 @cindex registers for local variables
6615 You can define a local register variable with a specified register
6619 register int *foo asm ("a5");
6623 Here @code{a5} is the name of the register that should be used. Note
6624 that this is the same syntax used for defining global register
6625 variables, but for a local variable it appears within a function.
6627 Naturally the register name is cpu-dependent, but this is not a
6628 problem, since specific registers are most often useful with explicit
6629 assembler instructions (@pxref{Extended Asm}). Both of these things
6630 generally require that you conditionalize your program according to
6633 In addition, operating systems on one type of cpu may differ in how they
6634 name the registers; then you need additional conditionals. For
6635 example, some 68000 operating systems call this register @code{%a5}.
6637 Defining such a register variable does not reserve the register; it
6638 remains available for other uses in places where flow control determines
6639 the variable's value is not live.
6641 This option does not guarantee that GCC generates code that has
6642 this variable in the register you specify at all times. You may not
6643 code an explicit reference to this register in the @emph{assembler
6644 instruction template} part of an @code{asm} statement and assume it
6645 always refers to this variable. However, using the variable as an
6646 @code{asm} @emph{operand} guarantees that the specified register is used
6649 Stores into local register variables may be deleted when they appear to be dead
6650 according to dataflow analysis. References to local register variables may
6651 be deleted or moved or simplified.
6653 As for global register variables, it's recommended that you choose a
6654 register that is normally saved and restored by function calls on
6655 your machine, so that library routines will not clobber it. A common
6656 pitfall is to initialize multiple call-clobbered registers with
6657 arbitrary expressions, where a function call or library call for an
6658 arithmetic operator overwrites a register value from a previous
6659 assignment, for example @code{r0} below:
6661 register int *p1 asm ("r0") = @dots{};
6662 register int *p2 asm ("r1") = @dots{};
6666 In those cases, a solution is to use a temporary variable for
6667 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
6669 @node Alternate Keywords
6670 @section Alternate Keywords
6671 @cindex alternate keywords
6672 @cindex keywords, alternate
6674 @option{-ansi} and the various @option{-std} options disable certain
6675 keywords. This causes trouble when you want to use GNU C extensions, or
6676 a general-purpose header file that should be usable by all programs,
6677 including ISO C programs. The keywords @code{asm}, @code{typeof} and
6678 @code{inline} are not available in programs compiled with
6679 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
6680 program compiled with @option{-std=c99} or @option{-std=c11}). The
6682 @code{restrict} is only available when @option{-std=gnu99} (which will
6683 eventually be the default) or @option{-std=c99} (or the equivalent
6684 @option{-std=iso9899:1999}), or an option for a later standard
6687 The way to solve these problems is to put @samp{__} at the beginning and
6688 end of each problematical keyword. For example, use @code{__asm__}
6689 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
6691 Other C compilers won't accept these alternative keywords; if you want to
6692 compile with another compiler, you can define the alternate keywords as
6693 macros to replace them with the customary keywords. It looks like this:
6701 @findex __extension__
6703 @option{-pedantic} and other options cause warnings for many GNU C extensions.
6705 prevent such warnings within one expression by writing
6706 @code{__extension__} before the expression. @code{__extension__} has no
6707 effect aside from this.
6709 @node Incomplete Enums
6710 @section Incomplete @code{enum} Types
6712 You can define an @code{enum} tag without specifying its possible values.
6713 This results in an incomplete type, much like what you get if you write
6714 @code{struct foo} without describing the elements. A later declaration
6715 that does specify the possible values completes the type.
6717 You can't allocate variables or storage using the type while it is
6718 incomplete. However, you can work with pointers to that type.
6720 This extension may not be very useful, but it makes the handling of
6721 @code{enum} more consistent with the way @code{struct} and @code{union}
6724 This extension is not supported by GNU C++.
6726 @node Function Names
6727 @section Function Names as Strings
6728 @cindex @code{__func__} identifier
6729 @cindex @code{__FUNCTION__} identifier
6730 @cindex @code{__PRETTY_FUNCTION__} identifier
6732 GCC provides three magic variables that hold the name of the current
6733 function, as a string. The first of these is @code{__func__}, which
6734 is part of the C99 standard:
6736 The identifier @code{__func__} is implicitly declared by the translator
6737 as if, immediately following the opening brace of each function
6738 definition, the declaration
6741 static const char __func__[] = "function-name";
6745 appeared, where function-name is the name of the lexically-enclosing
6746 function. This name is the unadorned name of the function.
6748 @code{__FUNCTION__} is another name for @code{__func__}. Older
6749 versions of GCC recognize only this name. However, it is not
6750 standardized. For maximum portability, we recommend you use
6751 @code{__func__}, but provide a fallback definition with the
6755 #if __STDC_VERSION__ < 199901L
6757 # define __func__ __FUNCTION__
6759 # define __func__ "<unknown>"
6764 In C, @code{__PRETTY_FUNCTION__} is yet another name for
6765 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
6766 the type signature of the function as well as its bare name. For
6767 example, this program:
6771 extern int printf (char *, ...);
6778 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
6779 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
6797 __PRETTY_FUNCTION__ = void a::sub(int)
6800 These identifiers are not preprocessor macros. In GCC 3.3 and
6801 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
6802 were treated as string literals; they could be used to initialize
6803 @code{char} arrays, and they could be concatenated with other string
6804 literals. GCC 3.4 and later treat them as variables, like
6805 @code{__func__}. In C++, @code{__FUNCTION__} and
6806 @code{__PRETTY_FUNCTION__} have always been variables.
6808 @node Return Address
6809 @section Getting the Return or Frame Address of a Function
6811 These functions may be used to get information about the callers of a
6814 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
6815 This function returns the return address of the current function, or of
6816 one of its callers. The @var{level} argument is number of frames to
6817 scan up the call stack. A value of @code{0} yields the return address
6818 of the current function, a value of @code{1} yields the return address
6819 of the caller of the current function, and so forth. When inlining
6820 the expected behavior is that the function returns the address of
6821 the function that is returned to. To work around this behavior use
6822 the @code{noinline} function attribute.
6824 The @var{level} argument must be a constant integer.
6826 On some machines it may be impossible to determine the return address of
6827 any function other than the current one; in such cases, or when the top
6828 of the stack has been reached, this function returns @code{0} or a
6829 random value. In addition, @code{__builtin_frame_address} may be used
6830 to determine if the top of the stack has been reached.
6832 Additional post-processing of the returned value may be needed, see
6833 @code{__builtin_extract_return_addr}.
6835 This function should only be used with a nonzero argument for debugging
6839 @deftypefn {Built-in Function} {void *} __builtin_extract_return_addr (void *@var{addr})
6840 The address as returned by @code{__builtin_return_address} may have to be fed
6841 through this function to get the actual encoded address. For example, on the
6842 31-bit S/390 platform the highest bit has to be masked out, or on SPARC
6843 platforms an offset has to be added for the true next instruction to be
6846 If no fixup is needed, this function simply passes through @var{addr}.
6849 @deftypefn {Built-in Function} {void *} __builtin_frob_return_address (void *@var{addr})
6850 This function does the reverse of @code{__builtin_extract_return_addr}.
6853 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
6854 This function is similar to @code{__builtin_return_address}, but it
6855 returns the address of the function frame rather than the return address
6856 of the function. Calling @code{__builtin_frame_address} with a value of
6857 @code{0} yields the frame address of the current function, a value of
6858 @code{1} yields the frame address of the caller of the current function,
6861 The frame is the area on the stack that holds local variables and saved
6862 registers. The frame address is normally the address of the first word
6863 pushed on to the stack by the function. However, the exact definition
6864 depends upon the processor and the calling convention. If the processor
6865 has a dedicated frame pointer register, and the function has a frame,
6866 then @code{__builtin_frame_address} returns the value of the frame
6869 On some machines it may be impossible to determine the frame address of
6870 any function other than the current one; in such cases, or when the top
6871 of the stack has been reached, this function returns @code{0} if
6872 the first frame pointer is properly initialized by the startup code.
6874 This function should only be used with a nonzero argument for debugging
6878 @node Vector Extensions
6879 @section Using Vector Instructions through Built-in Functions
6881 On some targets, the instruction set contains SIMD vector instructions which
6882 operate on multiple values contained in one large register at the same time.
6883 For example, on the i386 the MMX, 3DNow!@: and SSE extensions can be used
6886 The first step in using these extensions is to provide the necessary data
6887 types. This should be done using an appropriate @code{typedef}:
6890 typedef int v4si __attribute__ ((vector_size (16)));
6894 The @code{int} type specifies the base type, while the attribute specifies
6895 the vector size for the variable, measured in bytes. For example, the
6896 declaration above causes the compiler to set the mode for the @code{v4si}
6897 type to be 16 bytes wide and divided into @code{int} sized units. For
6898 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
6899 corresponding mode of @code{foo} is @acronym{V4SI}.
6901 The @code{vector_size} attribute is only applicable to integral and
6902 float scalars, although arrays, pointers, and function return values
6903 are allowed in conjunction with this construct. Only sizes that are
6904 a power of two are currently allowed.
6906 All the basic integer types can be used as base types, both as signed
6907 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
6908 @code{long long}. In addition, @code{float} and @code{double} can be
6909 used to build floating-point vector types.
6911 Specifying a combination that is not valid for the current architecture
6912 causes GCC to synthesize the instructions using a narrower mode.
6913 For example, if you specify a variable of type @code{V4SI} and your
6914 architecture does not allow for this specific SIMD type, GCC
6915 produces code that uses 4 @code{SIs}.
6917 The types defined in this manner can be used with a subset of normal C
6918 operations. Currently, GCC allows using the following operators
6919 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~, %}@.
6921 The operations behave like C++ @code{valarrays}. Addition is defined as
6922 the addition of the corresponding elements of the operands. For
6923 example, in the code below, each of the 4 elements in @var{a} is
6924 added to the corresponding 4 elements in @var{b} and the resulting
6925 vector is stored in @var{c}.
6928 typedef int v4si __attribute__ ((vector_size (16)));
6935 Subtraction, multiplication, division, and the logical operations
6936 operate in a similar manner. Likewise, the result of using the unary
6937 minus or complement operators on a vector type is a vector whose
6938 elements are the negative or complemented values of the corresponding
6939 elements in the operand.
6941 It is possible to use shifting operators @code{<<}, @code{>>} on
6942 integer-type vectors. The operation is defined as following: @code{@{a0,
6943 a1, @dots{}, an@} >> @{b0, b1, @dots{}, bn@} == @{a0 >> b0, a1 >> b1,
6944 @dots{}, an >> bn@}}@. Vector operands must have the same number of
6947 For convenience, it is allowed to use a binary vector operation
6948 where one operand is a scalar. In that case the compiler transforms
6949 the scalar operand into a vector where each element is the scalar from
6950 the operation. The transformation happens only if the scalar could be
6951 safely converted to the vector-element type.
6952 Consider the following code.
6955 typedef int v4si __attribute__ ((vector_size (16)));
6960 a = b + 1; /* a = b + @{1,1,1,1@}; */
6961 a = 2 * b; /* a = @{2,2,2,2@} * b; */
6963 a = l + a; /* Error, cannot convert long to int. */
6966 Vectors can be subscripted as if the vector were an array with
6967 the same number of elements and base type. Out of bound accesses
6968 invoke undefined behavior at run time. Warnings for out of bound
6969 accesses for vector subscription can be enabled with
6970 @option{-Warray-bounds}.
6972 Vector comparison is supported with standard comparison
6973 operators: @code{==, !=, <, <=, >, >=}. Comparison operands can be
6974 vector expressions of integer-type or real-type. Comparison between
6975 integer-type vectors and real-type vectors are not supported. The
6976 result of the comparison is a vector of the same width and number of
6977 elements as the comparison operands with a signed integral element
6980 Vectors are compared element-wise producing 0 when comparison is false
6981 and -1 (constant of the appropriate type where all bits are set)
6982 otherwise. Consider the following example.
6985 typedef int v4si __attribute__ ((vector_size (16)));
6987 v4si a = @{1,2,3,4@};
6988 v4si b = @{3,2,1,4@};
6991 c = a > b; /* The result would be @{0, 0,-1, 0@} */
6992 c = a == b; /* The result would be @{0,-1, 0,-1@} */
6995 Vector shuffling is available using functions
6996 @code{__builtin_shuffle (vec, mask)} and
6997 @code{__builtin_shuffle (vec0, vec1, mask)}.
6998 Both functions construct a permutation of elements from one or two
6999 vectors and return a vector of the same type as the input vector(s).
7000 The @var{mask} is an integral vector with the same width (@var{W})
7001 and element count (@var{N}) as the output vector.
7003 The elements of the input vectors are numbered in memory ordering of
7004 @var{vec0} beginning at 0 and @var{vec1} beginning at @var{N}. The
7005 elements of @var{mask} are considered modulo @var{N} in the single-operand
7006 case and modulo @math{2*@var{N}} in the two-operand case.
7008 Consider the following example,
7011 typedef int v4si __attribute__ ((vector_size (16)));
7013 v4si a = @{1,2,3,4@};
7014 v4si b = @{5,6,7,8@};
7015 v4si mask1 = @{0,1,1,3@};
7016 v4si mask2 = @{0,4,2,5@};
7019 res = __builtin_shuffle (a, mask1); /* res is @{1,2,2,4@} */
7020 res = __builtin_shuffle (a, b, mask2); /* res is @{1,5,3,6@} */
7023 Note that @code{__builtin_shuffle} is intentionally semantically
7024 compatible with the OpenCL @code{shuffle} and @code{shuffle2} functions.
7026 You can declare variables and use them in function calls and returns, as
7027 well as in assignments and some casts. You can specify a vector type as
7028 a return type for a function. Vector types can also be used as function
7029 arguments. It is possible to cast from one vector type to another,
7030 provided they are of the same size (in fact, you can also cast vectors
7031 to and from other datatypes of the same size).
7033 You cannot operate between vectors of different lengths or different
7034 signedness without a cast.
7038 @findex __builtin_offsetof
7040 GCC implements for both C and C++ a syntactic extension to implement
7041 the @code{offsetof} macro.
7045 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
7047 offsetof_member_designator:
7049 | offsetof_member_designator "." @code{identifier}
7050 | offsetof_member_designator "[" @code{expr} "]"
7053 This extension is sufficient such that
7056 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
7060 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
7061 may be dependent. In either case, @var{member} may consist of a single
7062 identifier, or a sequence of member accesses and array references.
7064 @node __sync Builtins
7065 @section Legacy __sync Built-in Functions for Atomic Memory Access
7067 The following built-in functions
7068 are intended to be compatible with those described
7069 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
7070 section 7.4. As such, they depart from the normal GCC practice of using
7071 the @samp{__builtin_} prefix, and further that they are overloaded such that
7072 they work on multiple types.
7074 The definition given in the Intel documentation allows only for the use of
7075 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
7076 counterparts. GCC allows any integral scalar or pointer type that is
7077 1, 2, 4 or 8 bytes in length.
7079 Not all operations are supported by all target processors. If a particular
7080 operation cannot be implemented on the target processor, a warning is
7081 generated and a call an external function is generated. The external
7082 function carries the same name as the built-in version,
7083 with an additional suffix
7084 @samp{_@var{n}} where @var{n} is the size of the data type.
7086 @c ??? Should we have a mechanism to suppress this warning? This is almost
7087 @c useful for implementing the operation under the control of an external
7090 In most cases, these built-in functions are considered a @dfn{full barrier}.
7092 no memory operand is moved across the operation, either forward or
7093 backward. Further, instructions are issued as necessary to prevent the
7094 processor from speculating loads across the operation and from queuing stores
7095 after the operation.
7097 All of the routines are described in the Intel documentation to take
7098 ``an optional list of variables protected by the memory barrier''. It's
7099 not clear what is meant by that; it could mean that @emph{only} the
7100 following variables are protected, or it could mean that these variables
7101 should in addition be protected. At present GCC ignores this list and
7102 protects all variables that are globally accessible. If in the future
7103 we make some use of this list, an empty list will continue to mean all
7104 globally accessible variables.
7107 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
7108 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
7109 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
7110 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
7111 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
7112 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
7113 @findex __sync_fetch_and_add
7114 @findex __sync_fetch_and_sub
7115 @findex __sync_fetch_and_or
7116 @findex __sync_fetch_and_and
7117 @findex __sync_fetch_and_xor
7118 @findex __sync_fetch_and_nand
7119 These built-in functions perform the operation suggested by the name, and
7120 returns the value that had previously been in memory. That is,
7123 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
7124 @{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; @} // nand
7127 @emph{Note:} GCC 4.4 and later implement @code{__sync_fetch_and_nand}
7128 as @code{*ptr = ~(tmp & value)} instead of @code{*ptr = ~tmp & value}.
7130 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
7131 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
7132 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
7133 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
7134 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
7135 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
7136 @findex __sync_add_and_fetch
7137 @findex __sync_sub_and_fetch
7138 @findex __sync_or_and_fetch
7139 @findex __sync_and_and_fetch
7140 @findex __sync_xor_and_fetch
7141 @findex __sync_nand_and_fetch
7142 These built-in functions perform the operation suggested by the name, and
7143 return the new value. That is,
7146 @{ *ptr @var{op}= value; return *ptr; @}
7147 @{ *ptr = ~(*ptr & value); return *ptr; @} // nand
7150 @emph{Note:} GCC 4.4 and later implement @code{__sync_nand_and_fetch}
7151 as @code{*ptr = ~(*ptr & value)} instead of
7152 @code{*ptr = ~*ptr & value}.
7154 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
7155 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
7156 @findex __sync_bool_compare_and_swap
7157 @findex __sync_val_compare_and_swap
7158 These built-in functions perform an atomic compare and swap.
7159 That is, if the current
7160 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
7163 The ``bool'' version returns true if the comparison is successful and
7164 @var{newval} is written. The ``val'' version returns the contents
7165 of @code{*@var{ptr}} before the operation.
7167 @item __sync_synchronize (...)
7168 @findex __sync_synchronize
7169 This built-in function issues a full memory barrier.
7171 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
7172 @findex __sync_lock_test_and_set
7173 This built-in function, as described by Intel, is not a traditional test-and-set
7174 operation, but rather an atomic exchange operation. It writes @var{value}
7175 into @code{*@var{ptr}}, and returns the previous contents of
7178 Many targets have only minimal support for such locks, and do not support
7179 a full exchange operation. In this case, a target may support reduced
7180 functionality here by which the @emph{only} valid value to store is the
7181 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
7182 is implementation defined.
7184 This built-in function is not a full barrier,
7185 but rather an @dfn{acquire barrier}.
7186 This means that references after the operation cannot move to (or be
7187 speculated to) before the operation, but previous memory stores may not
7188 be globally visible yet, and previous memory loads may not yet be
7191 @item void __sync_lock_release (@var{type} *ptr, ...)
7192 @findex __sync_lock_release
7193 This built-in function releases the lock acquired by
7194 @code{__sync_lock_test_and_set}.
7195 Normally this means writing the constant 0 to @code{*@var{ptr}}.
7197 This built-in function is not a full barrier,
7198 but rather a @dfn{release barrier}.
7199 This means that all previous memory stores are globally visible, and all
7200 previous memory loads have been satisfied, but following memory reads
7201 are not prevented from being speculated to before the barrier.
7204 @node __atomic Builtins
7205 @section Built-in functions for memory model aware atomic operations
7207 The following built-in functions approximately match the requirements for
7208 C++11 memory model. Many are similar to the @samp{__sync} prefixed built-in
7209 functions, but all also have a memory model parameter. These are all
7210 identified by being prefixed with @samp{__atomic}, and most are overloaded
7211 such that they work with multiple types.
7213 GCC allows any integral scalar or pointer type that is 1, 2, 4, or 8
7214 bytes in length. 16-byte integral types are also allowed if
7215 @samp{__int128} (@pxref{__int128}) is supported by the architecture.
7217 Target architectures are encouraged to provide their own patterns for
7218 each of these built-in functions. If no target is provided, the original
7219 non-memory model set of @samp{__sync} atomic built-in functions are
7220 utilized, along with any required synchronization fences surrounding it in
7221 order to achieve the proper behavior. Execution in this case is subject
7222 to the same restrictions as those built-in functions.
7224 If there is no pattern or mechanism to provide a lock free instruction
7225 sequence, a call is made to an external routine with the same parameters
7226 to be resolved at run time.
7228 The four non-arithmetic functions (load, store, exchange, and
7229 compare_exchange) all have a generic version as well. This generic
7230 version works on any data type. If the data type size maps to one
7231 of the integral sizes that may have lock free support, the generic
7232 version utilizes the lock free built-in function. Otherwise an
7233 external call is left to be resolved at run time. This external call is
7234 the same format with the addition of a @samp{size_t} parameter inserted
7235 as the first parameter indicating the size of the object being pointed to.
7236 All objects must be the same size.
7238 There are 6 different memory models that can be specified. These map
7239 to the same names in the C++11 standard. Refer there or to the
7240 @uref{http://gcc.gnu.org/wiki/Atomic/GCCMM/AtomicSync,GCC wiki on
7241 atomic synchronization} for more detailed definitions. These memory
7242 models integrate both barriers to code motion as well as synchronization
7243 requirements with other threads. These are listed in approximately
7244 ascending order of strength. It is also possible to use target specific
7245 flags for memory model flags, like Hardware Lock Elision.
7248 @item __ATOMIC_RELAXED
7249 No barriers or synchronization.
7250 @item __ATOMIC_CONSUME
7251 Data dependency only for both barrier and synchronization with another
7253 @item __ATOMIC_ACQUIRE
7254 Barrier to hoisting of code and synchronizes with release (or stronger)
7255 semantic stores from another thread.
7256 @item __ATOMIC_RELEASE
7257 Barrier to sinking of code and synchronizes with acquire (or stronger)
7258 semantic loads from another thread.
7259 @item __ATOMIC_ACQ_REL
7260 Full barrier in both directions and synchronizes with acquire loads and
7261 release stores in another thread.
7262 @item __ATOMIC_SEQ_CST
7263 Full barrier in both directions and synchronizes with acquire loads and
7264 release stores in all threads.
7267 When implementing patterns for these built-in functions, the memory model
7268 parameter can be ignored as long as the pattern implements the most
7269 restrictive @code{__ATOMIC_SEQ_CST} model. Any of the other memory models
7270 execute correctly with this memory model but they may not execute as
7271 efficiently as they could with a more appropriate implementation of the
7272 relaxed requirements.
7274 Note that the C++11 standard allows for the memory model parameter to be
7275 determined at run time rather than at compile time. These built-in
7276 functions map any run-time value to @code{__ATOMIC_SEQ_CST} rather
7277 than invoke a runtime library call or inline a switch statement. This is
7278 standard compliant, safe, and the simplest approach for now.
7280 The memory model parameter is a signed int, but only the lower 8 bits are
7281 reserved for the memory model. The remainder of the signed int is reserved
7282 for future use and should be 0. Use of the predefined atomic values
7283 ensures proper usage.
7285 @deftypefn {Built-in Function} @var{type} __atomic_load_n (@var{type} *ptr, int memmodel)
7286 This built-in function implements an atomic load operation. It returns the
7287 contents of @code{*@var{ptr}}.
7289 The valid memory model variants are
7290 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
7291 and @code{__ATOMIC_CONSUME}.
7295 @deftypefn {Built-in Function} void __atomic_load (@var{type} *ptr, @var{type} *ret, int memmodel)
7296 This is the generic version of an atomic load. It returns the
7297 contents of @code{*@var{ptr}} in @code{*@var{ret}}.
7301 @deftypefn {Built-in Function} void __atomic_store_n (@var{type} *ptr, @var{type} val, int memmodel)
7302 This built-in function implements an atomic store operation. It writes
7303 @code{@var{val}} into @code{*@var{ptr}}.
7305 The valid memory model variants are
7306 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and @code{__ATOMIC_RELEASE}.
7310 @deftypefn {Built-in Function} void __atomic_store (@var{type} *ptr, @var{type} *val, int memmodel)
7311 This is the generic version of an atomic store. It stores the value
7312 of @code{*@var{val}} into @code{*@var{ptr}}.
7316 @deftypefn {Built-in Function} @var{type} __atomic_exchange_n (@var{type} *ptr, @var{type} val, int memmodel)
7317 This built-in function implements an atomic exchange operation. It writes
7318 @var{val} into @code{*@var{ptr}}, and returns the previous contents of
7321 The valid memory model variants are
7322 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
7323 @code{__ATOMIC_RELEASE}, and @code{__ATOMIC_ACQ_REL}.
7327 @deftypefn {Built-in Function} void __atomic_exchange (@var{type} *ptr, @var{type} *val, @var{type} *ret, int memmodel)
7328 This is the generic version of an atomic exchange. It stores the
7329 contents of @code{*@var{val}} into @code{*@var{ptr}}. The original value
7330 of @code{*@var{ptr}} is copied into @code{*@var{ret}}.
7334 @deftypefn {Built-in Function} bool __atomic_compare_exchange_n (@var{type} *ptr, @var{type} *expected, @var{type} desired, bool weak, int success_memmodel, int failure_memmodel)
7335 This built-in function implements an atomic compare and exchange operation.
7336 This compares the contents of @code{*@var{ptr}} with the contents of
7337 @code{*@var{expected}} and if equal, writes @var{desired} into
7338 @code{*@var{ptr}}. If they are not equal, the current contents of
7339 @code{*@var{ptr}} is written into @code{*@var{expected}}. @var{weak} is true
7340 for weak compare_exchange, and false for the strong variation. Many targets
7341 only offer the strong variation and ignore the parameter. When in doubt, use
7342 the strong variation.
7344 True is returned if @var{desired} is written into
7345 @code{*@var{ptr}} and the execution is considered to conform to the
7346 memory model specified by @var{success_memmodel}. There are no
7347 restrictions on what memory model can be used here.
7349 False is returned otherwise, and the execution is considered to conform
7350 to @var{failure_memmodel}. This memory model cannot be
7351 @code{__ATOMIC_RELEASE} nor @code{__ATOMIC_ACQ_REL}. It also cannot be a
7352 stronger model than that specified by @var{success_memmodel}.
7356 @deftypefn {Built-in Function} bool __atomic_compare_exchange (@var{type} *ptr, @var{type} *expected, @var{type} *desired, bool weak, int success_memmodel, int failure_memmodel)
7357 This built-in function implements the generic version of
7358 @code{__atomic_compare_exchange}. The function is virtually identical to
7359 @code{__atomic_compare_exchange_n}, except the desired value is also a
7364 @deftypefn {Built-in Function} @var{type} __atomic_add_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7365 @deftypefnx {Built-in Function} @var{type} __atomic_sub_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7366 @deftypefnx {Built-in Function} @var{type} __atomic_and_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7367 @deftypefnx {Built-in Function} @var{type} __atomic_xor_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7368 @deftypefnx {Built-in Function} @var{type} __atomic_or_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7369 @deftypefnx {Built-in Function} @var{type} __atomic_nand_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7370 These built-in functions perform the operation suggested by the name, and
7371 return the result of the operation. That is,
7374 @{ *ptr @var{op}= val; return *ptr; @}
7377 All memory models are valid.
7381 @deftypefn {Built-in Function} @var{type} __atomic_fetch_add (@var{type} *ptr, @var{type} val, int memmodel)
7382 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_sub (@var{type} *ptr, @var{type} val, int memmodel)
7383 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_and (@var{type} *ptr, @var{type} val, int memmodel)
7384 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_xor (@var{type} *ptr, @var{type} val, int memmodel)
7385 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_or (@var{type} *ptr, @var{type} val, int memmodel)
7386 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_nand (@var{type} *ptr, @var{type} val, int memmodel)
7387 These built-in functions perform the operation suggested by the name, and
7388 return the value that had previously been in @code{*@var{ptr}}. That is,
7391 @{ tmp = *ptr; *ptr @var{op}= val; return tmp; @}
7394 All memory models are valid.
7398 @deftypefn {Built-in Function} bool __atomic_test_and_set (void *ptr, int memmodel)
7400 This built-in function performs an atomic test-and-set operation on
7401 the byte at @code{*@var{ptr}}. The byte is set to some implementation
7402 defined nonzero ``set'' value and the return value is @code{true} if and only
7403 if the previous contents were ``set''.
7405 All memory models are valid.
7409 @deftypefn {Built-in Function} void __atomic_clear (bool *ptr, int memmodel)
7411 This built-in function performs an atomic clear operation on
7412 @code{*@var{ptr}}. After the operation, @code{*@var{ptr}} contains 0.
7414 The valid memory model variants are
7415 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and
7416 @code{__ATOMIC_RELEASE}.
7420 @deftypefn {Built-in Function} void __atomic_thread_fence (int memmodel)
7422 This built-in function acts as a synchronization fence between threads
7423 based on the specified memory model.
7425 All memory orders are valid.
7429 @deftypefn {Built-in Function} void __atomic_signal_fence (int memmodel)
7431 This built-in function acts as a synchronization fence between a thread
7432 and signal handlers based in the same thread.
7434 All memory orders are valid.
7438 @deftypefn {Built-in Function} bool __atomic_always_lock_free (size_t size, void *ptr)
7440 This built-in function returns true if objects of @var{size} bytes always
7441 generate lock free atomic instructions for the target architecture.
7442 @var{size} must resolve to a compile-time constant and the result also
7443 resolves to a compile-time constant.
7445 @var{ptr} is an optional pointer to the object that may be used to determine
7446 alignment. A value of 0 indicates typical alignment should be used. The
7447 compiler may also ignore this parameter.
7450 if (_atomic_always_lock_free (sizeof (long long), 0))
7455 @deftypefn {Built-in Function} bool __atomic_is_lock_free (size_t size, void *ptr)
7457 This built-in function returns true if objects of @var{size} bytes always
7458 generate lock free atomic instructions for the target architecture. If
7459 it is not known to be lock free a call is made to a runtime routine named
7460 @code{__atomic_is_lock_free}.
7462 @var{ptr} is an optional pointer to the object that may be used to determine
7463 alignment. A value of 0 indicates typical alignment should be used. The
7464 compiler may also ignore this parameter.
7467 @node Object Size Checking
7468 @section Object Size Checking Built-in Functions
7469 @findex __builtin_object_size
7470 @findex __builtin___memcpy_chk
7471 @findex __builtin___mempcpy_chk
7472 @findex __builtin___memmove_chk
7473 @findex __builtin___memset_chk
7474 @findex __builtin___strcpy_chk
7475 @findex __builtin___stpcpy_chk
7476 @findex __builtin___strncpy_chk
7477 @findex __builtin___strcat_chk
7478 @findex __builtin___strncat_chk
7479 @findex __builtin___sprintf_chk
7480 @findex __builtin___snprintf_chk
7481 @findex __builtin___vsprintf_chk
7482 @findex __builtin___vsnprintf_chk
7483 @findex __builtin___printf_chk
7484 @findex __builtin___vprintf_chk
7485 @findex __builtin___fprintf_chk
7486 @findex __builtin___vfprintf_chk
7488 GCC implements a limited buffer overflow protection mechanism
7489 that can prevent some buffer overflow attacks.
7491 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
7492 is a built-in construct that returns a constant number of bytes from
7493 @var{ptr} to the end of the object @var{ptr} pointer points to
7494 (if known at compile time). @code{__builtin_object_size} never evaluates
7495 its arguments for side-effects. If there are any side-effects in them, it
7496 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
7497 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
7498 point to and all of them are known at compile time, the returned number
7499 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
7500 0 and minimum if nonzero. If it is not possible to determine which objects
7501 @var{ptr} points to at compile time, @code{__builtin_object_size} should
7502 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
7503 for @var{type} 2 or 3.
7505 @var{type} is an integer constant from 0 to 3. If the least significant
7506 bit is clear, objects are whole variables, if it is set, a closest
7507 surrounding subobject is considered the object a pointer points to.
7508 The second bit determines if maximum or minimum of remaining bytes
7512 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
7513 char *p = &var.buf1[1], *q = &var.b;
7515 /* Here the object p points to is var. */
7516 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
7517 /* The subobject p points to is var.buf1. */
7518 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
7519 /* The object q points to is var. */
7520 assert (__builtin_object_size (q, 0)
7521 == (char *) (&var + 1) - (char *) &var.b);
7522 /* The subobject q points to is var.b. */
7523 assert (__builtin_object_size (q, 1) == sizeof (var.b));
7527 There are built-in functions added for many common string operation
7528 functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk}
7529 built-in is provided. This built-in has an additional last argument,
7530 which is the number of bytes remaining in object the @var{dest}
7531 argument points to or @code{(size_t) -1} if the size is not known.
7533 The built-in functions are optimized into the normal string functions
7534 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
7535 it is known at compile time that the destination object will not
7536 be overflown. If the compiler can determine at compile time the
7537 object will be always overflown, it issues a warning.
7539 The intended use can be e.g.@:
7543 #define bos0(dest) __builtin_object_size (dest, 0)
7544 #define memcpy(dest, src, n) \
7545 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
7549 /* It is unknown what object p points to, so this is optimized
7550 into plain memcpy - no checking is possible. */
7551 memcpy (p, "abcde", n);
7552 /* Destination is known and length too. It is known at compile
7553 time there will be no overflow. */
7554 memcpy (&buf[5], "abcde", 5);
7555 /* Destination is known, but the length is not known at compile time.
7556 This will result in __memcpy_chk call that can check for overflow
7558 memcpy (&buf[5], "abcde", n);
7559 /* Destination is known and it is known at compile time there will
7560 be overflow. There will be a warning and __memcpy_chk call that
7561 will abort the program at run time. */
7562 memcpy (&buf[6], "abcde", 5);
7565 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
7566 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
7567 @code{strcat} and @code{strncat}.
7569 There are also checking built-in functions for formatted output functions.
7571 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
7572 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
7573 const char *fmt, ...);
7574 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
7576 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
7577 const char *fmt, va_list ap);
7580 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
7581 etc.@: functions and can contain implementation specific flags on what
7582 additional security measures the checking function might take, such as
7583 handling @code{%n} differently.
7585 The @var{os} argument is the object size @var{s} points to, like in the
7586 other built-in functions. There is a small difference in the behavior
7587 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
7588 optimized into the non-checking functions only if @var{flag} is 0, otherwise
7589 the checking function is called with @var{os} argument set to
7592 In addition to this, there are checking built-in functions
7593 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
7594 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
7595 These have just one additional argument, @var{flag}, right before
7596 format string @var{fmt}. If the compiler is able to optimize them to
7597 @code{fputc} etc.@: functions, it does, otherwise the checking function
7598 is called and the @var{flag} argument passed to it.
7600 @node Other Builtins
7601 @section Other Built-in Functions Provided by GCC
7602 @cindex built-in functions
7603 @findex __builtin_fpclassify
7604 @findex __builtin_isfinite
7605 @findex __builtin_isnormal
7606 @findex __builtin_isgreater
7607 @findex __builtin_isgreaterequal
7608 @findex __builtin_isinf_sign
7609 @findex __builtin_isless
7610 @findex __builtin_islessequal
7611 @findex __builtin_islessgreater
7612 @findex __builtin_isunordered
7613 @findex __builtin_powi
7614 @findex __builtin_powif
7615 @findex __builtin_powil
7773 @findex fprintf_unlocked
7775 @findex fputs_unlocked
7892 @findex printf_unlocked
7924 @findex significandf
7925 @findex significandl
7996 GCC provides a large number of built-in functions other than the ones
7997 mentioned above. Some of these are for internal use in the processing
7998 of exceptions or variable-length argument lists and are not
7999 documented here because they may change from time to time; we do not
8000 recommend general use of these functions.
8002 The remaining functions are provided for optimization purposes.
8004 @opindex fno-builtin
8005 GCC includes built-in versions of many of the functions in the standard
8006 C library. The versions prefixed with @code{__builtin_} are always
8007 treated as having the same meaning as the C library function even if you
8008 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
8009 Many of these functions are only optimized in certain cases; if they are
8010 not optimized in a particular case, a call to the library function is
8015 Outside strict ISO C mode (@option{-ansi}, @option{-std=c90},
8016 @option{-std=c99} or @option{-std=c11}), the functions
8017 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
8018 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
8019 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
8020 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
8021 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
8022 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
8023 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
8024 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
8025 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
8026 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
8027 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
8028 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
8029 @code{signbitd64}, @code{signbitd128}, @code{significandf},
8030 @code{significandl}, @code{significand}, @code{sincosf},
8031 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
8032 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
8033 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
8034 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
8036 may be handled as built-in functions.
8037 All these functions have corresponding versions
8038 prefixed with @code{__builtin_}, which may be used even in strict C90
8041 The ISO C99 functions
8042 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
8043 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
8044 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
8045 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
8046 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
8047 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
8048 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
8049 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
8050 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
8051 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
8052 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
8053 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
8054 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
8055 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
8056 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
8057 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
8058 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
8059 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
8060 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
8061 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
8062 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
8063 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
8064 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
8065 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
8066 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
8067 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
8068 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
8069 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
8070 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
8071 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
8072 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
8073 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
8074 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
8075 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
8076 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
8077 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
8078 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
8079 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
8080 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
8081 are handled as built-in functions
8082 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
8084 There are also built-in versions of the ISO C99 functions
8085 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
8086 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
8087 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
8088 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
8089 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
8090 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
8091 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
8092 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
8093 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
8094 that are recognized in any mode since ISO C90 reserves these names for
8095 the purpose to which ISO C99 puts them. All these functions have
8096 corresponding versions prefixed with @code{__builtin_}.
8098 The ISO C94 functions
8099 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
8100 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
8101 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
8103 are handled as built-in functions
8104 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
8106 The ISO C90 functions
8107 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
8108 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
8109 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
8110 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
8111 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
8112 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
8113 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
8114 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
8115 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
8116 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
8117 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
8118 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
8119 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
8120 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
8121 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
8122 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
8123 are all recognized as built-in functions unless
8124 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
8125 is specified for an individual function). All of these functions have
8126 corresponding versions prefixed with @code{__builtin_}.
8128 GCC provides built-in versions of the ISO C99 floating-point comparison
8129 macros that avoid raising exceptions for unordered operands. They have
8130 the same names as the standard macros ( @code{isgreater},
8131 @code{isgreaterequal}, @code{isless}, @code{islessequal},
8132 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
8133 prefixed. We intend for a library implementor to be able to simply
8134 @code{#define} each standard macro to its built-in equivalent.
8135 In the same fashion, GCC provides @code{fpclassify}, @code{isfinite},
8136 @code{isinf_sign} and @code{isnormal} built-ins used with
8137 @code{__builtin_} prefixed. The @code{isinf} and @code{isnan}
8138 built-in functions appear both with and without the @code{__builtin_} prefix.
8140 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
8142 You can use the built-in function @code{__builtin_types_compatible_p} to
8143 determine whether two types are the same.
8145 This built-in function returns 1 if the unqualified versions of the
8146 types @var{type1} and @var{type2} (which are types, not expressions) are
8147 compatible, 0 otherwise. The result of this built-in function can be
8148 used in integer constant expressions.
8150 This built-in function ignores top level qualifiers (e.g., @code{const},
8151 @code{volatile}). For example, @code{int} is equivalent to @code{const
8154 The type @code{int[]} and @code{int[5]} are compatible. On the other
8155 hand, @code{int} and @code{char *} are not compatible, even if the size
8156 of their types, on the particular architecture are the same. Also, the
8157 amount of pointer indirection is taken into account when determining
8158 similarity. Consequently, @code{short *} is not similar to
8159 @code{short **}. Furthermore, two types that are typedefed are
8160 considered compatible if their underlying types are compatible.
8162 An @code{enum} type is not considered to be compatible with another
8163 @code{enum} type even if both are compatible with the same integer
8164 type; this is what the C standard specifies.
8165 For example, @code{enum @{foo, bar@}} is not similar to
8166 @code{enum @{hot, dog@}}.
8168 You typically use this function in code whose execution varies
8169 depending on the arguments' types. For example:
8174 typeof (x) tmp = (x); \
8175 if (__builtin_types_compatible_p (typeof (x), long double)) \
8176 tmp = foo_long_double (tmp); \
8177 else if (__builtin_types_compatible_p (typeof (x), double)) \
8178 tmp = foo_double (tmp); \
8179 else if (__builtin_types_compatible_p (typeof (x), float)) \
8180 tmp = foo_float (tmp); \
8187 @emph{Note:} This construct is only available for C@.
8191 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
8193 You can use the built-in function @code{__builtin_choose_expr} to
8194 evaluate code depending on the value of a constant expression. This
8195 built-in function returns @var{exp1} if @var{const_exp}, which is an
8196 integer constant expression, is nonzero. Otherwise it returns @var{exp2}.
8198 This built-in function is analogous to the @samp{? :} operator in C,
8199 except that the expression returned has its type unaltered by promotion
8200 rules. Also, the built-in function does not evaluate the expression
8201 that is not chosen. For example, if @var{const_exp} evaluates to true,
8202 @var{exp2} is not evaluated even if it has side-effects.
8204 This built-in function can return an lvalue if the chosen argument is an
8207 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
8208 type. Similarly, if @var{exp2} is returned, its return type is the same
8215 __builtin_choose_expr ( \
8216 __builtin_types_compatible_p (typeof (x), double), \
8218 __builtin_choose_expr ( \
8219 __builtin_types_compatible_p (typeof (x), float), \
8221 /* @r{The void expression results in a compile-time error} \
8222 @r{when assigning the result to something.} */ \
8226 @emph{Note:} This construct is only available for C@. Furthermore, the
8227 unused expression (@var{exp1} or @var{exp2} depending on the value of
8228 @var{const_exp}) may still generate syntax errors. This may change in
8233 @deftypefn {Built-in Function} @var{type} __builtin_complex (@var{real}, @var{imag})
8235 The built-in function @code{__builtin_complex} is provided for use in
8236 implementing the ISO C11 macros @code{CMPLXF}, @code{CMPLX} and
8237 @code{CMPLXL}. @var{real} and @var{imag} must have the same type, a
8238 real binary floating-point type, and the result has the corresponding
8239 complex type with real and imaginary parts @var{real} and @var{imag}.
8240 Unlike @samp{@var{real} + I * @var{imag}}, this works even when
8241 infinities, NaNs and negative zeros are involved.
8245 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
8246 You can use the built-in function @code{__builtin_constant_p} to
8247 determine if a value is known to be constant at compile time and hence
8248 that GCC can perform constant-folding on expressions involving that
8249 value. The argument of the function is the value to test. The function
8250 returns the integer 1 if the argument is known to be a compile-time
8251 constant and 0 if it is not known to be a compile-time constant. A
8252 return of 0 does not indicate that the value is @emph{not} a constant,
8253 but merely that GCC cannot prove it is a constant with the specified
8254 value of the @option{-O} option.
8256 You typically use this function in an embedded application where
8257 memory is a critical resource. If you have some complex calculation,
8258 you may want it to be folded if it involves constants, but need to call
8259 a function if it does not. For example:
8262 #define Scale_Value(X) \
8263 (__builtin_constant_p (X) \
8264 ? ((X) * SCALE + OFFSET) : Scale (X))
8267 You may use this built-in function in either a macro or an inline
8268 function. However, if you use it in an inlined function and pass an
8269 argument of the function as the argument to the built-in, GCC
8270 never returns 1 when you call the inline function with a string constant
8271 or compound literal (@pxref{Compound Literals}) and does not return 1
8272 when you pass a constant numeric value to the inline function unless you
8273 specify the @option{-O} option.
8275 You may also use @code{__builtin_constant_p} in initializers for static
8276 data. For instance, you can write
8279 static const int table[] = @{
8280 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
8286 This is an acceptable initializer even if @var{EXPRESSION} is not a
8287 constant expression, including the case where
8288 @code{__builtin_constant_p} returns 1 because @var{EXPRESSION} can be
8289 folded to a constant but @var{EXPRESSION} contains operands that are
8290 not otherwise permitted in a static initializer (for example,
8291 @code{0 && foo ()}). GCC must be more conservative about evaluating the
8292 built-in in this case, because it has no opportunity to perform
8295 Previous versions of GCC did not accept this built-in in data
8296 initializers. The earliest version where it is completely safe is
8300 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
8301 @opindex fprofile-arcs
8302 You may use @code{__builtin_expect} to provide the compiler with
8303 branch prediction information. In general, you should prefer to
8304 use actual profile feedback for this (@option{-fprofile-arcs}), as
8305 programmers are notoriously bad at predicting how their programs
8306 actually perform. However, there are applications in which this
8307 data is hard to collect.
8309 The return value is the value of @var{exp}, which should be an integral
8310 expression. The semantics of the built-in are that it is expected that
8311 @var{exp} == @var{c}. For example:
8314 if (__builtin_expect (x, 0))
8319 indicates that we do not expect to call @code{foo}, since
8320 we expect @code{x} to be zero. Since you are limited to integral
8321 expressions for @var{exp}, you should use constructions such as
8324 if (__builtin_expect (ptr != NULL, 1))
8329 when testing pointer or floating-point values.
8332 @deftypefn {Built-in Function} void __builtin_trap (void)
8333 This function causes the program to exit abnormally. GCC implements
8334 this function by using a target-dependent mechanism (such as
8335 intentionally executing an illegal instruction) or by calling
8336 @code{abort}. The mechanism used may vary from release to release so
8337 you should not rely on any particular implementation.
8340 @deftypefn {Built-in Function} void __builtin_unreachable (void)
8341 If control flow reaches the point of the @code{__builtin_unreachable},
8342 the program is undefined. It is useful in situations where the
8343 compiler cannot deduce the unreachability of the code.
8345 One such case is immediately following an @code{asm} statement that
8346 either never terminates, or one that transfers control elsewhere
8347 and never returns. In this example, without the
8348 @code{__builtin_unreachable}, GCC issues a warning that control
8349 reaches the end of a non-void function. It also generates code
8350 to return after the @code{asm}.
8353 int f (int c, int v)
8361 asm("jmp error_handler");
8362 __builtin_unreachable ();
8368 Because the @code{asm} statement unconditionally transfers control out
8369 of the function, control never reaches the end of the function
8370 body. The @code{__builtin_unreachable} is in fact unreachable and
8371 communicates this fact to the compiler.
8373 Another use for @code{__builtin_unreachable} is following a call a
8374 function that never returns but that is not declared
8375 @code{__attribute__((noreturn))}, as in this example:
8378 void function_that_never_returns (void);
8388 function_that_never_returns ();
8389 __builtin_unreachable ();
8396 @deftypefn {Built-in Function} void *__builtin_assume_aligned (const void *@var{exp}, size_t @var{align}, ...)
8397 This function returns its first argument, and allows the compiler
8398 to assume that the returned pointer is at least @var{align} bytes
8399 aligned. This built-in can have either two or three arguments,
8400 if it has three, the third argument should have integer type, and
8401 if it is nonzero means misalignment offset. For example:
8404 void *x = __builtin_assume_aligned (arg, 16);
8408 means that the compiler can assume @code{x}, set to @code{arg}, is at least
8409 16-byte aligned, while:
8412 void *x = __builtin_assume_aligned (arg, 32, 8);
8416 means that the compiler can assume for @code{x}, set to @code{arg}, that
8417 @code{(char *) x - 8} is 32-byte aligned.
8420 @deftypefn {Built-in Function} int __builtin_LINE ()
8421 This function is the equivalent to the preprocessor @code{__LINE__}
8422 macro and returns the line number of the invocation of the built-in.
8425 @deftypefn {Built-in Function} int __builtin_FUNCTION ()
8426 This function is the equivalent to the preprocessor @code{__FUNCTION__}
8427 macro and returns the function name the invocation of the built-in is in.
8430 @deftypefn {Built-in Function} int __builtin_FILE ()
8431 This function is the equivalent to the preprocessor @code{__FILE__}
8432 macro and returns the file name the invocation of the built-in is in.
8435 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
8436 This function is used to flush the processor's instruction cache for
8437 the region of memory between @var{begin} inclusive and @var{end}
8438 exclusive. Some targets require that the instruction cache be
8439 flushed, after modifying memory containing code, in order to obtain
8440 deterministic behavior.
8442 If the target does not require instruction cache flushes,
8443 @code{__builtin___clear_cache} has no effect. Otherwise either
8444 instructions are emitted in-line to clear the instruction cache or a
8445 call to the @code{__clear_cache} function in libgcc is made.
8448 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
8449 This function is used to minimize cache-miss latency by moving data into
8450 a cache before it is accessed.
8451 You can insert calls to @code{__builtin_prefetch} into code for which
8452 you know addresses of data in memory that is likely to be accessed soon.
8453 If the target supports them, data prefetch instructions are generated.
8454 If the prefetch is done early enough before the access then the data will
8455 be in the cache by the time it is accessed.
8457 The value of @var{addr} is the address of the memory to prefetch.
8458 There are two optional arguments, @var{rw} and @var{locality}.
8459 The value of @var{rw} is a compile-time constant one or zero; one
8460 means that the prefetch is preparing for a write to the memory address
8461 and zero, the default, means that the prefetch is preparing for a read.
8462 The value @var{locality} must be a compile-time constant integer between
8463 zero and three. A value of zero means that the data has no temporal
8464 locality, so it need not be left in the cache after the access. A value
8465 of three means that the data has a high degree of temporal locality and
8466 should be left in all levels of cache possible. Values of one and two
8467 mean, respectively, a low or moderate degree of temporal locality. The
8471 for (i = 0; i < n; i++)
8474 __builtin_prefetch (&a[i+j], 1, 1);
8475 __builtin_prefetch (&b[i+j], 0, 1);
8480 Data prefetch does not generate faults if @var{addr} is invalid, but
8481 the address expression itself must be valid. For example, a prefetch
8482 of @code{p->next} does not fault if @code{p->next} is not a valid
8483 address, but evaluation faults if @code{p} is not a valid address.
8485 If the target does not support data prefetch, the address expression
8486 is evaluated if it includes side effects but no other code is generated
8487 and GCC does not issue a warning.
8490 @deftypefn {Built-in Function} double __builtin_huge_val (void)
8491 Returns a positive infinity, if supported by the floating-point format,
8492 else @code{DBL_MAX}. This function is suitable for implementing the
8493 ISO C macro @code{HUGE_VAL}.
8496 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
8497 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
8500 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
8501 Similar to @code{__builtin_huge_val}, except the return
8502 type is @code{long double}.
8505 @deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...)
8506 This built-in implements the C99 fpclassify functionality. The first
8507 five int arguments should be the target library's notion of the
8508 possible FP classes and are used for return values. They must be
8509 constant values and they must appear in this order: @code{FP_NAN},
8510 @code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and
8511 @code{FP_ZERO}. The ellipsis is for exactly one floating-point value
8512 to classify. GCC treats the last argument as type-generic, which
8513 means it does not do default promotion from float to double.
8516 @deftypefn {Built-in Function} double __builtin_inf (void)
8517 Similar to @code{__builtin_huge_val}, except a warning is generated
8518 if the target floating-point format does not support infinities.
8521 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
8522 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
8525 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
8526 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
8529 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
8530 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
8533 @deftypefn {Built-in Function} float __builtin_inff (void)
8534 Similar to @code{__builtin_inf}, except the return type is @code{float}.
8535 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
8538 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
8539 Similar to @code{__builtin_inf}, except the return
8540 type is @code{long double}.
8543 @deftypefn {Built-in Function} int __builtin_isinf_sign (...)
8544 Similar to @code{isinf}, except the return value is negative for
8545 an argument of @code{-Inf}. Note while the parameter list is an
8546 ellipsis, this function only accepts exactly one floating-point
8547 argument. GCC treats this parameter as type-generic, which means it
8548 does not do default promotion from float to double.
8551 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
8552 This is an implementation of the ISO C99 function @code{nan}.
8554 Since ISO C99 defines this function in terms of @code{strtod}, which we
8555 do not implement, a description of the parsing is in order. The string
8556 is parsed as by @code{strtol}; that is, the base is recognized by
8557 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
8558 in the significand such that the least significant bit of the number
8559 is at the least significant bit of the significand. The number is
8560 truncated to fit the significand field provided. The significand is
8561 forced to be a quiet NaN@.
8563 This function, if given a string literal all of which would have been
8564 consumed by @code{strtol}, is evaluated early enough that it is considered a
8565 compile-time constant.
8568 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
8569 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
8572 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
8573 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
8576 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
8577 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
8580 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
8581 Similar to @code{__builtin_nan}, except the return type is @code{float}.
8584 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
8585 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
8588 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
8589 Similar to @code{__builtin_nan}, except the significand is forced
8590 to be a signaling NaN@. The @code{nans} function is proposed by
8591 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
8594 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
8595 Similar to @code{__builtin_nans}, except the return type is @code{float}.
8598 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
8599 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
8602 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
8603 Returns one plus the index of the least significant 1-bit of @var{x}, or
8604 if @var{x} is zero, returns zero.
8607 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
8608 Returns the number of leading 0-bits in @var{x}, starting at the most
8609 significant bit position. If @var{x} is 0, the result is undefined.
8612 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
8613 Returns the number of trailing 0-bits in @var{x}, starting at the least
8614 significant bit position. If @var{x} is 0, the result is undefined.
8617 @deftypefn {Built-in Function} int __builtin_clrsb (int x)
8618 Returns the number of leading redundant sign bits in @var{x}, i.e.@: the
8619 number of bits following the most significant bit that are identical
8620 to it. There are no special cases for 0 or other values.
8623 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
8624 Returns the number of 1-bits in @var{x}.
8627 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
8628 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
8632 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
8633 Similar to @code{__builtin_ffs}, except the argument type is
8634 @code{unsigned long}.
8637 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
8638 Similar to @code{__builtin_clz}, except the argument type is
8639 @code{unsigned long}.
8642 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
8643 Similar to @code{__builtin_ctz}, except the argument type is
8644 @code{unsigned long}.
8647 @deftypefn {Built-in Function} int __builtin_clrsbl (long)
8648 Similar to @code{__builtin_clrsb}, except the argument type is
8652 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
8653 Similar to @code{__builtin_popcount}, except the argument type is
8654 @code{unsigned long}.
8657 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
8658 Similar to @code{__builtin_parity}, except the argument type is
8659 @code{unsigned long}.
8662 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
8663 Similar to @code{__builtin_ffs}, except the argument type is
8664 @code{unsigned long long}.
8667 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
8668 Similar to @code{__builtin_clz}, except the argument type is
8669 @code{unsigned long long}.
8672 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
8673 Similar to @code{__builtin_ctz}, except the argument type is
8674 @code{unsigned long long}.
8677 @deftypefn {Built-in Function} int __builtin_clrsbll (long long)
8678 Similar to @code{__builtin_clrsb}, except the argument type is
8682 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
8683 Similar to @code{__builtin_popcount}, except the argument type is
8684 @code{unsigned long long}.
8687 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
8688 Similar to @code{__builtin_parity}, except the argument type is
8689 @code{unsigned long long}.
8692 @deftypefn {Built-in Function} double __builtin_powi (double, int)
8693 Returns the first argument raised to the power of the second. Unlike the
8694 @code{pow} function no guarantees about precision and rounding are made.
8697 @deftypefn {Built-in Function} float __builtin_powif (float, int)
8698 Similar to @code{__builtin_powi}, except the argument and return types
8702 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
8703 Similar to @code{__builtin_powi}, except the argument and return types
8704 are @code{long double}.
8707 @deftypefn {Built-in Function} uint16_t __builtin_bswap16 (uint16_t x)
8708 Returns @var{x} with the order of the bytes reversed; for example,
8709 @code{0xaabb} becomes @code{0xbbaa}. Byte here always means
8713 @deftypefn {Built-in Function} uint32_t __builtin_bswap32 (uint32_t x)
8714 Similar to @code{__builtin_bswap16}, except the argument and return types
8718 @deftypefn {Built-in Function} uint64_t __builtin_bswap64 (uint64_t x)
8719 Similar to @code{__builtin_bswap32}, except the argument and return types
8723 @node Target Builtins
8724 @section Built-in Functions Specific to Particular Target Machines
8726 On some target machines, GCC supports many built-in functions specific
8727 to those machines. Generally these generate calls to specific machine
8728 instructions, but allow the compiler to schedule those calls.
8731 * Alpha Built-in Functions::
8732 * ARM iWMMXt Built-in Functions::
8733 * ARM NEON Intrinsics::
8734 * AVR Built-in Functions::
8735 * Blackfin Built-in Functions::
8736 * FR-V Built-in Functions::
8737 * X86 Built-in Functions::
8738 * MIPS DSP Built-in Functions::
8739 * MIPS Paired-Single Support::
8740 * MIPS Loongson Built-in Functions::
8741 * Other MIPS Built-in Functions::
8742 * picoChip Built-in Functions::
8743 * PowerPC Built-in Functions::
8744 * PowerPC AltiVec/VSX Built-in Functions::
8745 * RX Built-in Functions::
8746 * SH Built-in Functions::
8747 * SPARC VIS Built-in Functions::
8748 * SPU Built-in Functions::
8749 * TI C6X Built-in Functions::
8750 * TILE-Gx Built-in Functions::
8751 * TILEPro Built-in Functions::
8754 @node Alpha Built-in Functions
8755 @subsection Alpha Built-in Functions
8757 These built-in functions are available for the Alpha family of
8758 processors, depending on the command-line switches used.
8760 The following built-in functions are always available. They
8761 all generate the machine instruction that is part of the name.
8764 long __builtin_alpha_implver (void)
8765 long __builtin_alpha_rpcc (void)
8766 long __builtin_alpha_amask (long)
8767 long __builtin_alpha_cmpbge (long, long)
8768 long __builtin_alpha_extbl (long, long)
8769 long __builtin_alpha_extwl (long, long)
8770 long __builtin_alpha_extll (long, long)
8771 long __builtin_alpha_extql (long, long)
8772 long __builtin_alpha_extwh (long, long)
8773 long __builtin_alpha_extlh (long, long)
8774 long __builtin_alpha_extqh (long, long)
8775 long __builtin_alpha_insbl (long, long)
8776 long __builtin_alpha_inswl (long, long)
8777 long __builtin_alpha_insll (long, long)
8778 long __builtin_alpha_insql (long, long)
8779 long __builtin_alpha_inswh (long, long)
8780 long __builtin_alpha_inslh (long, long)
8781 long __builtin_alpha_insqh (long, long)
8782 long __builtin_alpha_mskbl (long, long)
8783 long __builtin_alpha_mskwl (long, long)
8784 long __builtin_alpha_mskll (long, long)
8785 long __builtin_alpha_mskql (long, long)
8786 long __builtin_alpha_mskwh (long, long)
8787 long __builtin_alpha_msklh (long, long)
8788 long __builtin_alpha_mskqh (long, long)
8789 long __builtin_alpha_umulh (long, long)
8790 long __builtin_alpha_zap (long, long)
8791 long __builtin_alpha_zapnot (long, long)
8794 The following built-in functions are always with @option{-mmax}
8795 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
8796 later. They all generate the machine instruction that is part
8800 long __builtin_alpha_pklb (long)
8801 long __builtin_alpha_pkwb (long)
8802 long __builtin_alpha_unpkbl (long)
8803 long __builtin_alpha_unpkbw (long)
8804 long __builtin_alpha_minub8 (long, long)
8805 long __builtin_alpha_minsb8 (long, long)
8806 long __builtin_alpha_minuw4 (long, long)
8807 long __builtin_alpha_minsw4 (long, long)
8808 long __builtin_alpha_maxub8 (long, long)
8809 long __builtin_alpha_maxsb8 (long, long)
8810 long __builtin_alpha_maxuw4 (long, long)
8811 long __builtin_alpha_maxsw4 (long, long)
8812 long __builtin_alpha_perr (long, long)
8815 The following built-in functions are always with @option{-mcix}
8816 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
8817 later. They all generate the machine instruction that is part
8821 long __builtin_alpha_cttz (long)
8822 long __builtin_alpha_ctlz (long)
8823 long __builtin_alpha_ctpop (long)
8826 The following built-in functions are available on systems that use the OSF/1
8827 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
8828 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
8829 @code{rdval} and @code{wrval}.
8832 void *__builtin_thread_pointer (void)
8833 void __builtin_set_thread_pointer (void *)
8836 @node ARM iWMMXt Built-in Functions
8837 @subsection ARM iWMMXt Built-in Functions
8839 These built-in functions are available for the ARM family of
8840 processors when the @option{-mcpu=iwmmxt} switch is used:
8843 typedef int v2si __attribute__ ((vector_size (8)));
8844 typedef short v4hi __attribute__ ((vector_size (8)));
8845 typedef char v8qi __attribute__ ((vector_size (8)));
8847 int __builtin_arm_getwcgr0 (void)
8848 void __builtin_arm_setwcgr0 (int)
8849 int __builtin_arm_getwcgr1 (void)
8850 void __builtin_arm_setwcgr1 (int)
8851 int __builtin_arm_getwcgr2 (void)
8852 void __builtin_arm_setwcgr2 (int)
8853 int __builtin_arm_getwcgr3 (void)
8854 void __builtin_arm_setwcgr3 (int)
8855 int __builtin_arm_textrmsb (v8qi, int)
8856 int __builtin_arm_textrmsh (v4hi, int)
8857 int __builtin_arm_textrmsw (v2si, int)
8858 int __builtin_arm_textrmub (v8qi, int)
8859 int __builtin_arm_textrmuh (v4hi, int)
8860 int __builtin_arm_textrmuw (v2si, int)
8861 v8qi __builtin_arm_tinsrb (v8qi, int, int)
8862 v4hi __builtin_arm_tinsrh (v4hi, int, int)
8863 v2si __builtin_arm_tinsrw (v2si, int, int)
8864 long long __builtin_arm_tmia (long long, int, int)
8865 long long __builtin_arm_tmiabb (long long, int, int)
8866 long long __builtin_arm_tmiabt (long long, int, int)
8867 long long __builtin_arm_tmiaph (long long, int, int)
8868 long long __builtin_arm_tmiatb (long long, int, int)
8869 long long __builtin_arm_tmiatt (long long, int, int)
8870 int __builtin_arm_tmovmskb (v8qi)
8871 int __builtin_arm_tmovmskh (v4hi)
8872 int __builtin_arm_tmovmskw (v2si)
8873 long long __builtin_arm_waccb (v8qi)
8874 long long __builtin_arm_wacch (v4hi)
8875 long long __builtin_arm_waccw (v2si)
8876 v8qi __builtin_arm_waddb (v8qi, v8qi)
8877 v8qi __builtin_arm_waddbss (v8qi, v8qi)
8878 v8qi __builtin_arm_waddbus (v8qi, v8qi)
8879 v4hi __builtin_arm_waddh (v4hi, v4hi)
8880 v4hi __builtin_arm_waddhss (v4hi, v4hi)
8881 v4hi __builtin_arm_waddhus (v4hi, v4hi)
8882 v2si __builtin_arm_waddw (v2si, v2si)
8883 v2si __builtin_arm_waddwss (v2si, v2si)
8884 v2si __builtin_arm_waddwus (v2si, v2si)
8885 v8qi __builtin_arm_walign (v8qi, v8qi, int)
8886 long long __builtin_arm_wand(long long, long long)
8887 long long __builtin_arm_wandn (long long, long long)
8888 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
8889 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
8890 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
8891 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
8892 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
8893 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
8894 v2si __builtin_arm_wcmpeqw (v2si, v2si)
8895 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
8896 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
8897 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
8898 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
8899 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
8900 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
8901 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
8902 long long __builtin_arm_wmacsz (v4hi, v4hi)
8903 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
8904 long long __builtin_arm_wmacuz (v4hi, v4hi)
8905 v4hi __builtin_arm_wmadds (v4hi, v4hi)
8906 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
8907 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
8908 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
8909 v2si __builtin_arm_wmaxsw (v2si, v2si)
8910 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
8911 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
8912 v2si __builtin_arm_wmaxuw (v2si, v2si)
8913 v8qi __builtin_arm_wminsb (v8qi, v8qi)
8914 v4hi __builtin_arm_wminsh (v4hi, v4hi)
8915 v2si __builtin_arm_wminsw (v2si, v2si)
8916 v8qi __builtin_arm_wminub (v8qi, v8qi)
8917 v4hi __builtin_arm_wminuh (v4hi, v4hi)
8918 v2si __builtin_arm_wminuw (v2si, v2si)
8919 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
8920 v4hi __builtin_arm_wmulul (v4hi, v4hi)
8921 v4hi __builtin_arm_wmulum (v4hi, v4hi)
8922 long long __builtin_arm_wor (long long, long long)
8923 v2si __builtin_arm_wpackdss (long long, long long)
8924 v2si __builtin_arm_wpackdus (long long, long long)
8925 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
8926 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
8927 v4hi __builtin_arm_wpackwss (v2si, v2si)
8928 v4hi __builtin_arm_wpackwus (v2si, v2si)
8929 long long __builtin_arm_wrord (long long, long long)
8930 long long __builtin_arm_wrordi (long long, int)
8931 v4hi __builtin_arm_wrorh (v4hi, long long)
8932 v4hi __builtin_arm_wrorhi (v4hi, int)
8933 v2si __builtin_arm_wrorw (v2si, long long)
8934 v2si __builtin_arm_wrorwi (v2si, int)
8935 v2si __builtin_arm_wsadb (v2si, v8qi, v8qi)
8936 v2si __builtin_arm_wsadbz (v8qi, v8qi)
8937 v2si __builtin_arm_wsadh (v2si, v4hi, v4hi)
8938 v2si __builtin_arm_wsadhz (v4hi, v4hi)
8939 v4hi __builtin_arm_wshufh (v4hi, int)
8940 long long __builtin_arm_wslld (long long, long long)
8941 long long __builtin_arm_wslldi (long long, int)
8942 v4hi __builtin_arm_wsllh (v4hi, long long)
8943 v4hi __builtin_arm_wsllhi (v4hi, int)
8944 v2si __builtin_arm_wsllw (v2si, long long)
8945 v2si __builtin_arm_wsllwi (v2si, int)
8946 long long __builtin_arm_wsrad (long long, long long)
8947 long long __builtin_arm_wsradi (long long, int)
8948 v4hi __builtin_arm_wsrah (v4hi, long long)
8949 v4hi __builtin_arm_wsrahi (v4hi, int)
8950 v2si __builtin_arm_wsraw (v2si, long long)
8951 v2si __builtin_arm_wsrawi (v2si, int)
8952 long long __builtin_arm_wsrld (long long, long long)
8953 long long __builtin_arm_wsrldi (long long, int)
8954 v4hi __builtin_arm_wsrlh (v4hi, long long)
8955 v4hi __builtin_arm_wsrlhi (v4hi, int)
8956 v2si __builtin_arm_wsrlw (v2si, long long)
8957 v2si __builtin_arm_wsrlwi (v2si, int)
8958 v8qi __builtin_arm_wsubb (v8qi, v8qi)
8959 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
8960 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
8961 v4hi __builtin_arm_wsubh (v4hi, v4hi)
8962 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
8963 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
8964 v2si __builtin_arm_wsubw (v2si, v2si)
8965 v2si __builtin_arm_wsubwss (v2si, v2si)
8966 v2si __builtin_arm_wsubwus (v2si, v2si)
8967 v4hi __builtin_arm_wunpckehsb (v8qi)
8968 v2si __builtin_arm_wunpckehsh (v4hi)
8969 long long __builtin_arm_wunpckehsw (v2si)
8970 v4hi __builtin_arm_wunpckehub (v8qi)
8971 v2si __builtin_arm_wunpckehuh (v4hi)
8972 long long __builtin_arm_wunpckehuw (v2si)
8973 v4hi __builtin_arm_wunpckelsb (v8qi)
8974 v2si __builtin_arm_wunpckelsh (v4hi)
8975 long long __builtin_arm_wunpckelsw (v2si)
8976 v4hi __builtin_arm_wunpckelub (v8qi)
8977 v2si __builtin_arm_wunpckeluh (v4hi)
8978 long long __builtin_arm_wunpckeluw (v2si)
8979 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
8980 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
8981 v2si __builtin_arm_wunpckihw (v2si, v2si)
8982 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
8983 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
8984 v2si __builtin_arm_wunpckilw (v2si, v2si)
8985 long long __builtin_arm_wxor (long long, long long)
8986 long long __builtin_arm_wzero ()
8989 @node ARM NEON Intrinsics
8990 @subsection ARM NEON Intrinsics
8992 These built-in intrinsics for the ARM Advanced SIMD extension are available
8993 when the @option{-mfpu=neon} switch is used:
8995 @include arm-neon-intrinsics.texi
8997 @node AVR Built-in Functions
8998 @subsection AVR Built-in Functions
9000 For each built-in function for AVR, there is an equally named,
9001 uppercase built-in macro defined. That way users can easily query if
9002 or if not a specific built-in is implemented or not. For example, if
9003 @code{__builtin_avr_nop} is available the macro
9004 @code{__BUILTIN_AVR_NOP} is defined to @code{1} and undefined otherwise.
9006 The following built-in functions map to the respective machine
9007 instruction, i.e.@: @code{nop}, @code{sei}, @code{cli}, @code{sleep},
9008 @code{wdr}, @code{swap}, @code{fmul}, @code{fmuls}
9009 resp. @code{fmulsu}. The three @code{fmul*} built-ins are implemented
9010 as library call if no hardware multiplier is available.
9013 void __builtin_avr_nop (void)
9014 void __builtin_avr_sei (void)
9015 void __builtin_avr_cli (void)
9016 void __builtin_avr_sleep (void)
9017 void __builtin_avr_wdr (void)
9018 unsigned char __builtin_avr_swap (unsigned char)
9019 unsigned int __builtin_avr_fmul (unsigned char, unsigned char)
9020 int __builtin_avr_fmuls (char, char)
9021 int __builtin_avr_fmulsu (char, unsigned char)
9024 In order to delay execution for a specific number of cycles, GCC
9027 void __builtin_avr_delay_cycles (unsigned long ticks)
9031 @code{ticks} is the number of ticks to delay execution. Note that this
9032 built-in does not take into account the effect of interrupts that
9033 might increase delay time. @code{ticks} must be a compile-time
9034 integer constant; delays with a variable number of cycles are not supported.
9037 char __builtin_avr_flash_segment (const __memx void*)
9041 This built-in takes a byte address to the 24-bit
9042 @ref{AVR Named Address Spaces,address space} @code{__memx} and returns
9043 the number of the flash segment (the 64 KiB chunk) where the address
9044 points to. Counting starts at @code{0}.
9045 If the address does not point to flash memory, return @code{-1}.
9048 unsigned char __builtin_avr_insert_bits (unsigned long map, unsigned char bits, unsigned char val)
9052 Insert bits from @var{bits} into @var{val} and return the resulting
9053 value. The nibbles of @var{map} determine how the insertion is
9054 performed: Let @var{X} be the @var{n}-th nibble of @var{map}
9056 @item If @var{X} is @code{0xf},
9057 then the @var{n}-th bit of @var{val} is returned unaltered.
9059 @item If X is in the range 0@dots{}7,
9060 then the @var{n}-th result bit is set to the @var{X}-th bit of @var{bits}
9062 @item If X is in the range 8@dots{}@code{0xe},
9063 then the @var{n}-th result bit is undefined.
9067 One typical use case for this built-in is adjusting input and
9068 output values to non-contiguous port layouts. Some examples:
9071 // same as val, bits is unused
9072 __builtin_avr_insert_bits (0xffffffff, bits, val)
9076 // same as bits, val is unused
9077 __builtin_avr_insert_bits (0x76543210, bits, val)
9081 // same as rotating bits by 4
9082 __builtin_avr_insert_bits (0x32107654, bits, 0)
9086 // high nibble of result is the high nibble of val
9087 // low nibble of result is the low nibble of bits
9088 __builtin_avr_insert_bits (0xffff3210, bits, val)
9092 // reverse the bit order of bits
9093 __builtin_avr_insert_bits (0x01234567, bits, 0)
9096 @node Blackfin Built-in Functions
9097 @subsection Blackfin Built-in Functions
9099 Currently, there are two Blackfin-specific built-in functions. These are
9100 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
9101 using inline assembly; by using these built-in functions the compiler can
9102 automatically add workarounds for hardware errata involving these
9103 instructions. These functions are named as follows:
9106 void __builtin_bfin_csync (void)
9107 void __builtin_bfin_ssync (void)
9110 @node FR-V Built-in Functions
9111 @subsection FR-V Built-in Functions
9113 GCC provides many FR-V-specific built-in functions. In general,
9114 these functions are intended to be compatible with those described
9115 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
9116 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
9117 @code{__MBTOHE}, the GCC forms of which pass 128-bit values by
9118 pointer rather than by value.
9120 Most of the functions are named after specific FR-V instructions.
9121 Such functions are said to be ``directly mapped'' and are summarized
9122 here in tabular form.
9126 * Directly-mapped Integer Functions::
9127 * Directly-mapped Media Functions::
9128 * Raw read/write Functions::
9129 * Other Built-in Functions::
9132 @node Argument Types
9133 @subsubsection Argument Types
9135 The arguments to the built-in functions can be divided into three groups:
9136 register numbers, compile-time constants and run-time values. In order
9137 to make this classification clear at a glance, the arguments and return
9138 values are given the following pseudo types:
9140 @multitable @columnfractions .20 .30 .15 .35
9141 @item Pseudo type @tab Real C type @tab Constant? @tab Description
9142 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
9143 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
9144 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
9145 @item @code{uw2} @tab @code{unsigned long long} @tab No
9146 @tab an unsigned doubleword
9147 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
9148 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
9149 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
9150 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
9153 These pseudo types are not defined by GCC, they are simply a notational
9154 convenience used in this manual.
9156 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
9157 and @code{sw2} are evaluated at run time. They correspond to
9158 register operands in the underlying FR-V instructions.
9160 @code{const} arguments represent immediate operands in the underlying
9161 FR-V instructions. They must be compile-time constants.
9163 @code{acc} arguments are evaluated at compile time and specify the number
9164 of an accumulator register. For example, an @code{acc} argument of 2
9165 selects the ACC2 register.
9167 @code{iacc} arguments are similar to @code{acc} arguments but specify the
9168 number of an IACC register. See @pxref{Other Built-in Functions}
9171 @node Directly-mapped Integer Functions
9172 @subsubsection Directly-mapped Integer Functions
9174 The functions listed below map directly to FR-V I-type instructions.
9176 @multitable @columnfractions .45 .32 .23
9177 @item Function prototype @tab Example usage @tab Assembly output
9178 @item @code{sw1 __ADDSS (sw1, sw1)}
9179 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
9180 @tab @code{ADDSS @var{a},@var{b},@var{c}}
9181 @item @code{sw1 __SCAN (sw1, sw1)}
9182 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
9183 @tab @code{SCAN @var{a},@var{b},@var{c}}
9184 @item @code{sw1 __SCUTSS (sw1)}
9185 @tab @code{@var{b} = __SCUTSS (@var{a})}
9186 @tab @code{SCUTSS @var{a},@var{b}}
9187 @item @code{sw1 __SLASS (sw1, sw1)}
9188 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
9189 @tab @code{SLASS @var{a},@var{b},@var{c}}
9190 @item @code{void __SMASS (sw1, sw1)}
9191 @tab @code{__SMASS (@var{a}, @var{b})}
9192 @tab @code{SMASS @var{a},@var{b}}
9193 @item @code{void __SMSSS (sw1, sw1)}
9194 @tab @code{__SMSSS (@var{a}, @var{b})}
9195 @tab @code{SMSSS @var{a},@var{b}}
9196 @item @code{void __SMU (sw1, sw1)}
9197 @tab @code{__SMU (@var{a}, @var{b})}
9198 @tab @code{SMU @var{a},@var{b}}
9199 @item @code{sw2 __SMUL (sw1, sw1)}
9200 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
9201 @tab @code{SMUL @var{a},@var{b},@var{c}}
9202 @item @code{sw1 __SUBSS (sw1, sw1)}
9203 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
9204 @tab @code{SUBSS @var{a},@var{b},@var{c}}
9205 @item @code{uw2 __UMUL (uw1, uw1)}
9206 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
9207 @tab @code{UMUL @var{a},@var{b},@var{c}}
9210 @node Directly-mapped Media Functions
9211 @subsubsection Directly-mapped Media Functions
9213 The functions listed below map directly to FR-V M-type instructions.
9215 @multitable @columnfractions .45 .32 .23
9216 @item Function prototype @tab Example usage @tab Assembly output
9217 @item @code{uw1 __MABSHS (sw1)}
9218 @tab @code{@var{b} = __MABSHS (@var{a})}
9219 @tab @code{MABSHS @var{a},@var{b}}
9220 @item @code{void __MADDACCS (acc, acc)}
9221 @tab @code{__MADDACCS (@var{b}, @var{a})}
9222 @tab @code{MADDACCS @var{a},@var{b}}
9223 @item @code{sw1 __MADDHSS (sw1, sw1)}
9224 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
9225 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
9226 @item @code{uw1 __MADDHUS (uw1, uw1)}
9227 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
9228 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
9229 @item @code{uw1 __MAND (uw1, uw1)}
9230 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
9231 @tab @code{MAND @var{a},@var{b},@var{c}}
9232 @item @code{void __MASACCS (acc, acc)}
9233 @tab @code{__MASACCS (@var{b}, @var{a})}
9234 @tab @code{MASACCS @var{a},@var{b}}
9235 @item @code{uw1 __MAVEH (uw1, uw1)}
9236 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
9237 @tab @code{MAVEH @var{a},@var{b},@var{c}}
9238 @item @code{uw2 __MBTOH (uw1)}
9239 @tab @code{@var{b} = __MBTOH (@var{a})}
9240 @tab @code{MBTOH @var{a},@var{b}}
9241 @item @code{void __MBTOHE (uw1 *, uw1)}
9242 @tab @code{__MBTOHE (&@var{b}, @var{a})}
9243 @tab @code{MBTOHE @var{a},@var{b}}
9244 @item @code{void __MCLRACC (acc)}
9245 @tab @code{__MCLRACC (@var{a})}
9246 @tab @code{MCLRACC @var{a}}
9247 @item @code{void __MCLRACCA (void)}
9248 @tab @code{__MCLRACCA ()}
9249 @tab @code{MCLRACCA}
9250 @item @code{uw1 __Mcop1 (uw1, uw1)}
9251 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
9252 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
9253 @item @code{uw1 __Mcop2 (uw1, uw1)}
9254 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
9255 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
9256 @item @code{uw1 __MCPLHI (uw2, const)}
9257 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
9258 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
9259 @item @code{uw1 __MCPLI (uw2, const)}
9260 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
9261 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
9262 @item @code{void __MCPXIS (acc, sw1, sw1)}
9263 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
9264 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
9265 @item @code{void __MCPXIU (acc, uw1, uw1)}
9266 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
9267 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
9268 @item @code{void __MCPXRS (acc, sw1, sw1)}
9269 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
9270 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
9271 @item @code{void __MCPXRU (acc, uw1, uw1)}
9272 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
9273 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
9274 @item @code{uw1 __MCUT (acc, uw1)}
9275 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
9276 @tab @code{MCUT @var{a},@var{b},@var{c}}
9277 @item @code{uw1 __MCUTSS (acc, sw1)}
9278 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
9279 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
9280 @item @code{void __MDADDACCS (acc, acc)}
9281 @tab @code{__MDADDACCS (@var{b}, @var{a})}
9282 @tab @code{MDADDACCS @var{a},@var{b}}
9283 @item @code{void __MDASACCS (acc, acc)}
9284 @tab @code{__MDASACCS (@var{b}, @var{a})}
9285 @tab @code{MDASACCS @var{a},@var{b}}
9286 @item @code{uw2 __MDCUTSSI (acc, const)}
9287 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
9288 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
9289 @item @code{uw2 __MDPACKH (uw2, uw2)}
9290 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
9291 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
9292 @item @code{uw2 __MDROTLI (uw2, const)}
9293 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
9294 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
9295 @item @code{void __MDSUBACCS (acc, acc)}
9296 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
9297 @tab @code{MDSUBACCS @var{a},@var{b}}
9298 @item @code{void __MDUNPACKH (uw1 *, uw2)}
9299 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
9300 @tab @code{MDUNPACKH @var{a},@var{b}}
9301 @item @code{uw2 __MEXPDHD (uw1, const)}
9302 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
9303 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
9304 @item @code{uw1 __MEXPDHW (uw1, const)}
9305 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
9306 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
9307 @item @code{uw1 __MHDSETH (uw1, const)}
9308 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
9309 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
9310 @item @code{sw1 __MHDSETS (const)}
9311 @tab @code{@var{b} = __MHDSETS (@var{a})}
9312 @tab @code{MHDSETS #@var{a},@var{b}}
9313 @item @code{uw1 __MHSETHIH (uw1, const)}
9314 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
9315 @tab @code{MHSETHIH #@var{a},@var{b}}
9316 @item @code{sw1 __MHSETHIS (sw1, const)}
9317 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
9318 @tab @code{MHSETHIS #@var{a},@var{b}}
9319 @item @code{uw1 __MHSETLOH (uw1, const)}
9320 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
9321 @tab @code{MHSETLOH #@var{a},@var{b}}
9322 @item @code{sw1 __MHSETLOS (sw1, const)}
9323 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
9324 @tab @code{MHSETLOS #@var{a},@var{b}}
9325 @item @code{uw1 __MHTOB (uw2)}
9326 @tab @code{@var{b} = __MHTOB (@var{a})}
9327 @tab @code{MHTOB @var{a},@var{b}}
9328 @item @code{void __MMACHS (acc, sw1, sw1)}
9329 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
9330 @tab @code{MMACHS @var{a},@var{b},@var{c}}
9331 @item @code{void __MMACHU (acc, uw1, uw1)}
9332 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
9333 @tab @code{MMACHU @var{a},@var{b},@var{c}}
9334 @item @code{void __MMRDHS (acc, sw1, sw1)}
9335 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
9336 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
9337 @item @code{void __MMRDHU (acc, uw1, uw1)}
9338 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
9339 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
9340 @item @code{void __MMULHS (acc, sw1, sw1)}
9341 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
9342 @tab @code{MMULHS @var{a},@var{b},@var{c}}
9343 @item @code{void __MMULHU (acc, uw1, uw1)}
9344 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
9345 @tab @code{MMULHU @var{a},@var{b},@var{c}}
9346 @item @code{void __MMULXHS (acc, sw1, sw1)}
9347 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
9348 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
9349 @item @code{void __MMULXHU (acc, uw1, uw1)}
9350 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
9351 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
9352 @item @code{uw1 __MNOT (uw1)}
9353 @tab @code{@var{b} = __MNOT (@var{a})}
9354 @tab @code{MNOT @var{a},@var{b}}
9355 @item @code{uw1 __MOR (uw1, uw1)}
9356 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
9357 @tab @code{MOR @var{a},@var{b},@var{c}}
9358 @item @code{uw1 __MPACKH (uh, uh)}
9359 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
9360 @tab @code{MPACKH @var{a},@var{b},@var{c}}
9361 @item @code{sw2 __MQADDHSS (sw2, sw2)}
9362 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
9363 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
9364 @item @code{uw2 __MQADDHUS (uw2, uw2)}
9365 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
9366 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
9367 @item @code{void __MQCPXIS (acc, sw2, sw2)}
9368 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
9369 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
9370 @item @code{void __MQCPXIU (acc, uw2, uw2)}
9371 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
9372 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
9373 @item @code{void __MQCPXRS (acc, sw2, sw2)}
9374 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
9375 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
9376 @item @code{void __MQCPXRU (acc, uw2, uw2)}
9377 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
9378 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
9379 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
9380 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
9381 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
9382 @item @code{sw2 __MQLMTHS (sw2, sw2)}
9383 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
9384 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
9385 @item @code{void __MQMACHS (acc, sw2, sw2)}
9386 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
9387 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
9388 @item @code{void __MQMACHU (acc, uw2, uw2)}
9389 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
9390 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
9391 @item @code{void __MQMACXHS (acc, sw2, sw2)}
9392 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
9393 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
9394 @item @code{void __MQMULHS (acc, sw2, sw2)}
9395 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
9396 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
9397 @item @code{void __MQMULHU (acc, uw2, uw2)}
9398 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
9399 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
9400 @item @code{void __MQMULXHS (acc, sw2, sw2)}
9401 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
9402 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
9403 @item @code{void __MQMULXHU (acc, uw2, uw2)}
9404 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
9405 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
9406 @item @code{sw2 __MQSATHS (sw2, sw2)}
9407 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
9408 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
9409 @item @code{uw2 __MQSLLHI (uw2, int)}
9410 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
9411 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
9412 @item @code{sw2 __MQSRAHI (sw2, int)}
9413 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
9414 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
9415 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
9416 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
9417 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
9418 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
9419 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
9420 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
9421 @item @code{void __MQXMACHS (acc, sw2, sw2)}
9422 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
9423 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
9424 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
9425 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
9426 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
9427 @item @code{uw1 __MRDACC (acc)}
9428 @tab @code{@var{b} = __MRDACC (@var{a})}
9429 @tab @code{MRDACC @var{a},@var{b}}
9430 @item @code{uw1 __MRDACCG (acc)}
9431 @tab @code{@var{b} = __MRDACCG (@var{a})}
9432 @tab @code{MRDACCG @var{a},@var{b}}
9433 @item @code{uw1 __MROTLI (uw1, const)}
9434 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
9435 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
9436 @item @code{uw1 __MROTRI (uw1, const)}
9437 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
9438 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
9439 @item @code{sw1 __MSATHS (sw1, sw1)}
9440 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
9441 @tab @code{MSATHS @var{a},@var{b},@var{c}}
9442 @item @code{uw1 __MSATHU (uw1, uw1)}
9443 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
9444 @tab @code{MSATHU @var{a},@var{b},@var{c}}
9445 @item @code{uw1 __MSLLHI (uw1, const)}
9446 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
9447 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
9448 @item @code{sw1 __MSRAHI (sw1, const)}
9449 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
9450 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
9451 @item @code{uw1 __MSRLHI (uw1, const)}
9452 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
9453 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
9454 @item @code{void __MSUBACCS (acc, acc)}
9455 @tab @code{__MSUBACCS (@var{b}, @var{a})}
9456 @tab @code{MSUBACCS @var{a},@var{b}}
9457 @item @code{sw1 __MSUBHSS (sw1, sw1)}
9458 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
9459 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
9460 @item @code{uw1 __MSUBHUS (uw1, uw1)}
9461 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
9462 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
9463 @item @code{void __MTRAP (void)}
9464 @tab @code{__MTRAP ()}
9466 @item @code{uw2 __MUNPACKH (uw1)}
9467 @tab @code{@var{b} = __MUNPACKH (@var{a})}
9468 @tab @code{MUNPACKH @var{a},@var{b}}
9469 @item @code{uw1 __MWCUT (uw2, uw1)}
9470 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
9471 @tab @code{MWCUT @var{a},@var{b},@var{c}}
9472 @item @code{void __MWTACC (acc, uw1)}
9473 @tab @code{__MWTACC (@var{b}, @var{a})}
9474 @tab @code{MWTACC @var{a},@var{b}}
9475 @item @code{void __MWTACCG (acc, uw1)}
9476 @tab @code{__MWTACCG (@var{b}, @var{a})}
9477 @tab @code{MWTACCG @var{a},@var{b}}
9478 @item @code{uw1 __MXOR (uw1, uw1)}
9479 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
9480 @tab @code{MXOR @var{a},@var{b},@var{c}}
9483 @node Raw read/write Functions
9484 @subsubsection Raw read/write Functions
9486 This sections describes built-in functions related to read and write
9487 instructions to access memory. These functions generate
9488 @code{membar} instructions to flush the I/O load and stores where
9489 appropriate, as described in Fujitsu's manual described above.
9493 @item unsigned char __builtin_read8 (void *@var{data})
9494 @item unsigned short __builtin_read16 (void *@var{data})
9495 @item unsigned long __builtin_read32 (void *@var{data})
9496 @item unsigned long long __builtin_read64 (void *@var{data})
9498 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
9499 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
9500 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
9501 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
9504 @node Other Built-in Functions
9505 @subsubsection Other Built-in Functions
9507 This section describes built-in functions that are not named after
9508 a specific FR-V instruction.
9511 @item sw2 __IACCreadll (iacc @var{reg})
9512 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
9513 for future expansion and must be 0.
9515 @item sw1 __IACCreadl (iacc @var{reg})
9516 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
9517 Other values of @var{reg} are rejected as invalid.
9519 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
9520 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
9521 is reserved for future expansion and must be 0.
9523 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
9524 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
9525 is 1. Other values of @var{reg} are rejected as invalid.
9527 @item void __data_prefetch0 (const void *@var{x})
9528 Use the @code{dcpl} instruction to load the contents of address @var{x}
9529 into the data cache.
9531 @item void __data_prefetch (const void *@var{x})
9532 Use the @code{nldub} instruction to load the contents of address @var{x}
9533 into the data cache. The instruction is issued in slot I1@.
9536 @node X86 Built-in Functions
9537 @subsection X86 Built-in Functions
9539 These built-in functions are available for the i386 and x86-64 family
9540 of computers, depending on the command-line switches used.
9542 If you specify command-line switches such as @option{-msse},
9543 the compiler could use the extended instruction sets even if the built-ins
9544 are not used explicitly in the program. For this reason, applications
9545 that perform run-time CPU detection must compile separate files for each
9546 supported architecture, using the appropriate flags. In particular,
9547 the file containing the CPU detection code should be compiled without
9550 The following machine modes are available for use with MMX built-in functions
9551 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
9552 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
9553 vector of eight 8-bit integers. Some of the built-in functions operate on
9554 MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode.
9556 If 3DNow!@: extensions are enabled, @code{V2SF} is used as a mode for a vector
9557 of two 32-bit floating-point values.
9559 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
9560 floating-point values. Some instructions use a vector of four 32-bit
9561 integers, these use @code{V4SI}. Finally, some instructions operate on an
9562 entire vector register, interpreting it as a 128-bit integer, these use mode
9565 In 64-bit mode, the x86-64 family of processors uses additional built-in
9566 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
9567 floating point and @code{TC} 128-bit complex floating-point values.
9569 The following floating-point built-in functions are available in 64-bit
9570 mode. All of them implement the function that is part of the name.
9573 __float128 __builtin_fabsq (__float128)
9574 __float128 __builtin_copysignq (__float128, __float128)
9577 The following built-in function is always available.
9580 @item void __builtin_ia32_pause (void)
9581 Generates the @code{pause} machine instruction with a compiler memory
9585 The following floating-point built-in functions are made available in the
9589 @item __float128 __builtin_infq (void)
9590 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
9591 @findex __builtin_infq
9593 @item __float128 __builtin_huge_valq (void)
9594 Similar to @code{__builtin_huge_val}, except the return type is @code{__float128}.
9595 @findex __builtin_huge_valq
9598 The following built-in functions are always available and can be used to
9599 check the target platform type.
9601 @deftypefn {Built-in Function} void __builtin_cpu_init (void)
9602 This function runs the CPU detection code to check the type of CPU and the
9603 features supported. This built-in function needs to be invoked along with the built-in functions
9604 to check CPU type and features, @code{__builtin_cpu_is} and
9605 @code{__builtin_cpu_supports}, only when used in a function that is
9606 executed before any constructors are called. The CPU detection code is
9607 automatically executed in a very high priority constructor.
9609 For example, this function has to be used in @code{ifunc} resolvers that
9610 check for CPU type using the built-in functions @code{__builtin_cpu_is}
9611 and @code{__builtin_cpu_supports}, or in constructors on targets that
9612 don't support constructor priority.
9615 static void (*resolve_memcpy (void)) (void)
9617 // ifunc resolvers fire before constructors, explicitly call the init
9619 __builtin_cpu_init ();
9620 if (__builtin_cpu_supports ("ssse3"))
9621 return ssse3_memcpy; // super fast memcpy with ssse3 instructions.
9623 return default_memcpy;
9626 void *memcpy (void *, const void *, size_t)
9627 __attribute__ ((ifunc ("resolve_memcpy")));
9632 @deftypefn {Built-in Function} int __builtin_cpu_is (const char *@var{cpuname})
9633 This function returns a positive integer if the run-time CPU
9634 is of type @var{cpuname}
9635 and returns @code{0} otherwise. The following CPU names can be detected:
9651 Intel Core i7 Nehalem CPU.
9654 Intel Core i7 Westmere CPU.
9657 Intel Core i7 Sandy Bridge CPU.
9666 AMD Family 10h Barcelona CPU.
9669 AMD Family 10h Shanghai CPU.
9672 AMD Family 10h Istanbul CPU.
9681 AMD Family 15h Bulldozer version 1.
9684 AMD Family 15h Bulldozer version 2.
9687 AMD Family 15h Bulldozer version 3.
9695 if (__builtin_cpu_is ("corei7"))
9697 do_corei7 (); // Core i7 specific implementation.
9701 do_generic (); // Generic implementation.
9706 @deftypefn {Built-in Function} int __builtin_cpu_supports (const char *@var{feature})
9707 This function returns a positive integer if the run-time CPU
9708 supports @var{feature}
9709 and returns @code{0} otherwise. The following features can be detected:
9727 SSE4.1 instructions.
9729 SSE4.2 instructions.
9738 if (__builtin_cpu_supports ("popcnt"))
9740 asm("popcnt %1,%0" : "=r"(count) : "rm"(n) : "cc");
9744 count = generic_countbits (n); //generic implementation.
9750 The following built-in functions are made available by @option{-mmmx}.
9751 All of them generate the machine instruction that is part of the name.
9754 v8qi __builtin_ia32_paddb (v8qi, v8qi)
9755 v4hi __builtin_ia32_paddw (v4hi, v4hi)
9756 v2si __builtin_ia32_paddd (v2si, v2si)
9757 v8qi __builtin_ia32_psubb (v8qi, v8qi)
9758 v4hi __builtin_ia32_psubw (v4hi, v4hi)
9759 v2si __builtin_ia32_psubd (v2si, v2si)
9760 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
9761 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
9762 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
9763 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
9764 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
9765 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
9766 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
9767 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
9768 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
9769 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
9770 di __builtin_ia32_pand (di, di)
9771 di __builtin_ia32_pandn (di,di)
9772 di __builtin_ia32_por (di, di)
9773 di __builtin_ia32_pxor (di, di)
9774 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
9775 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
9776 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
9777 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
9778 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
9779 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
9780 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
9781 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
9782 v2si __builtin_ia32_punpckhdq (v2si, v2si)
9783 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
9784 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
9785 v2si __builtin_ia32_punpckldq (v2si, v2si)
9786 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
9787 v4hi __builtin_ia32_packssdw (v2si, v2si)
9788 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
9790 v4hi __builtin_ia32_psllw (v4hi, v4hi)
9791 v2si __builtin_ia32_pslld (v2si, v2si)
9792 v1di __builtin_ia32_psllq (v1di, v1di)
9793 v4hi __builtin_ia32_psrlw (v4hi, v4hi)
9794 v2si __builtin_ia32_psrld (v2si, v2si)
9795 v1di __builtin_ia32_psrlq (v1di, v1di)
9796 v4hi __builtin_ia32_psraw (v4hi, v4hi)
9797 v2si __builtin_ia32_psrad (v2si, v2si)
9798 v4hi __builtin_ia32_psllwi (v4hi, int)
9799 v2si __builtin_ia32_pslldi (v2si, int)
9800 v1di __builtin_ia32_psllqi (v1di, int)
9801 v4hi __builtin_ia32_psrlwi (v4hi, int)
9802 v2si __builtin_ia32_psrldi (v2si, int)
9803 v1di __builtin_ia32_psrlqi (v1di, int)
9804 v4hi __builtin_ia32_psrawi (v4hi, int)
9805 v2si __builtin_ia32_psradi (v2si, int)
9809 The following built-in functions are made available either with
9810 @option{-msse}, or with a combination of @option{-m3dnow} and
9811 @option{-march=athlon}. All of them generate the machine
9812 instruction that is part of the name.
9815 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
9816 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
9817 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
9818 v1di __builtin_ia32_psadbw (v8qi, v8qi)
9819 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
9820 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
9821 v8qi __builtin_ia32_pminub (v8qi, v8qi)
9822 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
9823 int __builtin_ia32_pextrw (v4hi, int)
9824 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
9825 int __builtin_ia32_pmovmskb (v8qi)
9826 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
9827 void __builtin_ia32_movntq (di *, di)
9828 void __builtin_ia32_sfence (void)
9831 The following built-in functions are available when @option{-msse} is used.
9832 All of them generate the machine instruction that is part of the name.
9835 int __builtin_ia32_comieq (v4sf, v4sf)
9836 int __builtin_ia32_comineq (v4sf, v4sf)
9837 int __builtin_ia32_comilt (v4sf, v4sf)
9838 int __builtin_ia32_comile (v4sf, v4sf)
9839 int __builtin_ia32_comigt (v4sf, v4sf)
9840 int __builtin_ia32_comige (v4sf, v4sf)
9841 int __builtin_ia32_ucomieq (v4sf, v4sf)
9842 int __builtin_ia32_ucomineq (v4sf, v4sf)
9843 int __builtin_ia32_ucomilt (v4sf, v4sf)
9844 int __builtin_ia32_ucomile (v4sf, v4sf)
9845 int __builtin_ia32_ucomigt (v4sf, v4sf)
9846 int __builtin_ia32_ucomige (v4sf, v4sf)
9847 v4sf __builtin_ia32_addps (v4sf, v4sf)
9848 v4sf __builtin_ia32_subps (v4sf, v4sf)
9849 v4sf __builtin_ia32_mulps (v4sf, v4sf)
9850 v4sf __builtin_ia32_divps (v4sf, v4sf)
9851 v4sf __builtin_ia32_addss (v4sf, v4sf)
9852 v4sf __builtin_ia32_subss (v4sf, v4sf)
9853 v4sf __builtin_ia32_mulss (v4sf, v4sf)
9854 v4sf __builtin_ia32_divss (v4sf, v4sf)
9855 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
9856 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
9857 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
9858 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
9859 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
9860 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
9861 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
9862 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
9863 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
9864 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
9865 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
9866 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
9867 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
9868 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
9869 v4si __builtin_ia32_cmpless (v4sf, v4sf)
9870 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
9871 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
9872 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
9873 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
9874 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
9875 v4sf __builtin_ia32_maxps (v4sf, v4sf)
9876 v4sf __builtin_ia32_maxss (v4sf, v4sf)
9877 v4sf __builtin_ia32_minps (v4sf, v4sf)
9878 v4sf __builtin_ia32_minss (v4sf, v4sf)
9879 v4sf __builtin_ia32_andps (v4sf, v4sf)
9880 v4sf __builtin_ia32_andnps (v4sf, v4sf)
9881 v4sf __builtin_ia32_orps (v4sf, v4sf)
9882 v4sf __builtin_ia32_xorps (v4sf, v4sf)
9883 v4sf __builtin_ia32_movss (v4sf, v4sf)
9884 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
9885 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
9886 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
9887 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
9888 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
9889 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
9890 v2si __builtin_ia32_cvtps2pi (v4sf)
9891 int __builtin_ia32_cvtss2si (v4sf)
9892 v2si __builtin_ia32_cvttps2pi (v4sf)
9893 int __builtin_ia32_cvttss2si (v4sf)
9894 v4sf __builtin_ia32_rcpps (v4sf)
9895 v4sf __builtin_ia32_rsqrtps (v4sf)
9896 v4sf __builtin_ia32_sqrtps (v4sf)
9897 v4sf __builtin_ia32_rcpss (v4sf)
9898 v4sf __builtin_ia32_rsqrtss (v4sf)
9899 v4sf __builtin_ia32_sqrtss (v4sf)
9900 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
9901 void __builtin_ia32_movntps (float *, v4sf)
9902 int __builtin_ia32_movmskps (v4sf)
9905 The following built-in functions are available when @option{-msse} is used.
9908 @item v4sf __builtin_ia32_loadaps (float *)
9909 Generates the @code{movaps} machine instruction as a load from memory.
9910 @item void __builtin_ia32_storeaps (float *, v4sf)
9911 Generates the @code{movaps} machine instruction as a store to memory.
9912 @item v4sf __builtin_ia32_loadups (float *)
9913 Generates the @code{movups} machine instruction as a load from memory.
9914 @item void __builtin_ia32_storeups (float *, v4sf)
9915 Generates the @code{movups} machine instruction as a store to memory.
9916 @item v4sf __builtin_ia32_loadsss (float *)
9917 Generates the @code{movss} machine instruction as a load from memory.
9918 @item void __builtin_ia32_storess (float *, v4sf)
9919 Generates the @code{movss} machine instruction as a store to memory.
9920 @item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)
9921 Generates the @code{movhps} machine instruction as a load from memory.
9922 @item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)
9923 Generates the @code{movlps} machine instruction as a load from memory
9924 @item void __builtin_ia32_storehps (v2sf *, v4sf)
9925 Generates the @code{movhps} machine instruction as a store to memory.
9926 @item void __builtin_ia32_storelps (v2sf *, v4sf)
9927 Generates the @code{movlps} machine instruction as a store to memory.
9930 The following built-in functions are available when @option{-msse2} is used.
9931 All of them generate the machine instruction that is part of the name.
9934 int __builtin_ia32_comisdeq (v2df, v2df)
9935 int __builtin_ia32_comisdlt (v2df, v2df)
9936 int __builtin_ia32_comisdle (v2df, v2df)
9937 int __builtin_ia32_comisdgt (v2df, v2df)
9938 int __builtin_ia32_comisdge (v2df, v2df)
9939 int __builtin_ia32_comisdneq (v2df, v2df)
9940 int __builtin_ia32_ucomisdeq (v2df, v2df)
9941 int __builtin_ia32_ucomisdlt (v2df, v2df)
9942 int __builtin_ia32_ucomisdle (v2df, v2df)
9943 int __builtin_ia32_ucomisdgt (v2df, v2df)
9944 int __builtin_ia32_ucomisdge (v2df, v2df)
9945 int __builtin_ia32_ucomisdneq (v2df, v2df)
9946 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
9947 v2df __builtin_ia32_cmpltpd (v2df, v2df)
9948 v2df __builtin_ia32_cmplepd (v2df, v2df)
9949 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
9950 v2df __builtin_ia32_cmpgepd (v2df, v2df)
9951 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
9952 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
9953 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
9954 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
9955 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
9956 v2df __builtin_ia32_cmpngepd (v2df, v2df)
9957 v2df __builtin_ia32_cmpordpd (v2df, v2df)
9958 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
9959 v2df __builtin_ia32_cmpltsd (v2df, v2df)
9960 v2df __builtin_ia32_cmplesd (v2df, v2df)
9961 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
9962 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
9963 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
9964 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
9965 v2df __builtin_ia32_cmpordsd (v2df, v2df)
9966 v2di __builtin_ia32_paddq (v2di, v2di)
9967 v2di __builtin_ia32_psubq (v2di, v2di)
9968 v2df __builtin_ia32_addpd (v2df, v2df)
9969 v2df __builtin_ia32_subpd (v2df, v2df)
9970 v2df __builtin_ia32_mulpd (v2df, v2df)
9971 v2df __builtin_ia32_divpd (v2df, v2df)
9972 v2df __builtin_ia32_addsd (v2df, v2df)
9973 v2df __builtin_ia32_subsd (v2df, v2df)
9974 v2df __builtin_ia32_mulsd (v2df, v2df)
9975 v2df __builtin_ia32_divsd (v2df, v2df)
9976 v2df __builtin_ia32_minpd (v2df, v2df)
9977 v2df __builtin_ia32_maxpd (v2df, v2df)
9978 v2df __builtin_ia32_minsd (v2df, v2df)
9979 v2df __builtin_ia32_maxsd (v2df, v2df)
9980 v2df __builtin_ia32_andpd (v2df, v2df)
9981 v2df __builtin_ia32_andnpd (v2df, v2df)
9982 v2df __builtin_ia32_orpd (v2df, v2df)
9983 v2df __builtin_ia32_xorpd (v2df, v2df)
9984 v2df __builtin_ia32_movsd (v2df, v2df)
9985 v2df __builtin_ia32_unpckhpd (v2df, v2df)
9986 v2df __builtin_ia32_unpcklpd (v2df, v2df)
9987 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
9988 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
9989 v4si __builtin_ia32_paddd128 (v4si, v4si)
9990 v2di __builtin_ia32_paddq128 (v2di, v2di)
9991 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
9992 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
9993 v4si __builtin_ia32_psubd128 (v4si, v4si)
9994 v2di __builtin_ia32_psubq128 (v2di, v2di)
9995 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
9996 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
9997 v2di __builtin_ia32_pand128 (v2di, v2di)
9998 v2di __builtin_ia32_pandn128 (v2di, v2di)
9999 v2di __builtin_ia32_por128 (v2di, v2di)
10000 v2di __builtin_ia32_pxor128 (v2di, v2di)
10001 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
10002 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
10003 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
10004 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
10005 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
10006 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
10007 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
10008 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
10009 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
10010 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
10011 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
10012 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
10013 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
10014 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
10015 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
10016 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
10017 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
10018 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
10019 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
10020 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
10021 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi)
10022 v8hi __builtin_ia32_packssdw128 (v4si, v4si)
10023 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi)
10024 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
10025 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
10026 v2df __builtin_ia32_loadupd (double *)
10027 void __builtin_ia32_storeupd (double *, v2df)
10028 v2df __builtin_ia32_loadhpd (v2df, double const *)
10029 v2df __builtin_ia32_loadlpd (v2df, double const *)
10030 int __builtin_ia32_movmskpd (v2df)
10031 int __builtin_ia32_pmovmskb128 (v16qi)
10032 void __builtin_ia32_movnti (int *, int)
10033 void __builtin_ia32_movnti64 (long long int *, long long int)
10034 void __builtin_ia32_movntpd (double *, v2df)
10035 void __builtin_ia32_movntdq (v2df *, v2df)
10036 v4si __builtin_ia32_pshufd (v4si, int)
10037 v8hi __builtin_ia32_pshuflw (v8hi, int)
10038 v8hi __builtin_ia32_pshufhw (v8hi, int)
10039 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
10040 v2df __builtin_ia32_sqrtpd (v2df)
10041 v2df __builtin_ia32_sqrtsd (v2df)
10042 v2df __builtin_ia32_shufpd (v2df, v2df, int)
10043 v2df __builtin_ia32_cvtdq2pd (v4si)
10044 v4sf __builtin_ia32_cvtdq2ps (v4si)
10045 v4si __builtin_ia32_cvtpd2dq (v2df)
10046 v2si __builtin_ia32_cvtpd2pi (v2df)
10047 v4sf __builtin_ia32_cvtpd2ps (v2df)
10048 v4si __builtin_ia32_cvttpd2dq (v2df)
10049 v2si __builtin_ia32_cvttpd2pi (v2df)
10050 v2df __builtin_ia32_cvtpi2pd (v2si)
10051 int __builtin_ia32_cvtsd2si (v2df)
10052 int __builtin_ia32_cvttsd2si (v2df)
10053 long long __builtin_ia32_cvtsd2si64 (v2df)
10054 long long __builtin_ia32_cvttsd2si64 (v2df)
10055 v4si __builtin_ia32_cvtps2dq (v4sf)
10056 v2df __builtin_ia32_cvtps2pd (v4sf)
10057 v4si __builtin_ia32_cvttps2dq (v4sf)
10058 v2df __builtin_ia32_cvtsi2sd (v2df, int)
10059 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
10060 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
10061 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
10062 void __builtin_ia32_clflush (const void *)
10063 void __builtin_ia32_lfence (void)
10064 void __builtin_ia32_mfence (void)
10065 v16qi __builtin_ia32_loaddqu (const char *)
10066 void __builtin_ia32_storedqu (char *, v16qi)
10067 v1di __builtin_ia32_pmuludq (v2si, v2si)
10068 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
10069 v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
10070 v4si __builtin_ia32_pslld128 (v4si, v4si)
10071 v2di __builtin_ia32_psllq128 (v2di, v2di)
10072 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
10073 v4si __builtin_ia32_psrld128 (v4si, v4si)
10074 v2di __builtin_ia32_psrlq128 (v2di, v2di)
10075 v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
10076 v4si __builtin_ia32_psrad128 (v4si, v4si)
10077 v2di __builtin_ia32_pslldqi128 (v2di, int)
10078 v8hi __builtin_ia32_psllwi128 (v8hi, int)
10079 v4si __builtin_ia32_pslldi128 (v4si, int)
10080 v2di __builtin_ia32_psllqi128 (v2di, int)
10081 v2di __builtin_ia32_psrldqi128 (v2di, int)
10082 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
10083 v4si __builtin_ia32_psrldi128 (v4si, int)
10084 v2di __builtin_ia32_psrlqi128 (v2di, int)
10085 v8hi __builtin_ia32_psrawi128 (v8hi, int)
10086 v4si __builtin_ia32_psradi128 (v4si, int)
10087 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
10088 v2di __builtin_ia32_movq128 (v2di)
10091 The following built-in functions are available when @option{-msse3} is used.
10092 All of them generate the machine instruction that is part of the name.
10095 v2df __builtin_ia32_addsubpd (v2df, v2df)
10096 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
10097 v2df __builtin_ia32_haddpd (v2df, v2df)
10098 v4sf __builtin_ia32_haddps (v4sf, v4sf)
10099 v2df __builtin_ia32_hsubpd (v2df, v2df)
10100 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
10101 v16qi __builtin_ia32_lddqu (char const *)
10102 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
10103 v2df __builtin_ia32_movddup (v2df)
10104 v4sf __builtin_ia32_movshdup (v4sf)
10105 v4sf __builtin_ia32_movsldup (v4sf)
10106 void __builtin_ia32_mwait (unsigned int, unsigned int)
10109 The following built-in functions are available when @option{-msse3} is used.
10112 @item v2df __builtin_ia32_loadddup (double const *)
10113 Generates the @code{movddup} machine instruction as a load from memory.
10116 The following built-in functions are available when @option{-mssse3} is used.
10117 All of them generate the machine instruction that is part of the name
10118 with MMX registers.
10121 v2si __builtin_ia32_phaddd (v2si, v2si)
10122 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
10123 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
10124 v2si __builtin_ia32_phsubd (v2si, v2si)
10125 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
10126 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
10127 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi)
10128 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
10129 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
10130 v8qi __builtin_ia32_psignb (v8qi, v8qi)
10131 v2si __builtin_ia32_psignd (v2si, v2si)
10132 v4hi __builtin_ia32_psignw (v4hi, v4hi)
10133 v1di __builtin_ia32_palignr (v1di, v1di, int)
10134 v8qi __builtin_ia32_pabsb (v8qi)
10135 v2si __builtin_ia32_pabsd (v2si)
10136 v4hi __builtin_ia32_pabsw (v4hi)
10139 The following built-in functions are available when @option{-mssse3} is used.
10140 All of them generate the machine instruction that is part of the name
10141 with SSE registers.
10144 v4si __builtin_ia32_phaddd128 (v4si, v4si)
10145 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
10146 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
10147 v4si __builtin_ia32_phsubd128 (v4si, v4si)
10148 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
10149 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
10150 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
10151 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
10152 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
10153 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
10154 v4si __builtin_ia32_psignd128 (v4si, v4si)
10155 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
10156 v2di __builtin_ia32_palignr128 (v2di, v2di, int)
10157 v16qi __builtin_ia32_pabsb128 (v16qi)
10158 v4si __builtin_ia32_pabsd128 (v4si)
10159 v8hi __builtin_ia32_pabsw128 (v8hi)
10162 The following built-in functions are available when @option{-msse4.1} is
10163 used. All of them generate the machine instruction that is part of the
10167 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
10168 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
10169 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
10170 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
10171 v2df __builtin_ia32_dppd (v2df, v2df, const int)
10172 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
10173 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
10174 v2di __builtin_ia32_movntdqa (v2di *);
10175 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
10176 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
10177 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
10178 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
10179 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
10180 v8hi __builtin_ia32_phminposuw128 (v8hi)
10181 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
10182 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
10183 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
10184 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
10185 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
10186 v4si __builtin_ia32_pminsd128 (v4si, v4si)
10187 v4si __builtin_ia32_pminud128 (v4si, v4si)
10188 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
10189 v4si __builtin_ia32_pmovsxbd128 (v16qi)
10190 v2di __builtin_ia32_pmovsxbq128 (v16qi)
10191 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
10192 v2di __builtin_ia32_pmovsxdq128 (v4si)
10193 v4si __builtin_ia32_pmovsxwd128 (v8hi)
10194 v2di __builtin_ia32_pmovsxwq128 (v8hi)
10195 v4si __builtin_ia32_pmovzxbd128 (v16qi)
10196 v2di __builtin_ia32_pmovzxbq128 (v16qi)
10197 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
10198 v2di __builtin_ia32_pmovzxdq128 (v4si)
10199 v4si __builtin_ia32_pmovzxwd128 (v8hi)
10200 v2di __builtin_ia32_pmovzxwq128 (v8hi)
10201 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
10202 v4si __builtin_ia32_pmulld128 (v4si, v4si)
10203 int __builtin_ia32_ptestc128 (v2di, v2di)
10204 int __builtin_ia32_ptestnzc128 (v2di, v2di)
10205 int __builtin_ia32_ptestz128 (v2di, v2di)
10206 v2df __builtin_ia32_roundpd (v2df, const int)
10207 v4sf __builtin_ia32_roundps (v4sf, const int)
10208 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
10209 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
10212 The following built-in functions are available when @option{-msse4.1} is
10216 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
10217 Generates the @code{insertps} machine instruction.
10218 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
10219 Generates the @code{pextrb} machine instruction.
10220 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
10221 Generates the @code{pinsrb} machine instruction.
10222 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
10223 Generates the @code{pinsrd} machine instruction.
10224 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
10225 Generates the @code{pinsrq} machine instruction in 64bit mode.
10228 The following built-in functions are changed to generate new SSE4.1
10229 instructions when @option{-msse4.1} is used.
10232 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
10233 Generates the @code{extractps} machine instruction.
10234 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
10235 Generates the @code{pextrd} machine instruction.
10236 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
10237 Generates the @code{pextrq} machine instruction in 64bit mode.
10240 The following built-in functions are available when @option{-msse4.2} is
10241 used. All of them generate the machine instruction that is part of the
10245 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
10246 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
10247 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
10248 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
10249 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
10250 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
10251 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
10252 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
10253 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
10254 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
10255 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
10256 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
10257 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
10258 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
10259 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
10262 The following built-in functions are available when @option{-msse4.2} is
10266 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
10267 Generates the @code{crc32b} machine instruction.
10268 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
10269 Generates the @code{crc32w} machine instruction.
10270 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
10271 Generates the @code{crc32l} machine instruction.
10272 @item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)
10273 Generates the @code{crc32q} machine instruction.
10276 The following built-in functions are changed to generate new SSE4.2
10277 instructions when @option{-msse4.2} is used.
10280 @item int __builtin_popcount (unsigned int)
10281 Generates the @code{popcntl} machine instruction.
10282 @item int __builtin_popcountl (unsigned long)
10283 Generates the @code{popcntl} or @code{popcntq} machine instruction,
10284 depending on the size of @code{unsigned long}.
10285 @item int __builtin_popcountll (unsigned long long)
10286 Generates the @code{popcntq} machine instruction.
10289 The following built-in functions are available when @option{-mavx} is
10290 used. All of them generate the machine instruction that is part of the
10294 v4df __builtin_ia32_addpd256 (v4df,v4df)
10295 v8sf __builtin_ia32_addps256 (v8sf,v8sf)
10296 v4df __builtin_ia32_addsubpd256 (v4df,v4df)
10297 v8sf __builtin_ia32_addsubps256 (v8sf,v8sf)
10298 v4df __builtin_ia32_andnpd256 (v4df,v4df)
10299 v8sf __builtin_ia32_andnps256 (v8sf,v8sf)
10300 v4df __builtin_ia32_andpd256 (v4df,v4df)
10301 v8sf __builtin_ia32_andps256 (v8sf,v8sf)
10302 v4df __builtin_ia32_blendpd256 (v4df,v4df,int)
10303 v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int)
10304 v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df)
10305 v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf)
10306 v2df __builtin_ia32_cmppd (v2df,v2df,int)
10307 v4df __builtin_ia32_cmppd256 (v4df,v4df,int)
10308 v4sf __builtin_ia32_cmpps (v4sf,v4sf,int)
10309 v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int)
10310 v2df __builtin_ia32_cmpsd (v2df,v2df,int)
10311 v4sf __builtin_ia32_cmpss (v4sf,v4sf,int)
10312 v4df __builtin_ia32_cvtdq2pd256 (v4si)
10313 v8sf __builtin_ia32_cvtdq2ps256 (v8si)
10314 v4si __builtin_ia32_cvtpd2dq256 (v4df)
10315 v4sf __builtin_ia32_cvtpd2ps256 (v4df)
10316 v8si __builtin_ia32_cvtps2dq256 (v8sf)
10317 v4df __builtin_ia32_cvtps2pd256 (v4sf)
10318 v4si __builtin_ia32_cvttpd2dq256 (v4df)
10319 v8si __builtin_ia32_cvttps2dq256 (v8sf)
10320 v4df __builtin_ia32_divpd256 (v4df,v4df)
10321 v8sf __builtin_ia32_divps256 (v8sf,v8sf)
10322 v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int)
10323 v4df __builtin_ia32_haddpd256 (v4df,v4df)
10324 v8sf __builtin_ia32_haddps256 (v8sf,v8sf)
10325 v4df __builtin_ia32_hsubpd256 (v4df,v4df)
10326 v8sf __builtin_ia32_hsubps256 (v8sf,v8sf)
10327 v32qi __builtin_ia32_lddqu256 (pcchar)
10328 v32qi __builtin_ia32_loaddqu256 (pcchar)
10329 v4df __builtin_ia32_loadupd256 (pcdouble)
10330 v8sf __builtin_ia32_loadups256 (pcfloat)
10331 v2df __builtin_ia32_maskloadpd (pcv2df,v2df)
10332 v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df)
10333 v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf)
10334 v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf)
10335 void __builtin_ia32_maskstorepd (pv2df,v2df,v2df)
10336 void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df)
10337 void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf)
10338 void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf)
10339 v4df __builtin_ia32_maxpd256 (v4df,v4df)
10340 v8sf __builtin_ia32_maxps256 (v8sf,v8sf)
10341 v4df __builtin_ia32_minpd256 (v4df,v4df)
10342 v8sf __builtin_ia32_minps256 (v8sf,v8sf)
10343 v4df __builtin_ia32_movddup256 (v4df)
10344 int __builtin_ia32_movmskpd256 (v4df)
10345 int __builtin_ia32_movmskps256 (v8sf)
10346 v8sf __builtin_ia32_movshdup256 (v8sf)
10347 v8sf __builtin_ia32_movsldup256 (v8sf)
10348 v4df __builtin_ia32_mulpd256 (v4df,v4df)
10349 v8sf __builtin_ia32_mulps256 (v8sf,v8sf)
10350 v4df __builtin_ia32_orpd256 (v4df,v4df)
10351 v8sf __builtin_ia32_orps256 (v8sf,v8sf)
10352 v2df __builtin_ia32_pd_pd256 (v4df)
10353 v4df __builtin_ia32_pd256_pd (v2df)
10354 v4sf __builtin_ia32_ps_ps256 (v8sf)
10355 v8sf __builtin_ia32_ps256_ps (v4sf)
10356 int __builtin_ia32_ptestc256 (v4di,v4di,ptest)
10357 int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest)
10358 int __builtin_ia32_ptestz256 (v4di,v4di,ptest)
10359 v8sf __builtin_ia32_rcpps256 (v8sf)
10360 v4df __builtin_ia32_roundpd256 (v4df,int)
10361 v8sf __builtin_ia32_roundps256 (v8sf,int)
10362 v8sf __builtin_ia32_rsqrtps_nr256 (v8sf)
10363 v8sf __builtin_ia32_rsqrtps256 (v8sf)
10364 v4df __builtin_ia32_shufpd256 (v4df,v4df,int)
10365 v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int)
10366 v4si __builtin_ia32_si_si256 (v8si)
10367 v8si __builtin_ia32_si256_si (v4si)
10368 v4df __builtin_ia32_sqrtpd256 (v4df)
10369 v8sf __builtin_ia32_sqrtps_nr256 (v8sf)
10370 v8sf __builtin_ia32_sqrtps256 (v8sf)
10371 void __builtin_ia32_storedqu256 (pchar,v32qi)
10372 void __builtin_ia32_storeupd256 (pdouble,v4df)
10373 void __builtin_ia32_storeups256 (pfloat,v8sf)
10374 v4df __builtin_ia32_subpd256 (v4df,v4df)
10375 v8sf __builtin_ia32_subps256 (v8sf,v8sf)
10376 v4df __builtin_ia32_unpckhpd256 (v4df,v4df)
10377 v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf)
10378 v4df __builtin_ia32_unpcklpd256 (v4df,v4df)
10379 v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf)
10380 v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df)
10381 v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf)
10382 v4df __builtin_ia32_vbroadcastsd256 (pcdouble)
10383 v4sf __builtin_ia32_vbroadcastss (pcfloat)
10384 v8sf __builtin_ia32_vbroadcastss256 (pcfloat)
10385 v2df __builtin_ia32_vextractf128_pd256 (v4df,int)
10386 v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int)
10387 v4si __builtin_ia32_vextractf128_si256 (v8si,int)
10388 v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int)
10389 v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int)
10390 v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int)
10391 v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int)
10392 v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int)
10393 v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int)
10394 v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int)
10395 v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int)
10396 v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int)
10397 v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int)
10398 v2df __builtin_ia32_vpermilpd (v2df,int)
10399 v4df __builtin_ia32_vpermilpd256 (v4df,int)
10400 v4sf __builtin_ia32_vpermilps (v4sf,int)
10401 v8sf __builtin_ia32_vpermilps256 (v8sf,int)
10402 v2df __builtin_ia32_vpermilvarpd (v2df,v2di)
10403 v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di)
10404 v4sf __builtin_ia32_vpermilvarps (v4sf,v4si)
10405 v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si)
10406 int __builtin_ia32_vtestcpd (v2df,v2df,ptest)
10407 int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest)
10408 int __builtin_ia32_vtestcps (v4sf,v4sf,ptest)
10409 int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest)
10410 int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest)
10411 int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest)
10412 int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest)
10413 int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest)
10414 int __builtin_ia32_vtestzpd (v2df,v2df,ptest)
10415 int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest)
10416 int __builtin_ia32_vtestzps (v4sf,v4sf,ptest)
10417 int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest)
10418 void __builtin_ia32_vzeroall (void)
10419 void __builtin_ia32_vzeroupper (void)
10420 v4df __builtin_ia32_xorpd256 (v4df,v4df)
10421 v8sf __builtin_ia32_xorps256 (v8sf,v8sf)
10424 The following built-in functions are available when @option{-mavx2} is
10425 used. All of them generate the machine instruction that is part of the
10429 v32qi __builtin_ia32_mpsadbw256 (v32qi,v32qi,v32qi,int)
10430 v32qi __builtin_ia32_pabsb256 (v32qi)
10431 v16hi __builtin_ia32_pabsw256 (v16hi)
10432 v8si __builtin_ia32_pabsd256 (v8si)
10433 v16hi __builtin_ia32_packssdw256 (v8si,v8si)
10434 v32qi __builtin_ia32_packsswb256 (v16hi,v16hi)
10435 v16hi __builtin_ia32_packusdw256 (v8si,v8si)
10436 v32qi __builtin_ia32_packuswb256 (v16hi,v16hi)
10437 v32qi __builtin_ia32_paddb256 (v32qi,v32qi)
10438 v16hi __builtin_ia32_paddw256 (v16hi,v16hi)
10439 v8si __builtin_ia32_paddd256 (v8si,v8si)
10440 v4di __builtin_ia32_paddq256 (v4di,v4di)
10441 v32qi __builtin_ia32_paddsb256 (v32qi,v32qi)
10442 v16hi __builtin_ia32_paddsw256 (v16hi,v16hi)
10443 v32qi __builtin_ia32_paddusb256 (v32qi,v32qi)
10444 v16hi __builtin_ia32_paddusw256 (v16hi,v16hi)
10445 v4di __builtin_ia32_palignr256 (v4di,v4di,int)
10446 v4di __builtin_ia32_andsi256 (v4di,v4di)
10447 v4di __builtin_ia32_andnotsi256 (v4di,v4di)
10448 v32qi __builtin_ia32_pavgb256 (v32qi,v32qi)
10449 v16hi __builtin_ia32_pavgw256 (v16hi,v16hi)
10450 v32qi __builtin_ia32_pblendvb256 (v32qi,v32qi,v32qi)
10451 v16hi __builtin_ia32_pblendw256 (v16hi,v16hi,int)
10452 v32qi __builtin_ia32_pcmpeqb256 (v32qi,v32qi)
10453 v16hi __builtin_ia32_pcmpeqw256 (v16hi,v16hi)
10454 v8si __builtin_ia32_pcmpeqd256 (c8si,v8si)
10455 v4di __builtin_ia32_pcmpeqq256 (v4di,v4di)
10456 v32qi __builtin_ia32_pcmpgtb256 (v32qi,v32qi)
10457 v16hi __builtin_ia32_pcmpgtw256 (16hi,v16hi)
10458 v8si __builtin_ia32_pcmpgtd256 (v8si,v8si)
10459 v4di __builtin_ia32_pcmpgtq256 (v4di,v4di)
10460 v16hi __builtin_ia32_phaddw256 (v16hi,v16hi)
10461 v8si __builtin_ia32_phaddd256 (v8si,v8si)
10462 v16hi __builtin_ia32_phaddsw256 (v16hi,v16hi)
10463 v16hi __builtin_ia32_phsubw256 (v16hi,v16hi)
10464 v8si __builtin_ia32_phsubd256 (v8si,v8si)
10465 v16hi __builtin_ia32_phsubsw256 (v16hi,v16hi)
10466 v32qi __builtin_ia32_pmaddubsw256 (v32qi,v32qi)
10467 v16hi __builtin_ia32_pmaddwd256 (v16hi,v16hi)
10468 v32qi __builtin_ia32_pmaxsb256 (v32qi,v32qi)
10469 v16hi __builtin_ia32_pmaxsw256 (v16hi,v16hi)
10470 v8si __builtin_ia32_pmaxsd256 (v8si,v8si)
10471 v32qi __builtin_ia32_pmaxub256 (v32qi,v32qi)
10472 v16hi __builtin_ia32_pmaxuw256 (v16hi,v16hi)
10473 v8si __builtin_ia32_pmaxud256 (v8si,v8si)
10474 v32qi __builtin_ia32_pminsb256 (v32qi,v32qi)
10475 v16hi __builtin_ia32_pminsw256 (v16hi,v16hi)
10476 v8si __builtin_ia32_pminsd256 (v8si,v8si)
10477 v32qi __builtin_ia32_pminub256 (v32qi,v32qi)
10478 v16hi __builtin_ia32_pminuw256 (v16hi,v16hi)
10479 v8si __builtin_ia32_pminud256 (v8si,v8si)
10480 int __builtin_ia32_pmovmskb256 (v32qi)
10481 v16hi __builtin_ia32_pmovsxbw256 (v16qi)
10482 v8si __builtin_ia32_pmovsxbd256 (v16qi)
10483 v4di __builtin_ia32_pmovsxbq256 (v16qi)
10484 v8si __builtin_ia32_pmovsxwd256 (v8hi)
10485 v4di __builtin_ia32_pmovsxwq256 (v8hi)
10486 v4di __builtin_ia32_pmovsxdq256 (v4si)
10487 v16hi __builtin_ia32_pmovzxbw256 (v16qi)
10488 v8si __builtin_ia32_pmovzxbd256 (v16qi)
10489 v4di __builtin_ia32_pmovzxbq256 (v16qi)
10490 v8si __builtin_ia32_pmovzxwd256 (v8hi)
10491 v4di __builtin_ia32_pmovzxwq256 (v8hi)
10492 v4di __builtin_ia32_pmovzxdq256 (v4si)
10493 v4di __builtin_ia32_pmuldq256 (v8si,v8si)
10494 v16hi __builtin_ia32_pmulhrsw256 (v16hi, v16hi)
10495 v16hi __builtin_ia32_pmulhuw256 (v16hi,v16hi)
10496 v16hi __builtin_ia32_pmulhw256 (v16hi,v16hi)
10497 v16hi __builtin_ia32_pmullw256 (v16hi,v16hi)
10498 v8si __builtin_ia32_pmulld256 (v8si,v8si)
10499 v4di __builtin_ia32_pmuludq256 (v8si,v8si)
10500 v4di __builtin_ia32_por256 (v4di,v4di)
10501 v16hi __builtin_ia32_psadbw256 (v32qi,v32qi)
10502 v32qi __builtin_ia32_pshufb256 (v32qi,v32qi)
10503 v8si __builtin_ia32_pshufd256 (v8si,int)
10504 v16hi __builtin_ia32_pshufhw256 (v16hi,int)
10505 v16hi __builtin_ia32_pshuflw256 (v16hi,int)
10506 v32qi __builtin_ia32_psignb256 (v32qi,v32qi)
10507 v16hi __builtin_ia32_psignw256 (v16hi,v16hi)
10508 v8si __builtin_ia32_psignd256 (v8si,v8si)
10509 v4di __builtin_ia32_pslldqi256 (v4di,int)
10510 v16hi __builtin_ia32_psllwi256 (16hi,int)
10511 v16hi __builtin_ia32_psllw256(v16hi,v8hi)
10512 v8si __builtin_ia32_pslldi256 (v8si,int)
10513 v8si __builtin_ia32_pslld256(v8si,v4si)
10514 v4di __builtin_ia32_psllqi256 (v4di,int)
10515 v4di __builtin_ia32_psllq256(v4di,v2di)
10516 v16hi __builtin_ia32_psrawi256 (v16hi,int)
10517 v16hi __builtin_ia32_psraw256 (v16hi,v8hi)
10518 v8si __builtin_ia32_psradi256 (v8si,int)
10519 v8si __builtin_ia32_psrad256 (v8si,v4si)
10520 v4di __builtin_ia32_psrldqi256 (v4di, int)
10521 v16hi __builtin_ia32_psrlwi256 (v16hi,int)
10522 v16hi __builtin_ia32_psrlw256 (v16hi,v8hi)
10523 v8si __builtin_ia32_psrldi256 (v8si,int)
10524 v8si __builtin_ia32_psrld256 (v8si,v4si)
10525 v4di __builtin_ia32_psrlqi256 (v4di,int)
10526 v4di __builtin_ia32_psrlq256(v4di,v2di)
10527 v32qi __builtin_ia32_psubb256 (v32qi,v32qi)
10528 v32hi __builtin_ia32_psubw256 (v16hi,v16hi)
10529 v8si __builtin_ia32_psubd256 (v8si,v8si)
10530 v4di __builtin_ia32_psubq256 (v4di,v4di)
10531 v32qi __builtin_ia32_psubsb256 (v32qi,v32qi)
10532 v16hi __builtin_ia32_psubsw256 (v16hi,v16hi)
10533 v32qi __builtin_ia32_psubusb256 (v32qi,v32qi)
10534 v16hi __builtin_ia32_psubusw256 (v16hi,v16hi)
10535 v32qi __builtin_ia32_punpckhbw256 (v32qi,v32qi)
10536 v16hi __builtin_ia32_punpckhwd256 (v16hi,v16hi)
10537 v8si __builtin_ia32_punpckhdq256 (v8si,v8si)
10538 v4di __builtin_ia32_punpckhqdq256 (v4di,v4di)
10539 v32qi __builtin_ia32_punpcklbw256 (v32qi,v32qi)
10540 v16hi __builtin_ia32_punpcklwd256 (v16hi,v16hi)
10541 v8si __builtin_ia32_punpckldq256 (v8si,v8si)
10542 v4di __builtin_ia32_punpcklqdq256 (v4di,v4di)
10543 v4di __builtin_ia32_pxor256 (v4di,v4di)
10544 v4di __builtin_ia32_movntdqa256 (pv4di)
10545 v4sf __builtin_ia32_vbroadcastss_ps (v4sf)
10546 v8sf __builtin_ia32_vbroadcastss_ps256 (v4sf)
10547 v4df __builtin_ia32_vbroadcastsd_pd256 (v2df)
10548 v4di __builtin_ia32_vbroadcastsi256 (v2di)
10549 v4si __builtin_ia32_pblendd128 (v4si,v4si)
10550 v8si __builtin_ia32_pblendd256 (v8si,v8si)
10551 v32qi __builtin_ia32_pbroadcastb256 (v16qi)
10552 v16hi __builtin_ia32_pbroadcastw256 (v8hi)
10553 v8si __builtin_ia32_pbroadcastd256 (v4si)
10554 v4di __builtin_ia32_pbroadcastq256 (v2di)
10555 v16qi __builtin_ia32_pbroadcastb128 (v16qi)
10556 v8hi __builtin_ia32_pbroadcastw128 (v8hi)
10557 v4si __builtin_ia32_pbroadcastd128 (v4si)
10558 v2di __builtin_ia32_pbroadcastq128 (v2di)
10559 v8si __builtin_ia32_permvarsi256 (v8si,v8si)
10560 v4df __builtin_ia32_permdf256 (v4df,int)
10561 v8sf __builtin_ia32_permvarsf256 (v8sf,v8sf)
10562 v4di __builtin_ia32_permdi256 (v4di,int)
10563 v4di __builtin_ia32_permti256 (v4di,v4di,int)
10564 v4di __builtin_ia32_extract128i256 (v4di,int)
10565 v4di __builtin_ia32_insert128i256 (v4di,v2di,int)
10566 v8si __builtin_ia32_maskloadd256 (pcv8si,v8si)
10567 v4di __builtin_ia32_maskloadq256 (pcv4di,v4di)
10568 v4si __builtin_ia32_maskloadd (pcv4si,v4si)
10569 v2di __builtin_ia32_maskloadq (pcv2di,v2di)
10570 void __builtin_ia32_maskstored256 (pv8si,v8si,v8si)
10571 void __builtin_ia32_maskstoreq256 (pv4di,v4di,v4di)
10572 void __builtin_ia32_maskstored (pv4si,v4si,v4si)
10573 void __builtin_ia32_maskstoreq (pv2di,v2di,v2di)
10574 v8si __builtin_ia32_psllv8si (v8si,v8si)
10575 v4si __builtin_ia32_psllv4si (v4si,v4si)
10576 v4di __builtin_ia32_psllv4di (v4di,v4di)
10577 v2di __builtin_ia32_psllv2di (v2di,v2di)
10578 v8si __builtin_ia32_psrav8si (v8si,v8si)
10579 v4si __builtin_ia32_psrav4si (v4si,v4si)
10580 v8si __builtin_ia32_psrlv8si (v8si,v8si)
10581 v4si __builtin_ia32_psrlv4si (v4si,v4si)
10582 v4di __builtin_ia32_psrlv4di (v4di,v4di)
10583 v2di __builtin_ia32_psrlv2di (v2di,v2di)
10584 v2df __builtin_ia32_gathersiv2df (v2df, pcdouble,v4si,v2df,int)
10585 v4df __builtin_ia32_gathersiv4df (v4df, pcdouble,v4si,v4df,int)
10586 v2df __builtin_ia32_gatherdiv2df (v2df, pcdouble,v2di,v2df,int)
10587 v4df __builtin_ia32_gatherdiv4df (v4df, pcdouble,v4di,v4df,int)
10588 v4sf __builtin_ia32_gathersiv4sf (v4sf, pcfloat,v4si,v4sf,int)
10589 v8sf __builtin_ia32_gathersiv8sf (v8sf, pcfloat,v8si,v8sf,int)
10590 v4sf __builtin_ia32_gatherdiv4sf (v4sf, pcfloat,v2di,v4sf,int)
10591 v4sf __builtin_ia32_gatherdiv4sf256 (v4sf, pcfloat,v4di,v4sf,int)
10592 v2di __builtin_ia32_gathersiv2di (v2di, pcint64,v4si,v2di,int)
10593 v4di __builtin_ia32_gathersiv4di (v4di, pcint64,v4si,v4di,int)
10594 v2di __builtin_ia32_gatherdiv2di (v2di, pcint64,v2di,v2di,int)
10595 v4di __builtin_ia32_gatherdiv4di (v4di, pcint64,v4di,v4di,int)
10596 v4si __builtin_ia32_gathersiv4si (v4si, pcint,v4si,v4si,int)
10597 v8si __builtin_ia32_gathersiv8si (v8si, pcint,v8si,v8si,int)
10598 v4si __builtin_ia32_gatherdiv4si (v4si, pcint,v2di,v4si,int)
10599 v4si __builtin_ia32_gatherdiv4si256 (v4si, pcint,v4di,v4si,int)
10602 The following built-in functions are available when @option{-maes} is
10603 used. All of them generate the machine instruction that is part of the
10607 v2di __builtin_ia32_aesenc128 (v2di, v2di)
10608 v2di __builtin_ia32_aesenclast128 (v2di, v2di)
10609 v2di __builtin_ia32_aesdec128 (v2di, v2di)
10610 v2di __builtin_ia32_aesdeclast128 (v2di, v2di)
10611 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int)
10612 v2di __builtin_ia32_aesimc128 (v2di)
10615 The following built-in function is available when @option{-mpclmul} is
10619 @item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)
10620 Generates the @code{pclmulqdq} machine instruction.
10623 The following built-in function is available when @option{-mfsgsbase} is
10624 used. All of them generate the machine instruction that is part of the
10628 unsigned int __builtin_ia32_rdfsbase32 (void)
10629 unsigned long long __builtin_ia32_rdfsbase64 (void)
10630 unsigned int __builtin_ia32_rdgsbase32 (void)
10631 unsigned long long __builtin_ia32_rdgsbase64 (void)
10632 void _writefsbase_u32 (unsigned int)
10633 void _writefsbase_u64 (unsigned long long)
10634 void _writegsbase_u32 (unsigned int)
10635 void _writegsbase_u64 (unsigned long long)
10638 The following built-in function is available when @option{-mrdrnd} is
10639 used. All of them generate the machine instruction that is part of the
10643 unsigned int __builtin_ia32_rdrand16_step (unsigned short *)
10644 unsigned int __builtin_ia32_rdrand32_step (unsigned int *)
10645 unsigned int __builtin_ia32_rdrand64_step (unsigned long long *)
10648 The following built-in functions are available when @option{-msse4a} is used.
10649 All of them generate the machine instruction that is part of the name.
10652 void __builtin_ia32_movntsd (double *, v2df)
10653 void __builtin_ia32_movntss (float *, v4sf)
10654 v2di __builtin_ia32_extrq (v2di, v16qi)
10655 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
10656 v2di __builtin_ia32_insertq (v2di, v2di)
10657 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
10660 The following built-in functions are available when @option{-mxop} is used.
10662 v2df __builtin_ia32_vfrczpd (v2df)
10663 v4sf __builtin_ia32_vfrczps (v4sf)
10664 v2df __builtin_ia32_vfrczsd (v2df, v2df)
10665 v4sf __builtin_ia32_vfrczss (v4sf, v4sf)
10666 v4df __builtin_ia32_vfrczpd256 (v4df)
10667 v8sf __builtin_ia32_vfrczps256 (v8sf)
10668 v2di __builtin_ia32_vpcmov (v2di, v2di, v2di)
10669 v2di __builtin_ia32_vpcmov_v2di (v2di, v2di, v2di)
10670 v4si __builtin_ia32_vpcmov_v4si (v4si, v4si, v4si)
10671 v8hi __builtin_ia32_vpcmov_v8hi (v8hi, v8hi, v8hi)
10672 v16qi __builtin_ia32_vpcmov_v16qi (v16qi, v16qi, v16qi)
10673 v2df __builtin_ia32_vpcmov_v2df (v2df, v2df, v2df)
10674 v4sf __builtin_ia32_vpcmov_v4sf (v4sf, v4sf, v4sf)
10675 v4di __builtin_ia32_vpcmov_v4di256 (v4di, v4di, v4di)
10676 v8si __builtin_ia32_vpcmov_v8si256 (v8si, v8si, v8si)
10677 v16hi __builtin_ia32_vpcmov_v16hi256 (v16hi, v16hi, v16hi)
10678 v32qi __builtin_ia32_vpcmov_v32qi256 (v32qi, v32qi, v32qi)
10679 v4df __builtin_ia32_vpcmov_v4df256 (v4df, v4df, v4df)
10680 v8sf __builtin_ia32_vpcmov_v8sf256 (v8sf, v8sf, v8sf)
10681 v16qi __builtin_ia32_vpcomeqb (v16qi, v16qi)
10682 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
10683 v4si __builtin_ia32_vpcomeqd (v4si, v4si)
10684 v2di __builtin_ia32_vpcomeqq (v2di, v2di)
10685 v16qi __builtin_ia32_vpcomequb (v16qi, v16qi)
10686 v4si __builtin_ia32_vpcomequd (v4si, v4si)
10687 v2di __builtin_ia32_vpcomequq (v2di, v2di)
10688 v8hi __builtin_ia32_vpcomequw (v8hi, v8hi)
10689 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
10690 v16qi __builtin_ia32_vpcomfalseb (v16qi, v16qi)
10691 v4si __builtin_ia32_vpcomfalsed (v4si, v4si)
10692 v2di __builtin_ia32_vpcomfalseq (v2di, v2di)
10693 v16qi __builtin_ia32_vpcomfalseub (v16qi, v16qi)
10694 v4si __builtin_ia32_vpcomfalseud (v4si, v4si)
10695 v2di __builtin_ia32_vpcomfalseuq (v2di, v2di)
10696 v8hi __builtin_ia32_vpcomfalseuw (v8hi, v8hi)
10697 v8hi __builtin_ia32_vpcomfalsew (v8hi, v8hi)
10698 v16qi __builtin_ia32_vpcomgeb (v16qi, v16qi)
10699 v4si __builtin_ia32_vpcomged (v4si, v4si)
10700 v2di __builtin_ia32_vpcomgeq (v2di, v2di)
10701 v16qi __builtin_ia32_vpcomgeub (v16qi, v16qi)
10702 v4si __builtin_ia32_vpcomgeud (v4si, v4si)
10703 v2di __builtin_ia32_vpcomgeuq (v2di, v2di)
10704 v8hi __builtin_ia32_vpcomgeuw (v8hi, v8hi)
10705 v8hi __builtin_ia32_vpcomgew (v8hi, v8hi)
10706 v16qi __builtin_ia32_vpcomgtb (v16qi, v16qi)
10707 v4si __builtin_ia32_vpcomgtd (v4si, v4si)
10708 v2di __builtin_ia32_vpcomgtq (v2di, v2di)
10709 v16qi __builtin_ia32_vpcomgtub (v16qi, v16qi)
10710 v4si __builtin_ia32_vpcomgtud (v4si, v4si)
10711 v2di __builtin_ia32_vpcomgtuq (v2di, v2di)
10712 v8hi __builtin_ia32_vpcomgtuw (v8hi, v8hi)
10713 v8hi __builtin_ia32_vpcomgtw (v8hi, v8hi)
10714 v16qi __builtin_ia32_vpcomleb (v16qi, v16qi)
10715 v4si __builtin_ia32_vpcomled (v4si, v4si)
10716 v2di __builtin_ia32_vpcomleq (v2di, v2di)
10717 v16qi __builtin_ia32_vpcomleub (v16qi, v16qi)
10718 v4si __builtin_ia32_vpcomleud (v4si, v4si)
10719 v2di __builtin_ia32_vpcomleuq (v2di, v2di)
10720 v8hi __builtin_ia32_vpcomleuw (v8hi, v8hi)
10721 v8hi __builtin_ia32_vpcomlew (v8hi, v8hi)
10722 v16qi __builtin_ia32_vpcomltb (v16qi, v16qi)
10723 v4si __builtin_ia32_vpcomltd (v4si, v4si)
10724 v2di __builtin_ia32_vpcomltq (v2di, v2di)
10725 v16qi __builtin_ia32_vpcomltub (v16qi, v16qi)
10726 v4si __builtin_ia32_vpcomltud (v4si, v4si)
10727 v2di __builtin_ia32_vpcomltuq (v2di, v2di)
10728 v8hi __builtin_ia32_vpcomltuw (v8hi, v8hi)
10729 v8hi __builtin_ia32_vpcomltw (v8hi, v8hi)
10730 v16qi __builtin_ia32_vpcomneb (v16qi, v16qi)
10731 v4si __builtin_ia32_vpcomned (v4si, v4si)
10732 v2di __builtin_ia32_vpcomneq (v2di, v2di)
10733 v16qi __builtin_ia32_vpcomneub (v16qi, v16qi)
10734 v4si __builtin_ia32_vpcomneud (v4si, v4si)
10735 v2di __builtin_ia32_vpcomneuq (v2di, v2di)
10736 v8hi __builtin_ia32_vpcomneuw (v8hi, v8hi)
10737 v8hi __builtin_ia32_vpcomnew (v8hi, v8hi)
10738 v16qi __builtin_ia32_vpcomtrueb (v16qi, v16qi)
10739 v4si __builtin_ia32_vpcomtrued (v4si, v4si)
10740 v2di __builtin_ia32_vpcomtrueq (v2di, v2di)
10741 v16qi __builtin_ia32_vpcomtrueub (v16qi, v16qi)
10742 v4si __builtin_ia32_vpcomtrueud (v4si, v4si)
10743 v2di __builtin_ia32_vpcomtrueuq (v2di, v2di)
10744 v8hi __builtin_ia32_vpcomtrueuw (v8hi, v8hi)
10745 v8hi __builtin_ia32_vpcomtruew (v8hi, v8hi)
10746 v4si __builtin_ia32_vphaddbd (v16qi)
10747 v2di __builtin_ia32_vphaddbq (v16qi)
10748 v8hi __builtin_ia32_vphaddbw (v16qi)
10749 v2di __builtin_ia32_vphadddq (v4si)
10750 v4si __builtin_ia32_vphaddubd (v16qi)
10751 v2di __builtin_ia32_vphaddubq (v16qi)
10752 v8hi __builtin_ia32_vphaddubw (v16qi)
10753 v2di __builtin_ia32_vphaddudq (v4si)
10754 v4si __builtin_ia32_vphadduwd (v8hi)
10755 v2di __builtin_ia32_vphadduwq (v8hi)
10756 v4si __builtin_ia32_vphaddwd (v8hi)
10757 v2di __builtin_ia32_vphaddwq (v8hi)
10758 v8hi __builtin_ia32_vphsubbw (v16qi)
10759 v2di __builtin_ia32_vphsubdq (v4si)
10760 v4si __builtin_ia32_vphsubwd (v8hi)
10761 v4si __builtin_ia32_vpmacsdd (v4si, v4si, v4si)
10762 v2di __builtin_ia32_vpmacsdqh (v4si, v4si, v2di)
10763 v2di __builtin_ia32_vpmacsdql (v4si, v4si, v2di)
10764 v4si __builtin_ia32_vpmacssdd (v4si, v4si, v4si)
10765 v2di __builtin_ia32_vpmacssdqh (v4si, v4si, v2di)
10766 v2di __builtin_ia32_vpmacssdql (v4si, v4si, v2di)
10767 v4si __builtin_ia32_vpmacsswd (v8hi, v8hi, v4si)
10768 v8hi __builtin_ia32_vpmacssww (v8hi, v8hi, v8hi)
10769 v4si __builtin_ia32_vpmacswd (v8hi, v8hi, v4si)
10770 v8hi __builtin_ia32_vpmacsww (v8hi, v8hi, v8hi)
10771 v4si __builtin_ia32_vpmadcsswd (v8hi, v8hi, v4si)
10772 v4si __builtin_ia32_vpmadcswd (v8hi, v8hi, v4si)
10773 v16qi __builtin_ia32_vpperm (v16qi, v16qi, v16qi)
10774 v16qi __builtin_ia32_vprotb (v16qi, v16qi)
10775 v4si __builtin_ia32_vprotd (v4si, v4si)
10776 v2di __builtin_ia32_vprotq (v2di, v2di)
10777 v8hi __builtin_ia32_vprotw (v8hi, v8hi)
10778 v16qi __builtin_ia32_vpshab (v16qi, v16qi)
10779 v4si __builtin_ia32_vpshad (v4si, v4si)
10780 v2di __builtin_ia32_vpshaq (v2di, v2di)
10781 v8hi __builtin_ia32_vpshaw (v8hi, v8hi)
10782 v16qi __builtin_ia32_vpshlb (v16qi, v16qi)
10783 v4si __builtin_ia32_vpshld (v4si, v4si)
10784 v2di __builtin_ia32_vpshlq (v2di, v2di)
10785 v8hi __builtin_ia32_vpshlw (v8hi, v8hi)
10788 The following built-in functions are available when @option{-mfma4} is used.
10789 All of them generate the machine instruction that is part of the name
10790 with MMX registers.
10793 v2df __builtin_ia32_fmaddpd (v2df, v2df, v2df)
10794 v4sf __builtin_ia32_fmaddps (v4sf, v4sf, v4sf)
10795 v2df __builtin_ia32_fmaddsd (v2df, v2df, v2df)
10796 v4sf __builtin_ia32_fmaddss (v4sf, v4sf, v4sf)
10797 v2df __builtin_ia32_fmsubpd (v2df, v2df, v2df)
10798 v4sf __builtin_ia32_fmsubps (v4sf, v4sf, v4sf)
10799 v2df __builtin_ia32_fmsubsd (v2df, v2df, v2df)
10800 v4sf __builtin_ia32_fmsubss (v4sf, v4sf, v4sf)
10801 v2df __builtin_ia32_fnmaddpd (v2df, v2df, v2df)
10802 v4sf __builtin_ia32_fnmaddps (v4sf, v4sf, v4sf)
10803 v2df __builtin_ia32_fnmaddsd (v2df, v2df, v2df)
10804 v4sf __builtin_ia32_fnmaddss (v4sf, v4sf, v4sf)
10805 v2df __builtin_ia32_fnmsubpd (v2df, v2df, v2df)
10806 v4sf __builtin_ia32_fnmsubps (v4sf, v4sf, v4sf)
10807 v2df __builtin_ia32_fnmsubsd (v2df, v2df, v2df)
10808 v4sf __builtin_ia32_fnmsubss (v4sf, v4sf, v4sf)
10809 v2df __builtin_ia32_fmaddsubpd (v2df, v2df, v2df)
10810 v4sf __builtin_ia32_fmaddsubps (v4sf, v4sf, v4sf)
10811 v2df __builtin_ia32_fmsubaddpd (v2df, v2df, v2df)
10812 v4sf __builtin_ia32_fmsubaddps (v4sf, v4sf, v4sf)
10813 v4df __builtin_ia32_fmaddpd256 (v4df, v4df, v4df)
10814 v8sf __builtin_ia32_fmaddps256 (v8sf, v8sf, v8sf)
10815 v4df __builtin_ia32_fmsubpd256 (v4df, v4df, v4df)
10816 v8sf __builtin_ia32_fmsubps256 (v8sf, v8sf, v8sf)
10817 v4df __builtin_ia32_fnmaddpd256 (v4df, v4df, v4df)
10818 v8sf __builtin_ia32_fnmaddps256 (v8sf, v8sf, v8sf)
10819 v4df __builtin_ia32_fnmsubpd256 (v4df, v4df, v4df)
10820 v8sf __builtin_ia32_fnmsubps256 (v8sf, v8sf, v8sf)
10821 v4df __builtin_ia32_fmaddsubpd256 (v4df, v4df, v4df)
10822 v8sf __builtin_ia32_fmaddsubps256 (v8sf, v8sf, v8sf)
10823 v4df __builtin_ia32_fmsubaddpd256 (v4df, v4df, v4df)
10824 v8sf __builtin_ia32_fmsubaddps256 (v8sf, v8sf, v8sf)
10828 The following built-in functions are available when @option{-mlwp} is used.
10831 void __builtin_ia32_llwpcb16 (void *);
10832 void __builtin_ia32_llwpcb32 (void *);
10833 void __builtin_ia32_llwpcb64 (void *);
10834 void * __builtin_ia32_llwpcb16 (void);
10835 void * __builtin_ia32_llwpcb32 (void);
10836 void * __builtin_ia32_llwpcb64 (void);
10837 void __builtin_ia32_lwpval16 (unsigned short, unsigned int, unsigned short)
10838 void __builtin_ia32_lwpval32 (unsigned int, unsigned int, unsigned int)
10839 void __builtin_ia32_lwpval64 (unsigned __int64, unsigned int, unsigned int)
10840 unsigned char __builtin_ia32_lwpins16 (unsigned short, unsigned int, unsigned short)
10841 unsigned char __builtin_ia32_lwpins32 (unsigned int, unsigned int, unsigned int)
10842 unsigned char __builtin_ia32_lwpins64 (unsigned __int64, unsigned int, unsigned int)
10845 The following built-in functions are available when @option{-mbmi} is used.
10846 All of them generate the machine instruction that is part of the name.
10848 unsigned int __builtin_ia32_bextr_u32(unsigned int, unsigned int);
10849 unsigned long long __builtin_ia32_bextr_u64 (unsigned long long, unsigned long long);
10852 The following built-in functions are available when @option{-mbmi2} is used.
10853 All of them generate the machine instruction that is part of the name.
10855 unsigned int _bzhi_u32 (unsigned int, unsigned int)
10856 unsigned int _pdep_u32 (unsigned int, unsigned int)
10857 unsigned int _pext_u32 (unsigned int, unsigned int)
10858 unsigned long long _bzhi_u64 (unsigned long long, unsigned long long)
10859 unsigned long long _pdep_u64 (unsigned long long, unsigned long long)
10860 unsigned long long _pext_u64 (unsigned long long, unsigned long long)
10863 The following built-in functions are available when @option{-mlzcnt} is used.
10864 All of them generate the machine instruction that is part of the name.
10866 unsigned short __builtin_ia32_lzcnt_16(unsigned short);
10867 unsigned int __builtin_ia32_lzcnt_u32(unsigned int);
10868 unsigned long long __builtin_ia32_lzcnt_u64 (unsigned long long);
10871 The following built-in functions are available when @option{-mtbm} is used.
10872 Both of them generate the immediate form of the bextr machine instruction.
10874 unsigned int __builtin_ia32_bextri_u32 (unsigned int, const unsigned int);
10875 unsigned long long __builtin_ia32_bextri_u64 (unsigned long long, const unsigned long long);
10879 The following built-in functions are available when @option{-m3dnow} is used.
10880 All of them generate the machine instruction that is part of the name.
10883 void __builtin_ia32_femms (void)
10884 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
10885 v2si __builtin_ia32_pf2id (v2sf)
10886 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
10887 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
10888 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
10889 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
10890 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
10891 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
10892 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
10893 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
10894 v2sf __builtin_ia32_pfrcp (v2sf)
10895 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
10896 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
10897 v2sf __builtin_ia32_pfrsqrt (v2sf)
10898 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
10899 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
10900 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
10901 v2sf __builtin_ia32_pi2fd (v2si)
10902 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
10905 The following built-in functions are available when both @option{-m3dnow}
10906 and @option{-march=athlon} are used. All of them generate the machine
10907 instruction that is part of the name.
10910 v2si __builtin_ia32_pf2iw (v2sf)
10911 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
10912 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
10913 v2sf __builtin_ia32_pi2fw (v2si)
10914 v2sf __builtin_ia32_pswapdsf (v2sf)
10915 v2si __builtin_ia32_pswapdsi (v2si)
10918 @node MIPS DSP Built-in Functions
10919 @subsection MIPS DSP Built-in Functions
10921 The MIPS DSP Application-Specific Extension (ASE) includes new
10922 instructions that are designed to improve the performance of DSP and
10923 media applications. It provides instructions that operate on packed
10924 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
10926 GCC supports MIPS DSP operations using both the generic
10927 vector extensions (@pxref{Vector Extensions}) and a collection of
10928 MIPS-specific built-in functions. Both kinds of support are
10929 enabled by the @option{-mdsp} command-line option.
10931 Revision 2 of the ASE was introduced in the second half of 2006.
10932 This revision adds extra instructions to the original ASE, but is
10933 otherwise backwards-compatible with it. You can select revision 2
10934 using the command-line option @option{-mdspr2}; this option implies
10937 The SCOUNT and POS bits of the DSP control register are global. The
10938 WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and
10939 POS bits. During optimization, the compiler does not delete these
10940 instructions and it does not delete calls to functions containing
10941 these instructions.
10943 At present, GCC only provides support for operations on 32-bit
10944 vectors. The vector type associated with 8-bit integer data is
10945 usually called @code{v4i8}, the vector type associated with Q7
10946 is usually called @code{v4q7}, the vector type associated with 16-bit
10947 integer data is usually called @code{v2i16}, and the vector type
10948 associated with Q15 is usually called @code{v2q15}. They can be
10949 defined in C as follows:
10952 typedef signed char v4i8 __attribute__ ((vector_size(4)));
10953 typedef signed char v4q7 __attribute__ ((vector_size(4)));
10954 typedef short v2i16 __attribute__ ((vector_size(4)));
10955 typedef short v2q15 __attribute__ ((vector_size(4)));
10958 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
10959 initialized in the same way as aggregates. For example:
10962 v4i8 a = @{1, 2, 3, 4@};
10964 b = (v4i8) @{5, 6, 7, 8@};
10966 v2q15 c = @{0x0fcb, 0x3a75@};
10968 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
10971 @emph{Note:} The CPU's endianness determines the order in which values
10972 are packed. On little-endian targets, the first value is the least
10973 significant and the last value is the most significant. The opposite
10974 order applies to big-endian targets. For example, the code above
10975 sets the lowest byte of @code{a} to @code{1} on little-endian targets
10976 and @code{4} on big-endian targets.
10978 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
10979 representation. As shown in this example, the integer representation
10980 of a Q7 value can be obtained by multiplying the fractional value by
10981 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
10982 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
10985 The table below lists the @code{v4i8} and @code{v2q15} operations for which
10986 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
10987 and @code{c} and @code{d} are @code{v2q15} values.
10989 @multitable @columnfractions .50 .50
10990 @item C code @tab MIPS instruction
10991 @item @code{a + b} @tab @code{addu.qb}
10992 @item @code{c + d} @tab @code{addq.ph}
10993 @item @code{a - b} @tab @code{subu.qb}
10994 @item @code{c - d} @tab @code{subq.ph}
10997 The table below lists the @code{v2i16} operation for which
10998 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
10999 @code{v2i16} values.
11001 @multitable @columnfractions .50 .50
11002 @item C code @tab MIPS instruction
11003 @item @code{e * f} @tab @code{mul.ph}
11006 It is easier to describe the DSP built-in functions if we first define
11007 the following types:
11012 typedef unsigned int ui32;
11013 typedef long long a64;
11016 @code{q31} and @code{i32} are actually the same as @code{int}, but we
11017 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
11018 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
11019 @code{long long}, but we use @code{a64} to indicate values that are
11020 placed in one of the four DSP accumulators (@code{$ac0},
11021 @code{$ac1}, @code{$ac2} or @code{$ac3}).
11023 Also, some built-in functions prefer or require immediate numbers as
11024 parameters, because the corresponding DSP instructions accept both immediate
11025 numbers and register operands, or accept immediate numbers only. The
11026 immediate parameters are listed as follows.
11034 imm0_255: 0 to 255.
11035 imm_n32_31: -32 to 31.
11036 imm_n512_511: -512 to 511.
11039 The following built-in functions map directly to a particular MIPS DSP
11040 instruction. Please refer to the architecture specification
11041 for details on what each instruction does.
11044 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
11045 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
11046 q31 __builtin_mips_addq_s_w (q31, q31)
11047 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
11048 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
11049 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
11050 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
11051 q31 __builtin_mips_subq_s_w (q31, q31)
11052 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
11053 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
11054 i32 __builtin_mips_addsc (i32, i32)
11055 i32 __builtin_mips_addwc (i32, i32)
11056 i32 __builtin_mips_modsub (i32, i32)
11057 i32 __builtin_mips_raddu_w_qb (v4i8)
11058 v2q15 __builtin_mips_absq_s_ph (v2q15)
11059 q31 __builtin_mips_absq_s_w (q31)
11060 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
11061 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
11062 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
11063 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
11064 q31 __builtin_mips_preceq_w_phl (v2q15)
11065 q31 __builtin_mips_preceq_w_phr (v2q15)
11066 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
11067 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
11068 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
11069 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
11070 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
11071 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
11072 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
11073 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
11074 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
11075 v4i8 __builtin_mips_shll_qb (v4i8, i32)
11076 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
11077 v2q15 __builtin_mips_shll_ph (v2q15, i32)
11078 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
11079 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
11080 q31 __builtin_mips_shll_s_w (q31, imm0_31)
11081 q31 __builtin_mips_shll_s_w (q31, i32)
11082 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
11083 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
11084 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
11085 v2q15 __builtin_mips_shra_ph (v2q15, i32)
11086 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
11087 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
11088 q31 __builtin_mips_shra_r_w (q31, imm0_31)
11089 q31 __builtin_mips_shra_r_w (q31, i32)
11090 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
11091 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
11092 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
11093 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
11094 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
11095 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
11096 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
11097 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
11098 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
11099 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
11100 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
11101 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
11102 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
11103 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
11104 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
11105 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
11106 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
11107 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
11108 i32 __builtin_mips_bitrev (i32)
11109 i32 __builtin_mips_insv (i32, i32)
11110 v4i8 __builtin_mips_repl_qb (imm0_255)
11111 v4i8 __builtin_mips_repl_qb (i32)
11112 v2q15 __builtin_mips_repl_ph (imm_n512_511)
11113 v2q15 __builtin_mips_repl_ph (i32)
11114 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
11115 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
11116 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
11117 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
11118 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
11119 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
11120 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
11121 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
11122 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
11123 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
11124 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
11125 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
11126 i32 __builtin_mips_extr_w (a64, imm0_31)
11127 i32 __builtin_mips_extr_w (a64, i32)
11128 i32 __builtin_mips_extr_r_w (a64, imm0_31)
11129 i32 __builtin_mips_extr_s_h (a64, i32)
11130 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
11131 i32 __builtin_mips_extr_rs_w (a64, i32)
11132 i32 __builtin_mips_extr_s_h (a64, imm0_31)
11133 i32 __builtin_mips_extr_r_w (a64, i32)
11134 i32 __builtin_mips_extp (a64, imm0_31)
11135 i32 __builtin_mips_extp (a64, i32)
11136 i32 __builtin_mips_extpdp (a64, imm0_31)
11137 i32 __builtin_mips_extpdp (a64, i32)
11138 a64 __builtin_mips_shilo (a64, imm_n32_31)
11139 a64 __builtin_mips_shilo (a64, i32)
11140 a64 __builtin_mips_mthlip (a64, i32)
11141 void __builtin_mips_wrdsp (i32, imm0_63)
11142 i32 __builtin_mips_rddsp (imm0_63)
11143 i32 __builtin_mips_lbux (void *, i32)
11144 i32 __builtin_mips_lhx (void *, i32)
11145 i32 __builtin_mips_lwx (void *, i32)
11146 a64 __builtin_mips_ldx (void *, i32) [MIPS64 only]
11147 i32 __builtin_mips_bposge32 (void)
11148 a64 __builtin_mips_madd (a64, i32, i32);
11149 a64 __builtin_mips_maddu (a64, ui32, ui32);
11150 a64 __builtin_mips_msub (a64, i32, i32);
11151 a64 __builtin_mips_msubu (a64, ui32, ui32);
11152 a64 __builtin_mips_mult (i32, i32);
11153 a64 __builtin_mips_multu (ui32, ui32);
11156 The following built-in functions map directly to a particular MIPS DSP REV 2
11157 instruction. Please refer to the architecture specification
11158 for details on what each instruction does.
11161 v4q7 __builtin_mips_absq_s_qb (v4q7);
11162 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
11163 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
11164 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
11165 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
11166 i32 __builtin_mips_append (i32, i32, imm0_31);
11167 i32 __builtin_mips_balign (i32, i32, imm0_3);
11168 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
11169 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
11170 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
11171 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
11172 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
11173 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
11174 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
11175 q31 __builtin_mips_mulq_rs_w (q31, q31);
11176 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
11177 q31 __builtin_mips_mulq_s_w (q31, q31);
11178 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
11179 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
11180 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
11181 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
11182 i32 __builtin_mips_prepend (i32, i32, imm0_31);
11183 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
11184 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
11185 v4i8 __builtin_mips_shra_qb (v4i8, i32);
11186 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
11187 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
11188 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
11189 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
11190 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
11191 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
11192 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
11193 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
11194 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
11195 q31 __builtin_mips_addqh_w (q31, q31);
11196 q31 __builtin_mips_addqh_r_w (q31, q31);
11197 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
11198 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
11199 q31 __builtin_mips_subqh_w (q31, q31);
11200 q31 __builtin_mips_subqh_r_w (q31, q31);
11201 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
11202 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
11203 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
11204 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
11205 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
11206 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
11210 @node MIPS Paired-Single Support
11211 @subsection MIPS Paired-Single Support
11213 The MIPS64 architecture includes a number of instructions that
11214 operate on pairs of single-precision floating-point values.
11215 Each pair is packed into a 64-bit floating-point register,
11216 with one element being designated the ``upper half'' and
11217 the other being designated the ``lower half''.
11219 GCC supports paired-single operations using both the generic
11220 vector extensions (@pxref{Vector Extensions}) and a collection of
11221 MIPS-specific built-in functions. Both kinds of support are
11222 enabled by the @option{-mpaired-single} command-line option.
11224 The vector type associated with paired-single values is usually
11225 called @code{v2sf}. It can be defined in C as follows:
11228 typedef float v2sf __attribute__ ((vector_size (8)));
11231 @code{v2sf} values are initialized in the same way as aggregates.
11235 v2sf a = @{1.5, 9.1@};
11238 b = (v2sf) @{e, f@};
11241 @emph{Note:} The CPU's endianness determines which value is stored in
11242 the upper half of a register and which value is stored in the lower half.
11243 On little-endian targets, the first value is the lower one and the second
11244 value is the upper one. The opposite order applies to big-endian targets.
11245 For example, the code above sets the lower half of @code{a} to
11246 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
11248 @node MIPS Loongson Built-in Functions
11249 @subsection MIPS Loongson Built-in Functions
11251 GCC provides intrinsics to access the SIMD instructions provided by the
11252 ST Microelectronics Loongson-2E and -2F processors. These intrinsics,
11253 available after inclusion of the @code{loongson.h} header file,
11254 operate on the following 64-bit vector types:
11257 @item @code{uint8x8_t}, a vector of eight unsigned 8-bit integers;
11258 @item @code{uint16x4_t}, a vector of four unsigned 16-bit integers;
11259 @item @code{uint32x2_t}, a vector of two unsigned 32-bit integers;
11260 @item @code{int8x8_t}, a vector of eight signed 8-bit integers;
11261 @item @code{int16x4_t}, a vector of four signed 16-bit integers;
11262 @item @code{int32x2_t}, a vector of two signed 32-bit integers.
11265 The intrinsics provided are listed below; each is named after the
11266 machine instruction to which it corresponds, with suffixes added as
11267 appropriate to distinguish intrinsics that expand to the same machine
11268 instruction yet have different argument types. Refer to the architecture
11269 documentation for a description of the functionality of each
11273 int16x4_t packsswh (int32x2_t s, int32x2_t t);
11274 int8x8_t packsshb (int16x4_t s, int16x4_t t);
11275 uint8x8_t packushb (uint16x4_t s, uint16x4_t t);
11276 uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t);
11277 uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t);
11278 uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t);
11279 int32x2_t paddw_s (int32x2_t s, int32x2_t t);
11280 int16x4_t paddh_s (int16x4_t s, int16x4_t t);
11281 int8x8_t paddb_s (int8x8_t s, int8x8_t t);
11282 uint64_t paddd_u (uint64_t s, uint64_t t);
11283 int64_t paddd_s (int64_t s, int64_t t);
11284 int16x4_t paddsh (int16x4_t s, int16x4_t t);
11285 int8x8_t paddsb (int8x8_t s, int8x8_t t);
11286 uint16x4_t paddush (uint16x4_t s, uint16x4_t t);
11287 uint8x8_t paddusb (uint8x8_t s, uint8x8_t t);
11288 uint64_t pandn_ud (uint64_t s, uint64_t t);
11289 uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t);
11290 uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t);
11291 uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t);
11292 int64_t pandn_sd (int64_t s, int64_t t);
11293 int32x2_t pandn_sw (int32x2_t s, int32x2_t t);
11294 int16x4_t pandn_sh (int16x4_t s, int16x4_t t);
11295 int8x8_t pandn_sb (int8x8_t s, int8x8_t t);
11296 uint16x4_t pavgh (uint16x4_t s, uint16x4_t t);
11297 uint8x8_t pavgb (uint8x8_t s, uint8x8_t t);
11298 uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t);
11299 uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t);
11300 uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t);
11301 int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t);
11302 int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t);
11303 int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t);
11304 uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t);
11305 uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t);
11306 uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t);
11307 int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t);
11308 int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t);
11309 int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t);
11310 uint16x4_t pextrh_u (uint16x4_t s, int field);
11311 int16x4_t pextrh_s (int16x4_t s, int field);
11312 uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t);
11313 uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t);
11314 uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t);
11315 uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t);
11316 int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t);
11317 int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t);
11318 int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t);
11319 int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t);
11320 int32x2_t pmaddhw (int16x4_t s, int16x4_t t);
11321 int16x4_t pmaxsh (int16x4_t s, int16x4_t t);
11322 uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t);
11323 int16x4_t pminsh (int16x4_t s, int16x4_t t);
11324 uint8x8_t pminub (uint8x8_t s, uint8x8_t t);
11325 uint8x8_t pmovmskb_u (uint8x8_t s);
11326 int8x8_t pmovmskb_s (int8x8_t s);
11327 uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t);
11328 int16x4_t pmulhh (int16x4_t s, int16x4_t t);
11329 int16x4_t pmullh (int16x4_t s, int16x4_t t);
11330 int64_t pmuluw (uint32x2_t s, uint32x2_t t);
11331 uint8x8_t pasubub (uint8x8_t s, uint8x8_t t);
11332 uint16x4_t biadd (uint8x8_t s);
11333 uint16x4_t psadbh (uint8x8_t s, uint8x8_t t);
11334 uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order);
11335 int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order);
11336 uint16x4_t psllh_u (uint16x4_t s, uint8_t amount);
11337 int16x4_t psllh_s (int16x4_t s, uint8_t amount);
11338 uint32x2_t psllw_u (uint32x2_t s, uint8_t amount);
11339 int32x2_t psllw_s (int32x2_t s, uint8_t amount);
11340 uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount);
11341 int16x4_t psrlh_s (int16x4_t s, uint8_t amount);
11342 uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount);
11343 int32x2_t psrlw_s (int32x2_t s, uint8_t amount);
11344 uint16x4_t psrah_u (uint16x4_t s, uint8_t amount);
11345 int16x4_t psrah_s (int16x4_t s, uint8_t amount);
11346 uint32x2_t psraw_u (uint32x2_t s, uint8_t amount);
11347 int32x2_t psraw_s (int32x2_t s, uint8_t amount);
11348 uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t);
11349 uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t);
11350 uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t);
11351 int32x2_t psubw_s (int32x2_t s, int32x2_t t);
11352 int16x4_t psubh_s (int16x4_t s, int16x4_t t);
11353 int8x8_t psubb_s (int8x8_t s, int8x8_t t);
11354 uint64_t psubd_u (uint64_t s, uint64_t t);
11355 int64_t psubd_s (int64_t s, int64_t t);
11356 int16x4_t psubsh (int16x4_t s, int16x4_t t);
11357 int8x8_t psubsb (int8x8_t s, int8x8_t t);
11358 uint16x4_t psubush (uint16x4_t s, uint16x4_t t);
11359 uint8x8_t psubusb (uint8x8_t s, uint8x8_t t);
11360 uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t);
11361 uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t);
11362 uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t);
11363 int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t);
11364 int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t);
11365 int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t);
11366 uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t);
11367 uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t);
11368 uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t);
11369 int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t);
11370 int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t);
11371 int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t);
11375 * Paired-Single Arithmetic::
11376 * Paired-Single Built-in Functions::
11377 * MIPS-3D Built-in Functions::
11380 @node Paired-Single Arithmetic
11381 @subsubsection Paired-Single Arithmetic
11383 The table below lists the @code{v2sf} operations for which hardware
11384 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
11385 values and @code{x} is an integral value.
11387 @multitable @columnfractions .50 .50
11388 @item C code @tab MIPS instruction
11389 @item @code{a + b} @tab @code{add.ps}
11390 @item @code{a - b} @tab @code{sub.ps}
11391 @item @code{-a} @tab @code{neg.ps}
11392 @item @code{a * b} @tab @code{mul.ps}
11393 @item @code{a * b + c} @tab @code{madd.ps}
11394 @item @code{a * b - c} @tab @code{msub.ps}
11395 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
11396 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
11397 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
11400 Note that the multiply-accumulate instructions can be disabled
11401 using the command-line option @code{-mno-fused-madd}.
11403 @node Paired-Single Built-in Functions
11404 @subsubsection Paired-Single Built-in Functions
11406 The following paired-single functions map directly to a particular
11407 MIPS instruction. Please refer to the architecture specification
11408 for details on what each instruction does.
11411 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
11412 Pair lower lower (@code{pll.ps}).
11414 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
11415 Pair upper lower (@code{pul.ps}).
11417 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
11418 Pair lower upper (@code{plu.ps}).
11420 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
11421 Pair upper upper (@code{puu.ps}).
11423 @item v2sf __builtin_mips_cvt_ps_s (float, float)
11424 Convert pair to paired single (@code{cvt.ps.s}).
11426 @item float __builtin_mips_cvt_s_pl (v2sf)
11427 Convert pair lower to single (@code{cvt.s.pl}).
11429 @item float __builtin_mips_cvt_s_pu (v2sf)
11430 Convert pair upper to single (@code{cvt.s.pu}).
11432 @item v2sf __builtin_mips_abs_ps (v2sf)
11433 Absolute value (@code{abs.ps}).
11435 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
11436 Align variable (@code{alnv.ps}).
11438 @emph{Note:} The value of the third parameter must be 0 or 4
11439 modulo 8, otherwise the result is unpredictable. Please read the
11440 instruction description for details.
11443 The following multi-instruction functions are also available.
11444 In each case, @var{cond} can be any of the 16 floating-point conditions:
11445 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
11446 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
11447 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
11450 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11451 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11452 Conditional move based on floating-point comparison (@code{c.@var{cond}.ps},
11453 @code{movt.ps}/@code{movf.ps}).
11455 The @code{movt} functions return the value @var{x} computed by:
11458 c.@var{cond}.ps @var{cc},@var{a},@var{b}
11459 mov.ps @var{x},@var{c}
11460 movt.ps @var{x},@var{d},@var{cc}
11463 The @code{movf} functions are similar but use @code{movf.ps} instead
11466 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11467 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11468 Comparison of two paired-single values (@code{c.@var{cond}.ps},
11469 @code{bc1t}/@code{bc1f}).
11471 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
11472 and return either the upper or lower half of the result. For example:
11476 if (__builtin_mips_upper_c_eq_ps (a, b))
11477 upper_halves_are_equal ();
11479 upper_halves_are_unequal ();
11481 if (__builtin_mips_lower_c_eq_ps (a, b))
11482 lower_halves_are_equal ();
11484 lower_halves_are_unequal ();
11488 @node MIPS-3D Built-in Functions
11489 @subsubsection MIPS-3D Built-in Functions
11491 The MIPS-3D Application-Specific Extension (ASE) includes additional
11492 paired-single instructions that are designed to improve the performance
11493 of 3D graphics operations. Support for these instructions is controlled
11494 by the @option{-mips3d} command-line option.
11496 The functions listed below map directly to a particular MIPS-3D
11497 instruction. Please refer to the architecture specification for
11498 more details on what each instruction does.
11501 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
11502 Reduction add (@code{addr.ps}).
11504 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
11505 Reduction multiply (@code{mulr.ps}).
11507 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
11508 Convert paired single to paired word (@code{cvt.pw.ps}).
11510 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
11511 Convert paired word to paired single (@code{cvt.ps.pw}).
11513 @item float __builtin_mips_recip1_s (float)
11514 @itemx double __builtin_mips_recip1_d (double)
11515 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
11516 Reduced-precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
11518 @item float __builtin_mips_recip2_s (float, float)
11519 @itemx double __builtin_mips_recip2_d (double, double)
11520 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
11521 Reduced-precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
11523 @item float __builtin_mips_rsqrt1_s (float)
11524 @itemx double __builtin_mips_rsqrt1_d (double)
11525 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
11526 Reduced-precision reciprocal square root (sequence step 1)
11527 (@code{rsqrt1.@var{fmt}}).
11529 @item float __builtin_mips_rsqrt2_s (float, float)
11530 @itemx double __builtin_mips_rsqrt2_d (double, double)
11531 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
11532 Reduced-precision reciprocal square root (sequence step 2)
11533 (@code{rsqrt2.@var{fmt}}).
11536 The following multi-instruction functions are also available.
11537 In each case, @var{cond} can be any of the 16 floating-point conditions:
11538 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
11539 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
11540 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
11543 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
11544 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
11545 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
11546 @code{bc1t}/@code{bc1f}).
11548 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
11549 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
11554 if (__builtin_mips_cabs_eq_s (a, b))
11560 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11561 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11562 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
11563 @code{bc1t}/@code{bc1f}).
11565 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
11566 and return either the upper or lower half of the result. For example:
11570 if (__builtin_mips_upper_cabs_eq_ps (a, b))
11571 upper_halves_are_equal ();
11573 upper_halves_are_unequal ();
11575 if (__builtin_mips_lower_cabs_eq_ps (a, b))
11576 lower_halves_are_equal ();
11578 lower_halves_are_unequal ();
11581 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11582 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11583 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
11584 @code{movt.ps}/@code{movf.ps}).
11586 The @code{movt} functions return the value @var{x} computed by:
11589 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
11590 mov.ps @var{x},@var{c}
11591 movt.ps @var{x},@var{d},@var{cc}
11594 The @code{movf} functions are similar but use @code{movf.ps} instead
11597 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11598 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11599 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11600 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11601 Comparison of two paired-single values
11602 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
11603 @code{bc1any2t}/@code{bc1any2f}).
11605 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
11606 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
11607 result is true and the @code{all} forms return true if both results are true.
11612 if (__builtin_mips_any_c_eq_ps (a, b))
11617 if (__builtin_mips_all_c_eq_ps (a, b))
11623 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11624 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11625 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11626 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11627 Comparison of four paired-single values
11628 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
11629 @code{bc1any4t}/@code{bc1any4f}).
11631 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
11632 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
11633 The @code{any} forms return true if any of the four results are true
11634 and the @code{all} forms return true if all four results are true.
11639 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
11644 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
11651 @node picoChip Built-in Functions
11652 @subsection picoChip Built-in Functions
11654 GCC provides an interface to selected machine instructions from the
11655 picoChip instruction set.
11658 @item int __builtin_sbc (int @var{value})
11659 Sign bit count. Return the number of consecutive bits in @var{value}
11660 that have the same value as the sign bit. The result is the number of
11661 leading sign bits minus one, giving the number of redundant sign bits in
11664 @item int __builtin_byteswap (int @var{value})
11665 Byte swap. Return the result of swapping the upper and lower bytes of
11668 @item int __builtin_brev (int @var{value})
11669 Bit reversal. Return the result of reversing the bits in
11670 @var{value}. Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1,
11673 @item int __builtin_adds (int @var{x}, int @var{y})
11674 Saturating addition. Return the result of adding @var{x} and @var{y},
11675 storing the value 32767 if the result overflows.
11677 @item int __builtin_subs (int @var{x}, int @var{y})
11678 Saturating subtraction. Return the result of subtracting @var{y} from
11679 @var{x}, storing the value @minus{}32768 if the result overflows.
11681 @item void __builtin_halt (void)
11682 Halt. The processor stops execution. This built-in is useful for
11683 implementing assertions.
11687 @node Other MIPS Built-in Functions
11688 @subsection Other MIPS Built-in Functions
11690 GCC provides other MIPS-specific built-in functions:
11693 @item void __builtin_mips_cache (int @var{op}, const volatile void *@var{addr})
11694 Insert a @samp{cache} instruction with operands @var{op} and @var{addr}.
11695 GCC defines the preprocessor macro @code{___GCC_HAVE_BUILTIN_MIPS_CACHE}
11696 when this function is available.
11699 @node PowerPC Built-in Functions
11700 @subsection PowerPC Built-in Functions
11702 These built-in functions are available for the PowerPC family of
11705 float __builtin_recipdivf (float, float);
11706 float __builtin_rsqrtf (float);
11707 double __builtin_recipdiv (double, double);
11708 double __builtin_rsqrt (double);
11709 long __builtin_bpermd (long, long);
11710 uint64_t __builtin_ppc_get_timebase ();
11711 unsigned long __builtin_ppc_mftb ();
11714 The @code{vec_rsqrt}, @code{__builtin_rsqrt}, and
11715 @code{__builtin_rsqrtf} functions generate multiple instructions to
11716 implement the reciprocal sqrt functionality using reciprocal sqrt
11717 estimate instructions.
11719 The @code{__builtin_recipdiv}, and @code{__builtin_recipdivf}
11720 functions generate multiple instructions to implement division using
11721 the reciprocal estimate instructions.
11723 The @code{__builtin_ppc_get_timebase} and @code{__builtin_ppc_mftb}
11724 functions generate instructions to read the Time Base Register. The
11725 @code{__builtin_ppc_get_timebase} function may generate multiple
11726 instructions and always returns the 64 bits of the Time Base Register.
11727 The @code{__builtin_ppc_mftb} function always generates one instruction and
11728 returns the Time Base Register value as an unsigned long, throwing away
11729 the most significant word on 32-bit environments.
11731 @node PowerPC AltiVec/VSX Built-in Functions
11732 @subsection PowerPC AltiVec Built-in Functions
11734 GCC provides an interface for the PowerPC family of processors to access
11735 the AltiVec operations described in Motorola's AltiVec Programming
11736 Interface Manual. The interface is made available by including
11737 @code{<altivec.h>} and using @option{-maltivec} and
11738 @option{-mabi=altivec}. The interface supports the following vector
11742 vector unsigned char
11746 vector unsigned short
11747 vector signed short
11751 vector unsigned int
11757 If @option{-mvsx} is used the following additional vector types are
11761 vector unsigned long
11766 The long types are only implemented for 64-bit code generation, and
11767 the long type is only used in the floating point/integer conversion
11770 GCC's implementation of the high-level language interface available from
11771 C and C++ code differs from Motorola's documentation in several ways.
11776 A vector constant is a list of constant expressions within curly braces.
11779 A vector initializer requires no cast if the vector constant is of the
11780 same type as the variable it is initializing.
11783 If @code{signed} or @code{unsigned} is omitted, the signedness of the
11784 vector type is the default signedness of the base type. The default
11785 varies depending on the operating system, so a portable program should
11786 always specify the signedness.
11789 Compiling with @option{-maltivec} adds keywords @code{__vector},
11790 @code{vector}, @code{__pixel}, @code{pixel}, @code{__bool} and
11791 @code{bool}. When compiling ISO C, the context-sensitive substitution
11792 of the keywords @code{vector}, @code{pixel} and @code{bool} is
11793 disabled. To use them, you must include @code{<altivec.h>} instead.
11796 GCC allows using a @code{typedef} name as the type specifier for a
11800 For C, overloaded functions are implemented with macros so the following
11804 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
11808 Since @code{vec_add} is a macro, the vector constant in the example
11809 is treated as four separate arguments. Wrap the entire argument in
11810 parentheses for this to work.
11813 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
11814 Internally, GCC uses built-in functions to achieve the functionality in
11815 the aforementioned header file, but they are not supported and are
11816 subject to change without notice.
11818 The following interfaces are supported for the generic and specific
11819 AltiVec operations and the AltiVec predicates. In cases where there
11820 is a direct mapping between generic and specific operations, only the
11821 generic names are shown here, although the specific operations can also
11824 Arguments that are documented as @code{const int} require literal
11825 integral values within the range required for that operation.
11828 vector signed char vec_abs (vector signed char);
11829 vector signed short vec_abs (vector signed short);
11830 vector signed int vec_abs (vector signed int);
11831 vector float vec_abs (vector float);
11833 vector signed char vec_abss (vector signed char);
11834 vector signed short vec_abss (vector signed short);
11835 vector signed int vec_abss (vector signed int);
11837 vector signed char vec_add (vector bool char, vector signed char);
11838 vector signed char vec_add (vector signed char, vector bool char);
11839 vector signed char vec_add (vector signed char, vector signed char);
11840 vector unsigned char vec_add (vector bool char, vector unsigned char);
11841 vector unsigned char vec_add (vector unsigned char, vector bool char);
11842 vector unsigned char vec_add (vector unsigned char,
11843 vector unsigned char);
11844 vector signed short vec_add (vector bool short, vector signed short);
11845 vector signed short vec_add (vector signed short, vector bool short);
11846 vector signed short vec_add (vector signed short, vector signed short);
11847 vector unsigned short vec_add (vector bool short,
11848 vector unsigned short);
11849 vector unsigned short vec_add (vector unsigned short,
11850 vector bool short);
11851 vector unsigned short vec_add (vector unsigned short,
11852 vector unsigned short);
11853 vector signed int vec_add (vector bool int, vector signed int);
11854 vector signed int vec_add (vector signed int, vector bool int);
11855 vector signed int vec_add (vector signed int, vector signed int);
11856 vector unsigned int vec_add (vector bool int, vector unsigned int);
11857 vector unsigned int vec_add (vector unsigned int, vector bool int);
11858 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
11859 vector float vec_add (vector float, vector float);
11861 vector float vec_vaddfp (vector float, vector float);
11863 vector signed int vec_vadduwm (vector bool int, vector signed int);
11864 vector signed int vec_vadduwm (vector signed int, vector bool int);
11865 vector signed int vec_vadduwm (vector signed int, vector signed int);
11866 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
11867 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
11868 vector unsigned int vec_vadduwm (vector unsigned int,
11869 vector unsigned int);
11871 vector signed short vec_vadduhm (vector bool short,
11872 vector signed short);
11873 vector signed short vec_vadduhm (vector signed short,
11874 vector bool short);
11875 vector signed short vec_vadduhm (vector signed short,
11876 vector signed short);
11877 vector unsigned short vec_vadduhm (vector bool short,
11878 vector unsigned short);
11879 vector unsigned short vec_vadduhm (vector unsigned short,
11880 vector bool short);
11881 vector unsigned short vec_vadduhm (vector unsigned short,
11882 vector unsigned short);
11884 vector signed char vec_vaddubm (vector bool char, vector signed char);
11885 vector signed char vec_vaddubm (vector signed char, vector bool char);
11886 vector signed char vec_vaddubm (vector signed char, vector signed char);
11887 vector unsigned char vec_vaddubm (vector bool char,
11888 vector unsigned char);
11889 vector unsigned char vec_vaddubm (vector unsigned char,
11891 vector unsigned char vec_vaddubm (vector unsigned char,
11892 vector unsigned char);
11894 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
11896 vector unsigned char vec_adds (vector bool char, vector unsigned char);
11897 vector unsigned char vec_adds (vector unsigned char, vector bool char);
11898 vector unsigned char vec_adds (vector unsigned char,
11899 vector unsigned char);
11900 vector signed char vec_adds (vector bool char, vector signed char);
11901 vector signed char vec_adds (vector signed char, vector bool char);
11902 vector signed char vec_adds (vector signed char, vector signed char);
11903 vector unsigned short vec_adds (vector bool short,
11904 vector unsigned short);
11905 vector unsigned short vec_adds (vector unsigned short,
11906 vector bool short);
11907 vector unsigned short vec_adds (vector unsigned short,
11908 vector unsigned short);
11909 vector signed short vec_adds (vector bool short, vector signed short);
11910 vector signed short vec_adds (vector signed short, vector bool short);
11911 vector signed short vec_adds (vector signed short, vector signed short);
11912 vector unsigned int vec_adds (vector bool int, vector unsigned int);
11913 vector unsigned int vec_adds (vector unsigned int, vector bool int);
11914 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
11915 vector signed int vec_adds (vector bool int, vector signed int);
11916 vector signed int vec_adds (vector signed int, vector bool int);
11917 vector signed int vec_adds (vector signed int, vector signed int);
11919 vector signed int vec_vaddsws (vector bool int, vector signed int);
11920 vector signed int vec_vaddsws (vector signed int, vector bool int);
11921 vector signed int vec_vaddsws (vector signed int, vector signed int);
11923 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
11924 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
11925 vector unsigned int vec_vadduws (vector unsigned int,
11926 vector unsigned int);
11928 vector signed short vec_vaddshs (vector bool short,
11929 vector signed short);
11930 vector signed short vec_vaddshs (vector signed short,
11931 vector bool short);
11932 vector signed short vec_vaddshs (vector signed short,
11933 vector signed short);
11935 vector unsigned short vec_vadduhs (vector bool short,
11936 vector unsigned short);
11937 vector unsigned short vec_vadduhs (vector unsigned short,
11938 vector bool short);
11939 vector unsigned short vec_vadduhs (vector unsigned short,
11940 vector unsigned short);
11942 vector signed char vec_vaddsbs (vector bool char, vector signed char);
11943 vector signed char vec_vaddsbs (vector signed char, vector bool char);
11944 vector signed char vec_vaddsbs (vector signed char, vector signed char);
11946 vector unsigned char vec_vaddubs (vector bool char,
11947 vector unsigned char);
11948 vector unsigned char vec_vaddubs (vector unsigned char,
11950 vector unsigned char vec_vaddubs (vector unsigned char,
11951 vector unsigned char);
11953 vector float vec_and (vector float, vector float);
11954 vector float vec_and (vector float, vector bool int);
11955 vector float vec_and (vector bool int, vector float);
11956 vector bool int vec_and (vector bool int, vector bool int);
11957 vector signed int vec_and (vector bool int, vector signed int);
11958 vector signed int vec_and (vector signed int, vector bool int);
11959 vector signed int vec_and (vector signed int, vector signed int);
11960 vector unsigned int vec_and (vector bool int, vector unsigned int);
11961 vector unsigned int vec_and (vector unsigned int, vector bool int);
11962 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
11963 vector bool short vec_and (vector bool short, vector bool short);
11964 vector signed short vec_and (vector bool short, vector signed short);
11965 vector signed short vec_and (vector signed short, vector bool short);
11966 vector signed short vec_and (vector signed short, vector signed short);
11967 vector unsigned short vec_and (vector bool short,
11968 vector unsigned short);
11969 vector unsigned short vec_and (vector unsigned short,
11970 vector bool short);
11971 vector unsigned short vec_and (vector unsigned short,
11972 vector unsigned short);
11973 vector signed char vec_and (vector bool char, vector signed char);
11974 vector bool char vec_and (vector bool char, vector bool char);
11975 vector signed char vec_and (vector signed char, vector bool char);
11976 vector signed char vec_and (vector signed char, vector signed char);
11977 vector unsigned char vec_and (vector bool char, vector unsigned char);
11978 vector unsigned char vec_and (vector unsigned char, vector bool char);
11979 vector unsigned char vec_and (vector unsigned char,
11980 vector unsigned char);
11982 vector float vec_andc (vector float, vector float);
11983 vector float vec_andc (vector float, vector bool int);
11984 vector float vec_andc (vector bool int, vector float);
11985 vector bool int vec_andc (vector bool int, vector bool int);
11986 vector signed int vec_andc (vector bool int, vector signed int);
11987 vector signed int vec_andc (vector signed int, vector bool int);
11988 vector signed int vec_andc (vector signed int, vector signed int);
11989 vector unsigned int vec_andc (vector bool int, vector unsigned int);
11990 vector unsigned int vec_andc (vector unsigned int, vector bool int);
11991 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
11992 vector bool short vec_andc (vector bool short, vector bool short);
11993 vector signed short vec_andc (vector bool short, vector signed short);
11994 vector signed short vec_andc (vector signed short, vector bool short);
11995 vector signed short vec_andc (vector signed short, vector signed short);
11996 vector unsigned short vec_andc (vector bool short,
11997 vector unsigned short);
11998 vector unsigned short vec_andc (vector unsigned short,
11999 vector bool short);
12000 vector unsigned short vec_andc (vector unsigned short,
12001 vector unsigned short);
12002 vector signed char vec_andc (vector bool char, vector signed char);
12003 vector bool char vec_andc (vector bool char, vector bool char);
12004 vector signed char vec_andc (vector signed char, vector bool char);
12005 vector signed char vec_andc (vector signed char, vector signed char);
12006 vector unsigned char vec_andc (vector bool char, vector unsigned char);
12007 vector unsigned char vec_andc (vector unsigned char, vector bool char);
12008 vector unsigned char vec_andc (vector unsigned char,
12009 vector unsigned char);
12011 vector unsigned char vec_avg (vector unsigned char,
12012 vector unsigned char);
12013 vector signed char vec_avg (vector signed char, vector signed char);
12014 vector unsigned short vec_avg (vector unsigned short,
12015 vector unsigned short);
12016 vector signed short vec_avg (vector signed short, vector signed short);
12017 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
12018 vector signed int vec_avg (vector signed int, vector signed int);
12020 vector signed int vec_vavgsw (vector signed int, vector signed int);
12022 vector unsigned int vec_vavguw (vector unsigned int,
12023 vector unsigned int);
12025 vector signed short vec_vavgsh (vector signed short,
12026 vector signed short);
12028 vector unsigned short vec_vavguh (vector unsigned short,
12029 vector unsigned short);
12031 vector signed char vec_vavgsb (vector signed char, vector signed char);
12033 vector unsigned char vec_vavgub (vector unsigned char,
12034 vector unsigned char);
12036 vector float vec_copysign (vector float);
12038 vector float vec_ceil (vector float);
12040 vector signed int vec_cmpb (vector float, vector float);
12042 vector bool char vec_cmpeq (vector signed char, vector signed char);
12043 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
12044 vector bool short vec_cmpeq (vector signed short, vector signed short);
12045 vector bool short vec_cmpeq (vector unsigned short,
12046 vector unsigned short);
12047 vector bool int vec_cmpeq (vector signed int, vector signed int);
12048 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
12049 vector bool int vec_cmpeq (vector float, vector float);
12051 vector bool int vec_vcmpeqfp (vector float, vector float);
12053 vector bool int vec_vcmpequw (vector signed int, vector signed int);
12054 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
12056 vector bool short vec_vcmpequh (vector signed short,
12057 vector signed short);
12058 vector bool short vec_vcmpequh (vector unsigned short,
12059 vector unsigned short);
12061 vector bool char vec_vcmpequb (vector signed char, vector signed char);
12062 vector bool char vec_vcmpequb (vector unsigned char,
12063 vector unsigned char);
12065 vector bool int vec_cmpge (vector float, vector float);
12067 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
12068 vector bool char vec_cmpgt (vector signed char, vector signed char);
12069 vector bool short vec_cmpgt (vector unsigned short,
12070 vector unsigned short);
12071 vector bool short vec_cmpgt (vector signed short, vector signed short);
12072 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
12073 vector bool int vec_cmpgt (vector signed int, vector signed int);
12074 vector bool int vec_cmpgt (vector float, vector float);
12076 vector bool int vec_vcmpgtfp (vector float, vector float);
12078 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
12080 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
12082 vector bool short vec_vcmpgtsh (vector signed short,
12083 vector signed short);
12085 vector bool short vec_vcmpgtuh (vector unsigned short,
12086 vector unsigned short);
12088 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
12090 vector bool char vec_vcmpgtub (vector unsigned char,
12091 vector unsigned char);
12093 vector bool int vec_cmple (vector float, vector float);
12095 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
12096 vector bool char vec_cmplt (vector signed char, vector signed char);
12097 vector bool short vec_cmplt (vector unsigned short,
12098 vector unsigned short);
12099 vector bool short vec_cmplt (vector signed short, vector signed short);
12100 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
12101 vector bool int vec_cmplt (vector signed int, vector signed int);
12102 vector bool int vec_cmplt (vector float, vector float);
12104 vector float vec_ctf (vector unsigned int, const int);
12105 vector float vec_ctf (vector signed int, const int);
12107 vector float vec_vcfsx (vector signed int, const int);
12109 vector float vec_vcfux (vector unsigned int, const int);
12111 vector signed int vec_cts (vector float, const int);
12113 vector unsigned int vec_ctu (vector float, const int);
12115 void vec_dss (const int);
12117 void vec_dssall (void);
12119 void vec_dst (const vector unsigned char *, int, const int);
12120 void vec_dst (const vector signed char *, int, const int);
12121 void vec_dst (const vector bool char *, int, const int);
12122 void vec_dst (const vector unsigned short *, int, const int);
12123 void vec_dst (const vector signed short *, int, const int);
12124 void vec_dst (const vector bool short *, int, const int);
12125 void vec_dst (const vector pixel *, int, const int);
12126 void vec_dst (const vector unsigned int *, int, const int);
12127 void vec_dst (const vector signed int *, int, const int);
12128 void vec_dst (const vector bool int *, int, const int);
12129 void vec_dst (const vector float *, int, const int);
12130 void vec_dst (const unsigned char *, int, const int);
12131 void vec_dst (const signed char *, int, const int);
12132 void vec_dst (const unsigned short *, int, const int);
12133 void vec_dst (const short *, int, const int);
12134 void vec_dst (const unsigned int *, int, const int);
12135 void vec_dst (const int *, int, const int);
12136 void vec_dst (const unsigned long *, int, const int);
12137 void vec_dst (const long *, int, const int);
12138 void vec_dst (const float *, int, const int);
12140 void vec_dstst (const vector unsigned char *, int, const int);
12141 void vec_dstst (const vector signed char *, int, const int);
12142 void vec_dstst (const vector bool char *, int, const int);
12143 void vec_dstst (const vector unsigned short *, int, const int);
12144 void vec_dstst (const vector signed short *, int, const int);
12145 void vec_dstst (const vector bool short *, int, const int);
12146 void vec_dstst (const vector pixel *, int, const int);
12147 void vec_dstst (const vector unsigned int *, int, const int);
12148 void vec_dstst (const vector signed int *, int, const int);
12149 void vec_dstst (const vector bool int *, int, const int);
12150 void vec_dstst (const vector float *, int, const int);
12151 void vec_dstst (const unsigned char *, int, const int);
12152 void vec_dstst (const signed char *, int, const int);
12153 void vec_dstst (const unsigned short *, int, const int);
12154 void vec_dstst (const short *, int, const int);
12155 void vec_dstst (const unsigned int *, int, const int);
12156 void vec_dstst (const int *, int, const int);
12157 void vec_dstst (const unsigned long *, int, const int);
12158 void vec_dstst (const long *, int, const int);
12159 void vec_dstst (const float *, int, const int);
12161 void vec_dststt (const vector unsigned char *, int, const int);
12162 void vec_dststt (const vector signed char *, int, const int);
12163 void vec_dststt (const vector bool char *, int, const int);
12164 void vec_dststt (const vector unsigned short *, int, const int);
12165 void vec_dststt (const vector signed short *, int, const int);
12166 void vec_dststt (const vector bool short *, int, const int);
12167 void vec_dststt (const vector pixel *, int, const int);
12168 void vec_dststt (const vector unsigned int *, int, const int);
12169 void vec_dststt (const vector signed int *, int, const int);
12170 void vec_dststt (const vector bool int *, int, const int);
12171 void vec_dststt (const vector float *, int, const int);
12172 void vec_dststt (const unsigned char *, int, const int);
12173 void vec_dststt (const signed char *, int, const int);
12174 void vec_dststt (const unsigned short *, int, const int);
12175 void vec_dststt (const short *, int, const int);
12176 void vec_dststt (const unsigned int *, int, const int);
12177 void vec_dststt (const int *, int, const int);
12178 void vec_dststt (const unsigned long *, int, const int);
12179 void vec_dststt (const long *, int, const int);
12180 void vec_dststt (const float *, int, const int);
12182 void vec_dstt (const vector unsigned char *, int, const int);
12183 void vec_dstt (const vector signed char *, int, const int);
12184 void vec_dstt (const vector bool char *, int, const int);
12185 void vec_dstt (const vector unsigned short *, int, const int);
12186 void vec_dstt (const vector signed short *, int, const int);
12187 void vec_dstt (const vector bool short *, int, const int);
12188 void vec_dstt (const vector pixel *, int, const int);
12189 void vec_dstt (const vector unsigned int *, int, const int);
12190 void vec_dstt (const vector signed int *, int, const int);
12191 void vec_dstt (const vector bool int *, int, const int);
12192 void vec_dstt (const vector float *, int, const int);
12193 void vec_dstt (const unsigned char *, int, const int);
12194 void vec_dstt (const signed char *, int, const int);
12195 void vec_dstt (const unsigned short *, int, const int);
12196 void vec_dstt (const short *, int, const int);
12197 void vec_dstt (const unsigned int *, int, const int);
12198 void vec_dstt (const int *, int, const int);
12199 void vec_dstt (const unsigned long *, int, const int);
12200 void vec_dstt (const long *, int, const int);
12201 void vec_dstt (const float *, int, const int);
12203 vector float vec_expte (vector float);
12205 vector float vec_floor (vector float);
12207 vector float vec_ld (int, const vector float *);
12208 vector float vec_ld (int, const float *);
12209 vector bool int vec_ld (int, const vector bool int *);
12210 vector signed int vec_ld (int, const vector signed int *);
12211 vector signed int vec_ld (int, const int *);
12212 vector signed int vec_ld (int, const long *);
12213 vector unsigned int vec_ld (int, const vector unsigned int *);
12214 vector unsigned int vec_ld (int, const unsigned int *);
12215 vector unsigned int vec_ld (int, const unsigned long *);
12216 vector bool short vec_ld (int, const vector bool short *);
12217 vector pixel vec_ld (int, const vector pixel *);
12218 vector signed short vec_ld (int, const vector signed short *);
12219 vector signed short vec_ld (int, const short *);
12220 vector unsigned short vec_ld (int, const vector unsigned short *);
12221 vector unsigned short vec_ld (int, const unsigned short *);
12222 vector bool char vec_ld (int, const vector bool char *);
12223 vector signed char vec_ld (int, const vector signed char *);
12224 vector signed char vec_ld (int, const signed char *);
12225 vector unsigned char vec_ld (int, const vector unsigned char *);
12226 vector unsigned char vec_ld (int, const unsigned char *);
12228 vector signed char vec_lde (int, const signed char *);
12229 vector unsigned char vec_lde (int, const unsigned char *);
12230 vector signed short vec_lde (int, const short *);
12231 vector unsigned short vec_lde (int, const unsigned short *);
12232 vector float vec_lde (int, const float *);
12233 vector signed int vec_lde (int, const int *);
12234 vector unsigned int vec_lde (int, const unsigned int *);
12235 vector signed int vec_lde (int, const long *);
12236 vector unsigned int vec_lde (int, const unsigned long *);
12238 vector float vec_lvewx (int, float *);
12239 vector signed int vec_lvewx (int, int *);
12240 vector unsigned int vec_lvewx (int, unsigned int *);
12241 vector signed int vec_lvewx (int, long *);
12242 vector unsigned int vec_lvewx (int, unsigned long *);
12244 vector signed short vec_lvehx (int, short *);
12245 vector unsigned short vec_lvehx (int, unsigned short *);
12247 vector signed char vec_lvebx (int, char *);
12248 vector unsigned char vec_lvebx (int, unsigned char *);
12250 vector float vec_ldl (int, const vector float *);
12251 vector float vec_ldl (int, const float *);
12252 vector bool int vec_ldl (int, const vector bool int *);
12253 vector signed int vec_ldl (int, const vector signed int *);
12254 vector signed int vec_ldl (int, const int *);
12255 vector signed int vec_ldl (int, const long *);
12256 vector unsigned int vec_ldl (int, const vector unsigned int *);
12257 vector unsigned int vec_ldl (int, const unsigned int *);
12258 vector unsigned int vec_ldl (int, const unsigned long *);
12259 vector bool short vec_ldl (int, const vector bool short *);
12260 vector pixel vec_ldl (int, const vector pixel *);
12261 vector signed short vec_ldl (int, const vector signed short *);
12262 vector signed short vec_ldl (int, const short *);
12263 vector unsigned short vec_ldl (int, const vector unsigned short *);
12264 vector unsigned short vec_ldl (int, const unsigned short *);
12265 vector bool char vec_ldl (int, const vector bool char *);
12266 vector signed char vec_ldl (int, const vector signed char *);
12267 vector signed char vec_ldl (int, const signed char *);
12268 vector unsigned char vec_ldl (int, const vector unsigned char *);
12269 vector unsigned char vec_ldl (int, const unsigned char *);
12271 vector float vec_loge (vector float);
12273 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
12274 vector unsigned char vec_lvsl (int, const volatile signed char *);
12275 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
12276 vector unsigned char vec_lvsl (int, const volatile short *);
12277 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
12278 vector unsigned char vec_lvsl (int, const volatile int *);
12279 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
12280 vector unsigned char vec_lvsl (int, const volatile long *);
12281 vector unsigned char vec_lvsl (int, const volatile float *);
12283 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
12284 vector unsigned char vec_lvsr (int, const volatile signed char *);
12285 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
12286 vector unsigned char vec_lvsr (int, const volatile short *);
12287 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
12288 vector unsigned char vec_lvsr (int, const volatile int *);
12289 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
12290 vector unsigned char vec_lvsr (int, const volatile long *);
12291 vector unsigned char vec_lvsr (int, const volatile float *);
12293 vector float vec_madd (vector float, vector float, vector float);
12295 vector signed short vec_madds (vector signed short,
12296 vector signed short,
12297 vector signed short);
12299 vector unsigned char vec_max (vector bool char, vector unsigned char);
12300 vector unsigned char vec_max (vector unsigned char, vector bool char);
12301 vector unsigned char vec_max (vector unsigned char,
12302 vector unsigned char);
12303 vector signed char vec_max (vector bool char, vector signed char);
12304 vector signed char vec_max (vector signed char, vector bool char);
12305 vector signed char vec_max (vector signed char, vector signed char);
12306 vector unsigned short vec_max (vector bool short,
12307 vector unsigned short);
12308 vector unsigned short vec_max (vector unsigned short,
12309 vector bool short);
12310 vector unsigned short vec_max (vector unsigned short,
12311 vector unsigned short);
12312 vector signed short vec_max (vector bool short, vector signed short);
12313 vector signed short vec_max (vector signed short, vector bool short);
12314 vector signed short vec_max (vector signed short, vector signed short);
12315 vector unsigned int vec_max (vector bool int, vector unsigned int);
12316 vector unsigned int vec_max (vector unsigned int, vector bool int);
12317 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
12318 vector signed int vec_max (vector bool int, vector signed int);
12319 vector signed int vec_max (vector signed int, vector bool int);
12320 vector signed int vec_max (vector signed int, vector signed int);
12321 vector float vec_max (vector float, vector float);
12323 vector float vec_vmaxfp (vector float, vector float);
12325 vector signed int vec_vmaxsw (vector bool int, vector signed int);
12326 vector signed int vec_vmaxsw (vector signed int, vector bool int);
12327 vector signed int vec_vmaxsw (vector signed int, vector signed int);
12329 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
12330 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
12331 vector unsigned int vec_vmaxuw (vector unsigned int,
12332 vector unsigned int);
12334 vector signed short vec_vmaxsh (vector bool short, vector signed short);
12335 vector signed short vec_vmaxsh (vector signed short, vector bool short);
12336 vector signed short vec_vmaxsh (vector signed short,
12337 vector signed short);
12339 vector unsigned short vec_vmaxuh (vector bool short,
12340 vector unsigned short);
12341 vector unsigned short vec_vmaxuh (vector unsigned short,
12342 vector bool short);
12343 vector unsigned short vec_vmaxuh (vector unsigned short,
12344 vector unsigned short);
12346 vector signed char vec_vmaxsb (vector bool char, vector signed char);
12347 vector signed char vec_vmaxsb (vector signed char, vector bool char);
12348 vector signed char vec_vmaxsb (vector signed char, vector signed char);
12350 vector unsigned char vec_vmaxub (vector bool char,
12351 vector unsigned char);
12352 vector unsigned char vec_vmaxub (vector unsigned char,
12354 vector unsigned char vec_vmaxub (vector unsigned char,
12355 vector unsigned char);
12357 vector bool char vec_mergeh (vector bool char, vector bool char);
12358 vector signed char vec_mergeh (vector signed char, vector signed char);
12359 vector unsigned char vec_mergeh (vector unsigned char,
12360 vector unsigned char);
12361 vector bool short vec_mergeh (vector bool short, vector bool short);
12362 vector pixel vec_mergeh (vector pixel, vector pixel);
12363 vector signed short vec_mergeh (vector signed short,
12364 vector signed short);
12365 vector unsigned short vec_mergeh (vector unsigned short,
12366 vector unsigned short);
12367 vector float vec_mergeh (vector float, vector float);
12368 vector bool int vec_mergeh (vector bool int, vector bool int);
12369 vector signed int vec_mergeh (vector signed int, vector signed int);
12370 vector unsigned int vec_mergeh (vector unsigned int,
12371 vector unsigned int);
12373 vector float vec_vmrghw (vector float, vector float);
12374 vector bool int vec_vmrghw (vector bool int, vector bool int);
12375 vector signed int vec_vmrghw (vector signed int, vector signed int);
12376 vector unsigned int vec_vmrghw (vector unsigned int,
12377 vector unsigned int);
12379 vector bool short vec_vmrghh (vector bool short, vector bool short);
12380 vector signed short vec_vmrghh (vector signed short,
12381 vector signed short);
12382 vector unsigned short vec_vmrghh (vector unsigned short,
12383 vector unsigned short);
12384 vector pixel vec_vmrghh (vector pixel, vector pixel);
12386 vector bool char vec_vmrghb (vector bool char, vector bool char);
12387 vector signed char vec_vmrghb (vector signed char, vector signed char);
12388 vector unsigned char vec_vmrghb (vector unsigned char,
12389 vector unsigned char);
12391 vector bool char vec_mergel (vector bool char, vector bool char);
12392 vector signed char vec_mergel (vector signed char, vector signed char);
12393 vector unsigned char vec_mergel (vector unsigned char,
12394 vector unsigned char);
12395 vector bool short vec_mergel (vector bool short, vector bool short);
12396 vector pixel vec_mergel (vector pixel, vector pixel);
12397 vector signed short vec_mergel (vector signed short,
12398 vector signed short);
12399 vector unsigned short vec_mergel (vector unsigned short,
12400 vector unsigned short);
12401 vector float vec_mergel (vector float, vector float);
12402 vector bool int vec_mergel (vector bool int, vector bool int);
12403 vector signed int vec_mergel (vector signed int, vector signed int);
12404 vector unsigned int vec_mergel (vector unsigned int,
12405 vector unsigned int);
12407 vector float vec_vmrglw (vector float, vector float);
12408 vector signed int vec_vmrglw (vector signed int, vector signed int);
12409 vector unsigned int vec_vmrglw (vector unsigned int,
12410 vector unsigned int);
12411 vector bool int vec_vmrglw (vector bool int, vector bool int);
12413 vector bool short vec_vmrglh (vector bool short, vector bool short);
12414 vector signed short vec_vmrglh (vector signed short,
12415 vector signed short);
12416 vector unsigned short vec_vmrglh (vector unsigned short,
12417 vector unsigned short);
12418 vector pixel vec_vmrglh (vector pixel, vector pixel);
12420 vector bool char vec_vmrglb (vector bool char, vector bool char);
12421 vector signed char vec_vmrglb (vector signed char, vector signed char);
12422 vector unsigned char vec_vmrglb (vector unsigned char,
12423 vector unsigned char);
12425 vector unsigned short vec_mfvscr (void);
12427 vector unsigned char vec_min (vector bool char, vector unsigned char);
12428 vector unsigned char vec_min (vector unsigned char, vector bool char);
12429 vector unsigned char vec_min (vector unsigned char,
12430 vector unsigned char);
12431 vector signed char vec_min (vector bool char, vector signed char);
12432 vector signed char vec_min (vector signed char, vector bool char);
12433 vector signed char vec_min (vector signed char, vector signed char);
12434 vector unsigned short vec_min (vector bool short,
12435 vector unsigned short);
12436 vector unsigned short vec_min (vector unsigned short,
12437 vector bool short);
12438 vector unsigned short vec_min (vector unsigned short,
12439 vector unsigned short);
12440 vector signed short vec_min (vector bool short, vector signed short);
12441 vector signed short vec_min (vector signed short, vector bool short);
12442 vector signed short vec_min (vector signed short, vector signed short);
12443 vector unsigned int vec_min (vector bool int, vector unsigned int);
12444 vector unsigned int vec_min (vector unsigned int, vector bool int);
12445 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
12446 vector signed int vec_min (vector bool int, vector signed int);
12447 vector signed int vec_min (vector signed int, vector bool int);
12448 vector signed int vec_min (vector signed int, vector signed int);
12449 vector float vec_min (vector float, vector float);
12451 vector float vec_vminfp (vector float, vector float);
12453 vector signed int vec_vminsw (vector bool int, vector signed int);
12454 vector signed int vec_vminsw (vector signed int, vector bool int);
12455 vector signed int vec_vminsw (vector signed int, vector signed int);
12457 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
12458 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
12459 vector unsigned int vec_vminuw (vector unsigned int,
12460 vector unsigned int);
12462 vector signed short vec_vminsh (vector bool short, vector signed short);
12463 vector signed short vec_vminsh (vector signed short, vector bool short);
12464 vector signed short vec_vminsh (vector signed short,
12465 vector signed short);
12467 vector unsigned short vec_vminuh (vector bool short,
12468 vector unsigned short);
12469 vector unsigned short vec_vminuh (vector unsigned short,
12470 vector bool short);
12471 vector unsigned short vec_vminuh (vector unsigned short,
12472 vector unsigned short);
12474 vector signed char vec_vminsb (vector bool char, vector signed char);
12475 vector signed char vec_vminsb (vector signed char, vector bool char);
12476 vector signed char vec_vminsb (vector signed char, vector signed char);
12478 vector unsigned char vec_vminub (vector bool char,
12479 vector unsigned char);
12480 vector unsigned char vec_vminub (vector unsigned char,
12482 vector unsigned char vec_vminub (vector unsigned char,
12483 vector unsigned char);
12485 vector signed short vec_mladd (vector signed short,
12486 vector signed short,
12487 vector signed short);
12488 vector signed short vec_mladd (vector signed short,
12489 vector unsigned short,
12490 vector unsigned short);
12491 vector signed short vec_mladd (vector unsigned short,
12492 vector signed short,
12493 vector signed short);
12494 vector unsigned short vec_mladd (vector unsigned short,
12495 vector unsigned short,
12496 vector unsigned short);
12498 vector signed short vec_mradds (vector signed short,
12499 vector signed short,
12500 vector signed short);
12502 vector unsigned int vec_msum (vector unsigned char,
12503 vector unsigned char,
12504 vector unsigned int);
12505 vector signed int vec_msum (vector signed char,
12506 vector unsigned char,
12507 vector signed int);
12508 vector unsigned int vec_msum (vector unsigned short,
12509 vector unsigned short,
12510 vector unsigned int);
12511 vector signed int vec_msum (vector signed short,
12512 vector signed short,
12513 vector signed int);
12515 vector signed int vec_vmsumshm (vector signed short,
12516 vector signed short,
12517 vector signed int);
12519 vector unsigned int vec_vmsumuhm (vector unsigned short,
12520 vector unsigned short,
12521 vector unsigned int);
12523 vector signed int vec_vmsummbm (vector signed char,
12524 vector unsigned char,
12525 vector signed int);
12527 vector unsigned int vec_vmsumubm (vector unsigned char,
12528 vector unsigned char,
12529 vector unsigned int);
12531 vector unsigned int vec_msums (vector unsigned short,
12532 vector unsigned short,
12533 vector unsigned int);
12534 vector signed int vec_msums (vector signed short,
12535 vector signed short,
12536 vector signed int);
12538 vector signed int vec_vmsumshs (vector signed short,
12539 vector signed short,
12540 vector signed int);
12542 vector unsigned int vec_vmsumuhs (vector unsigned short,
12543 vector unsigned short,
12544 vector unsigned int);
12546 void vec_mtvscr (vector signed int);
12547 void vec_mtvscr (vector unsigned int);
12548 void vec_mtvscr (vector bool int);
12549 void vec_mtvscr (vector signed short);
12550 void vec_mtvscr (vector unsigned short);
12551 void vec_mtvscr (vector bool short);
12552 void vec_mtvscr (vector pixel);
12553 void vec_mtvscr (vector signed char);
12554 void vec_mtvscr (vector unsigned char);
12555 void vec_mtvscr (vector bool char);
12557 vector unsigned short vec_mule (vector unsigned char,
12558 vector unsigned char);
12559 vector signed short vec_mule (vector signed char,
12560 vector signed char);
12561 vector unsigned int vec_mule (vector unsigned short,
12562 vector unsigned short);
12563 vector signed int vec_mule (vector signed short, vector signed short);
12565 vector signed int vec_vmulesh (vector signed short,
12566 vector signed short);
12568 vector unsigned int vec_vmuleuh (vector unsigned short,
12569 vector unsigned short);
12571 vector signed short vec_vmulesb (vector signed char,
12572 vector signed char);
12574 vector unsigned short vec_vmuleub (vector unsigned char,
12575 vector unsigned char);
12577 vector unsigned short vec_mulo (vector unsigned char,
12578 vector unsigned char);
12579 vector signed short vec_mulo (vector signed char, vector signed char);
12580 vector unsigned int vec_mulo (vector unsigned short,
12581 vector unsigned short);
12582 vector signed int vec_mulo (vector signed short, vector signed short);
12584 vector signed int vec_vmulosh (vector signed short,
12585 vector signed short);
12587 vector unsigned int vec_vmulouh (vector unsigned short,
12588 vector unsigned short);
12590 vector signed short vec_vmulosb (vector signed char,
12591 vector signed char);
12593 vector unsigned short vec_vmuloub (vector unsigned char,
12594 vector unsigned char);
12596 vector float vec_nmsub (vector float, vector float, vector float);
12598 vector float vec_nor (vector float, vector float);
12599 vector signed int vec_nor (vector signed int, vector signed int);
12600 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
12601 vector bool int vec_nor (vector bool int, vector bool int);
12602 vector signed short vec_nor (vector signed short, vector signed short);
12603 vector unsigned short vec_nor (vector unsigned short,
12604 vector unsigned short);
12605 vector bool short vec_nor (vector bool short, vector bool short);
12606 vector signed char vec_nor (vector signed char, vector signed char);
12607 vector unsigned char vec_nor (vector unsigned char,
12608 vector unsigned char);
12609 vector bool char vec_nor (vector bool char, vector bool char);
12611 vector float vec_or (vector float, vector float);
12612 vector float vec_or (vector float, vector bool int);
12613 vector float vec_or (vector bool int, vector float);
12614 vector bool int vec_or (vector bool int, vector bool int);
12615 vector signed int vec_or (vector bool int, vector signed int);
12616 vector signed int vec_or (vector signed int, vector bool int);
12617 vector signed int vec_or (vector signed int, vector signed int);
12618 vector unsigned int vec_or (vector bool int, vector unsigned int);
12619 vector unsigned int vec_or (vector unsigned int, vector bool int);
12620 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
12621 vector bool short vec_or (vector bool short, vector bool short);
12622 vector signed short vec_or (vector bool short, vector signed short);
12623 vector signed short vec_or (vector signed short, vector bool short);
12624 vector signed short vec_or (vector signed short, vector signed short);
12625 vector unsigned short vec_or (vector bool short, vector unsigned short);
12626 vector unsigned short vec_or (vector unsigned short, vector bool short);
12627 vector unsigned short vec_or (vector unsigned short,
12628 vector unsigned short);
12629 vector signed char vec_or (vector bool char, vector signed char);
12630 vector bool char vec_or (vector bool char, vector bool char);
12631 vector signed char vec_or (vector signed char, vector bool char);
12632 vector signed char vec_or (vector signed char, vector signed char);
12633 vector unsigned char vec_or (vector bool char, vector unsigned char);
12634 vector unsigned char vec_or (vector unsigned char, vector bool char);
12635 vector unsigned char vec_or (vector unsigned char,
12636 vector unsigned char);
12638 vector signed char vec_pack (vector signed short, vector signed short);
12639 vector unsigned char vec_pack (vector unsigned short,
12640 vector unsigned short);
12641 vector bool char vec_pack (vector bool short, vector bool short);
12642 vector signed short vec_pack (vector signed int, vector signed int);
12643 vector unsigned short vec_pack (vector unsigned int,
12644 vector unsigned int);
12645 vector bool short vec_pack (vector bool int, vector bool int);
12647 vector bool short vec_vpkuwum (vector bool int, vector bool int);
12648 vector signed short vec_vpkuwum (vector signed int, vector signed int);
12649 vector unsigned short vec_vpkuwum (vector unsigned int,
12650 vector unsigned int);
12652 vector bool char vec_vpkuhum (vector bool short, vector bool short);
12653 vector signed char vec_vpkuhum (vector signed short,
12654 vector signed short);
12655 vector unsigned char vec_vpkuhum (vector unsigned short,
12656 vector unsigned short);
12658 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
12660 vector unsigned char vec_packs (vector unsigned short,
12661 vector unsigned short);
12662 vector signed char vec_packs (vector signed short, vector signed short);
12663 vector unsigned short vec_packs (vector unsigned int,
12664 vector unsigned int);
12665 vector signed short vec_packs (vector signed int, vector signed int);
12667 vector signed short vec_vpkswss (vector signed int, vector signed int);
12669 vector unsigned short vec_vpkuwus (vector unsigned int,
12670 vector unsigned int);
12672 vector signed char vec_vpkshss (vector signed short,
12673 vector signed short);
12675 vector unsigned char vec_vpkuhus (vector unsigned short,
12676 vector unsigned short);
12678 vector unsigned char vec_packsu (vector unsigned short,
12679 vector unsigned short);
12680 vector unsigned char vec_packsu (vector signed short,
12681 vector signed short);
12682 vector unsigned short vec_packsu (vector unsigned int,
12683 vector unsigned int);
12684 vector unsigned short vec_packsu (vector signed int, vector signed int);
12686 vector unsigned short vec_vpkswus (vector signed int,
12687 vector signed int);
12689 vector unsigned char vec_vpkshus (vector signed short,
12690 vector signed short);
12692 vector float vec_perm (vector float,
12694 vector unsigned char);
12695 vector signed int vec_perm (vector signed int,
12697 vector unsigned char);
12698 vector unsigned int vec_perm (vector unsigned int,
12699 vector unsigned int,
12700 vector unsigned char);
12701 vector bool int vec_perm (vector bool int,
12703 vector unsigned char);
12704 vector signed short vec_perm (vector signed short,
12705 vector signed short,
12706 vector unsigned char);
12707 vector unsigned short vec_perm (vector unsigned short,
12708 vector unsigned short,
12709 vector unsigned char);
12710 vector bool short vec_perm (vector bool short,
12712 vector unsigned char);
12713 vector pixel vec_perm (vector pixel,
12715 vector unsigned char);
12716 vector signed char vec_perm (vector signed char,
12717 vector signed char,
12718 vector unsigned char);
12719 vector unsigned char vec_perm (vector unsigned char,
12720 vector unsigned char,
12721 vector unsigned char);
12722 vector bool char vec_perm (vector bool char,
12724 vector unsigned char);
12726 vector float vec_re (vector float);
12728 vector signed char vec_rl (vector signed char,
12729 vector unsigned char);
12730 vector unsigned char vec_rl (vector unsigned char,
12731 vector unsigned char);
12732 vector signed short vec_rl (vector signed short, vector unsigned short);
12733 vector unsigned short vec_rl (vector unsigned short,
12734 vector unsigned short);
12735 vector signed int vec_rl (vector signed int, vector unsigned int);
12736 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
12738 vector signed int vec_vrlw (vector signed int, vector unsigned int);
12739 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
12741 vector signed short vec_vrlh (vector signed short,
12742 vector unsigned short);
12743 vector unsigned short vec_vrlh (vector unsigned short,
12744 vector unsigned short);
12746 vector signed char vec_vrlb (vector signed char, vector unsigned char);
12747 vector unsigned char vec_vrlb (vector unsigned char,
12748 vector unsigned char);
12750 vector float vec_round (vector float);
12752 vector float vec_recip (vector float, vector float);
12754 vector float vec_rsqrt (vector float);
12756 vector float vec_rsqrte (vector float);
12758 vector float vec_sel (vector float, vector float, vector bool int);
12759 vector float vec_sel (vector float, vector float, vector unsigned int);
12760 vector signed int vec_sel (vector signed int,
12763 vector signed int vec_sel (vector signed int,
12765 vector unsigned int);
12766 vector unsigned int vec_sel (vector unsigned int,
12767 vector unsigned int,
12769 vector unsigned int vec_sel (vector unsigned int,
12770 vector unsigned int,
12771 vector unsigned int);
12772 vector bool int vec_sel (vector bool int,
12775 vector bool int vec_sel (vector bool int,
12777 vector unsigned int);
12778 vector signed short vec_sel (vector signed short,
12779 vector signed short,
12780 vector bool short);
12781 vector signed short vec_sel (vector signed short,
12782 vector signed short,
12783 vector unsigned short);
12784 vector unsigned short vec_sel (vector unsigned short,
12785 vector unsigned short,
12786 vector bool short);
12787 vector unsigned short vec_sel (vector unsigned short,
12788 vector unsigned short,
12789 vector unsigned short);
12790 vector bool short vec_sel (vector bool short,
12792 vector bool short);
12793 vector bool short vec_sel (vector bool short,
12795 vector unsigned short);
12796 vector signed char vec_sel (vector signed char,
12797 vector signed char,
12799 vector signed char vec_sel (vector signed char,
12800 vector signed char,
12801 vector unsigned char);
12802 vector unsigned char vec_sel (vector unsigned char,
12803 vector unsigned char,
12805 vector unsigned char vec_sel (vector unsigned char,
12806 vector unsigned char,
12807 vector unsigned char);
12808 vector bool char vec_sel (vector bool char,
12811 vector bool char vec_sel (vector bool char,
12813 vector unsigned char);
12815 vector signed char vec_sl (vector signed char,
12816 vector unsigned char);
12817 vector unsigned char vec_sl (vector unsigned char,
12818 vector unsigned char);
12819 vector signed short vec_sl (vector signed short, vector unsigned short);
12820 vector unsigned short vec_sl (vector unsigned short,
12821 vector unsigned short);
12822 vector signed int vec_sl (vector signed int, vector unsigned int);
12823 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
12825 vector signed int vec_vslw (vector signed int, vector unsigned int);
12826 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
12828 vector signed short vec_vslh (vector signed short,
12829 vector unsigned short);
12830 vector unsigned short vec_vslh (vector unsigned short,
12831 vector unsigned short);
12833 vector signed char vec_vslb (vector signed char, vector unsigned char);
12834 vector unsigned char vec_vslb (vector unsigned char,
12835 vector unsigned char);
12837 vector float vec_sld (vector float, vector float, const int);
12838 vector signed int vec_sld (vector signed int,
12841 vector unsigned int vec_sld (vector unsigned int,
12842 vector unsigned int,
12844 vector bool int vec_sld (vector bool int,
12847 vector signed short vec_sld (vector signed short,
12848 vector signed short,
12850 vector unsigned short vec_sld (vector unsigned short,
12851 vector unsigned short,
12853 vector bool short vec_sld (vector bool short,
12856 vector pixel vec_sld (vector pixel,
12859 vector signed char vec_sld (vector signed char,
12860 vector signed char,
12862 vector unsigned char vec_sld (vector unsigned char,
12863 vector unsigned char,
12865 vector bool char vec_sld (vector bool char,
12869 vector signed int vec_sll (vector signed int,
12870 vector unsigned int);
12871 vector signed int vec_sll (vector signed int,
12872 vector unsigned short);
12873 vector signed int vec_sll (vector signed int,
12874 vector unsigned char);
12875 vector unsigned int vec_sll (vector unsigned int,
12876 vector unsigned int);
12877 vector unsigned int vec_sll (vector unsigned int,
12878 vector unsigned short);
12879 vector unsigned int vec_sll (vector unsigned int,
12880 vector unsigned char);
12881 vector bool int vec_sll (vector bool int,
12882 vector unsigned int);
12883 vector bool int vec_sll (vector bool int,
12884 vector unsigned short);
12885 vector bool int vec_sll (vector bool int,
12886 vector unsigned char);
12887 vector signed short vec_sll (vector signed short,
12888 vector unsigned int);
12889 vector signed short vec_sll (vector signed short,
12890 vector unsigned short);
12891 vector signed short vec_sll (vector signed short,
12892 vector unsigned char);
12893 vector unsigned short vec_sll (vector unsigned short,
12894 vector unsigned int);
12895 vector unsigned short vec_sll (vector unsigned short,
12896 vector unsigned short);
12897 vector unsigned short vec_sll (vector unsigned short,
12898 vector unsigned char);
12899 vector bool short vec_sll (vector bool short, vector unsigned int);
12900 vector bool short vec_sll (vector bool short, vector unsigned short);
12901 vector bool short vec_sll (vector bool short, vector unsigned char);
12902 vector pixel vec_sll (vector pixel, vector unsigned int);
12903 vector pixel vec_sll (vector pixel, vector unsigned short);
12904 vector pixel vec_sll (vector pixel, vector unsigned char);
12905 vector signed char vec_sll (vector signed char, vector unsigned int);
12906 vector signed char vec_sll (vector signed char, vector unsigned short);
12907 vector signed char vec_sll (vector signed char, vector unsigned char);
12908 vector unsigned char vec_sll (vector unsigned char,
12909 vector unsigned int);
12910 vector unsigned char vec_sll (vector unsigned char,
12911 vector unsigned short);
12912 vector unsigned char vec_sll (vector unsigned char,
12913 vector unsigned char);
12914 vector bool char vec_sll (vector bool char, vector unsigned int);
12915 vector bool char vec_sll (vector bool char, vector unsigned short);
12916 vector bool char vec_sll (vector bool char, vector unsigned char);
12918 vector float vec_slo (vector float, vector signed char);
12919 vector float vec_slo (vector float, vector unsigned char);
12920 vector signed int vec_slo (vector signed int, vector signed char);
12921 vector signed int vec_slo (vector signed int, vector unsigned char);
12922 vector unsigned int vec_slo (vector unsigned int, vector signed char);
12923 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
12924 vector signed short vec_slo (vector signed short, vector signed char);
12925 vector signed short vec_slo (vector signed short, vector unsigned char);
12926 vector unsigned short vec_slo (vector unsigned short,
12927 vector signed char);
12928 vector unsigned short vec_slo (vector unsigned short,
12929 vector unsigned char);
12930 vector pixel vec_slo (vector pixel, vector signed char);
12931 vector pixel vec_slo (vector pixel, vector unsigned char);
12932 vector signed char vec_slo (vector signed char, vector signed char);
12933 vector signed char vec_slo (vector signed char, vector unsigned char);
12934 vector unsigned char vec_slo (vector unsigned char, vector signed char);
12935 vector unsigned char vec_slo (vector unsigned char,
12936 vector unsigned char);
12938 vector signed char vec_splat (vector signed char, const int);
12939 vector unsigned char vec_splat (vector unsigned char, const int);
12940 vector bool char vec_splat (vector bool char, const int);
12941 vector signed short vec_splat (vector signed short, const int);
12942 vector unsigned short vec_splat (vector unsigned short, const int);
12943 vector bool short vec_splat (vector bool short, const int);
12944 vector pixel vec_splat (vector pixel, const int);
12945 vector float vec_splat (vector float, const int);
12946 vector signed int vec_splat (vector signed int, const int);
12947 vector unsigned int vec_splat (vector unsigned int, const int);
12948 vector bool int vec_splat (vector bool int, const int);
12950 vector float vec_vspltw (vector float, const int);
12951 vector signed int vec_vspltw (vector signed int, const int);
12952 vector unsigned int vec_vspltw (vector unsigned int, const int);
12953 vector bool int vec_vspltw (vector bool int, const int);
12955 vector bool short vec_vsplth (vector bool short, const int);
12956 vector signed short vec_vsplth (vector signed short, const int);
12957 vector unsigned short vec_vsplth (vector unsigned short, const int);
12958 vector pixel vec_vsplth (vector pixel, const int);
12960 vector signed char vec_vspltb (vector signed char, const int);
12961 vector unsigned char vec_vspltb (vector unsigned char, const int);
12962 vector bool char vec_vspltb (vector bool char, const int);
12964 vector signed char vec_splat_s8 (const int);
12966 vector signed short vec_splat_s16 (const int);
12968 vector signed int vec_splat_s32 (const int);
12970 vector unsigned char vec_splat_u8 (const int);
12972 vector unsigned short vec_splat_u16 (const int);
12974 vector unsigned int vec_splat_u32 (const int);
12976 vector signed char vec_sr (vector signed char, vector unsigned char);
12977 vector unsigned char vec_sr (vector unsigned char,
12978 vector unsigned char);
12979 vector signed short vec_sr (vector signed short,
12980 vector unsigned short);
12981 vector unsigned short vec_sr (vector unsigned short,
12982 vector unsigned short);
12983 vector signed int vec_sr (vector signed int, vector unsigned int);
12984 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
12986 vector signed int vec_vsrw (vector signed int, vector unsigned int);
12987 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
12989 vector signed short vec_vsrh (vector signed short,
12990 vector unsigned short);
12991 vector unsigned short vec_vsrh (vector unsigned short,
12992 vector unsigned short);
12994 vector signed char vec_vsrb (vector signed char, vector unsigned char);
12995 vector unsigned char vec_vsrb (vector unsigned char,
12996 vector unsigned char);
12998 vector signed char vec_sra (vector signed char, vector unsigned char);
12999 vector unsigned char vec_sra (vector unsigned char,
13000 vector unsigned char);
13001 vector signed short vec_sra (vector signed short,
13002 vector unsigned short);
13003 vector unsigned short vec_sra (vector unsigned short,
13004 vector unsigned short);
13005 vector signed int vec_sra (vector signed int, vector unsigned int);
13006 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
13008 vector signed int vec_vsraw (vector signed int, vector unsigned int);
13009 vector unsigned int vec_vsraw (vector unsigned int,
13010 vector unsigned int);
13012 vector signed short vec_vsrah (vector signed short,
13013 vector unsigned short);
13014 vector unsigned short vec_vsrah (vector unsigned short,
13015 vector unsigned short);
13017 vector signed char vec_vsrab (vector signed char, vector unsigned char);
13018 vector unsigned char vec_vsrab (vector unsigned char,
13019 vector unsigned char);
13021 vector signed int vec_srl (vector signed int, vector unsigned int);
13022 vector signed int vec_srl (vector signed int, vector unsigned short);
13023 vector signed int vec_srl (vector signed int, vector unsigned char);
13024 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
13025 vector unsigned int vec_srl (vector unsigned int,
13026 vector unsigned short);
13027 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
13028 vector bool int vec_srl (vector bool int, vector unsigned int);
13029 vector bool int vec_srl (vector bool int, vector unsigned short);
13030 vector bool int vec_srl (vector bool int, vector unsigned char);
13031 vector signed short vec_srl (vector signed short, vector unsigned int);
13032 vector signed short vec_srl (vector signed short,
13033 vector unsigned short);
13034 vector signed short vec_srl (vector signed short, vector unsigned char);
13035 vector unsigned short vec_srl (vector unsigned short,
13036 vector unsigned int);
13037 vector unsigned short vec_srl (vector unsigned short,
13038 vector unsigned short);
13039 vector unsigned short vec_srl (vector unsigned short,
13040 vector unsigned char);
13041 vector bool short vec_srl (vector bool short, vector unsigned int);
13042 vector bool short vec_srl (vector bool short, vector unsigned short);
13043 vector bool short vec_srl (vector bool short, vector unsigned char);
13044 vector pixel vec_srl (vector pixel, vector unsigned int);
13045 vector pixel vec_srl (vector pixel, vector unsigned short);
13046 vector pixel vec_srl (vector pixel, vector unsigned char);
13047 vector signed char vec_srl (vector signed char, vector unsigned int);
13048 vector signed char vec_srl (vector signed char, vector unsigned short);
13049 vector signed char vec_srl (vector signed char, vector unsigned char);
13050 vector unsigned char vec_srl (vector unsigned char,
13051 vector unsigned int);
13052 vector unsigned char vec_srl (vector unsigned char,
13053 vector unsigned short);
13054 vector unsigned char vec_srl (vector unsigned char,
13055 vector unsigned char);
13056 vector bool char vec_srl (vector bool char, vector unsigned int);
13057 vector bool char vec_srl (vector bool char, vector unsigned short);
13058 vector bool char vec_srl (vector bool char, vector unsigned char);
13060 vector float vec_sro (vector float, vector signed char);
13061 vector float vec_sro (vector float, vector unsigned char);
13062 vector signed int vec_sro (vector signed int, vector signed char);
13063 vector signed int vec_sro (vector signed int, vector unsigned char);
13064 vector unsigned int vec_sro (vector unsigned int, vector signed char);
13065 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
13066 vector signed short vec_sro (vector signed short, vector signed char);
13067 vector signed short vec_sro (vector signed short, vector unsigned char);
13068 vector unsigned short vec_sro (vector unsigned short,
13069 vector signed char);
13070 vector unsigned short vec_sro (vector unsigned short,
13071 vector unsigned char);
13072 vector pixel vec_sro (vector pixel, vector signed char);
13073 vector pixel vec_sro (vector pixel, vector unsigned char);
13074 vector signed char vec_sro (vector signed char, vector signed char);
13075 vector signed char vec_sro (vector signed char, vector unsigned char);
13076 vector unsigned char vec_sro (vector unsigned char, vector signed char);
13077 vector unsigned char vec_sro (vector unsigned char,
13078 vector unsigned char);
13080 void vec_st (vector float, int, vector float *);
13081 void vec_st (vector float, int, float *);
13082 void vec_st (vector signed int, int, vector signed int *);
13083 void vec_st (vector signed int, int, int *);
13084 void vec_st (vector unsigned int, int, vector unsigned int *);
13085 void vec_st (vector unsigned int, int, unsigned int *);
13086 void vec_st (vector bool int, int, vector bool int *);
13087 void vec_st (vector bool int, int, unsigned int *);
13088 void vec_st (vector bool int, int, int *);
13089 void vec_st (vector signed short, int, vector signed short *);
13090 void vec_st (vector signed short, int, short *);
13091 void vec_st (vector unsigned short, int, vector unsigned short *);
13092 void vec_st (vector unsigned short, int, unsigned short *);
13093 void vec_st (vector bool short, int, vector bool short *);
13094 void vec_st (vector bool short, int, unsigned short *);
13095 void vec_st (vector pixel, int, vector pixel *);
13096 void vec_st (vector pixel, int, unsigned short *);
13097 void vec_st (vector pixel, int, short *);
13098 void vec_st (vector bool short, int, short *);
13099 void vec_st (vector signed char, int, vector signed char *);
13100 void vec_st (vector signed char, int, signed char *);
13101 void vec_st (vector unsigned char, int, vector unsigned char *);
13102 void vec_st (vector unsigned char, int, unsigned char *);
13103 void vec_st (vector bool char, int, vector bool char *);
13104 void vec_st (vector bool char, int, unsigned char *);
13105 void vec_st (vector bool char, int, signed char *);
13107 void vec_ste (vector signed char, int, signed char *);
13108 void vec_ste (vector unsigned char, int, unsigned char *);
13109 void vec_ste (vector bool char, int, signed char *);
13110 void vec_ste (vector bool char, int, unsigned char *);
13111 void vec_ste (vector signed short, int, short *);
13112 void vec_ste (vector unsigned short, int, unsigned short *);
13113 void vec_ste (vector bool short, int, short *);
13114 void vec_ste (vector bool short, int, unsigned short *);
13115 void vec_ste (vector pixel, int, short *);
13116 void vec_ste (vector pixel, int, unsigned short *);
13117 void vec_ste (vector float, int, float *);
13118 void vec_ste (vector signed int, int, int *);
13119 void vec_ste (vector unsigned int, int, unsigned int *);
13120 void vec_ste (vector bool int, int, int *);
13121 void vec_ste (vector bool int, int, unsigned int *);
13123 void vec_stvewx (vector float, int, float *);
13124 void vec_stvewx (vector signed int, int, int *);
13125 void vec_stvewx (vector unsigned int, int, unsigned int *);
13126 void vec_stvewx (vector bool int, int, int *);
13127 void vec_stvewx (vector bool int, int, unsigned int *);
13129 void vec_stvehx (vector signed short, int, short *);
13130 void vec_stvehx (vector unsigned short, int, unsigned short *);
13131 void vec_stvehx (vector bool short, int, short *);
13132 void vec_stvehx (vector bool short, int, unsigned short *);
13133 void vec_stvehx (vector pixel, int, short *);
13134 void vec_stvehx (vector pixel, int, unsigned short *);
13136 void vec_stvebx (vector signed char, int, signed char *);
13137 void vec_stvebx (vector unsigned char, int, unsigned char *);
13138 void vec_stvebx (vector bool char, int, signed char *);
13139 void vec_stvebx (vector bool char, int, unsigned char *);
13141 void vec_stl (vector float, int, vector float *);
13142 void vec_stl (vector float, int, float *);
13143 void vec_stl (vector signed int, int, vector signed int *);
13144 void vec_stl (vector signed int, int, int *);
13145 void vec_stl (vector unsigned int, int, vector unsigned int *);
13146 void vec_stl (vector unsigned int, int, unsigned int *);
13147 void vec_stl (vector bool int, int, vector bool int *);
13148 void vec_stl (vector bool int, int, unsigned int *);
13149 void vec_stl (vector bool int, int, int *);
13150 void vec_stl (vector signed short, int, vector signed short *);
13151 void vec_stl (vector signed short, int, short *);
13152 void vec_stl (vector unsigned short, int, vector unsigned short *);
13153 void vec_stl (vector unsigned short, int, unsigned short *);
13154 void vec_stl (vector bool short, int, vector bool short *);
13155 void vec_stl (vector bool short, int, unsigned short *);
13156 void vec_stl (vector bool short, int, short *);
13157 void vec_stl (vector pixel, int, vector pixel *);
13158 void vec_stl (vector pixel, int, unsigned short *);
13159 void vec_stl (vector pixel, int, short *);
13160 void vec_stl (vector signed char, int, vector signed char *);
13161 void vec_stl (vector signed char, int, signed char *);
13162 void vec_stl (vector unsigned char, int, vector unsigned char *);
13163 void vec_stl (vector unsigned char, int, unsigned char *);
13164 void vec_stl (vector bool char, int, vector bool char *);
13165 void vec_stl (vector bool char, int, unsigned char *);
13166 void vec_stl (vector bool char, int, signed char *);
13168 vector signed char vec_sub (vector bool char, vector signed char);
13169 vector signed char vec_sub (vector signed char, vector bool char);
13170 vector signed char vec_sub (vector signed char, vector signed char);
13171 vector unsigned char vec_sub (vector bool char, vector unsigned char);
13172 vector unsigned char vec_sub (vector unsigned char, vector bool char);
13173 vector unsigned char vec_sub (vector unsigned char,
13174 vector unsigned char);
13175 vector signed short vec_sub (vector bool short, vector signed short);
13176 vector signed short vec_sub (vector signed short, vector bool short);
13177 vector signed short vec_sub (vector signed short, vector signed short);
13178 vector unsigned short vec_sub (vector bool short,
13179 vector unsigned short);
13180 vector unsigned short vec_sub (vector unsigned short,
13181 vector bool short);
13182 vector unsigned short vec_sub (vector unsigned short,
13183 vector unsigned short);
13184 vector signed int vec_sub (vector bool int, vector signed int);
13185 vector signed int vec_sub (vector signed int, vector bool int);
13186 vector signed int vec_sub (vector signed int, vector signed int);
13187 vector unsigned int vec_sub (vector bool int, vector unsigned int);
13188 vector unsigned int vec_sub (vector unsigned int, vector bool int);
13189 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
13190 vector float vec_sub (vector float, vector float);
13192 vector float vec_vsubfp (vector float, vector float);
13194 vector signed int vec_vsubuwm (vector bool int, vector signed int);
13195 vector signed int vec_vsubuwm (vector signed int, vector bool int);
13196 vector signed int vec_vsubuwm (vector signed int, vector signed int);
13197 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
13198 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
13199 vector unsigned int vec_vsubuwm (vector unsigned int,
13200 vector unsigned int);
13202 vector signed short vec_vsubuhm (vector bool short,
13203 vector signed short);
13204 vector signed short vec_vsubuhm (vector signed short,
13205 vector bool short);
13206 vector signed short vec_vsubuhm (vector signed short,
13207 vector signed short);
13208 vector unsigned short vec_vsubuhm (vector bool short,
13209 vector unsigned short);
13210 vector unsigned short vec_vsubuhm (vector unsigned short,
13211 vector bool short);
13212 vector unsigned short vec_vsubuhm (vector unsigned short,
13213 vector unsigned short);
13215 vector signed char vec_vsububm (vector bool char, vector signed char);
13216 vector signed char vec_vsububm (vector signed char, vector bool char);
13217 vector signed char vec_vsububm (vector signed char, vector signed char);
13218 vector unsigned char vec_vsububm (vector bool char,
13219 vector unsigned char);
13220 vector unsigned char vec_vsububm (vector unsigned char,
13222 vector unsigned char vec_vsububm (vector unsigned char,
13223 vector unsigned char);
13225 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
13227 vector unsigned char vec_subs (vector bool char, vector unsigned char);
13228 vector unsigned char vec_subs (vector unsigned char, vector bool char);
13229 vector unsigned char vec_subs (vector unsigned char,
13230 vector unsigned char);
13231 vector signed char vec_subs (vector bool char, vector signed char);
13232 vector signed char vec_subs (vector signed char, vector bool char);
13233 vector signed char vec_subs (vector signed char, vector signed char);
13234 vector unsigned short vec_subs (vector bool short,
13235 vector unsigned short);
13236 vector unsigned short vec_subs (vector unsigned short,
13237 vector bool short);
13238 vector unsigned short vec_subs (vector unsigned short,
13239 vector unsigned short);
13240 vector signed short vec_subs (vector bool short, vector signed short);
13241 vector signed short vec_subs (vector signed short, vector bool short);
13242 vector signed short vec_subs (vector signed short, vector signed short);
13243 vector unsigned int vec_subs (vector bool int, vector unsigned int);
13244 vector unsigned int vec_subs (vector unsigned int, vector bool int);
13245 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
13246 vector signed int vec_subs (vector bool int, vector signed int);
13247 vector signed int vec_subs (vector signed int, vector bool int);
13248 vector signed int vec_subs (vector signed int, vector signed int);
13250 vector signed int vec_vsubsws (vector bool int, vector signed int);
13251 vector signed int vec_vsubsws (vector signed int, vector bool int);
13252 vector signed int vec_vsubsws (vector signed int, vector signed int);
13254 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
13255 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
13256 vector unsigned int vec_vsubuws (vector unsigned int,
13257 vector unsigned int);
13259 vector signed short vec_vsubshs (vector bool short,
13260 vector signed short);
13261 vector signed short vec_vsubshs (vector signed short,
13262 vector bool short);
13263 vector signed short vec_vsubshs (vector signed short,
13264 vector signed short);
13266 vector unsigned short vec_vsubuhs (vector bool short,
13267 vector unsigned short);
13268 vector unsigned short vec_vsubuhs (vector unsigned short,
13269 vector bool short);
13270 vector unsigned short vec_vsubuhs (vector unsigned short,
13271 vector unsigned short);
13273 vector signed char vec_vsubsbs (vector bool char, vector signed char);
13274 vector signed char vec_vsubsbs (vector signed char, vector bool char);
13275 vector signed char vec_vsubsbs (vector signed char, vector signed char);
13277 vector unsigned char vec_vsububs (vector bool char,
13278 vector unsigned char);
13279 vector unsigned char vec_vsububs (vector unsigned char,
13281 vector unsigned char vec_vsububs (vector unsigned char,
13282 vector unsigned char);
13284 vector unsigned int vec_sum4s (vector unsigned char,
13285 vector unsigned int);
13286 vector signed int vec_sum4s (vector signed char, vector signed int);
13287 vector signed int vec_sum4s (vector signed short, vector signed int);
13289 vector signed int vec_vsum4shs (vector signed short, vector signed int);
13291 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
13293 vector unsigned int vec_vsum4ubs (vector unsigned char,
13294 vector unsigned int);
13296 vector signed int vec_sum2s (vector signed int, vector signed int);
13298 vector signed int vec_sums (vector signed int, vector signed int);
13300 vector float vec_trunc (vector float);
13302 vector signed short vec_unpackh (vector signed char);
13303 vector bool short vec_unpackh (vector bool char);
13304 vector signed int vec_unpackh (vector signed short);
13305 vector bool int vec_unpackh (vector bool short);
13306 vector unsigned int vec_unpackh (vector pixel);
13308 vector bool int vec_vupkhsh (vector bool short);
13309 vector signed int vec_vupkhsh (vector signed short);
13311 vector unsigned int vec_vupkhpx (vector pixel);
13313 vector bool short vec_vupkhsb (vector bool char);
13314 vector signed short vec_vupkhsb (vector signed char);
13316 vector signed short vec_unpackl (vector signed char);
13317 vector bool short vec_unpackl (vector bool char);
13318 vector unsigned int vec_unpackl (vector pixel);
13319 vector signed int vec_unpackl (vector signed short);
13320 vector bool int vec_unpackl (vector bool short);
13322 vector unsigned int vec_vupklpx (vector pixel);
13324 vector bool int vec_vupklsh (vector bool short);
13325 vector signed int vec_vupklsh (vector signed short);
13327 vector bool short vec_vupklsb (vector bool char);
13328 vector signed short vec_vupklsb (vector signed char);
13330 vector float vec_xor (vector float, vector float);
13331 vector float vec_xor (vector float, vector bool int);
13332 vector float vec_xor (vector bool int, vector float);
13333 vector bool int vec_xor (vector bool int, vector bool int);
13334 vector signed int vec_xor (vector bool int, vector signed int);
13335 vector signed int vec_xor (vector signed int, vector bool int);
13336 vector signed int vec_xor (vector signed int, vector signed int);
13337 vector unsigned int vec_xor (vector bool int, vector unsigned int);
13338 vector unsigned int vec_xor (vector unsigned int, vector bool int);
13339 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
13340 vector bool short vec_xor (vector bool short, vector bool short);
13341 vector signed short vec_xor (vector bool short, vector signed short);
13342 vector signed short vec_xor (vector signed short, vector bool short);
13343 vector signed short vec_xor (vector signed short, vector signed short);
13344 vector unsigned short vec_xor (vector bool short,
13345 vector unsigned short);
13346 vector unsigned short vec_xor (vector unsigned short,
13347 vector bool short);
13348 vector unsigned short vec_xor (vector unsigned short,
13349 vector unsigned short);
13350 vector signed char vec_xor (vector bool char, vector signed char);
13351 vector bool char vec_xor (vector bool char, vector bool char);
13352 vector signed char vec_xor (vector signed char, vector bool char);
13353 vector signed char vec_xor (vector signed char, vector signed char);
13354 vector unsigned char vec_xor (vector bool char, vector unsigned char);
13355 vector unsigned char vec_xor (vector unsigned char, vector bool char);
13356 vector unsigned char vec_xor (vector unsigned char,
13357 vector unsigned char);
13359 int vec_all_eq (vector signed char, vector bool char);
13360 int vec_all_eq (vector signed char, vector signed char);
13361 int vec_all_eq (vector unsigned char, vector bool char);
13362 int vec_all_eq (vector unsigned char, vector unsigned char);
13363 int vec_all_eq (vector bool char, vector bool char);
13364 int vec_all_eq (vector bool char, vector unsigned char);
13365 int vec_all_eq (vector bool char, vector signed char);
13366 int vec_all_eq (vector signed short, vector bool short);
13367 int vec_all_eq (vector signed short, vector signed short);
13368 int vec_all_eq (vector unsigned short, vector bool short);
13369 int vec_all_eq (vector unsigned short, vector unsigned short);
13370 int vec_all_eq (vector bool short, vector bool short);
13371 int vec_all_eq (vector bool short, vector unsigned short);
13372 int vec_all_eq (vector bool short, vector signed short);
13373 int vec_all_eq (vector pixel, vector pixel);
13374 int vec_all_eq (vector signed int, vector bool int);
13375 int vec_all_eq (vector signed int, vector signed int);
13376 int vec_all_eq (vector unsigned int, vector bool int);
13377 int vec_all_eq (vector unsigned int, vector unsigned int);
13378 int vec_all_eq (vector bool int, vector bool int);
13379 int vec_all_eq (vector bool int, vector unsigned int);
13380 int vec_all_eq (vector bool int, vector signed int);
13381 int vec_all_eq (vector float, vector float);
13383 int vec_all_ge (vector bool char, vector unsigned char);
13384 int vec_all_ge (vector unsigned char, vector bool char);
13385 int vec_all_ge (vector unsigned char, vector unsigned char);
13386 int vec_all_ge (vector bool char, vector signed char);
13387 int vec_all_ge (vector signed char, vector bool char);
13388 int vec_all_ge (vector signed char, vector signed char);
13389 int vec_all_ge (vector bool short, vector unsigned short);
13390 int vec_all_ge (vector unsigned short, vector bool short);
13391 int vec_all_ge (vector unsigned short, vector unsigned short);
13392 int vec_all_ge (vector signed short, vector signed short);
13393 int vec_all_ge (vector bool short, vector signed short);
13394 int vec_all_ge (vector signed short, vector bool short);
13395 int vec_all_ge (vector bool int, vector unsigned int);
13396 int vec_all_ge (vector unsigned int, vector bool int);
13397 int vec_all_ge (vector unsigned int, vector unsigned int);
13398 int vec_all_ge (vector bool int, vector signed int);
13399 int vec_all_ge (vector signed int, vector bool int);
13400 int vec_all_ge (vector signed int, vector signed int);
13401 int vec_all_ge (vector float, vector float);
13403 int vec_all_gt (vector bool char, vector unsigned char);
13404 int vec_all_gt (vector unsigned char, vector bool char);
13405 int vec_all_gt (vector unsigned char, vector unsigned char);
13406 int vec_all_gt (vector bool char, vector signed char);
13407 int vec_all_gt (vector signed char, vector bool char);
13408 int vec_all_gt (vector signed char, vector signed char);
13409 int vec_all_gt (vector bool short, vector unsigned short);
13410 int vec_all_gt (vector unsigned short, vector bool short);
13411 int vec_all_gt (vector unsigned short, vector unsigned short);
13412 int vec_all_gt (vector bool short, vector signed short);
13413 int vec_all_gt (vector signed short, vector bool short);
13414 int vec_all_gt (vector signed short, vector signed short);
13415 int vec_all_gt (vector bool int, vector unsigned int);
13416 int vec_all_gt (vector unsigned int, vector bool int);
13417 int vec_all_gt (vector unsigned int, vector unsigned int);
13418 int vec_all_gt (vector bool int, vector signed int);
13419 int vec_all_gt (vector signed int, vector bool int);
13420 int vec_all_gt (vector signed int, vector signed int);
13421 int vec_all_gt (vector float, vector float);
13423 int vec_all_in (vector float, vector float);
13425 int vec_all_le (vector bool char, vector unsigned char);
13426 int vec_all_le (vector unsigned char, vector bool char);
13427 int vec_all_le (vector unsigned char, vector unsigned char);
13428 int vec_all_le (vector bool char, vector signed char);
13429 int vec_all_le (vector signed char, vector bool char);
13430 int vec_all_le (vector signed char, vector signed char);
13431 int vec_all_le (vector bool short, vector unsigned short);
13432 int vec_all_le (vector unsigned short, vector bool short);
13433 int vec_all_le (vector unsigned short, vector unsigned short);
13434 int vec_all_le (vector bool short, vector signed short);
13435 int vec_all_le (vector signed short, vector bool short);
13436 int vec_all_le (vector signed short, vector signed short);
13437 int vec_all_le (vector bool int, vector unsigned int);
13438 int vec_all_le (vector unsigned int, vector bool int);
13439 int vec_all_le (vector unsigned int, vector unsigned int);
13440 int vec_all_le (vector bool int, vector signed int);
13441 int vec_all_le (vector signed int, vector bool int);
13442 int vec_all_le (vector signed int, vector signed int);
13443 int vec_all_le (vector float, vector float);
13445 int vec_all_lt (vector bool char, vector unsigned char);
13446 int vec_all_lt (vector unsigned char, vector bool char);
13447 int vec_all_lt (vector unsigned char, vector unsigned char);
13448 int vec_all_lt (vector bool char, vector signed char);
13449 int vec_all_lt (vector signed char, vector bool char);
13450 int vec_all_lt (vector signed char, vector signed char);
13451 int vec_all_lt (vector bool short, vector unsigned short);
13452 int vec_all_lt (vector unsigned short, vector bool short);
13453 int vec_all_lt (vector unsigned short, vector unsigned short);
13454 int vec_all_lt (vector bool short, vector signed short);
13455 int vec_all_lt (vector signed short, vector bool short);
13456 int vec_all_lt (vector signed short, vector signed short);
13457 int vec_all_lt (vector bool int, vector unsigned int);
13458 int vec_all_lt (vector unsigned int, vector bool int);
13459 int vec_all_lt (vector unsigned int, vector unsigned int);
13460 int vec_all_lt (vector bool int, vector signed int);
13461 int vec_all_lt (vector signed int, vector bool int);
13462 int vec_all_lt (vector signed int, vector signed int);
13463 int vec_all_lt (vector float, vector float);
13465 int vec_all_nan (vector float);
13467 int vec_all_ne (vector signed char, vector bool char);
13468 int vec_all_ne (vector signed char, vector signed char);
13469 int vec_all_ne (vector unsigned char, vector bool char);
13470 int vec_all_ne (vector unsigned char, vector unsigned char);
13471 int vec_all_ne (vector bool char, vector bool char);
13472 int vec_all_ne (vector bool char, vector unsigned char);
13473 int vec_all_ne (vector bool char, vector signed char);
13474 int vec_all_ne (vector signed short, vector bool short);
13475 int vec_all_ne (vector signed short, vector signed short);
13476 int vec_all_ne (vector unsigned short, vector bool short);
13477 int vec_all_ne (vector unsigned short, vector unsigned short);
13478 int vec_all_ne (vector bool short, vector bool short);
13479 int vec_all_ne (vector bool short, vector unsigned short);
13480 int vec_all_ne (vector bool short, vector signed short);
13481 int vec_all_ne (vector pixel, vector pixel);
13482 int vec_all_ne (vector signed int, vector bool int);
13483 int vec_all_ne (vector signed int, vector signed int);
13484 int vec_all_ne (vector unsigned int, vector bool int);
13485 int vec_all_ne (vector unsigned int, vector unsigned int);
13486 int vec_all_ne (vector bool int, vector bool int);
13487 int vec_all_ne (vector bool int, vector unsigned int);
13488 int vec_all_ne (vector bool int, vector signed int);
13489 int vec_all_ne (vector float, vector float);
13491 int vec_all_nge (vector float, vector float);
13493 int vec_all_ngt (vector float, vector float);
13495 int vec_all_nle (vector float, vector float);
13497 int vec_all_nlt (vector float, vector float);
13499 int vec_all_numeric (vector float);
13501 int vec_any_eq (vector signed char, vector bool char);
13502 int vec_any_eq (vector signed char, vector signed char);
13503 int vec_any_eq (vector unsigned char, vector bool char);
13504 int vec_any_eq (vector unsigned char, vector unsigned char);
13505 int vec_any_eq (vector bool char, vector bool char);
13506 int vec_any_eq (vector bool char, vector unsigned char);
13507 int vec_any_eq (vector bool char, vector signed char);
13508 int vec_any_eq (vector signed short, vector bool short);
13509 int vec_any_eq (vector signed short, vector signed short);
13510 int vec_any_eq (vector unsigned short, vector bool short);
13511 int vec_any_eq (vector unsigned short, vector unsigned short);
13512 int vec_any_eq (vector bool short, vector bool short);
13513 int vec_any_eq (vector bool short, vector unsigned short);
13514 int vec_any_eq (vector bool short, vector signed short);
13515 int vec_any_eq (vector pixel, vector pixel);
13516 int vec_any_eq (vector signed int, vector bool int);
13517 int vec_any_eq (vector signed int, vector signed int);
13518 int vec_any_eq (vector unsigned int, vector bool int);
13519 int vec_any_eq (vector unsigned int, vector unsigned int);
13520 int vec_any_eq (vector bool int, vector bool int);
13521 int vec_any_eq (vector bool int, vector unsigned int);
13522 int vec_any_eq (vector bool int, vector signed int);
13523 int vec_any_eq (vector float, vector float);
13525 int vec_any_ge (vector signed char, vector bool char);
13526 int vec_any_ge (vector unsigned char, vector bool char);
13527 int vec_any_ge (vector unsigned char, vector unsigned char);
13528 int vec_any_ge (vector signed char, vector signed char);
13529 int vec_any_ge (vector bool char, vector unsigned char);
13530 int vec_any_ge (vector bool char, vector signed char);
13531 int vec_any_ge (vector unsigned short, vector bool short);
13532 int vec_any_ge (vector unsigned short, vector unsigned short);
13533 int vec_any_ge (vector signed short, vector signed short);
13534 int vec_any_ge (vector signed short, vector bool short);
13535 int vec_any_ge (vector bool short, vector unsigned short);
13536 int vec_any_ge (vector bool short, vector signed short);
13537 int vec_any_ge (vector signed int, vector bool int);
13538 int vec_any_ge (vector unsigned int, vector bool int);
13539 int vec_any_ge (vector unsigned int, vector unsigned int);
13540 int vec_any_ge (vector signed int, vector signed int);
13541 int vec_any_ge (vector bool int, vector unsigned int);
13542 int vec_any_ge (vector bool int, vector signed int);
13543 int vec_any_ge (vector float, vector float);
13545 int vec_any_gt (vector bool char, vector unsigned char);
13546 int vec_any_gt (vector unsigned char, vector bool char);
13547 int vec_any_gt (vector unsigned char, vector unsigned char);
13548 int vec_any_gt (vector bool char, vector signed char);
13549 int vec_any_gt (vector signed char, vector bool char);
13550 int vec_any_gt (vector signed char, vector signed char);
13551 int vec_any_gt (vector bool short, vector unsigned short);
13552 int vec_any_gt (vector unsigned short, vector bool short);
13553 int vec_any_gt (vector unsigned short, vector unsigned short);
13554 int vec_any_gt (vector bool short, vector signed short);
13555 int vec_any_gt (vector signed short, vector bool short);
13556 int vec_any_gt (vector signed short, vector signed short);
13557 int vec_any_gt (vector bool int, vector unsigned int);
13558 int vec_any_gt (vector unsigned int, vector bool int);
13559 int vec_any_gt (vector unsigned int, vector unsigned int);
13560 int vec_any_gt (vector bool int, vector signed int);
13561 int vec_any_gt (vector signed int, vector bool int);
13562 int vec_any_gt (vector signed int, vector signed int);
13563 int vec_any_gt (vector float, vector float);
13565 int vec_any_le (vector bool char, vector unsigned char);
13566 int vec_any_le (vector unsigned char, vector bool char);
13567 int vec_any_le (vector unsigned char, vector unsigned char);
13568 int vec_any_le (vector bool char, vector signed char);
13569 int vec_any_le (vector signed char, vector bool char);
13570 int vec_any_le (vector signed char, vector signed char);
13571 int vec_any_le (vector bool short, vector unsigned short);
13572 int vec_any_le (vector unsigned short, vector bool short);
13573 int vec_any_le (vector unsigned short, vector unsigned short);
13574 int vec_any_le (vector bool short, vector signed short);
13575 int vec_any_le (vector signed short, vector bool short);
13576 int vec_any_le (vector signed short, vector signed short);
13577 int vec_any_le (vector bool int, vector unsigned int);
13578 int vec_any_le (vector unsigned int, vector bool int);
13579 int vec_any_le (vector unsigned int, vector unsigned int);
13580 int vec_any_le (vector bool int, vector signed int);
13581 int vec_any_le (vector signed int, vector bool int);
13582 int vec_any_le (vector signed int, vector signed int);
13583 int vec_any_le (vector float, vector float);
13585 int vec_any_lt (vector bool char, vector unsigned char);
13586 int vec_any_lt (vector unsigned char, vector bool char);
13587 int vec_any_lt (vector unsigned char, vector unsigned char);
13588 int vec_any_lt (vector bool char, vector signed char);
13589 int vec_any_lt (vector signed char, vector bool char);
13590 int vec_any_lt (vector signed char, vector signed char);
13591 int vec_any_lt (vector bool short, vector unsigned short);
13592 int vec_any_lt (vector unsigned short, vector bool short);
13593 int vec_any_lt (vector unsigned short, vector unsigned short);
13594 int vec_any_lt (vector bool short, vector signed short);
13595 int vec_any_lt (vector signed short, vector bool short);
13596 int vec_any_lt (vector signed short, vector signed short);
13597 int vec_any_lt (vector bool int, vector unsigned int);
13598 int vec_any_lt (vector unsigned int, vector bool int);
13599 int vec_any_lt (vector unsigned int, vector unsigned int);
13600 int vec_any_lt (vector bool int, vector signed int);
13601 int vec_any_lt (vector signed int, vector bool int);
13602 int vec_any_lt (vector signed int, vector signed int);
13603 int vec_any_lt (vector float, vector float);
13605 int vec_any_nan (vector float);
13607 int vec_any_ne (vector signed char, vector bool char);
13608 int vec_any_ne (vector signed char, vector signed char);
13609 int vec_any_ne (vector unsigned char, vector bool char);
13610 int vec_any_ne (vector unsigned char, vector unsigned char);
13611 int vec_any_ne (vector bool char, vector bool char);
13612 int vec_any_ne (vector bool char, vector unsigned char);
13613 int vec_any_ne (vector bool char, vector signed char);
13614 int vec_any_ne (vector signed short, vector bool short);
13615 int vec_any_ne (vector signed short, vector signed short);
13616 int vec_any_ne (vector unsigned short, vector bool short);
13617 int vec_any_ne (vector unsigned short, vector unsigned short);
13618 int vec_any_ne (vector bool short, vector bool short);
13619 int vec_any_ne (vector bool short, vector unsigned short);
13620 int vec_any_ne (vector bool short, vector signed short);
13621 int vec_any_ne (vector pixel, vector pixel);
13622 int vec_any_ne (vector signed int, vector bool int);
13623 int vec_any_ne (vector signed int, vector signed int);
13624 int vec_any_ne (vector unsigned int, vector bool int);
13625 int vec_any_ne (vector unsigned int, vector unsigned int);
13626 int vec_any_ne (vector bool int, vector bool int);
13627 int vec_any_ne (vector bool int, vector unsigned int);
13628 int vec_any_ne (vector bool int, vector signed int);
13629 int vec_any_ne (vector float, vector float);
13631 int vec_any_nge (vector float, vector float);
13633 int vec_any_ngt (vector float, vector float);
13635 int vec_any_nle (vector float, vector float);
13637 int vec_any_nlt (vector float, vector float);
13639 int vec_any_numeric (vector float);
13641 int vec_any_out (vector float, vector float);
13644 If the vector/scalar (VSX) instruction set is available, the following
13645 additional functions are available:
13648 vector double vec_abs (vector double);
13649 vector double vec_add (vector double, vector double);
13650 vector double vec_and (vector double, vector double);
13651 vector double vec_and (vector double, vector bool long);
13652 vector double vec_and (vector bool long, vector double);
13653 vector double vec_andc (vector double, vector double);
13654 vector double vec_andc (vector double, vector bool long);
13655 vector double vec_andc (vector bool long, vector double);
13656 vector double vec_ceil (vector double);
13657 vector bool long vec_cmpeq (vector double, vector double);
13658 vector bool long vec_cmpge (vector double, vector double);
13659 vector bool long vec_cmpgt (vector double, vector double);
13660 vector bool long vec_cmple (vector double, vector double);
13661 vector bool long vec_cmplt (vector double, vector double);
13662 vector float vec_div (vector float, vector float);
13663 vector double vec_div (vector double, vector double);
13664 vector double vec_floor (vector double);
13665 vector double vec_ld (int, const vector double *);
13666 vector double vec_ld (int, const double *);
13667 vector double vec_ldl (int, const vector double *);
13668 vector double vec_ldl (int, const double *);
13669 vector unsigned char vec_lvsl (int, const volatile double *);
13670 vector unsigned char vec_lvsr (int, const volatile double *);
13671 vector double vec_madd (vector double, vector double, vector double);
13672 vector double vec_max (vector double, vector double);
13673 vector double vec_min (vector double, vector double);
13674 vector float vec_msub (vector float, vector float, vector float);
13675 vector double vec_msub (vector double, vector double, vector double);
13676 vector float vec_mul (vector float, vector float);
13677 vector double vec_mul (vector double, vector double);
13678 vector float vec_nearbyint (vector float);
13679 vector double vec_nearbyint (vector double);
13680 vector float vec_nmadd (vector float, vector float, vector float);
13681 vector double vec_nmadd (vector double, vector double, vector double);
13682 vector double vec_nmsub (vector double, vector double, vector double);
13683 vector double vec_nor (vector double, vector double);
13684 vector double vec_or (vector double, vector double);
13685 vector double vec_or (vector double, vector bool long);
13686 vector double vec_or (vector bool long, vector double);
13687 vector double vec_perm (vector double,
13689 vector unsigned char);
13690 vector double vec_rint (vector double);
13691 vector double vec_recip (vector double, vector double);
13692 vector double vec_rsqrt (vector double);
13693 vector double vec_rsqrte (vector double);
13694 vector double vec_sel (vector double, vector double, vector bool long);
13695 vector double vec_sel (vector double, vector double, vector unsigned long);
13696 vector double vec_sub (vector double, vector double);
13697 vector float vec_sqrt (vector float);
13698 vector double vec_sqrt (vector double);
13699 void vec_st (vector double, int, vector double *);
13700 void vec_st (vector double, int, double *);
13701 vector double vec_trunc (vector double);
13702 vector double vec_xor (vector double, vector double);
13703 vector double vec_xor (vector double, vector bool long);
13704 vector double vec_xor (vector bool long, vector double);
13705 int vec_all_eq (vector double, vector double);
13706 int vec_all_ge (vector double, vector double);
13707 int vec_all_gt (vector double, vector double);
13708 int vec_all_le (vector double, vector double);
13709 int vec_all_lt (vector double, vector double);
13710 int vec_all_nan (vector double);
13711 int vec_all_ne (vector double, vector double);
13712 int vec_all_nge (vector double, vector double);
13713 int vec_all_ngt (vector double, vector double);
13714 int vec_all_nle (vector double, vector double);
13715 int vec_all_nlt (vector double, vector double);
13716 int vec_all_numeric (vector double);
13717 int vec_any_eq (vector double, vector double);
13718 int vec_any_ge (vector double, vector double);
13719 int vec_any_gt (vector double, vector double);
13720 int vec_any_le (vector double, vector double);
13721 int vec_any_lt (vector double, vector double);
13722 int vec_any_nan (vector double);
13723 int vec_any_ne (vector double, vector double);
13724 int vec_any_nge (vector double, vector double);
13725 int vec_any_ngt (vector double, vector double);
13726 int vec_any_nle (vector double, vector double);
13727 int vec_any_nlt (vector double, vector double);
13728 int vec_any_numeric (vector double);
13730 vector double vec_vsx_ld (int, const vector double *);
13731 vector double vec_vsx_ld (int, const double *);
13732 vector float vec_vsx_ld (int, const vector float *);
13733 vector float vec_vsx_ld (int, const float *);
13734 vector bool int vec_vsx_ld (int, const vector bool int *);
13735 vector signed int vec_vsx_ld (int, const vector signed int *);
13736 vector signed int vec_vsx_ld (int, const int *);
13737 vector signed int vec_vsx_ld (int, const long *);
13738 vector unsigned int vec_vsx_ld (int, const vector unsigned int *);
13739 vector unsigned int vec_vsx_ld (int, const unsigned int *);
13740 vector unsigned int vec_vsx_ld (int, const unsigned long *);
13741 vector bool short vec_vsx_ld (int, const vector bool short *);
13742 vector pixel vec_vsx_ld (int, const vector pixel *);
13743 vector signed short vec_vsx_ld (int, const vector signed short *);
13744 vector signed short vec_vsx_ld (int, const short *);
13745 vector unsigned short vec_vsx_ld (int, const vector unsigned short *);
13746 vector unsigned short vec_vsx_ld (int, const unsigned short *);
13747 vector bool char vec_vsx_ld (int, const vector bool char *);
13748 vector signed char vec_vsx_ld (int, const vector signed char *);
13749 vector signed char vec_vsx_ld (int, const signed char *);
13750 vector unsigned char vec_vsx_ld (int, const vector unsigned char *);
13751 vector unsigned char vec_vsx_ld (int, const unsigned char *);
13753 void vec_vsx_st (vector double, int, vector double *);
13754 void vec_vsx_st (vector double, int, double *);
13755 void vec_vsx_st (vector float, int, vector float *);
13756 void vec_vsx_st (vector float, int, float *);
13757 void vec_vsx_st (vector signed int, int, vector signed int *);
13758 void vec_vsx_st (vector signed int, int, int *);
13759 void vec_vsx_st (vector unsigned int, int, vector unsigned int *);
13760 void vec_vsx_st (vector unsigned int, int, unsigned int *);
13761 void vec_vsx_st (vector bool int, int, vector bool int *);
13762 void vec_vsx_st (vector bool int, int, unsigned int *);
13763 void vec_vsx_st (vector bool int, int, int *);
13764 void vec_vsx_st (vector signed short, int, vector signed short *);
13765 void vec_vsx_st (vector signed short, int, short *);
13766 void vec_vsx_st (vector unsigned short, int, vector unsigned short *);
13767 void vec_vsx_st (vector unsigned short, int, unsigned short *);
13768 void vec_vsx_st (vector bool short, int, vector bool short *);
13769 void vec_vsx_st (vector bool short, int, unsigned short *);
13770 void vec_vsx_st (vector pixel, int, vector pixel *);
13771 void vec_vsx_st (vector pixel, int, unsigned short *);
13772 void vec_vsx_st (vector pixel, int, short *);
13773 void vec_vsx_st (vector bool short, int, short *);
13774 void vec_vsx_st (vector signed char, int, vector signed char *);
13775 void vec_vsx_st (vector signed char, int, signed char *);
13776 void vec_vsx_st (vector unsigned char, int, vector unsigned char *);
13777 void vec_vsx_st (vector unsigned char, int, unsigned char *);
13778 void vec_vsx_st (vector bool char, int, vector bool char *);
13779 void vec_vsx_st (vector bool char, int, unsigned char *);
13780 void vec_vsx_st (vector bool char, int, signed char *);
13783 Note that the @samp{vec_ld} and @samp{vec_st} built-in functions always
13784 generate the AltiVec @samp{LVX} and @samp{STVX} instructions even
13785 if the VSX instruction set is available. The @samp{vec_vsx_ld} and
13786 @samp{vec_vsx_st} built-in functions always generate the VSX @samp{LXVD2X},
13787 @samp{LXVW4X}, @samp{STXVD2X}, and @samp{STXVW4X} instructions.
13789 @node SH Built-in Functions
13790 @subsection SH Built-in Functions
13791 The following built-in functions are supported on the SH1, SH2, SH3 and SH4
13792 families of processors:
13794 @deftypefn {Built-in Function} {void} __builtin_set_thread_pointer (void *@var{ptr})
13795 Sets the @samp{GBR} register to the specified value @var{ptr}. This is usually
13796 used by system code that manages threads and execution contexts. The compiler
13797 normally does not generate code that modifies the contents of @samp{GBR} and
13798 thus the value is preserved across function calls. Changing the @samp{GBR}
13799 value in user code must be done with caution, since the compiler might use
13800 @samp{GBR} in order to access thread local variables.
13804 @deftypefn {Built-in Function} {void *} __builtin_thread_pointer (void)
13805 Returns the value that is currently set in the @samp{GBR} register.
13806 Memory loads and stores that use the thread pointer as a base address are
13807 turned into @samp{GBR} based displacement loads and stores, if possible.
13815 int get_tcb_value (void)
13817 // Generate @samp{mov.l @@(8,gbr),r0} instruction
13818 return ((my_tcb*)__builtin_thread_pointer ())->c;
13824 @node RX Built-in Functions
13825 @subsection RX Built-in Functions
13826 GCC supports some of the RX instructions which cannot be expressed in
13827 the C programming language via the use of built-in functions. The
13828 following functions are supported:
13830 @deftypefn {Built-in Function} void __builtin_rx_brk (void)
13831 Generates the @code{brk} machine instruction.
13834 @deftypefn {Built-in Function} void __builtin_rx_clrpsw (int)
13835 Generates the @code{clrpsw} machine instruction to clear the specified
13836 bit in the processor status word.
13839 @deftypefn {Built-in Function} void __builtin_rx_int (int)
13840 Generates the @code{int} machine instruction to generate an interrupt
13841 with the specified value.
13844 @deftypefn {Built-in Function} void __builtin_rx_machi (int, int)
13845 Generates the @code{machi} machine instruction to add the result of
13846 multiplying the top 16 bits of the two arguments into the
13850 @deftypefn {Built-in Function} void __builtin_rx_maclo (int, int)
13851 Generates the @code{maclo} machine instruction to add the result of
13852 multiplying the bottom 16 bits of the two arguments into the
13856 @deftypefn {Built-in Function} void __builtin_rx_mulhi (int, int)
13857 Generates the @code{mulhi} machine instruction to place the result of
13858 multiplying the top 16 bits of the two arguments into the
13862 @deftypefn {Built-in Function} void __builtin_rx_mullo (int, int)
13863 Generates the @code{mullo} machine instruction to place the result of
13864 multiplying the bottom 16 bits of the two arguments into the
13868 @deftypefn {Built-in Function} int __builtin_rx_mvfachi (void)
13869 Generates the @code{mvfachi} machine instruction to read the top
13870 32 bits of the accumulator.
13873 @deftypefn {Built-in Function} int __builtin_rx_mvfacmi (void)
13874 Generates the @code{mvfacmi} machine instruction to read the middle
13875 32 bits of the accumulator.
13878 @deftypefn {Built-in Function} int __builtin_rx_mvfc (int)
13879 Generates the @code{mvfc} machine instruction which reads the control
13880 register specified in its argument and returns its value.
13883 @deftypefn {Built-in Function} void __builtin_rx_mvtachi (int)
13884 Generates the @code{mvtachi} machine instruction to set the top
13885 32 bits of the accumulator.
13888 @deftypefn {Built-in Function} void __builtin_rx_mvtaclo (int)
13889 Generates the @code{mvtaclo} machine instruction to set the bottom
13890 32 bits of the accumulator.
13893 @deftypefn {Built-in Function} void __builtin_rx_mvtc (int reg, int val)
13894 Generates the @code{mvtc} machine instruction which sets control
13895 register number @code{reg} to @code{val}.
13898 @deftypefn {Built-in Function} void __builtin_rx_mvtipl (int)
13899 Generates the @code{mvtipl} machine instruction set the interrupt
13903 @deftypefn {Built-in Function} void __builtin_rx_racw (int)
13904 Generates the @code{racw} machine instruction to round the accumulator
13905 according to the specified mode.
13908 @deftypefn {Built-in Function} int __builtin_rx_revw (int)
13909 Generates the @code{revw} machine instruction which swaps the bytes in
13910 the argument so that bits 0--7 now occupy bits 8--15 and vice versa,
13911 and also bits 16--23 occupy bits 24--31 and vice versa.
13914 @deftypefn {Built-in Function} void __builtin_rx_rmpa (void)
13915 Generates the @code{rmpa} machine instruction which initiates a
13916 repeated multiply and accumulate sequence.
13919 @deftypefn {Built-in Function} void __builtin_rx_round (float)
13920 Generates the @code{round} machine instruction which returns the
13921 floating-point argument rounded according to the current rounding mode
13922 set in the floating-point status word register.
13925 @deftypefn {Built-in Function} int __builtin_rx_sat (int)
13926 Generates the @code{sat} machine instruction which returns the
13927 saturated value of the argument.
13930 @deftypefn {Built-in Function} void __builtin_rx_setpsw (int)
13931 Generates the @code{setpsw} machine instruction to set the specified
13932 bit in the processor status word.
13935 @deftypefn {Built-in Function} void __builtin_rx_wait (void)
13936 Generates the @code{wait} machine instruction.
13939 @node SPARC VIS Built-in Functions
13940 @subsection SPARC VIS Built-in Functions
13942 GCC supports SIMD operations on the SPARC using both the generic vector
13943 extensions (@pxref{Vector Extensions}) as well as built-in functions for
13944 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
13945 switch, the VIS extension is exposed as the following built-in functions:
13948 typedef int v1si __attribute__ ((vector_size (4)));
13949 typedef int v2si __attribute__ ((vector_size (8)));
13950 typedef short v4hi __attribute__ ((vector_size (8)));
13951 typedef short v2hi __attribute__ ((vector_size (4)));
13952 typedef unsigned char v8qi __attribute__ ((vector_size (8)));
13953 typedef unsigned char v4qi __attribute__ ((vector_size (4)));
13955 void __builtin_vis_write_gsr (int64_t);
13956 int64_t __builtin_vis_read_gsr (void);
13958 void * __builtin_vis_alignaddr (void *, long);
13959 void * __builtin_vis_alignaddrl (void *, long);
13960 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
13961 v2si __builtin_vis_faligndatav2si (v2si, v2si);
13962 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
13963 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
13965 v4hi __builtin_vis_fexpand (v4qi);
13967 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
13968 v4hi __builtin_vis_fmul8x16au (v4qi, v2hi);
13969 v4hi __builtin_vis_fmul8x16al (v4qi, v2hi);
13970 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
13971 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
13972 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
13973 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
13975 v4qi __builtin_vis_fpack16 (v4hi);
13976 v8qi __builtin_vis_fpack32 (v2si, v8qi);
13977 v2hi __builtin_vis_fpackfix (v2si);
13978 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
13980 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
13982 long __builtin_vis_edge8 (void *, void *);
13983 long __builtin_vis_edge8l (void *, void *);
13984 long __builtin_vis_edge16 (void *, void *);
13985 long __builtin_vis_edge16l (void *, void *);
13986 long __builtin_vis_edge32 (void *, void *);
13987 long __builtin_vis_edge32l (void *, void *);
13989 long __builtin_vis_fcmple16 (v4hi, v4hi);
13990 long __builtin_vis_fcmple32 (v2si, v2si);
13991 long __builtin_vis_fcmpne16 (v4hi, v4hi);
13992 long __builtin_vis_fcmpne32 (v2si, v2si);
13993 long __builtin_vis_fcmpgt16 (v4hi, v4hi);
13994 long __builtin_vis_fcmpgt32 (v2si, v2si);
13995 long __builtin_vis_fcmpeq16 (v4hi, v4hi);
13996 long __builtin_vis_fcmpeq32 (v2si, v2si);
13998 v4hi __builtin_vis_fpadd16 (v4hi, v4hi);
13999 v2hi __builtin_vis_fpadd16s (v2hi, v2hi);
14000 v2si __builtin_vis_fpadd32 (v2si, v2si);
14001 v1si __builtin_vis_fpadd32s (v1si, v1si);
14002 v4hi __builtin_vis_fpsub16 (v4hi, v4hi);
14003 v2hi __builtin_vis_fpsub16s (v2hi, v2hi);
14004 v2si __builtin_vis_fpsub32 (v2si, v2si);
14005 v1si __builtin_vis_fpsub32s (v1si, v1si);
14007 long __builtin_vis_array8 (long, long);
14008 long __builtin_vis_array16 (long, long);
14009 long __builtin_vis_array32 (long, long);
14012 When you use the @option{-mvis2} switch, the VIS version 2.0 built-in
14013 functions also become available:
14016 long __builtin_vis_bmask (long, long);
14017 int64_t __builtin_vis_bshuffledi (int64_t, int64_t);
14018 v2si __builtin_vis_bshufflev2si (v2si, v2si);
14019 v4hi __builtin_vis_bshufflev2si (v4hi, v4hi);
14020 v8qi __builtin_vis_bshufflev2si (v8qi, v8qi);
14022 long __builtin_vis_edge8n (void *, void *);
14023 long __builtin_vis_edge8ln (void *, void *);
14024 long __builtin_vis_edge16n (void *, void *);
14025 long __builtin_vis_edge16ln (void *, void *);
14026 long __builtin_vis_edge32n (void *, void *);
14027 long __builtin_vis_edge32ln (void *, void *);
14030 When you use the @option{-mvis3} switch, the VIS version 3.0 built-in
14031 functions also become available:
14034 void __builtin_vis_cmask8 (long);
14035 void __builtin_vis_cmask16 (long);
14036 void __builtin_vis_cmask32 (long);
14038 v4hi __builtin_vis_fchksm16 (v4hi, v4hi);
14040 v4hi __builtin_vis_fsll16 (v4hi, v4hi);
14041 v4hi __builtin_vis_fslas16 (v4hi, v4hi);
14042 v4hi __builtin_vis_fsrl16 (v4hi, v4hi);
14043 v4hi __builtin_vis_fsra16 (v4hi, v4hi);
14044 v2si __builtin_vis_fsll16 (v2si, v2si);
14045 v2si __builtin_vis_fslas16 (v2si, v2si);
14046 v2si __builtin_vis_fsrl16 (v2si, v2si);
14047 v2si __builtin_vis_fsra16 (v2si, v2si);
14049 long __builtin_vis_pdistn (v8qi, v8qi);
14051 v4hi __builtin_vis_fmean16 (v4hi, v4hi);
14053 int64_t __builtin_vis_fpadd64 (int64_t, int64_t);
14054 int64_t __builtin_vis_fpsub64 (int64_t, int64_t);
14056 v4hi __builtin_vis_fpadds16 (v4hi, v4hi);
14057 v2hi __builtin_vis_fpadds16s (v2hi, v2hi);
14058 v4hi __builtin_vis_fpsubs16 (v4hi, v4hi);
14059 v2hi __builtin_vis_fpsubs16s (v2hi, v2hi);
14060 v2si __builtin_vis_fpadds32 (v2si, v2si);
14061 v1si __builtin_vis_fpadds32s (v1si, v1si);
14062 v2si __builtin_vis_fpsubs32 (v2si, v2si);
14063 v1si __builtin_vis_fpsubs32s (v1si, v1si);
14065 long __builtin_vis_fucmple8 (v8qi, v8qi);
14066 long __builtin_vis_fucmpne8 (v8qi, v8qi);
14067 long __builtin_vis_fucmpgt8 (v8qi, v8qi);
14068 long __builtin_vis_fucmpeq8 (v8qi, v8qi);
14070 float __builtin_vis_fhadds (float, float);
14071 double __builtin_vis_fhaddd (double, double);
14072 float __builtin_vis_fhsubs (float, float);
14073 double __builtin_vis_fhsubd (double, double);
14074 float __builtin_vis_fnhadds (float, float);
14075 double __builtin_vis_fnhaddd (double, double);
14077 int64_t __builtin_vis_umulxhi (int64_t, int64_t);
14078 int64_t __builtin_vis_xmulx (int64_t, int64_t);
14079 int64_t __builtin_vis_xmulxhi (int64_t, int64_t);
14082 @node SPU Built-in Functions
14083 @subsection SPU Built-in Functions
14085 GCC provides extensions for the SPU processor as described in the
14086 Sony/Toshiba/IBM SPU Language Extensions Specification, which can be
14087 found at @uref{http://cell.scei.co.jp/} or
14088 @uref{http://www.ibm.com/developerworks/power/cell/}. GCC's
14089 implementation differs in several ways.
14094 The optional extension of specifying vector constants in parentheses is
14098 A vector initializer requires no cast if the vector constant is of the
14099 same type as the variable it is initializing.
14102 If @code{signed} or @code{unsigned} is omitted, the signedness of the
14103 vector type is the default signedness of the base type. The default
14104 varies depending on the operating system, so a portable program should
14105 always specify the signedness.
14108 By default, the keyword @code{__vector} is added. The macro
14109 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
14113 GCC allows using a @code{typedef} name as the type specifier for a
14117 For C, overloaded functions are implemented with macros so the following
14121 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
14125 Since @code{spu_add} is a macro, the vector constant in the example
14126 is treated as four separate arguments. Wrap the entire argument in
14127 parentheses for this to work.
14130 The extended version of @code{__builtin_expect} is not supported.
14134 @emph{Note:} Only the interface described in the aforementioned
14135 specification is supported. Internally, GCC uses built-in functions to
14136 implement the required functionality, but these are not supported and
14137 are subject to change without notice.
14139 @node TI C6X Built-in Functions
14140 @subsection TI C6X Built-in Functions
14142 GCC provides intrinsics to access certain instructions of the TI C6X
14143 processors. These intrinsics, listed below, are available after
14144 inclusion of the @code{c6x_intrinsics.h} header file. They map directly
14145 to C6X instructions.
14149 int _sadd (int, int)
14150 int _ssub (int, int)
14151 int _sadd2 (int, int)
14152 int _ssub2 (int, int)
14153 long long _mpy2 (int, int)
14154 long long _smpy2 (int, int)
14155 int _add4 (int, int)
14156 int _sub4 (int, int)
14157 int _saddu4 (int, int)
14159 int _smpy (int, int)
14160 int _smpyh (int, int)
14161 int _smpyhl (int, int)
14162 int _smpylh (int, int)
14164 int _sshl (int, int)
14165 int _subc (int, int)
14167 int _avg2 (int, int)
14168 int _avgu4 (int, int)
14170 int _clrr (int, int)
14171 int _extr (int, int)
14172 int _extru (int, int)
14178 @node TILE-Gx Built-in Functions
14179 @subsection TILE-Gx Built-in Functions
14181 GCC provides intrinsics to access every instruction of the TILE-Gx
14182 processor. The intrinsics are of the form:
14186 unsigned long long __insn_@var{op} (...)
14190 Where @var{op} is the name of the instruction. Refer to the ISA manual
14191 for the complete list of instructions.
14193 GCC also provides intrinsics to directly access the network registers.
14194 The intrinsics are:
14198 unsigned long long __tile_idn0_receive (void)
14199 unsigned long long __tile_idn1_receive (void)
14200 unsigned long long __tile_udn0_receive (void)
14201 unsigned long long __tile_udn1_receive (void)
14202 unsigned long long __tile_udn2_receive (void)
14203 unsigned long long __tile_udn3_receive (void)
14204 void __tile_idn_send (unsigned long long)
14205 void __tile_udn_send (unsigned long long)
14209 The intrinsic @code{void __tile_network_barrier (void)} is used to
14210 guarantee that no network operations before it are reordered with
14213 @node TILEPro Built-in Functions
14214 @subsection TILEPro Built-in Functions
14216 GCC provides intrinsics to access every instruction of the TILEPro
14217 processor. The intrinsics are of the form:
14221 unsigned __insn_@var{op} (...)
14226 where @var{op} is the name of the instruction. Refer to the ISA manual
14227 for the complete list of instructions.
14229 GCC also provides intrinsics to directly access the network registers.
14230 The intrinsics are:
14234 unsigned __tile_idn0_receive (void)
14235 unsigned __tile_idn1_receive (void)
14236 unsigned __tile_sn_receive (void)
14237 unsigned __tile_udn0_receive (void)
14238 unsigned __tile_udn1_receive (void)
14239 unsigned __tile_udn2_receive (void)
14240 unsigned __tile_udn3_receive (void)
14241 void __tile_idn_send (unsigned)
14242 void __tile_sn_send (unsigned)
14243 void __tile_udn_send (unsigned)
14247 The intrinsic @code{void __tile_network_barrier (void)} is used to
14248 guarantee that no network operations before it are reordered with
14251 @node Target Format Checks
14252 @section Format Checks Specific to Particular Target Machines
14254 For some target machines, GCC supports additional options to the
14256 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
14259 * Solaris Format Checks::
14260 * Darwin Format Checks::
14263 @node Solaris Format Checks
14264 @subsection Solaris Format Checks
14266 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
14267 check. @code{cmn_err} accepts a subset of the standard @code{printf}
14268 conversions, and the two-argument @code{%b} conversion for displaying
14269 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
14271 @node Darwin Format Checks
14272 @subsection Darwin Format Checks
14274 Darwin targets support the @code{CFString} (or @code{__CFString__}) in the format
14275 attribute context. Declarations made with such attribution are parsed for correct syntax
14276 and format argument types. However, parsing of the format string itself is currently undefined
14277 and is not carried out by this version of the compiler.
14279 Additionally, @code{CFStringRefs} (defined by the @code{CoreFoundation} headers) may
14280 also be used as format arguments. Note that the relevant headers are only likely to be
14281 available on Darwin (OSX) installations. On such installations, the XCode and system
14282 documentation provide descriptions of @code{CFString}, @code{CFStringRefs} and
14283 associated functions.
14286 @section Pragmas Accepted by GCC
14288 @cindex @code{#pragma}
14290 GCC supports several types of pragmas, primarily in order to compile
14291 code originally written for other compilers. Note that in general
14292 we do not recommend the use of pragmas; @xref{Function Attributes},
14293 for further explanation.
14299 * RS/6000 and PowerPC Pragmas::
14301 * Solaris Pragmas::
14302 * Symbol-Renaming Pragmas::
14303 * Structure-Packing Pragmas::
14305 * Diagnostic Pragmas::
14306 * Visibility Pragmas::
14307 * Push/Pop Macro Pragmas::
14308 * Function Specific Option Pragmas::
14312 @subsection ARM Pragmas
14314 The ARM target defines pragmas for controlling the default addition of
14315 @code{long_call} and @code{short_call} attributes to functions.
14316 @xref{Function Attributes}, for information about the effects of these
14321 @cindex pragma, long_calls
14322 Set all subsequent functions to have the @code{long_call} attribute.
14324 @item no_long_calls
14325 @cindex pragma, no_long_calls
14326 Set all subsequent functions to have the @code{short_call} attribute.
14328 @item long_calls_off
14329 @cindex pragma, long_calls_off
14330 Do not affect the @code{long_call} or @code{short_call} attributes of
14331 subsequent functions.
14335 @subsection M32C Pragmas
14338 @item GCC memregs @var{number}
14339 @cindex pragma, memregs
14340 Overrides the command-line option @code{-memregs=} for the current
14341 file. Use with care! This pragma must be before any function in the
14342 file, and mixing different memregs values in different objects may
14343 make them incompatible. This pragma is useful when a
14344 performance-critical function uses a memreg for temporary values,
14345 as it may allow you to reduce the number of memregs used.
14347 @item ADDRESS @var{name} @var{address}
14348 @cindex pragma, address
14349 For any declared symbols matching @var{name}, this does three things
14350 to that symbol: it forces the symbol to be located at the given
14351 address (a number), it forces the symbol to be volatile, and it
14352 changes the symbol's scope to be static. This pragma exists for
14353 compatibility with other compilers, but note that the common
14354 @code{1234H} numeric syntax is not supported (use @code{0x1234}
14358 #pragma ADDRESS port3 0x103
14365 @subsection MeP Pragmas
14369 @item custom io_volatile (on|off)
14370 @cindex pragma, custom io_volatile
14371 Overrides the command-line option @code{-mio-volatile} for the current
14372 file. Note that for compatibility with future GCC releases, this
14373 option should only be used once before any @code{io} variables in each
14376 @item GCC coprocessor available @var{registers}
14377 @cindex pragma, coprocessor available
14378 Specifies which coprocessor registers are available to the register
14379 allocator. @var{registers} may be a single register, register range
14380 separated by ellipses, or comma-separated list of those. Example:
14383 #pragma GCC coprocessor available $c0...$c10, $c28
14386 @item GCC coprocessor call_saved @var{registers}
14387 @cindex pragma, coprocessor call_saved
14388 Specifies which coprocessor registers are to be saved and restored by
14389 any function using them. @var{registers} may be a single register,
14390 register range separated by ellipses, or comma-separated list of
14394 #pragma GCC coprocessor call_saved $c4...$c6, $c31
14397 @item GCC coprocessor subclass '(A|B|C|D)' = @var{registers}
14398 @cindex pragma, coprocessor subclass
14399 Creates and defines a register class. These register classes can be
14400 used by inline @code{asm} constructs. @var{registers} may be a single
14401 register, register range separated by ellipses, or comma-separated
14402 list of those. Example:
14405 #pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6
14407 asm ("cpfoo %0" : "=B" (x));
14410 @item GCC disinterrupt @var{name} , @var{name} @dots{}
14411 @cindex pragma, disinterrupt
14412 For the named functions, the compiler adds code to disable interrupts
14413 for the duration of those functions. If any functions so named
14414 are not encountered in the source, a warning is emitted that the pragma is
14415 not used. Examples:
14418 #pragma disinterrupt foo
14419 #pragma disinterrupt bar, grill
14420 int foo () @{ @dots{} @}
14423 @item GCC call @var{name} , @var{name} @dots{}
14424 @cindex pragma, call
14425 For the named functions, the compiler always uses a register-indirect
14426 call model when calling the named functions. Examples:
14435 @node RS/6000 and PowerPC Pragmas
14436 @subsection RS/6000 and PowerPC Pragmas
14438 The RS/6000 and PowerPC targets define one pragma for controlling
14439 whether or not the @code{longcall} attribute is added to function
14440 declarations by default. This pragma overrides the @option{-mlongcall}
14441 option, but not the @code{longcall} and @code{shortcall} attributes.
14442 @xref{RS/6000 and PowerPC Options}, for more information about when long
14443 calls are and are not necessary.
14447 @cindex pragma, longcall
14448 Apply the @code{longcall} attribute to all subsequent function
14452 Do not apply the @code{longcall} attribute to subsequent function
14456 @c Describe h8300 pragmas here.
14457 @c Describe sh pragmas here.
14458 @c Describe v850 pragmas here.
14460 @node Darwin Pragmas
14461 @subsection Darwin Pragmas
14463 The following pragmas are available for all architectures running the
14464 Darwin operating system. These are useful for compatibility with other
14468 @item mark @var{tokens}@dots{}
14469 @cindex pragma, mark
14470 This pragma is accepted, but has no effect.
14472 @item options align=@var{alignment}
14473 @cindex pragma, options align
14474 This pragma sets the alignment of fields in structures. The values of
14475 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
14476 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
14477 properly; to restore the previous setting, use @code{reset} for the
14480 @item segment @var{tokens}@dots{}
14481 @cindex pragma, segment
14482 This pragma is accepted, but has no effect.
14484 @item unused (@var{var} [, @var{var}]@dots{})
14485 @cindex pragma, unused
14486 This pragma declares variables to be possibly unused. GCC does not
14487 produce warnings for the listed variables. The effect is similar to
14488 that of the @code{unused} attribute, except that this pragma may appear
14489 anywhere within the variables' scopes.
14492 @node Solaris Pragmas
14493 @subsection Solaris Pragmas
14495 The Solaris target supports @code{#pragma redefine_extname}
14496 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
14497 @code{#pragma} directives for compatibility with the system compiler.
14500 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
14501 @cindex pragma, align
14503 Increase the minimum alignment of each @var{variable} to @var{alignment}.
14504 This is the same as GCC's @code{aligned} attribute @pxref{Variable
14505 Attributes}). Macro expansion occurs on the arguments to this pragma
14506 when compiling C and Objective-C@. It does not currently occur when
14507 compiling C++, but this is a bug which may be fixed in a future
14510 @item fini (@var{function} [, @var{function}]...)
14511 @cindex pragma, fini
14513 This pragma causes each listed @var{function} to be called after
14514 main, or during shared module unloading, by adding a call to the
14515 @code{.fini} section.
14517 @item init (@var{function} [, @var{function}]...)
14518 @cindex pragma, init
14520 This pragma causes each listed @var{function} to be called during
14521 initialization (before @code{main}) or during shared module loading, by
14522 adding a call to the @code{.init} section.
14526 @node Symbol-Renaming Pragmas
14527 @subsection Symbol-Renaming Pragmas
14529 For compatibility with the Solaris system headers, GCC
14530 supports two @code{#pragma} directives that change the name used in
14531 assembly for a given declaration. To get this effect
14532 on all platforms supported by GCC, use the asm labels extension (@pxref{Asm
14536 @item redefine_extname @var{oldname} @var{newname}
14537 @cindex pragma, redefine_extname
14539 This pragma gives the C function @var{oldname} the assembly symbol
14540 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
14541 is defined if this pragma is available (currently on all platforms).
14544 This pragma and the asm labels extension interact in a complicated
14545 manner. Here are some corner cases you may want to be aware of.
14548 @item Both pragmas silently apply only to declarations with external
14549 linkage. Asm labels do not have this restriction.
14551 @item In C++, both pragmas silently apply only to declarations with
14552 ``C'' linkage. Again, asm labels do not have this restriction.
14554 @item If any of the three ways of changing the assembly name of a
14555 declaration is applied to a declaration whose assembly name has
14556 already been determined (either by a previous use of one of these
14557 features, or because the compiler needed the assembly name in order to
14558 generate code), and the new name is different, a warning issues and
14559 the name does not change.
14561 @item The @var{oldname} used by @code{#pragma redefine_extname} is
14562 always the C-language name.
14565 @node Structure-Packing Pragmas
14566 @subsection Structure-Packing Pragmas
14568 For compatibility with Microsoft Windows compilers, GCC supports a
14569 set of @code{#pragma} directives that change the maximum alignment of
14570 members of structures (other than zero-width bit-fields), unions, and
14571 classes subsequently defined. The @var{n} value below always is required
14572 to be a small power of two and specifies the new alignment in bytes.
14575 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
14576 @item @code{#pragma pack()} sets the alignment to the one that was in
14577 effect when compilation started (see also command-line option
14578 @option{-fpack-struct[=@var{n}]} @pxref{Code Gen Options}).
14579 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
14580 setting on an internal stack and then optionally sets the new alignment.
14581 @item @code{#pragma pack(pop)} restores the alignment setting to the one
14582 saved at the top of the internal stack (and removes that stack entry).
14583 Note that @code{#pragma pack([@var{n}])} does not influence this internal
14584 stack; thus it is possible to have @code{#pragma pack(push)} followed by
14585 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
14586 @code{#pragma pack(pop)}.
14589 Some targets, e.g.@: i386 and PowerPC, support the @code{ms_struct}
14590 @code{#pragma} which lays out a structure as the documented
14591 @code{__attribute__ ((ms_struct))}.
14593 @item @code{#pragma ms_struct on} turns on the layout for structures
14595 @item @code{#pragma ms_struct off} turns off the layout for structures
14597 @item @code{#pragma ms_struct reset} goes back to the default layout.
14601 @subsection Weak Pragmas
14603 For compatibility with SVR4, GCC supports a set of @code{#pragma}
14604 directives for declaring symbols to be weak, and defining weak
14608 @item #pragma weak @var{symbol}
14609 @cindex pragma, weak
14610 This pragma declares @var{symbol} to be weak, as if the declaration
14611 had the attribute of the same name. The pragma may appear before
14612 or after the declaration of @var{symbol}. It is not an error for
14613 @var{symbol} to never be defined at all.
14615 @item #pragma weak @var{symbol1} = @var{symbol2}
14616 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
14617 It is an error if @var{symbol2} is not defined in the current
14621 @node Diagnostic Pragmas
14622 @subsection Diagnostic Pragmas
14624 GCC allows the user to selectively enable or disable certain types of
14625 diagnostics, and change the kind of the diagnostic. For example, a
14626 project's policy might require that all sources compile with
14627 @option{-Werror} but certain files might have exceptions allowing
14628 specific types of warnings. Or, a project might selectively enable
14629 diagnostics and treat them as errors depending on which preprocessor
14630 macros are defined.
14633 @item #pragma GCC diagnostic @var{kind} @var{option}
14634 @cindex pragma, diagnostic
14636 Modifies the disposition of a diagnostic. Note that not all
14637 diagnostics are modifiable; at the moment only warnings (normally
14638 controlled by @samp{-W@dots{}}) can be controlled, and not all of them.
14639 Use @option{-fdiagnostics-show-option} to determine which diagnostics
14640 are controllable and which option controls them.
14642 @var{kind} is @samp{error} to treat this diagnostic as an error,
14643 @samp{warning} to treat it like a warning (even if @option{-Werror} is
14644 in effect), or @samp{ignored} if the diagnostic is to be ignored.
14645 @var{option} is a double quoted string that matches the command-line
14649 #pragma GCC diagnostic warning "-Wformat"
14650 #pragma GCC diagnostic error "-Wformat"
14651 #pragma GCC diagnostic ignored "-Wformat"
14654 Note that these pragmas override any command-line options. GCC keeps
14655 track of the location of each pragma, and issues diagnostics according
14656 to the state as of that point in the source file. Thus, pragmas occurring
14657 after a line do not affect diagnostics caused by that line.
14659 @item #pragma GCC diagnostic push
14660 @itemx #pragma GCC diagnostic pop
14662 Causes GCC to remember the state of the diagnostics as of each
14663 @code{push}, and restore to that point at each @code{pop}. If a
14664 @code{pop} has no matching @code{push}, the command-line options are
14668 #pragma GCC diagnostic error "-Wuninitialized"
14669 foo(a); /* error is given for this one */
14670 #pragma GCC diagnostic push
14671 #pragma GCC diagnostic ignored "-Wuninitialized"
14672 foo(b); /* no diagnostic for this one */
14673 #pragma GCC diagnostic pop
14674 foo(c); /* error is given for this one */
14675 #pragma GCC diagnostic pop
14676 foo(d); /* depends on command-line options */
14681 GCC also offers a simple mechanism for printing messages during
14685 @item #pragma message @var{string}
14686 @cindex pragma, diagnostic
14688 Prints @var{string} as a compiler message on compilation. The message
14689 is informational only, and is neither a compilation warning nor an error.
14692 #pragma message "Compiling " __FILE__ "..."
14695 @var{string} may be parenthesized, and is printed with location
14696 information. For example,
14699 #define DO_PRAGMA(x) _Pragma (#x)
14700 #define TODO(x) DO_PRAGMA(message ("TODO - " #x))
14702 TODO(Remember to fix this)
14706 prints @samp{/tmp/file.c:4: note: #pragma message:
14707 TODO - Remember to fix this}.
14711 @node Visibility Pragmas
14712 @subsection Visibility Pragmas
14715 @item #pragma GCC visibility push(@var{visibility})
14716 @itemx #pragma GCC visibility pop
14717 @cindex pragma, visibility
14719 This pragma allows the user to set the visibility for multiple
14720 declarations without having to give each a visibility attribute
14721 @xref{Function Attributes}, for more information about visibility and
14722 the attribute syntax.
14724 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
14725 declarations. Class members and template specializations are not
14726 affected; if you want to override the visibility for a particular
14727 member or instantiation, you must use an attribute.
14732 @node Push/Pop Macro Pragmas
14733 @subsection Push/Pop Macro Pragmas
14735 For compatibility with Microsoft Windows compilers, GCC supports
14736 @samp{#pragma push_macro(@var{"macro_name"})}
14737 and @samp{#pragma pop_macro(@var{"macro_name"})}.
14740 @item #pragma push_macro(@var{"macro_name"})
14741 @cindex pragma, push_macro
14742 This pragma saves the value of the macro named as @var{macro_name} to
14743 the top of the stack for this macro.
14745 @item #pragma pop_macro(@var{"macro_name"})
14746 @cindex pragma, pop_macro
14747 This pragma sets the value of the macro named as @var{macro_name} to
14748 the value on top of the stack for this macro. If the stack for
14749 @var{macro_name} is empty, the value of the macro remains unchanged.
14756 #pragma push_macro("X")
14759 #pragma pop_macro("X")
14764 In this example, the definition of X as 1 is saved by @code{#pragma
14765 push_macro} and restored by @code{#pragma pop_macro}.
14767 @node Function Specific Option Pragmas
14768 @subsection Function Specific Option Pragmas
14771 @item #pragma GCC target (@var{"string"}...)
14772 @cindex pragma GCC target
14774 This pragma allows you to set target specific options for functions
14775 defined later in the source file. One or more strings can be
14776 specified. Each function that is defined after this point is as
14777 if @code{attribute((target("STRING")))} was specified for that
14778 function. The parenthesis around the options is optional.
14779 @xref{Function Attributes}, for more information about the
14780 @code{target} attribute and the attribute syntax.
14782 The @code{#pragma GCC target} attribute is not implemented in GCC versions earlier
14783 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC back ends. At
14784 present, it is not implemented for other back ends.
14788 @item #pragma GCC optimize (@var{"string"}...)
14789 @cindex pragma GCC optimize
14791 This pragma allows you to set global optimization options for functions
14792 defined later in the source file. One or more strings can be
14793 specified. Each function that is defined after this point is as
14794 if @code{attribute((optimize("STRING")))} was specified for that
14795 function. The parenthesis around the options is optional.
14796 @xref{Function Attributes}, for more information about the
14797 @code{optimize} attribute and the attribute syntax.
14799 The @samp{#pragma GCC optimize} pragma is not implemented in GCC
14800 versions earlier than 4.4.
14804 @item #pragma GCC push_options
14805 @itemx #pragma GCC pop_options
14806 @cindex pragma GCC push_options
14807 @cindex pragma GCC pop_options
14809 These pragmas maintain a stack of the current target and optimization
14810 options. It is intended for include files where you temporarily want
14811 to switch to using a different @samp{#pragma GCC target} or
14812 @samp{#pragma GCC optimize} and then to pop back to the previous
14815 The @samp{#pragma GCC push_options} and @samp{#pragma GCC pop_options}
14816 pragmas are not implemented in GCC versions earlier than 4.4.
14820 @item #pragma GCC reset_options
14821 @cindex pragma GCC reset_options
14823 This pragma clears the current @code{#pragma GCC target} and
14824 @code{#pragma GCC optimize} to use the default switches as specified
14825 on the command line.
14827 The @samp{#pragma GCC reset_options} pragma is not implemented in GCC
14828 versions earlier than 4.4.
14831 @node Unnamed Fields
14832 @section Unnamed struct/union fields within structs/unions
14833 @cindex @code{struct}
14834 @cindex @code{union}
14836 As permitted by ISO C11 and for compatibility with other compilers,
14837 GCC allows you to define
14838 a structure or union that contains, as fields, structures and unions
14839 without names. For example:
14853 In this example, you are able to access members of the unnamed
14854 union with code like @samp{foo.b}. Note that only unnamed structs and
14855 unions are allowed, you may not have, for example, an unnamed
14858 You must never create such structures that cause ambiguous field definitions.
14859 For example, in this structure:
14871 it is ambiguous which @code{a} is being referred to with @samp{foo.a}.
14872 The compiler gives errors for such constructs.
14874 @opindex fms-extensions
14875 Unless @option{-fms-extensions} is used, the unnamed field must be a
14876 structure or union definition without a tag (for example, @samp{struct
14877 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
14878 also be a definition with a tag such as @samp{struct foo @{ int a;
14879 @};}, a reference to a previously defined structure or union such as
14880 @samp{struct foo;}, or a reference to a @code{typedef} name for a
14881 previously defined structure or union type.
14883 @opindex fplan9-extensions
14884 The option @option{-fplan9-extensions} enables
14885 @option{-fms-extensions} as well as two other extensions. First, a
14886 pointer to a structure is automatically converted to a pointer to an
14887 anonymous field for assignments and function calls. For example:
14890 struct s1 @{ int a; @};
14891 struct s2 @{ struct s1; @};
14892 extern void f1 (struct s1 *);
14893 void f2 (struct s2 *p) @{ f1 (p); @}
14897 In the call to @code{f1} inside @code{f2}, the pointer @code{p} is
14898 converted into a pointer to the anonymous field.
14900 Second, when the type of an anonymous field is a @code{typedef} for a
14901 @code{struct} or @code{union}, code may refer to the field using the
14902 name of the @code{typedef}.
14905 typedef struct @{ int a; @} s1;
14906 struct s2 @{ s1; @};
14907 s1 f1 (struct s2 *p) @{ return p->s1; @}
14910 These usages are only permitted when they are not ambiguous.
14913 @section Thread-Local Storage
14914 @cindex Thread-Local Storage
14915 @cindex @acronym{TLS}
14916 @cindex @code{__thread}
14918 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
14919 are allocated such that there is one instance of the variable per extant
14920 thread. The runtime model GCC uses to implement this originates
14921 in the IA-64 processor-specific ABI, but has since been migrated
14922 to other processors as well. It requires significant support from
14923 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
14924 system libraries (@file{libc.so} and @file{libpthread.so}), so it
14925 is not available everywhere.
14927 At the user level, the extension is visible with a new storage
14928 class keyword: @code{__thread}. For example:
14932 extern __thread struct state s;
14933 static __thread char *p;
14936 The @code{__thread} specifier may be used alone, with the @code{extern}
14937 or @code{static} specifiers, but with no other storage class specifier.
14938 When used with @code{extern} or @code{static}, @code{__thread} must appear
14939 immediately after the other storage class specifier.
14941 The @code{__thread} specifier may be applied to any global, file-scoped
14942 static, function-scoped static, or static data member of a class. It may
14943 not be applied to block-scoped automatic or non-static data member.
14945 When the address-of operator is applied to a thread-local variable, it is
14946 evaluated at run time and returns the address of the current thread's
14947 instance of that variable. An address so obtained may be used by any
14948 thread. When a thread terminates, any pointers to thread-local variables
14949 in that thread become invalid.
14951 No static initialization may refer to the address of a thread-local variable.
14953 In C++, if an initializer is present for a thread-local variable, it must
14954 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
14957 See @uref{http://www.akkadia.org/drepper/tls.pdf,
14958 ELF Handling For Thread-Local Storage} for a detailed explanation of
14959 the four thread-local storage addressing models, and how the runtime
14960 is expected to function.
14963 * C99 Thread-Local Edits::
14964 * C++98 Thread-Local Edits::
14967 @node C99 Thread-Local Edits
14968 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
14970 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
14971 that document the exact semantics of the language extension.
14975 @cite{5.1.2 Execution environments}
14977 Add new text after paragraph 1
14980 Within either execution environment, a @dfn{thread} is a flow of
14981 control within a program. It is implementation defined whether
14982 or not there may be more than one thread associated with a program.
14983 It is implementation defined how threads beyond the first are
14984 created, the name and type of the function called at thread
14985 startup, and how threads may be terminated. However, objects
14986 with thread storage duration shall be initialized before thread
14991 @cite{6.2.4 Storage durations of objects}
14993 Add new text before paragraph 3
14996 An object whose identifier is declared with the storage-class
14997 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
14998 Its lifetime is the entire execution of the thread, and its
14999 stored value is initialized only once, prior to thread startup.
15003 @cite{6.4.1 Keywords}
15005 Add @code{__thread}.
15008 @cite{6.7.1 Storage-class specifiers}
15010 Add @code{__thread} to the list of storage class specifiers in
15013 Change paragraph 2 to
15016 With the exception of @code{__thread}, at most one storage-class
15017 specifier may be given [@dots{}]. The @code{__thread} specifier may
15018 be used alone, or immediately following @code{extern} or
15022 Add new text after paragraph 6
15025 The declaration of an identifier for a variable that has
15026 block scope that specifies @code{__thread} shall also
15027 specify either @code{extern} or @code{static}.
15029 The @code{__thread} specifier shall be used only with
15034 @node C++98 Thread-Local Edits
15035 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
15037 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
15038 that document the exact semantics of the language extension.
15042 @b{[intro.execution]}
15044 New text after paragraph 4
15047 A @dfn{thread} is a flow of control within the abstract machine.
15048 It is implementation defined whether or not there may be more than
15052 New text after paragraph 7
15055 It is unspecified whether additional action must be taken to
15056 ensure when and whether side effects are visible to other threads.
15062 Add @code{__thread}.
15065 @b{[basic.start.main]}
15067 Add after paragraph 5
15070 The thread that begins execution at the @code{main} function is called
15071 the @dfn{main thread}. It is implementation defined how functions
15072 beginning threads other than the main thread are designated or typed.
15073 A function so designated, as well as the @code{main} function, is called
15074 a @dfn{thread startup function}. It is implementation defined what
15075 happens if a thread startup function returns. It is implementation
15076 defined what happens to other threads when any thread calls @code{exit}.
15080 @b{[basic.start.init]}
15082 Add after paragraph 4
15085 The storage for an object of thread storage duration shall be
15086 statically initialized before the first statement of the thread startup
15087 function. An object of thread storage duration shall not require
15088 dynamic initialization.
15092 @b{[basic.start.term]}
15094 Add after paragraph 3
15097 The type of an object with thread storage duration shall not have a
15098 non-trivial destructor, nor shall it be an array type whose elements
15099 (directly or indirectly) have non-trivial destructors.
15105 Add ``thread storage duration'' to the list in paragraph 1.
15110 Thread, static, and automatic storage durations are associated with
15111 objects introduced by declarations [@dots{}].
15114 Add @code{__thread} to the list of specifiers in paragraph 3.
15117 @b{[basic.stc.thread]}
15119 New section before @b{[basic.stc.static]}
15122 The keyword @code{__thread} applied to a non-local object gives the
15123 object thread storage duration.
15125 A local variable or class data member declared both @code{static}
15126 and @code{__thread} gives the variable or member thread storage
15131 @b{[basic.stc.static]}
15136 All objects that have neither thread storage duration, dynamic
15137 storage duration nor are local [@dots{}].
15143 Add @code{__thread} to the list in paragraph 1.
15148 With the exception of @code{__thread}, at most one
15149 @var{storage-class-specifier} shall appear in a given
15150 @var{decl-specifier-seq}. The @code{__thread} specifier may
15151 be used alone, or immediately following the @code{extern} or
15152 @code{static} specifiers. [@dots{}]
15155 Add after paragraph 5
15158 The @code{__thread} specifier can be applied only to the names of objects
15159 and to anonymous unions.
15165 Add after paragraph 6
15168 Non-@code{static} members shall not be @code{__thread}.
15172 @node Binary constants
15173 @section Binary constants using the @samp{0b} prefix
15174 @cindex Binary constants using the @samp{0b} prefix
15176 Integer constants can be written as binary constants, consisting of a
15177 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
15178 @samp{0B}. This is particularly useful in environments that operate a
15179 lot on the bit level (like microcontrollers).
15181 The following statements are identical:
15190 The type of these constants follows the same rules as for octal or
15191 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
15194 @node C++ Extensions
15195 @chapter Extensions to the C++ Language
15196 @cindex extensions, C++ language
15197 @cindex C++ language extensions
15199 The GNU compiler provides these extensions to the C++ language (and you
15200 can also use most of the C language extensions in your C++ programs). If you
15201 want to write code that checks whether these features are available, you can
15202 test for the GNU compiler the same way as for C programs: check for a
15203 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
15204 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
15205 Predefined Macros,cpp,The GNU C Preprocessor}).
15208 * C++ Volatiles:: What constitutes an access to a volatile object.
15209 * Restricted Pointers:: C99 restricted pointers and references.
15210 * Vague Linkage:: Where G++ puts inlines, vtables and such.
15211 * C++ Interface:: You can use a single C++ header file for both
15212 declarations and definitions.
15213 * Template Instantiation:: Methods for ensuring that exactly one copy of
15214 each needed template instantiation is emitted.
15215 * Bound member functions:: You can extract a function pointer to the
15216 method denoted by a @samp{->*} or @samp{.*} expression.
15217 * C++ Attributes:: Variable, function, and type attributes for C++ only.
15218 * Namespace Association:: Strong using-directives for namespace association.
15219 * Type Traits:: Compiler support for type traits
15220 * Java Exceptions:: Tweaking exception handling to work with Java.
15221 * Deprecated Features:: Things will disappear from G++.
15222 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
15225 @node C++ Volatiles
15226 @section When is a Volatile C++ Object Accessed?
15227 @cindex accessing volatiles
15228 @cindex volatile read
15229 @cindex volatile write
15230 @cindex volatile access
15232 The C++ standard differs from the C standard in its treatment of
15233 volatile objects. It fails to specify what constitutes a volatile
15234 access, except to say that C++ should behave in a similar manner to C
15235 with respect to volatiles, where possible. However, the different
15236 lvalueness of expressions between C and C++ complicate the behavior.
15237 G++ behaves the same as GCC for volatile access, @xref{C
15238 Extensions,,Volatiles}, for a description of GCC's behavior.
15240 The C and C++ language specifications differ when an object is
15241 accessed in a void context:
15244 volatile int *src = @var{somevalue};
15248 The C++ standard specifies that such expressions do not undergo lvalue
15249 to rvalue conversion, and that the type of the dereferenced object may
15250 be incomplete. The C++ standard does not specify explicitly that it
15251 is lvalue to rvalue conversion that is responsible for causing an
15252 access. There is reason to believe that it is, because otherwise
15253 certain simple expressions become undefined. However, because it
15254 would surprise most programmers, G++ treats dereferencing a pointer to
15255 volatile object of complete type as GCC would do for an equivalent
15256 type in C@. When the object has incomplete type, G++ issues a
15257 warning; if you wish to force an error, you must force a conversion to
15258 rvalue with, for instance, a static cast.
15260 When using a reference to volatile, G++ does not treat equivalent
15261 expressions as accesses to volatiles, but instead issues a warning that
15262 no volatile is accessed. The rationale for this is that otherwise it
15263 becomes difficult to determine where volatile access occur, and not
15264 possible to ignore the return value from functions returning volatile
15265 references. Again, if you wish to force a read, cast the reference to
15268 G++ implements the same behavior as GCC does when assigning to a
15269 volatile object---there is no reread of the assigned-to object, the
15270 assigned rvalue is reused. Note that in C++ assignment expressions
15271 are lvalues, and if used as an lvalue, the volatile object is
15272 referred to. For instance, @var{vref} refers to @var{vobj}, as
15273 expected, in the following example:
15277 volatile int &vref = vobj = @var{something};
15280 @node Restricted Pointers
15281 @section Restricting Pointer Aliasing
15282 @cindex restricted pointers
15283 @cindex restricted references
15284 @cindex restricted this pointer
15286 As with the C front end, G++ understands the C99 feature of restricted pointers,
15287 specified with the @code{__restrict__}, or @code{__restrict} type
15288 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
15289 language flag, @code{restrict} is not a keyword in C++.
15291 In addition to allowing restricted pointers, you can specify restricted
15292 references, which indicate that the reference is not aliased in the local
15296 void fn (int *__restrict__ rptr, int &__restrict__ rref)
15303 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
15304 @var{rref} refers to a (different) unaliased integer.
15306 You may also specify whether a member function's @var{this} pointer is
15307 unaliased by using @code{__restrict__} as a member function qualifier.
15310 void T::fn () __restrict__
15317 Within the body of @code{T::fn}, @var{this} has the effective
15318 definition @code{T *__restrict__ const this}. Notice that the
15319 interpretation of a @code{__restrict__} member function qualifier is
15320 different to that of @code{const} or @code{volatile} qualifier, in that it
15321 is applied to the pointer rather than the object. This is consistent with
15322 other compilers that implement restricted pointers.
15324 As with all outermost parameter qualifiers, @code{__restrict__} is
15325 ignored in function definition matching. This means you only need to
15326 specify @code{__restrict__} in a function definition, rather than
15327 in a function prototype as well.
15329 @node Vague Linkage
15330 @section Vague Linkage
15331 @cindex vague linkage
15333 There are several constructs in C++ that require space in the object
15334 file but are not clearly tied to a single translation unit. We say that
15335 these constructs have ``vague linkage''. Typically such constructs are
15336 emitted wherever they are needed, though sometimes we can be more
15340 @item Inline Functions
15341 Inline functions are typically defined in a header file which can be
15342 included in many different compilations. Hopefully they can usually be
15343 inlined, but sometimes an out-of-line copy is necessary, if the address
15344 of the function is taken or if inlining fails. In general, we emit an
15345 out-of-line copy in all translation units where one is needed. As an
15346 exception, we only emit inline virtual functions with the vtable, since
15347 it always requires a copy.
15349 Local static variables and string constants used in an inline function
15350 are also considered to have vague linkage, since they must be shared
15351 between all inlined and out-of-line instances of the function.
15355 C++ virtual functions are implemented in most compilers using a lookup
15356 table, known as a vtable. The vtable contains pointers to the virtual
15357 functions provided by a class, and each object of the class contains a
15358 pointer to its vtable (or vtables, in some multiple-inheritance
15359 situations). If the class declares any non-inline, non-pure virtual
15360 functions, the first one is chosen as the ``key method'' for the class,
15361 and the vtable is only emitted in the translation unit where the key
15364 @emph{Note:} If the chosen key method is later defined as inline, the
15365 vtable is still emitted in every translation unit that defines it.
15366 Make sure that any inline virtuals are declared inline in the class
15367 body, even if they are not defined there.
15369 @item @code{type_info} objects
15370 @cindex @code{type_info}
15372 C++ requires information about types to be written out in order to
15373 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
15374 For polymorphic classes (classes with virtual functions), the @samp{type_info}
15375 object is written out along with the vtable so that @samp{dynamic_cast}
15376 can determine the dynamic type of a class object at run time. For all
15377 other types, we write out the @samp{type_info} object when it is used: when
15378 applying @samp{typeid} to an expression, throwing an object, or
15379 referring to a type in a catch clause or exception specification.
15381 @item Template Instantiations
15382 Most everything in this section also applies to template instantiations,
15383 but there are other options as well.
15384 @xref{Template Instantiation,,Where's the Template?}.
15388 When used with GNU ld version 2.8 or later on an ELF system such as
15389 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
15390 these constructs will be discarded at link time. This is known as
15393 On targets that don't support COMDAT, but do support weak symbols, GCC
15394 uses them. This way one copy overrides all the others, but
15395 the unused copies still take up space in the executable.
15397 For targets that do not support either COMDAT or weak symbols,
15398 most entities with vague linkage are emitted as local symbols to
15399 avoid duplicate definition errors from the linker. This does not happen
15400 for local statics in inlines, however, as having multiple copies
15401 almost certainly breaks things.
15403 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
15404 another way to control placement of these constructs.
15406 @node C++ Interface
15407 @section #pragma interface and implementation
15409 @cindex interface and implementation headers, C++
15410 @cindex C++ interface and implementation headers
15411 @cindex pragmas, interface and implementation
15413 @code{#pragma interface} and @code{#pragma implementation} provide the
15414 user with a way of explicitly directing the compiler to emit entities
15415 with vague linkage (and debugging information) in a particular
15418 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
15419 most cases, because of COMDAT support and the ``key method'' heuristic
15420 mentioned in @ref{Vague Linkage}. Using them can actually cause your
15421 program to grow due to unnecessary out-of-line copies of inline
15422 functions. Currently (3.4) the only benefit of these
15423 @code{#pragma}s is reduced duplication of debugging information, and
15424 that should be addressed soon on DWARF 2 targets with the use of
15428 @item #pragma interface
15429 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
15430 @kindex #pragma interface
15431 Use this directive in @emph{header files} that define object classes, to save
15432 space in most of the object files that use those classes. Normally,
15433 local copies of certain information (backup copies of inline member
15434 functions, debugging information, and the internal tables that implement
15435 virtual functions) must be kept in each object file that includes class
15436 definitions. You can use this pragma to avoid such duplication. When a
15437 header file containing @samp{#pragma interface} is included in a
15438 compilation, this auxiliary information is not generated (unless
15439 the main input source file itself uses @samp{#pragma implementation}).
15440 Instead, the object files contain references to be resolved at link
15443 The second form of this directive is useful for the case where you have
15444 multiple headers with the same name in different directories. If you
15445 use this form, you must specify the same string to @samp{#pragma
15448 @item #pragma implementation
15449 @itemx #pragma implementation "@var{objects}.h"
15450 @kindex #pragma implementation
15451 Use this pragma in a @emph{main input file}, when you want full output from
15452 included header files to be generated (and made globally visible). The
15453 included header file, in turn, should use @samp{#pragma interface}.
15454 Backup copies of inline member functions, debugging information, and the
15455 internal tables used to implement virtual functions are all generated in
15456 implementation files.
15458 @cindex implied @code{#pragma implementation}
15459 @cindex @code{#pragma implementation}, implied
15460 @cindex naming convention, implementation headers
15461 If you use @samp{#pragma implementation} with no argument, it applies to
15462 an include file with the same basename@footnote{A file's @dfn{basename}
15463 is the name stripped of all leading path information and of trailing
15464 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
15465 file. For example, in @file{allclass.cc}, giving just
15466 @samp{#pragma implementation}
15467 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
15469 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
15470 an implementation file whenever you would include it from
15471 @file{allclass.cc} even if you never specified @samp{#pragma
15472 implementation}. This was deemed to be more trouble than it was worth,
15473 however, and disabled.
15475 Use the string argument if you want a single implementation file to
15476 include code from multiple header files. (You must also use
15477 @samp{#include} to include the header file; @samp{#pragma
15478 implementation} only specifies how to use the file---it doesn't actually
15481 There is no way to split up the contents of a single header file into
15482 multiple implementation files.
15485 @cindex inlining and C++ pragmas
15486 @cindex C++ pragmas, effect on inlining
15487 @cindex pragmas in C++, effect on inlining
15488 @samp{#pragma implementation} and @samp{#pragma interface} also have an
15489 effect on function inlining.
15491 If you define a class in a header file marked with @samp{#pragma
15492 interface}, the effect on an inline function defined in that class is
15493 similar to an explicit @code{extern} declaration---the compiler emits
15494 no code at all to define an independent version of the function. Its
15495 definition is used only for inlining with its callers.
15497 @opindex fno-implement-inlines
15498 Conversely, when you include the same header file in a main source file
15499 that declares it as @samp{#pragma implementation}, the compiler emits
15500 code for the function itself; this defines a version of the function
15501 that can be found via pointers (or by callers compiled without
15502 inlining). If all calls to the function can be inlined, you can avoid
15503 emitting the function by compiling with @option{-fno-implement-inlines}.
15504 If any calls are not inlined, you will get linker errors.
15506 @node Template Instantiation
15507 @section Where's the Template?
15508 @cindex template instantiation
15510 C++ templates are the first language feature to require more
15511 intelligence from the environment than one usually finds on a UNIX
15512 system. Somehow the compiler and linker have to make sure that each
15513 template instance occurs exactly once in the executable if it is needed,
15514 and not at all otherwise. There are two basic approaches to this
15515 problem, which are referred to as the Borland model and the Cfront model.
15518 @item Borland model
15519 Borland C++ solved the template instantiation problem by adding the code
15520 equivalent of common blocks to their linker; the compiler emits template
15521 instances in each translation unit that uses them, and the linker
15522 collapses them together. The advantage of this model is that the linker
15523 only has to consider the object files themselves; there is no external
15524 complexity to worry about. This disadvantage is that compilation time
15525 is increased because the template code is being compiled repeatedly.
15526 Code written for this model tends to include definitions of all
15527 templates in the header file, since they must be seen to be
15531 The AT&T C++ translator, Cfront, solved the template instantiation
15532 problem by creating the notion of a template repository, an
15533 automatically maintained place where template instances are stored. A
15534 more modern version of the repository works as follows: As individual
15535 object files are built, the compiler places any template definitions and
15536 instantiations encountered in the repository. At link time, the link
15537 wrapper adds in the objects in the repository and compiles any needed
15538 instances that were not previously emitted. The advantages of this
15539 model are more optimal compilation speed and the ability to use the
15540 system linker; to implement the Borland model a compiler vendor also
15541 needs to replace the linker. The disadvantages are vastly increased
15542 complexity, and thus potential for error; for some code this can be
15543 just as transparent, but in practice it can been very difficult to build
15544 multiple programs in one directory and one program in multiple
15545 directories. Code written for this model tends to separate definitions
15546 of non-inline member templates into a separate file, which should be
15547 compiled separately.
15550 When used with GNU ld version 2.8 or later on an ELF system such as
15551 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
15552 Borland model. On other systems, G++ implements neither automatic
15555 You have the following options for dealing with template instantiations:
15560 Compile your template-using code with @option{-frepo}. The compiler
15561 generates files with the extension @samp{.rpo} listing all of the
15562 template instantiations used in the corresponding object files that
15563 could be instantiated there; the link wrapper, @samp{collect2},
15564 then updates the @samp{.rpo} files to tell the compiler where to place
15565 those instantiations and rebuild any affected object files. The
15566 link-time overhead is negligible after the first pass, as the compiler
15567 continues to place the instantiations in the same files.
15569 This is your best option for application code written for the Borland
15570 model, as it just works. Code written for the Cfront model
15571 needs to be modified so that the template definitions are available at
15572 one or more points of instantiation; usually this is as simple as adding
15573 @code{#include <tmethods.cc>} to the end of each template header.
15575 For library code, if you want the library to provide all of the template
15576 instantiations it needs, just try to link all of its object files
15577 together; the link will fail, but cause the instantiations to be
15578 generated as a side effect. Be warned, however, that this may cause
15579 conflicts if multiple libraries try to provide the same instantiations.
15580 For greater control, use explicit instantiation as described in the next
15584 @opindex fno-implicit-templates
15585 Compile your code with @option{-fno-implicit-templates} to disable the
15586 implicit generation of template instances, and explicitly instantiate
15587 all the ones you use. This approach requires more knowledge of exactly
15588 which instances you need than do the others, but it's less
15589 mysterious and allows greater control. You can scatter the explicit
15590 instantiations throughout your program, perhaps putting them in the
15591 translation units where the instances are used or the translation units
15592 that define the templates themselves; you can put all of the explicit
15593 instantiations you need into one big file; or you can create small files
15600 template class Foo<int>;
15601 template ostream& operator <<
15602 (ostream&, const Foo<int>&);
15606 for each of the instances you need, and create a template instantiation
15607 library from those.
15609 If you are using Cfront-model code, you can probably get away with not
15610 using @option{-fno-implicit-templates} when compiling files that don't
15611 @samp{#include} the member template definitions.
15613 If you use one big file to do the instantiations, you may want to
15614 compile it without @option{-fno-implicit-templates} so you get all of the
15615 instances required by your explicit instantiations (but not by any
15616 other files) without having to specify them as well.
15618 The ISO C++ 2011 standard allows forward declaration of explicit
15619 instantiations (with @code{extern}). G++ supports explicit instantiation
15620 declarations in C++98 mode and has extended the template instantiation
15621 syntax to support instantiation of the compiler support data for a
15622 template class (i.e.@: the vtable) without instantiating any of its
15623 members (with @code{inline}), and instantiation of only the static data
15624 members of a template class, without the support data or member
15625 functions (with (@code{static}):
15628 extern template int max (int, int);
15629 inline template class Foo<int>;
15630 static template class Foo<int>;
15634 Do nothing. Pretend G++ does implement automatic instantiation
15635 management. Code written for the Borland model works fine, but
15636 each translation unit contains instances of each of the templates it
15637 uses. In a large program, this can lead to an unacceptable amount of code
15641 @node Bound member functions
15642 @section Extracting the function pointer from a bound pointer to member function
15644 @cindex pointer to member function
15645 @cindex bound pointer to member function
15647 In C++, pointer to member functions (PMFs) are implemented using a wide
15648 pointer of sorts to handle all the possible call mechanisms; the PMF
15649 needs to store information about how to adjust the @samp{this} pointer,
15650 and if the function pointed to is virtual, where to find the vtable, and
15651 where in the vtable to look for the member function. If you are using
15652 PMFs in an inner loop, you should really reconsider that decision. If
15653 that is not an option, you can extract the pointer to the function that
15654 would be called for a given object/PMF pair and call it directly inside
15655 the inner loop, to save a bit of time.
15657 Note that you still pay the penalty for the call through a
15658 function pointer; on most modern architectures, such a call defeats the
15659 branch prediction features of the CPU@. This is also true of normal
15660 virtual function calls.
15662 The syntax for this extension is
15666 extern int (A::*fp)();
15667 typedef int (*fptr)(A *);
15669 fptr p = (fptr)(a.*fp);
15672 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
15673 no object is needed to obtain the address of the function. They can be
15674 converted to function pointers directly:
15677 fptr p1 = (fptr)(&A::foo);
15680 @opindex Wno-pmf-conversions
15681 You must specify @option{-Wno-pmf-conversions} to use this extension.
15683 @node C++ Attributes
15684 @section C++-Specific Variable, Function, and Type Attributes
15686 Some attributes only make sense for C++ programs.
15689 @item abi_tag ("@var{tag}", ...)
15690 @cindex @code{abi_tag} attribute
15691 The @code{abi_tag} attribute can be applied to a function or class
15692 declaration. It modifies the mangled name of the function or class to
15693 incorporate the tag name, in order to distinguish the function or
15694 class from an earlier version with a different ABI; perhaps the class
15695 has changed size, or the function has a different return type that is
15696 not encoded in the mangled name.
15698 The argument can be a list of strings of arbitrary length. The
15699 strings are sorted on output, so the order of the list is
15702 A redeclaration of a function or class must not add new ABI tags,
15703 since doing so would change the mangled name.
15705 The @option{-Wabi-tag} flag enables a warning about a class which does
15706 not have all the ABI tags used by its subobjects and virtual functions; for users with code
15707 that needs to coexist with an earlier ABI, using this option can help
15708 to find all affected types that need to be tagged.
15710 @item init_priority (@var{priority})
15711 @cindex @code{init_priority} attribute
15714 In Standard C++, objects defined at namespace scope are guaranteed to be
15715 initialized in an order in strict accordance with that of their definitions
15716 @emph{in a given translation unit}. No guarantee is made for initializations
15717 across translation units. However, GNU C++ allows users to control the
15718 order of initialization of objects defined at namespace scope with the
15719 @code{init_priority} attribute by specifying a relative @var{priority},
15720 a constant integral expression currently bounded between 101 and 65535
15721 inclusive. Lower numbers indicate a higher priority.
15723 In the following example, @code{A} would normally be created before
15724 @code{B}, but the @code{init_priority} attribute reverses that order:
15727 Some_Class A __attribute__ ((init_priority (2000)));
15728 Some_Class B __attribute__ ((init_priority (543)));
15732 Note that the particular values of @var{priority} do not matter; only their
15735 @item java_interface
15736 @cindex @code{java_interface} attribute
15738 This type attribute informs C++ that the class is a Java interface. It may
15739 only be applied to classes declared within an @code{extern "Java"} block.
15740 Calls to methods declared in this interface are dispatched using GCJ's
15741 interface table mechanism, instead of regular virtual table dispatch.
15745 See also @ref{Namespace Association}.
15747 @node Namespace Association
15748 @section Namespace Association
15750 @strong{Caution:} The semantics of this extension are equivalent
15751 to C++ 2011 inline namespaces. Users should use inline namespaces
15752 instead as this extension will be removed in future versions of G++.
15754 A using-directive with @code{__attribute ((strong))} is stronger
15755 than a normal using-directive in two ways:
15759 Templates from the used namespace can be specialized and explicitly
15760 instantiated as though they were members of the using namespace.
15763 The using namespace is considered an associated namespace of all
15764 templates in the used namespace for purposes of argument-dependent
15768 The used namespace must be nested within the using namespace so that
15769 normal unqualified lookup works properly.
15771 This is useful for composing a namespace transparently from
15772 implementation namespaces. For example:
15777 template <class T> struct A @{ @};
15779 using namespace debug __attribute ((__strong__));
15780 template <> struct A<int> @{ @}; // @r{ok to specialize}
15782 template <class T> void f (A<T>);
15787 f (std::A<float>()); // @r{lookup finds} std::f
15793 @section Type Traits
15795 The C++ front end implements syntactic extensions that allow
15796 compile-time determination of
15797 various characteristics of a type (or of a
15801 @item __has_nothrow_assign (type)
15802 If @code{type} is const qualified or is a reference type then the trait is
15803 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
15804 is true, else if @code{type} is a cv class or union type with copy assignment
15805 operators that are known not to throw an exception then the trait is true,
15806 else it is false. Requires: @code{type} shall be a complete type,
15807 (possibly cv-qualified) @code{void}, or an array of unknown bound.
15809 @item __has_nothrow_copy (type)
15810 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
15811 @code{type} is a cv class or union type with copy constructors that
15812 are known not to throw an exception then the trait is true, else it is false.
15813 Requires: @code{type} shall be a complete type, (possibly cv-qualified)
15814 @code{void}, or an array of unknown bound.
15816 @item __has_nothrow_constructor (type)
15817 If @code{__has_trivial_constructor (type)} is true then the trait is
15818 true, else if @code{type} is a cv class or union type (or array
15819 thereof) with a default constructor that is known not to throw an
15820 exception then the trait is true, else it is false. Requires:
15821 @code{type} shall be a complete type, (possibly cv-qualified)
15822 @code{void}, or an array of unknown bound.
15824 @item __has_trivial_assign (type)
15825 If @code{type} is const qualified or is a reference type then the trait is
15826 false. Otherwise if @code{__is_pod (type)} is true then the trait is
15827 true, else if @code{type} is a cv class or union type with a trivial
15828 copy assignment ([class.copy]) then the trait is true, else it is
15829 false. Requires: @code{type} shall be a complete type, (possibly
15830 cv-qualified) @code{void}, or an array of unknown bound.
15832 @item __has_trivial_copy (type)
15833 If @code{__is_pod (type)} is true or @code{type} is a reference type
15834 then the trait is true, else if @code{type} is a cv class or union type
15835 with a trivial copy constructor ([class.copy]) then the trait
15836 is true, else it is false. Requires: @code{type} shall be a complete
15837 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15839 @item __has_trivial_constructor (type)
15840 If @code{__is_pod (type)} is true then the trait is true, else if
15841 @code{type} is a cv class or union type (or array thereof) with a
15842 trivial default constructor ([class.ctor]) then the trait is true,
15843 else it is false. Requires: @code{type} shall be a complete
15844 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15846 @item __has_trivial_destructor (type)
15847 If @code{__is_pod (type)} is true or @code{type} is a reference type then
15848 the trait is true, else if @code{type} is a cv class or union type (or
15849 array thereof) with a trivial destructor ([class.dtor]) then the trait
15850 is true, else it is false. Requires: @code{type} shall be a complete
15851 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15853 @item __has_virtual_destructor (type)
15854 If @code{type} is a class type with a virtual destructor
15855 ([class.dtor]) then the trait is true, else it is false. Requires:
15856 @code{type} shall be a complete type, (possibly cv-qualified)
15857 @code{void}, or an array of unknown bound.
15859 @item __is_abstract (type)
15860 If @code{type} is an abstract class ([class.abstract]) then the trait
15861 is true, else it is false. Requires: @code{type} shall be a complete
15862 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15864 @item __is_base_of (base_type, derived_type)
15865 If @code{base_type} is a base class of @code{derived_type}
15866 ([class.derived]) then the trait is true, otherwise it is false.
15867 Top-level cv qualifications of @code{base_type} and
15868 @code{derived_type} are ignored. For the purposes of this trait, a
15869 class type is considered is own base. Requires: if @code{__is_class
15870 (base_type)} and @code{__is_class (derived_type)} are true and
15871 @code{base_type} and @code{derived_type} are not the same type
15872 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
15873 type. Diagnostic is produced if this requirement is not met.
15875 @item __is_class (type)
15876 If @code{type} is a cv class type, and not a union type
15877 ([basic.compound]) the trait is true, else it is false.
15879 @item __is_empty (type)
15880 If @code{__is_class (type)} is false then the trait is false.
15881 Otherwise @code{type} is considered empty if and only if: @code{type}
15882 has no non-static data members, or all non-static data members, if
15883 any, are bit-fields of length 0, and @code{type} has no virtual
15884 members, and @code{type} has no virtual base classes, and @code{type}
15885 has no base classes @code{base_type} for which
15886 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
15887 be a complete type, (possibly cv-qualified) @code{void}, or an array
15890 @item __is_enum (type)
15891 If @code{type} is a cv enumeration type ([basic.compound]) the trait is
15892 true, else it is false.
15894 @item __is_literal_type (type)
15895 If @code{type} is a literal type ([basic.types]) the trait is
15896 true, else it is false. Requires: @code{type} shall be a complete type,
15897 (possibly cv-qualified) @code{void}, or an array of unknown bound.
15899 @item __is_pod (type)
15900 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
15901 else it is false. Requires: @code{type} shall be a complete type,
15902 (possibly cv-qualified) @code{void}, or an array of unknown bound.
15904 @item __is_polymorphic (type)
15905 If @code{type} is a polymorphic class ([class.virtual]) then the trait
15906 is true, else it is false. Requires: @code{type} shall be a complete
15907 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15909 @item __is_standard_layout (type)
15910 If @code{type} is a standard-layout type ([basic.types]) the trait is
15911 true, else it is false. Requires: @code{type} shall be a complete
15912 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15914 @item __is_trivial (type)
15915 If @code{type} is a trivial type ([basic.types]) the trait is
15916 true, else it is false. Requires: @code{type} shall be a complete
15917 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
15919 @item __is_union (type)
15920 If @code{type} is a cv union type ([basic.compound]) the trait is
15921 true, else it is false.
15923 @item __underlying_type (type)
15924 The underlying type of @code{type}. Requires: @code{type} shall be
15925 an enumeration type ([dcl.enum]).
15929 @node Java Exceptions
15930 @section Java Exceptions
15932 The Java language uses a slightly different exception handling model
15933 from C++. Normally, GNU C++ automatically detects when you are
15934 writing C++ code that uses Java exceptions, and handle them
15935 appropriately. However, if C++ code only needs to execute destructors
15936 when Java exceptions are thrown through it, GCC guesses incorrectly.
15937 Sample problematic code is:
15940 struct S @{ ~S(); @};
15941 extern void bar(); // @r{is written in Java, and may throw exceptions}
15950 The usual effect of an incorrect guess is a link failure, complaining of
15951 a missing routine called @samp{__gxx_personality_v0}.
15953 You can inform the compiler that Java exceptions are to be used in a
15954 translation unit, irrespective of what it might think, by writing
15955 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
15956 @samp{#pragma} must appear before any functions that throw or catch
15957 exceptions, or run destructors when exceptions are thrown through them.
15959 You cannot mix Java and C++ exceptions in the same translation unit. It
15960 is believed to be safe to throw a C++ exception from one file through
15961 another file compiled for the Java exception model, or vice versa, but
15962 there may be bugs in this area.
15964 @node Deprecated Features
15965 @section Deprecated Features
15967 In the past, the GNU C++ compiler was extended to experiment with new
15968 features, at a time when the C++ language was still evolving. Now that
15969 the C++ standard is complete, some of those features are superseded by
15970 superior alternatives. Using the old features might cause a warning in
15971 some cases that the feature will be dropped in the future. In other
15972 cases, the feature might be gone already.
15974 While the list below is not exhaustive, it documents some of the options
15975 that are now deprecated:
15978 @item -fexternal-templates
15979 @itemx -falt-external-templates
15980 These are two of the many ways for G++ to implement template
15981 instantiation. @xref{Template Instantiation}. The C++ standard clearly
15982 defines how template definitions have to be organized across
15983 implementation units. G++ has an implicit instantiation mechanism that
15984 should work just fine for standard-conforming code.
15986 @item -fstrict-prototype
15987 @itemx -fno-strict-prototype
15988 Previously it was possible to use an empty prototype parameter list to
15989 indicate an unspecified number of parameters (like C), rather than no
15990 parameters, as C++ demands. This feature has been removed, except where
15991 it is required for backwards compatibility. @xref{Backwards Compatibility}.
15994 G++ allows a virtual function returning @samp{void *} to be overridden
15995 by one returning a different pointer type. This extension to the
15996 covariant return type rules is now deprecated and will be removed from a
15999 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
16000 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
16001 and are now removed from G++. Code using these operators should be
16002 modified to use @code{std::min} and @code{std::max} instead.
16004 The named return value extension has been deprecated, and is now
16007 The use of initializer lists with new expressions has been deprecated,
16008 and is now removed from G++.
16010 Floating and complex non-type template parameters have been deprecated,
16011 and are now removed from G++.
16013 The implicit typename extension has been deprecated and is now
16016 The use of default arguments in function pointers, function typedefs
16017 and other places where they are not permitted by the standard is
16018 deprecated and will be removed from a future version of G++.
16020 G++ allows floating-point literals to appear in integral constant expressions,
16021 e.g.@: @samp{ enum E @{ e = int(2.2 * 3.7) @} }
16022 This extension is deprecated and will be removed from a future version.
16024 G++ allows static data members of const floating-point type to be declared
16025 with an initializer in a class definition. The standard only allows
16026 initializers for static members of const integral types and const
16027 enumeration types so this extension has been deprecated and will be removed
16028 from a future version.
16030 @node Backwards Compatibility
16031 @section Backwards Compatibility
16032 @cindex Backwards Compatibility
16033 @cindex ARM [Annotated C++ Reference Manual]
16035 Now that there is a definitive ISO standard C++, G++ has a specification
16036 to adhere to. The C++ language evolved over time, and features that
16037 used to be acceptable in previous drafts of the standard, such as the ARM
16038 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
16039 compilation of C++ written to such drafts, G++ contains some backwards
16040 compatibilities. @emph{All such backwards compatibility features are
16041 liable to disappear in future versions of G++.} They should be considered
16042 deprecated. @xref{Deprecated Features}.
16046 If a variable is declared at for scope, it used to remain in scope until
16047 the end of the scope that contained the for statement (rather than just
16048 within the for scope). G++ retains this, but issues a warning, if such a
16049 variable is accessed outside the for scope.
16051 @item Implicit C language
16052 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
16053 scope to set the language. On such systems, all header files are
16054 implicitly scoped inside a C language scope. Also, an empty prototype
16055 @code{()} is treated as an unspecified number of arguments, rather
16056 than no arguments, as C++ demands.