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 * __int128:: 128-bit integers---@code{__int128}.
34 * Long Long:: Double-word integers---@code{long long int}.
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 * Empty Structures:: Structures with no members.
44 * Variable Length:: Arrays whose length is computed at run time.
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 * Case Ranges:: `case 1 ... 9' and such.
54 * Cast to Union:: Casting to union type from any member of the union.
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 * x86 specific memory model extensions for transactional memory:: x86 memory models.
83 * Object Size Checking:: Built-in functions for limited buffer overflow
85 * Cilk Plus Builtins:: Built-in functions for the Cilk Plus language extension.
86 * Other Builtins:: Other built-in functions.
87 * Target Builtins:: Built-in functions specific to particular targets.
88 * Target Format Checks:: Format checks specific to particular targets.
89 * Pragmas:: Pragmas accepted by GCC.
90 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
91 * Thread-Local:: Per-thread variables.
92 * Binary constants:: Binary constants using the @samp{0b} prefix.
96 @section Statements and Declarations in Expressions
97 @cindex statements inside expressions
98 @cindex declarations inside expressions
99 @cindex expressions containing statements
100 @cindex macros, statements in expressions
102 @c the above section title wrapped and causes an underfull hbox.. i
103 @c changed it from "within" to "in". --mew 4feb93
104 A compound statement enclosed in parentheses may appear as an expression
105 in GNU C@. This allows you to use loops, switches, and local variables
106 within an expression.
108 Recall that a compound statement is a sequence of statements surrounded
109 by braces; in this construct, parentheses go around the braces. For
113 (@{ int y = foo (); int z;
120 is a valid (though slightly more complex than necessary) expression
121 for the absolute value of @code{foo ()}.
123 The last thing in the compound statement should be an expression
124 followed by a semicolon; the value of this subexpression serves as the
125 value of the entire construct. (If you use some other kind of statement
126 last within the braces, the construct has type @code{void}, and thus
127 effectively no value.)
129 This feature is especially useful in making macro definitions ``safe'' (so
130 that they evaluate each operand exactly once). For example, the
131 ``maximum'' function is commonly defined as a macro in standard C as
135 #define max(a,b) ((a) > (b) ? (a) : (b))
139 @cindex side effects, macro argument
140 But this definition computes either @var{a} or @var{b} twice, with bad
141 results if the operand has side effects. In GNU C, if you know the
142 type of the operands (here taken as @code{int}), you can define
143 the macro safely as follows:
146 #define maxint(a,b) \
147 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
150 Embedded statements are not allowed in constant expressions, such as
151 the value of an enumeration constant, the width of a bit-field, or
152 the initial value of a static variable.
154 If you don't know the type of the operand, you can still do this, but you
155 must use @code{typeof} (@pxref{Typeof}).
157 In G++, the result value of a statement expression undergoes array and
158 function pointer decay, and is returned by value to the enclosing
159 expression. For instance, if @code{A} is a class, then
168 constructs a temporary @code{A} object to hold the result of the
169 statement expression, and that is used to invoke @code{Foo}.
170 Therefore the @code{this} pointer observed by @code{Foo} is not the
173 In a statement expression, any temporaries created within a statement
174 are destroyed at that statement's end. This makes statement
175 expressions inside macros slightly different from function calls. In
176 the latter case temporaries introduced during argument evaluation are
177 destroyed at the end of the statement that includes the function
178 call. In the statement expression case they are destroyed during
179 the statement expression. For instance,
182 #define macro(a) (@{__typeof__(a) b = (a); b + 3; @})
183 template<typename T> T function(T a) @{ T b = a; return b + 3; @}
193 has different places where temporaries are destroyed. For the
194 @code{macro} case, the temporary @code{X} is destroyed just after
195 the initialization of @code{b}. In the @code{function} case that
196 temporary is destroyed when the function returns.
198 These considerations mean that it is probably a bad idea to use
199 statement expressions of this form in header files that are designed to
200 work with C++. (Note that some versions of the GNU C Library contained
201 header files using statement expressions that lead to precisely this
204 Jumping into a statement expression with @code{goto} or using a
205 @code{switch} statement outside the statement expression with a
206 @code{case} or @code{default} label inside the statement expression is
207 not permitted. Jumping into a statement expression with a computed
208 @code{goto} (@pxref{Labels as Values}) has undefined behavior.
209 Jumping out of a statement expression is permitted, but if the
210 statement expression is part of a larger expression then it is
211 unspecified which other subexpressions of that expression have been
212 evaluated except where the language definition requires certain
213 subexpressions to be evaluated before or after the statement
214 expression. In any case, as with a function call, the evaluation of a
215 statement expression is not interleaved with the evaluation of other
216 parts of the containing expression. For example,
219 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz();
223 calls @code{foo} and @code{bar1} and does not call @code{baz} but
224 may or may not call @code{bar2}. If @code{bar2} is called, it is
225 called after @code{foo} and before @code{bar1}.
228 @section Locally Declared Labels
230 @cindex macros, local labels
232 GCC allows you to declare @dfn{local labels} in any nested block
233 scope. A local label is just like an ordinary label, but you can
234 only reference it (with a @code{goto} statement, or by taking its
235 address) within the block in which it is declared.
237 A local label declaration looks like this:
240 __label__ @var{label};
247 __label__ @var{label1}, @var{label2}, /* @r{@dots{}} */;
250 Local label declarations must come at the beginning of the block,
251 before any ordinary declarations or statements.
253 The label declaration defines the label @emph{name}, but does not define
254 the label itself. You must do this in the usual way, with
255 @code{@var{label}:}, within the statements of the statement expression.
257 The local label feature is useful for complex macros. If a macro
258 contains nested loops, a @code{goto} can be useful for breaking out of
259 them. However, an ordinary label whose scope is the whole function
260 cannot be used: if the macro can be expanded several times in one
261 function, the label is multiply defined in that function. A
262 local label avoids this problem. For example:
265 #define SEARCH(value, array, target) \
268 typeof (target) _SEARCH_target = (target); \
269 typeof (*(array)) *_SEARCH_array = (array); \
272 for (i = 0; i < max; i++) \
273 for (j = 0; j < max; j++) \
274 if (_SEARCH_array[i][j] == _SEARCH_target) \
275 @{ (value) = i; goto found; @} \
281 This could also be written using a statement expression:
284 #define SEARCH(array, target) \
287 typeof (target) _SEARCH_target = (target); \
288 typeof (*(array)) *_SEARCH_array = (array); \
291 for (i = 0; i < max; i++) \
292 for (j = 0; j < max; j++) \
293 if (_SEARCH_array[i][j] == _SEARCH_target) \
294 @{ value = i; goto found; @} \
301 Local label declarations also make the labels they declare visible to
302 nested functions, if there are any. @xref{Nested Functions}, for details.
304 @node Labels as Values
305 @section Labels as Values
306 @cindex labels as values
307 @cindex computed gotos
308 @cindex goto with computed label
309 @cindex address of a label
311 You can get the address of a label defined in the current function
312 (or a containing function) with the unary operator @samp{&&}. The
313 value has type @code{void *}. This value is a constant and can be used
314 wherever a constant of that type is valid. For example:
322 To use these values, you need to be able to jump to one. This is done
323 with the computed goto statement@footnote{The analogous feature in
324 Fortran is called an assigned goto, but that name seems inappropriate in
325 C, where one can do more than simply store label addresses in label
326 variables.}, @code{goto *@var{exp};}. For example,
333 Any expression of type @code{void *} is allowed.
335 One way of using these constants is in initializing a static array that
336 serves as a jump table:
339 static void *array[] = @{ &&foo, &&bar, &&hack @};
343 Then you can select a label with indexing, like this:
350 Note that this does not check whether the subscript is in bounds---array
351 indexing in C never does that.
353 Such an array of label values serves a purpose much like that of the
354 @code{switch} statement. The @code{switch} statement is cleaner, so
355 use that rather than an array unless the problem does not fit a
356 @code{switch} statement very well.
358 Another use of label values is in an interpreter for threaded code.
359 The labels within the interpreter function can be stored in the
360 threaded code for super-fast dispatching.
362 You may not use this mechanism to jump to code in a different function.
363 If you do that, totally unpredictable things happen. The best way to
364 avoid this is to store the label address only in automatic variables and
365 never pass it as an argument.
367 An alternate way to write the above example is
370 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
372 goto *(&&foo + array[i]);
376 This is more friendly to code living in shared libraries, as it reduces
377 the number of dynamic relocations that are needed, and by consequence,
378 allows the data to be read-only.
380 The @code{&&foo} expressions for the same label might have different
381 values if the containing function is inlined or cloned. If a program
382 relies on them being always the same,
383 @code{__attribute__((__noinline__,__noclone__))} should be used to
384 prevent inlining and cloning. If @code{&&foo} is used in a static
385 variable initializer, inlining and cloning is forbidden.
387 @node Nested Functions
388 @section Nested Functions
389 @cindex nested functions
390 @cindex downward funargs
393 A @dfn{nested function} is a function defined inside another function.
394 Nested functions are supported as an extension in GNU C, but are not
395 supported by GNU C++.
397 The nested function's name is local to the block where it is defined.
398 For example, here we define a nested function named @code{square}, and
403 foo (double a, double b)
405 double square (double z) @{ return z * z; @}
407 return square (a) + square (b);
412 The nested function can access all the variables of the containing
413 function that are visible at the point of its definition. This is
414 called @dfn{lexical scoping}. For example, here we show a nested
415 function which uses an inherited variable named @code{offset}:
419 bar (int *array, int offset, int size)
421 int access (int *array, int index)
422 @{ return array[index + offset]; @}
425 for (i = 0; i < size; i++)
426 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
431 Nested function definitions are permitted within functions in the places
432 where variable definitions are allowed; that is, in any block, mixed
433 with the other declarations and statements in the block.
435 It is possible to call the nested function from outside the scope of its
436 name by storing its address or passing the address to another function:
439 hack (int *array, int size)
441 void store (int index, int value)
442 @{ array[index] = value; @}
444 intermediate (store, size);
448 Here, the function @code{intermediate} receives the address of
449 @code{store} as an argument. If @code{intermediate} calls @code{store},
450 the arguments given to @code{store} are used to store into @code{array}.
451 But this technique works only so long as the containing function
452 (@code{hack}, in this example) does not exit.
454 If you try to call the nested function through its address after the
455 containing function exits, all hell breaks loose. If you try
456 to call it after a containing scope level exits, and if it refers
457 to some of the variables that are no longer in scope, you may be lucky,
458 but it's not wise to take the risk. If, however, the nested function
459 does not refer to anything that has gone out of scope, you should be
462 GCC implements taking the address of a nested function using a technique
463 called @dfn{trampolines}. This technique was described in
464 @cite{Lexical Closures for C++} (Thomas M. Breuel, USENIX
465 C++ Conference Proceedings, October 17-21, 1988).
467 A nested function can jump to a label inherited from a containing
468 function, provided the label is explicitly declared in the containing
469 function (@pxref{Local Labels}). Such a jump returns instantly to the
470 containing function, exiting the nested function that did the
471 @code{goto} and any intermediate functions as well. Here is an example:
475 bar (int *array, int offset, int size)
478 int access (int *array, int index)
482 return array[index + offset];
486 for (i = 0; i < size; i++)
487 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */
491 /* @r{Control comes here from @code{access}
492 if it detects an error.} */
499 A nested function always has no linkage. Declaring one with
500 @code{extern} or @code{static} is erroneous. If you need to declare the nested function
501 before its definition, use @code{auto} (which is otherwise meaningless
502 for function declarations).
505 bar (int *array, int offset, int size)
508 auto int access (int *, int);
510 int access (int *array, int index)
514 return array[index + offset];
520 @node Constructing Calls
521 @section Constructing Function Calls
522 @cindex constructing calls
523 @cindex forwarding calls
525 Using the built-in functions described below, you can record
526 the arguments a function received, and call another function
527 with the same arguments, without knowing the number or types
530 You can also record the return value of that function call,
531 and later return that value, without knowing what data type
532 the function tried to return (as long as your caller expects
535 However, these built-in functions may interact badly with some
536 sophisticated features or other extensions of the language. It
537 is, therefore, not recommended to use them outside very simple
538 functions acting as mere forwarders for their arguments.
540 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
541 This built-in function returns a pointer to data
542 describing how to perform a call with the same arguments as are passed
543 to the current function.
545 The function saves the arg pointer register, structure value address,
546 and all registers that might be used to pass arguments to a function
547 into a block of memory allocated on the stack. Then it returns the
548 address of that block.
551 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
552 This built-in function invokes @var{function}
553 with a copy of the parameters described by @var{arguments}
556 The value of @var{arguments} should be the value returned by
557 @code{__builtin_apply_args}. The argument @var{size} specifies the size
558 of the stack argument data, in bytes.
560 This function returns a pointer to data describing
561 how to return whatever value is returned by @var{function}. The data
562 is saved in a block of memory allocated on the stack.
564 It is not always simple to compute the proper value for @var{size}. The
565 value is used by @code{__builtin_apply} to compute the amount of data
566 that should be pushed on the stack and copied from the incoming argument
570 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
571 This built-in function returns the value described by @var{result} from
572 the containing function. You should specify, for @var{result}, a value
573 returned by @code{__builtin_apply}.
576 @deftypefn {Built-in Function} {} __builtin_va_arg_pack ()
577 This built-in function represents all anonymous arguments of an inline
578 function. It can be used only in inline functions that are always
579 inlined, never compiled as a separate function, such as those using
580 @code{__attribute__ ((__always_inline__))} or
581 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
582 It must be only passed as last argument to some other function
583 with variable arguments. This is useful for writing small wrapper
584 inlines for variable argument functions, when using preprocessor
585 macros is undesirable. For example:
587 extern int myprintf (FILE *f, const char *format, ...);
588 extern inline __attribute__ ((__gnu_inline__)) int
589 myprintf (FILE *f, const char *format, ...)
591 int r = fprintf (f, "myprintf: ");
594 int s = fprintf (f, format, __builtin_va_arg_pack ());
602 @deftypefn {Built-in Function} {size_t} __builtin_va_arg_pack_len ()
603 This built-in function returns the number of anonymous arguments of
604 an inline function. It can be used only in inline functions that
605 are always inlined, never compiled as a separate function, such
606 as those using @code{__attribute__ ((__always_inline__))} or
607 @code{__attribute__ ((__gnu_inline__))} extern inline functions.
608 For example following does link- or run-time checking of open
609 arguments for optimized code:
612 extern inline __attribute__((__gnu_inline__)) int
613 myopen (const char *path, int oflag, ...)
615 if (__builtin_va_arg_pack_len () > 1)
616 warn_open_too_many_arguments ();
618 if (__builtin_constant_p (oflag))
620 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1)
622 warn_open_missing_mode ();
623 return __open_2 (path, oflag);
625 return open (path, oflag, __builtin_va_arg_pack ());
628 if (__builtin_va_arg_pack_len () < 1)
629 return __open_2 (path, oflag);
631 return open (path, oflag, __builtin_va_arg_pack ());
638 @section Referring to a Type with @code{typeof}
641 @cindex macros, types of arguments
643 Another way to refer to the type of an expression is with @code{typeof}.
644 The syntax of using of this keyword looks like @code{sizeof}, but the
645 construct acts semantically like a type name defined with @code{typedef}.
647 There are two ways of writing the argument to @code{typeof}: with an
648 expression or with a type. Here is an example with an expression:
655 This assumes that @code{x} is an array of pointers to functions;
656 the type described is that of the values of the functions.
658 Here is an example with a typename as the argument:
665 Here the type described is that of pointers to @code{int}.
667 If you are writing a header file that must work when included in ISO C
668 programs, write @code{__typeof__} instead of @code{typeof}.
669 @xref{Alternate Keywords}.
671 A @code{typeof} construct can be used anywhere a typedef name can be
672 used. For example, you can use it in a declaration, in a cast, or inside
673 of @code{sizeof} or @code{typeof}.
675 The operand of @code{typeof} is evaluated for its side effects if and
676 only if it is an expression of variably modified type or the name of
679 @code{typeof} is often useful in conjunction with
680 statement expressions (@pxref{Statement Exprs}).
681 Here is how the two together can
682 be used to define a safe ``maximum'' macro which operates on any
683 arithmetic type and evaluates each of its arguments exactly once:
687 (@{ typeof (a) _a = (a); \
688 typeof (b) _b = (b); \
689 _a > _b ? _a : _b; @})
692 @cindex underscores in variables in macros
693 @cindex @samp{_} in variables in macros
694 @cindex local variables in macros
695 @cindex variables, local, in macros
696 @cindex macros, local variables in
698 The reason for using names that start with underscores for the local
699 variables is to avoid conflicts with variable names that occur within the
700 expressions that are substituted for @code{a} and @code{b}. Eventually we
701 hope to design a new form of declaration syntax that allows you to declare
702 variables whose scopes start only after their initializers; this will be a
703 more reliable way to prevent such conflicts.
706 Some more examples of the use of @code{typeof}:
710 This declares @code{y} with the type of what @code{x} points to.
717 This declares @code{y} as an array of such values.
724 This declares @code{y} as an array of pointers to characters:
727 typeof (typeof (char *)[4]) y;
731 It is equivalent to the following traditional C declaration:
737 To see the meaning of the declaration using @code{typeof}, and why it
738 might be a useful way to write, rewrite it with these macros:
741 #define pointer(T) typeof(T *)
742 #define array(T, N) typeof(T [N])
746 Now the declaration can be rewritten this way:
749 array (pointer (char), 4) y;
753 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
754 pointers to @code{char}.
757 @emph{Compatibility Note:} In addition to @code{typeof}, GCC 2 supported
758 a more limited extension that permitted one to write
761 typedef @var{T} = @var{expr};
765 with the effect of declaring @var{T} to have the type of the expression
766 @var{expr}. This extension does not work with GCC 3 (versions between
767 3.0 and 3.2 crash; 3.2.1 and later give an error). Code that
768 relies on it should be rewritten to use @code{typeof}:
771 typedef typeof(@var{expr}) @var{T};
775 This works with all versions of GCC@.
778 @section Conditionals with Omitted Operands
779 @cindex conditional expressions, extensions
780 @cindex omitted middle-operands
781 @cindex middle-operands, omitted
782 @cindex extensions, @code{?:}
783 @cindex @code{?:} extensions
785 The middle operand in a conditional expression may be omitted. Then
786 if the first operand is nonzero, its value is the value of the conditional
789 Therefore, the expression
796 has the value of @code{x} if that is nonzero; otherwise, the value of
799 This example is perfectly equivalent to
805 @cindex side effect in @code{?:}
806 @cindex @code{?:} side effect
808 In this simple case, the ability to omit the middle operand is not
809 especially useful. When it becomes useful is when the first operand does,
810 or may (if it is a macro argument), contain a side effect. Then repeating
811 the operand in the middle would perform the side effect twice. Omitting
812 the middle operand uses the value already computed without the undesirable
813 effects of recomputing it.
816 @section 128-bit integers
817 @cindex @code{__int128} data types
819 As an extension the integer scalar type @code{__int128} is supported for
820 targets which have an integer mode wide enough to hold 128 bits.
821 Simply write @code{__int128} for a signed 128-bit integer, or
822 @code{unsigned __int128} for an unsigned 128-bit integer. There is no
823 support in GCC for expressing an integer constant of type @code{__int128}
824 for targets with @code{long long} integer less than 128 bits wide.
827 @section Double-Word Integers
828 @cindex @code{long long} data types
829 @cindex double-word arithmetic
830 @cindex multiprecision arithmetic
831 @cindex @code{LL} integer suffix
832 @cindex @code{ULL} integer suffix
834 ISO C99 supports data types for integers that are at least 64 bits wide,
835 and as an extension GCC supports them in C90 mode and in C++.
836 Simply write @code{long long int} for a signed integer, or
837 @code{unsigned long long int} for an unsigned integer. To make an
838 integer constant of type @code{long long int}, add the suffix @samp{LL}
839 to the integer. To make an integer constant of type @code{unsigned long
840 long int}, add the suffix @samp{ULL} to the integer.
842 You can use these types in arithmetic like any other integer types.
843 Addition, subtraction, and bitwise boolean operations on these types
844 are open-coded on all types of machines. Multiplication is open-coded
845 if the machine supports a fullword-to-doubleword widening multiply
846 instruction. Division and shifts are open-coded only on machines that
847 provide special support. The operations that are not open-coded use
848 special library routines that come with GCC@.
850 There may be pitfalls when you use @code{long long} types for function
851 arguments without function prototypes. If a function
852 expects type @code{int} for its argument, and you pass a value of type
853 @code{long long int}, confusion results because the caller and the
854 subroutine disagree about the number of bytes for the argument.
855 Likewise, if the function expects @code{long long int} and you pass
856 @code{int}. The best way to avoid such problems is to use prototypes.
859 @section Complex Numbers
860 @cindex complex numbers
861 @cindex @code{_Complex} keyword
862 @cindex @code{__complex__} keyword
864 ISO C99 supports complex floating data types, and as an extension GCC
865 supports them in C90 mode and in C++. GCC also supports complex integer data
866 types which are not part of ISO C99. You can declare complex types
867 using the keyword @code{_Complex}. As an extension, the older GNU
868 keyword @code{__complex__} is also supported.
870 For example, @samp{_Complex double x;} declares @code{x} as a
871 variable whose real part and imaginary part are both of type
872 @code{double}. @samp{_Complex short int y;} declares @code{y} to
873 have real and imaginary parts of type @code{short int}; this is not
874 likely to be useful, but it shows that the set of complex types is
877 To write a constant with a complex data type, use the suffix @samp{i} or
878 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
879 has type @code{_Complex float} and @code{3i} has type
880 @code{_Complex int}. Such a constant always has a pure imaginary
881 value, but you can form any complex value you like by adding one to a
882 real constant. This is a GNU extension; if you have an ISO C99
883 conforming C library (such as the GNU C Library), and want to construct complex
884 constants of floating type, you should include @code{<complex.h>} and
885 use the macros @code{I} or @code{_Complex_I} instead.
887 @cindex @code{__real__} keyword
888 @cindex @code{__imag__} keyword
889 To extract the real part of a complex-valued expression @var{exp}, write
890 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
891 extract the imaginary part. This is a GNU extension; for values of
892 floating type, you should use the ISO C99 functions @code{crealf},
893 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
894 @code{cimagl}, declared in @code{<complex.h>} and also provided as
895 built-in functions by GCC@.
897 @cindex complex conjugation
898 The operator @samp{~} performs complex conjugation when used on a value
899 with a complex type. This is a GNU extension; for values of
900 floating type, you should use the ISO C99 functions @code{conjf},
901 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
902 provided as built-in functions by GCC@.
904 GCC can allocate complex automatic variables in a noncontiguous
905 fashion; it's even possible for the real part to be in a register while
906 the imaginary part is on the stack (or vice versa). Only the DWARF 2
907 debug info format can represent this, so use of DWARF 2 is recommended.
908 If you are using the stabs debug info format, GCC describes a noncontiguous
909 complex variable as if it were two separate variables of noncomplex type.
910 If the variable's actual name is @code{foo}, the two fictitious
911 variables are named @code{foo$real} and @code{foo$imag}. You can
912 examine and set these two fictitious variables with your debugger.
915 @section Additional Floating Types
916 @cindex additional floating types
917 @cindex @code{__float80} data type
918 @cindex @code{__float128} data type
919 @cindex @code{w} floating point suffix
920 @cindex @code{q} floating point suffix
921 @cindex @code{W} floating point suffix
922 @cindex @code{Q} floating point suffix
924 As an extension, GNU C supports additional floating
925 types, @code{__float80} and @code{__float128} to support 80-bit
926 (@code{XFmode}) and 128-bit (@code{TFmode}) floating types.
927 Support for additional types includes the arithmetic operators:
928 add, subtract, multiply, divide; unary arithmetic operators;
929 relational operators; equality operators; and conversions to and from
930 integer and other floating types. Use a suffix @samp{w} or @samp{W}
931 in a literal constant of type @code{__float80} and @samp{q} or @samp{Q}
932 for @code{_float128}. You can declare complex types using the
933 corresponding internal complex type, @code{XCmode} for @code{__float80}
934 type and @code{TCmode} for @code{__float128} type:
937 typedef _Complex float __attribute__((mode(TC))) _Complex128;
938 typedef _Complex float __attribute__((mode(XC))) _Complex80;
941 Not all targets support additional floating-point types. @code{__float80}
942 and @code{__float128} types are supported on i386, x86_64 and IA-64 targets.
943 The @code{__float128} type is supported on hppa HP-UX targets.
946 @section Half-Precision Floating Point
947 @cindex half-precision floating point
948 @cindex @code{__fp16} data type
950 On ARM targets, GCC supports half-precision (16-bit) floating point via
951 the @code{__fp16} type. You must enable this type explicitly
952 with the @option{-mfp16-format} command-line option in order to use it.
954 ARM supports two incompatible representations for half-precision
955 floating-point values. You must choose one of the representations and
956 use it consistently in your program.
958 Specifying @option{-mfp16-format=ieee} selects the IEEE 754-2008 format.
959 This format can represent normalized values in the range of @math{2^{-14}} to 65504.
960 There are 11 bits of significand precision, approximately 3
963 Specifying @option{-mfp16-format=alternative} selects the ARM
964 alternative format. This representation is similar to the IEEE
965 format, but does not support infinities or NaNs. Instead, the range
966 of exponents is extended, so that this format can represent normalized
967 values in the range of @math{2^{-14}} to 131008.
969 The @code{__fp16} type is a storage format only. For purposes
970 of arithmetic and other operations, @code{__fp16} values in C or C++
971 expressions are automatically promoted to @code{float}. In addition,
972 you cannot declare a function with a return value or parameters
973 of type @code{__fp16}.
975 Note that conversions from @code{double} to @code{__fp16}
976 involve an intermediate conversion to @code{float}. Because
977 of rounding, this can sometimes produce a different result than a
980 ARM provides hardware support for conversions between
981 @code{__fp16} and @code{float} values
982 as an extension to VFP and NEON (Advanced SIMD). GCC generates
983 code using these hardware instructions if you compile with
984 options to select an FPU that provides them;
985 for example, @option{-mfpu=neon-fp16 -mfloat-abi=softfp},
986 in addition to the @option{-mfp16-format} option to select
987 a half-precision format.
989 Language-level support for the @code{__fp16} data type is
990 independent of whether GCC generates code using hardware floating-point
991 instructions. In cases where hardware support is not specified, GCC
992 implements conversions between @code{__fp16} and @code{float} values
996 @section Decimal Floating Types
997 @cindex decimal floating types
998 @cindex @code{_Decimal32} data type
999 @cindex @code{_Decimal64} data type
1000 @cindex @code{_Decimal128} data type
1001 @cindex @code{df} integer suffix
1002 @cindex @code{dd} integer suffix
1003 @cindex @code{dl} integer suffix
1004 @cindex @code{DF} integer suffix
1005 @cindex @code{DD} integer suffix
1006 @cindex @code{DL} integer suffix
1008 As an extension, GNU C supports decimal floating types as
1009 defined in the N1312 draft of ISO/IEC WDTR24732. Support for decimal
1010 floating types in GCC will evolve as the draft technical report changes.
1011 Calling conventions for any target might also change. Not all targets
1012 support decimal floating types.
1014 The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and
1015 @code{_Decimal128}. They use a radix of ten, unlike the floating types
1016 @code{float}, @code{double}, and @code{long double} whose radix is not
1017 specified by the C standard but is usually two.
1019 Support for decimal floating types includes the arithmetic operators
1020 add, subtract, multiply, divide; unary arithmetic operators;
1021 relational operators; equality operators; and conversions to and from
1022 integer and other floating types. Use a suffix @samp{df} or
1023 @samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd}
1024 or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for
1027 GCC support of decimal float as specified by the draft technical report
1032 When the value of a decimal floating type cannot be represented in the
1033 integer type to which it is being converted, the result is undefined
1034 rather than the result value specified by the draft technical report.
1037 GCC does not provide the C library functionality associated with
1038 @file{math.h}, @file{fenv.h}, @file{stdio.h}, @file{stdlib.h}, and
1039 @file{wchar.h}, which must come from a separate C library implementation.
1040 Because of this the GNU C compiler does not define macro
1041 @code{__STDC_DEC_FP__} to indicate that the implementation conforms to
1042 the technical report.
1045 Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128}
1046 are supported by the DWARF 2 debug information format.
1052 ISO C99 supports floating-point numbers written not only in the usual
1053 decimal notation, such as @code{1.55e1}, but also numbers such as
1054 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1055 supports this in C90 mode (except in some cases when strictly
1056 conforming) and in C++. In that format the
1057 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1058 mandatory. The exponent is a decimal number that indicates the power of
1059 2 by which the significant part is multiplied. Thus @samp{0x1.f} is
1066 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1067 is the same as @code{1.55e1}.
1069 Unlike for floating-point numbers in the decimal notation the exponent
1070 is always required in the hexadecimal notation. Otherwise the compiler
1071 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1072 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1073 extension for floating-point constants of type @code{float}.
1076 @section Fixed-Point Types
1077 @cindex fixed-point types
1078 @cindex @code{_Fract} data type
1079 @cindex @code{_Accum} data type
1080 @cindex @code{_Sat} data type
1081 @cindex @code{hr} fixed-suffix
1082 @cindex @code{r} fixed-suffix
1083 @cindex @code{lr} fixed-suffix
1084 @cindex @code{llr} fixed-suffix
1085 @cindex @code{uhr} fixed-suffix
1086 @cindex @code{ur} fixed-suffix
1087 @cindex @code{ulr} fixed-suffix
1088 @cindex @code{ullr} fixed-suffix
1089 @cindex @code{hk} fixed-suffix
1090 @cindex @code{k} fixed-suffix
1091 @cindex @code{lk} fixed-suffix
1092 @cindex @code{llk} fixed-suffix
1093 @cindex @code{uhk} fixed-suffix
1094 @cindex @code{uk} fixed-suffix
1095 @cindex @code{ulk} fixed-suffix
1096 @cindex @code{ullk} fixed-suffix
1097 @cindex @code{HR} fixed-suffix
1098 @cindex @code{R} fixed-suffix
1099 @cindex @code{LR} fixed-suffix
1100 @cindex @code{LLR} fixed-suffix
1101 @cindex @code{UHR} fixed-suffix
1102 @cindex @code{UR} fixed-suffix
1103 @cindex @code{ULR} fixed-suffix
1104 @cindex @code{ULLR} fixed-suffix
1105 @cindex @code{HK} fixed-suffix
1106 @cindex @code{K} fixed-suffix
1107 @cindex @code{LK} fixed-suffix
1108 @cindex @code{LLK} fixed-suffix
1109 @cindex @code{UHK} fixed-suffix
1110 @cindex @code{UK} fixed-suffix
1111 @cindex @code{ULK} fixed-suffix
1112 @cindex @code{ULLK} fixed-suffix
1114 As an extension, GNU C supports fixed-point types as
1115 defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point
1116 types in GCC will evolve as the draft technical report changes.
1117 Calling conventions for any target might also change. Not all targets
1118 support fixed-point types.
1120 The fixed-point types are
1121 @code{short _Fract},
1124 @code{long long _Fract},
1125 @code{unsigned short _Fract},
1126 @code{unsigned _Fract},
1127 @code{unsigned long _Fract},
1128 @code{unsigned long long _Fract},
1129 @code{_Sat short _Fract},
1131 @code{_Sat long _Fract},
1132 @code{_Sat long long _Fract},
1133 @code{_Sat unsigned short _Fract},
1134 @code{_Sat unsigned _Fract},
1135 @code{_Sat unsigned long _Fract},
1136 @code{_Sat unsigned long long _Fract},
1137 @code{short _Accum},
1140 @code{long long _Accum},
1141 @code{unsigned short _Accum},
1142 @code{unsigned _Accum},
1143 @code{unsigned long _Accum},
1144 @code{unsigned long long _Accum},
1145 @code{_Sat short _Accum},
1147 @code{_Sat long _Accum},
1148 @code{_Sat long long _Accum},
1149 @code{_Sat unsigned short _Accum},
1150 @code{_Sat unsigned _Accum},
1151 @code{_Sat unsigned long _Accum},
1152 @code{_Sat unsigned long long _Accum}.
1154 Fixed-point data values contain fractional and optional integral parts.
1155 The format of fixed-point data varies and depends on the target machine.
1157 Support for fixed-point types includes:
1160 prefix and postfix increment and decrement operators (@code{++}, @code{--})
1162 unary arithmetic operators (@code{+}, @code{-}, @code{!})
1164 binary arithmetic operators (@code{+}, @code{-}, @code{*}, @code{/})
1166 binary shift operators (@code{<<}, @code{>>})
1168 relational operators (@code{<}, @code{<=}, @code{>=}, @code{>})
1170 equality operators (@code{==}, @code{!=})
1172 assignment operators (@code{+=}, @code{-=}, @code{*=}, @code{/=},
1173 @code{<<=}, @code{>>=})
1175 conversions to and from integer, floating-point, or fixed-point types
1178 Use a suffix in a fixed-point literal constant:
1180 @item @samp{hr} or @samp{HR} for @code{short _Fract} and
1181 @code{_Sat short _Fract}
1182 @item @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract}
1183 @item @samp{lr} or @samp{LR} for @code{long _Fract} and
1184 @code{_Sat long _Fract}
1185 @item @samp{llr} or @samp{LLR} for @code{long long _Fract} and
1186 @code{_Sat long long _Fract}
1187 @item @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and
1188 @code{_Sat unsigned short _Fract}
1189 @item @samp{ur} or @samp{UR} for @code{unsigned _Fract} and
1190 @code{_Sat unsigned _Fract}
1191 @item @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and
1192 @code{_Sat unsigned long _Fract}
1193 @item @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract}
1194 and @code{_Sat unsigned long long _Fract}
1195 @item @samp{hk} or @samp{HK} for @code{short _Accum} and
1196 @code{_Sat short _Accum}
1197 @item @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum}
1198 @item @samp{lk} or @samp{LK} for @code{long _Accum} and
1199 @code{_Sat long _Accum}
1200 @item @samp{llk} or @samp{LLK} for @code{long long _Accum} and
1201 @code{_Sat long long _Accum}
1202 @item @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and
1203 @code{_Sat unsigned short _Accum}
1204 @item @samp{uk} or @samp{UK} for @code{unsigned _Accum} and
1205 @code{_Sat unsigned _Accum}
1206 @item @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and
1207 @code{_Sat unsigned long _Accum}
1208 @item @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum}
1209 and @code{_Sat unsigned long long _Accum}
1212 GCC support of fixed-point types as specified by the draft technical report
1217 Pragmas to control overflow and rounding behaviors are not implemented.
1220 Fixed-point types are supported by the DWARF 2 debug information format.
1222 @node Named Address Spaces
1223 @section Named Address Spaces
1224 @cindex Named Address Spaces
1226 As an extension, GNU C supports named address spaces as
1227 defined in the N1275 draft of ISO/IEC DTR 18037. Support for named
1228 address spaces in GCC will evolve as the draft technical report
1229 changes. Calling conventions for any target might also change. At
1230 present, only the AVR, SPU, M32C, and RL78 targets support address
1231 spaces other than the generic address space.
1233 Address space identifiers may be used exactly like any other C type
1234 qualifier (e.g., @code{const} or @code{volatile}). See the N1275
1235 document for more details.
1237 @anchor{AVR Named Address Spaces}
1238 @subsection AVR Named Address Spaces
1240 On the AVR target, there are several address spaces that can be used
1241 in order to put read-only data into the flash memory and access that
1242 data by means of the special instructions @code{LPM} or @code{ELPM}
1243 needed to read from flash.
1245 Per default, any data including read-only data is located in RAM
1246 (the generic address space) so that non-generic address spaces are
1247 needed to locate read-only data in flash memory
1248 @emph{and} to generate the right instructions to access this data
1249 without using (inline) assembler code.
1253 @cindex @code{__flash} AVR Named Address Spaces
1254 The @code{__flash} qualifier locates data in the
1255 @code{.progmem.data} section. Data is read using the @code{LPM}
1256 instruction. Pointers to this address space are 16 bits wide.
1263 @cindex @code{__flash1} AVR Named Address Spaces
1264 @cindex @code{__flash2} AVR Named Address Spaces
1265 @cindex @code{__flash3} AVR Named Address Spaces
1266 @cindex @code{__flash4} AVR Named Address Spaces
1267 @cindex @code{__flash5} AVR Named Address Spaces
1268 These are 16-bit address spaces locating data in section
1269 @code{.progmem@var{N}.data} where @var{N} refers to
1270 address space @code{__flash@var{N}}.
1271 The compiler sets the @code{RAMPZ} segment register appropriately
1272 before reading data by means of the @code{ELPM} instruction.
1275 @cindex @code{__memx} AVR Named Address Spaces
1276 This is a 24-bit address space that linearizes flash and RAM:
1277 If the high bit of the address is set, data is read from
1278 RAM using the lower two bytes as RAM address.
1279 If the high bit of the address is clear, data is read from flash
1280 with @code{RAMPZ} set according to the high byte of the address.
1281 @xref{AVR Built-in Functions,,@code{__builtin_avr_flash_segment}}.
1283 Objects in this address space are located in @code{.progmemx.data}.
1289 char my_read (const __flash char ** p)
1291 /* p is a pointer to RAM that points to a pointer to flash.
1292 The first indirection of p reads that flash pointer
1293 from RAM and the second indirection reads a char from this
1299 /* Locate array[] in flash memory */
1300 const __flash int array[] = @{ 3, 5, 7, 11, 13, 17, 19 @};
1306 /* Return 17 by reading from flash memory */
1307 return array[array[i]];
1312 For each named address space supported by avr-gcc there is an equally
1313 named but uppercase built-in macro defined.
1314 The purpose is to facilitate testing if respective address space
1315 support is available or not:
1319 const __flash int var = 1;
1326 #include <avr/pgmspace.h> /* From AVR-LibC */
1328 const int var PROGMEM = 1;
1332 return (int) pgm_read_word (&var);
1334 #endif /* __FLASH */
1338 Notice that attribute @ref{AVR Variable Attributes,,@code{progmem}}
1339 locates data in flash but
1340 accesses to these data read from generic address space, i.e.@:
1342 so that you need special accessors like @code{pgm_read_byte}
1343 from @w{@uref{http://nongnu.org/avr-libc/user-manual/,AVR-LibC}}
1344 together with attribute @code{progmem}.
1347 @b{Limitations and caveats}
1351 Reading across the 64@tie{}KiB section boundary of
1352 the @code{__flash} or @code{__flash@var{N}} address spaces
1353 shows undefined behavior. The only address space that
1354 supports reading across the 64@tie{}KiB flash segment boundaries is
1358 If you use one of the @code{__flash@var{N}} address spaces
1359 you must arrange your linker script to locate the
1360 @code{.progmem@var{N}.data} sections according to your needs.
1363 Any data or pointers to the non-generic address spaces must
1364 be qualified as @code{const}, i.e.@: as read-only data.
1365 This still applies if the data in one of these address
1366 spaces like software version number or calibration lookup table are intended to
1367 be changed after load time by, say, a boot loader. In this case
1368 the right qualification is @code{const} @code{volatile} so that the compiler
1369 must not optimize away known values or insert them
1370 as immediates into operands of instructions.
1373 The following code initializes a variable @code{pfoo}
1374 located in static storage with a 24-bit address:
1376 extern const __memx char foo;
1377 const __memx void *pfoo = &foo;
1381 Such code requires at least binutils 2.23, see
1382 @w{@uref{http://sourceware.org/PR13503,PR13503}}.
1386 @subsection M32C Named Address Spaces
1387 @cindex @code{__far} M32C Named Address Spaces
1389 On the M32C target, with the R8C and M16C CPU variants, variables
1390 qualified with @code{__far} are accessed using 32-bit addresses in
1391 order to access memory beyond the first 64@tie{}Ki bytes. If
1392 @code{__far} is used with the M32CM or M32C CPU variants, it has no
1395 @subsection RL78 Named Address Spaces
1396 @cindex @code{__far} RL78 Named Address Spaces
1398 On the RL78 target, variables qualified with @code{__far} are accessed
1399 with 32-bit pointers (20-bit addresses) rather than the default 16-bit
1400 addresses. Non-far variables are assumed to appear in the topmost
1401 64@tie{}KiB of the address space.
1403 @subsection SPU Named Address Spaces
1404 @cindex @code{__ea} SPU Named Address Spaces
1406 On the SPU target variables may be declared as
1407 belonging to another address space by qualifying the type with the
1408 @code{__ea} address space identifier:
1415 The compiler generates special code to access the variable @code{i}.
1416 It may use runtime library
1417 support, or generate special machine instructions to access that address
1421 @section Arrays of Length Zero
1422 @cindex arrays of length zero
1423 @cindex zero-length arrays
1424 @cindex length-zero arrays
1425 @cindex flexible array members
1427 Zero-length arrays are allowed in GNU C@. They are very useful as the
1428 last element of a structure that is really a header for a variable-length
1437 struct line *thisline = (struct line *)
1438 malloc (sizeof (struct line) + this_length);
1439 thisline->length = this_length;
1442 In ISO C90, you would have to give @code{contents} a length of 1, which
1443 means either you waste space or complicate the argument to @code{malloc}.
1445 In ISO C99, you would use a @dfn{flexible array member}, which is
1446 slightly different in syntax and semantics:
1450 Flexible array members are written as @code{contents[]} without
1454 Flexible array members have incomplete type, and so the @code{sizeof}
1455 operator may not be applied. As a quirk of the original implementation
1456 of zero-length arrays, @code{sizeof} evaluates to zero.
1459 Flexible array members may only appear as the last member of a
1460 @code{struct} that is otherwise non-empty.
1463 A structure containing a flexible array member, or a union containing
1464 such a structure (possibly recursively), may not be a member of a
1465 structure or an element of an array. (However, these uses are
1466 permitted by GCC as extensions.)
1469 GCC versions before 3.0 allowed zero-length arrays to be statically
1470 initialized, as if they were flexible arrays. In addition to those
1471 cases that were useful, it also allowed initializations in situations
1472 that would corrupt later data. Non-empty initialization of zero-length
1473 arrays is now treated like any case where there are more initializer
1474 elements than the array holds, in that a suitable warning about ``excess
1475 elements in array'' is given, and the excess elements (all of them, in
1476 this case) are ignored.
1478 Instead GCC allows static initialization of flexible array members.
1479 This is equivalent to defining a new structure containing the original
1480 structure followed by an array of sufficient size to contain the data.
1481 E.g.@: in the following, @code{f1} is constructed as if it were declared
1487 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1490 struct f1 f1; int data[3];
1491 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1495 The convenience of this extension is that @code{f1} has the desired
1496 type, eliminating the need to consistently refer to @code{f2.f1}.
1498 This has symmetry with normal static arrays, in that an array of
1499 unknown size is also written with @code{[]}.
1501 Of course, this extension only makes sense if the extra data comes at
1502 the end of a top-level object, as otherwise we would be overwriting
1503 data at subsequent offsets. To avoid undue complication and confusion
1504 with initialization of deeply nested arrays, we simply disallow any
1505 non-empty initialization except when the structure is the top-level
1506 object. For example:
1509 struct foo @{ int x; int y[]; @};
1510 struct bar @{ struct foo z; @};
1512 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1513 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1514 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1515 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1518 @node Empty Structures
1519 @section Structures With No Members
1520 @cindex empty structures
1521 @cindex zero-size structures
1523 GCC permits a C structure to have no members:
1530 The structure has size zero. In C++, empty structures are part
1531 of the language. G++ treats empty structures as if they had a single
1532 member of type @code{char}.
1534 @node Variable Length
1535 @section Arrays of Variable Length
1536 @cindex variable-length arrays
1537 @cindex arrays of variable length
1540 Variable-length automatic arrays are allowed in ISO C99, and as an
1541 extension GCC accepts them in C90 mode and in C++. These arrays are
1542 declared like any other automatic arrays, but with a length that is not
1543 a constant expression. The storage is allocated at the point of
1544 declaration and deallocated when the block scope containing the declaration
1550 concat_fopen (char *s1, char *s2, char *mode)
1552 char str[strlen (s1) + strlen (s2) + 1];
1555 return fopen (str, mode);
1559 @cindex scope of a variable length array
1560 @cindex variable-length array scope
1561 @cindex deallocating variable length arrays
1562 Jumping or breaking out of the scope of the array name deallocates the
1563 storage. Jumping into the scope is not allowed; you get an error
1566 @cindex @code{alloca} vs variable-length arrays
1567 You can use the function @code{alloca} to get an effect much like
1568 variable-length arrays. The function @code{alloca} is available in
1569 many other C implementations (but not in all). On the other hand,
1570 variable-length arrays are more elegant.
1572 There are other differences between these two methods. Space allocated
1573 with @code{alloca} exists until the containing @emph{function} returns.
1574 The space for a variable-length array is deallocated as soon as the array
1575 name's scope ends. (If you use both variable-length arrays and
1576 @code{alloca} in the same function, deallocation of a variable-length array
1577 also deallocates anything more recently allocated with @code{alloca}.)
1579 You can also use variable-length arrays as arguments to functions:
1583 tester (int len, char data[len][len])
1589 The length of an array is computed once when the storage is allocated
1590 and is remembered for the scope of the array in case you access it with
1593 If you want to pass the array first and the length afterward, you can
1594 use a forward declaration in the parameter list---another GNU extension.
1598 tester (int len; char data[len][len], int len)
1604 @cindex parameter forward declaration
1605 The @samp{int len} before the semicolon is a @dfn{parameter forward
1606 declaration}, and it serves the purpose of making the name @code{len}
1607 known when the declaration of @code{data} is parsed.
1609 You can write any number of such parameter forward declarations in the
1610 parameter list. They can be separated by commas or semicolons, but the
1611 last one must end with a semicolon, which is followed by the ``real''
1612 parameter declarations. Each forward declaration must match a ``real''
1613 declaration in parameter name and data type. ISO C99 does not support
1614 parameter forward declarations.
1616 @node Variadic Macros
1617 @section Macros with a Variable Number of Arguments.
1618 @cindex variable number of arguments
1619 @cindex macro with variable arguments
1620 @cindex rest argument (in macro)
1621 @cindex variadic macros
1623 In the ISO C standard of 1999, a macro can be declared to accept a
1624 variable number of arguments much as a function can. The syntax for
1625 defining the macro is similar to that of a function. Here is an
1629 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1633 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1634 such a macro, it represents the zero or more tokens until the closing
1635 parenthesis that ends the invocation, including any commas. This set of
1636 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1637 wherever it appears. See the CPP manual for more information.
1639 GCC has long supported variadic macros, and used a different syntax that
1640 allowed you to give a name to the variable arguments just like any other
1641 argument. Here is an example:
1644 #define debug(format, args...) fprintf (stderr, format, args)
1648 This is in all ways equivalent to the ISO C example above, but arguably
1649 more readable and descriptive.
1651 GNU CPP has two further variadic macro extensions, and permits them to
1652 be used with either of the above forms of macro definition.
1654 In standard C, you are not allowed to leave the variable argument out
1655 entirely; but you are allowed to pass an empty argument. For example,
1656 this invocation is invalid in ISO C, because there is no comma after
1663 GNU CPP permits you to completely omit the variable arguments in this
1664 way. In the above examples, the compiler would complain, though since
1665 the expansion of the macro still has the extra comma after the format
1668 To help solve this problem, CPP behaves specially for variable arguments
1669 used with the token paste operator, @samp{##}. If instead you write
1672 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1676 and if the variable arguments are omitted or empty, the @samp{##}
1677 operator causes the preprocessor to remove the comma before it. If you
1678 do provide some variable arguments in your macro invocation, GNU CPP
1679 does not complain about the paste operation and instead places the
1680 variable arguments after the comma. Just like any other pasted macro
1681 argument, these arguments are not macro expanded.
1683 @node Escaped Newlines
1684 @section Slightly Looser Rules for Escaped Newlines
1685 @cindex escaped newlines
1686 @cindex newlines (escaped)
1688 Recently, the preprocessor has relaxed its treatment of escaped
1689 newlines. Previously, the newline had to immediately follow a
1690 backslash. The current implementation allows whitespace in the form
1691 of spaces, horizontal and vertical tabs, and form feeds between the
1692 backslash and the subsequent newline. The preprocessor issues a
1693 warning, but treats it as a valid escaped newline and combines the two
1694 lines to form a single logical line. This works within comments and
1695 tokens, as well as between tokens. Comments are @emph{not} treated as
1696 whitespace for the purposes of this relaxation, since they have not
1697 yet been replaced with spaces.
1700 @section Non-Lvalue Arrays May Have Subscripts
1701 @cindex subscripting
1702 @cindex arrays, non-lvalue
1704 @cindex subscripting and function values
1705 In ISO C99, arrays that are not lvalues still decay to pointers, and
1706 may be subscripted, although they may not be modified or used after
1707 the next sequence point and the unary @samp{&} operator may not be
1708 applied to them. As an extension, GNU C allows such arrays to be
1709 subscripted in C90 mode, though otherwise they do not decay to
1710 pointers outside C99 mode. For example,
1711 this is valid in GNU C though not valid in C90:
1715 struct foo @{int a[4];@};
1721 return f().a[index];
1727 @section Arithmetic on @code{void}- and Function-Pointers
1728 @cindex void pointers, arithmetic
1729 @cindex void, size of pointer to
1730 @cindex function pointers, arithmetic
1731 @cindex function, size of pointer to
1733 In GNU C, addition and subtraction operations are supported on pointers to
1734 @code{void} and on pointers to functions. This is done by treating the
1735 size of a @code{void} or of a function as 1.
1737 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1738 and on function types, and returns 1.
1740 @opindex Wpointer-arith
1741 The option @option{-Wpointer-arith} requests a warning if these extensions
1745 @section Non-Constant Initializers
1746 @cindex initializers, non-constant
1747 @cindex non-constant initializers
1749 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1750 automatic variable are not required to be constant expressions in GNU C@.
1751 Here is an example of an initializer with run-time varying elements:
1754 foo (float f, float g)
1756 float beat_freqs[2] = @{ f-g, f+g @};
1761 @node Compound Literals
1762 @section Compound Literals
1763 @cindex constructor expressions
1764 @cindex initializations in expressions
1765 @cindex structures, constructor expression
1766 @cindex expressions, constructor
1767 @cindex compound literals
1768 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1770 ISO C99 supports compound literals. A compound literal looks like
1771 a cast containing an initializer. Its value is an object of the
1772 type specified in the cast, containing the elements specified in
1773 the initializer; it is an lvalue. As an extension, GCC supports
1774 compound literals in C90 mode and in C++, though the semantics are
1775 somewhat different in C++.
1777 Usually, the specified type is a structure. Assume that
1778 @code{struct foo} and @code{structure} are declared as shown:
1781 struct foo @{int a; char b[2];@} structure;
1785 Here is an example of constructing a @code{struct foo} with a compound literal:
1788 structure = ((struct foo) @{x + y, 'a', 0@});
1792 This is equivalent to writing the following:
1796 struct foo temp = @{x + y, 'a', 0@};
1801 You can also construct an array, though this is dangerous in C++, as
1802 explained below. If all the elements of the compound literal are
1803 (made up of) simple constant expressions, suitable for use in
1804 initializers of objects of static storage duration, then the compound
1805 literal can be coerced to a pointer to its first element and used in
1806 such an initializer, as shown here:
1809 char **foo = (char *[]) @{ "x", "y", "z" @};
1812 Compound literals for scalar types and union types are
1813 also allowed, but then the compound literal is equivalent
1816 As a GNU extension, GCC allows initialization of objects with static storage
1817 duration by compound literals (which is not possible in ISO C99, because
1818 the initializer is not a constant).
1819 It is handled as if the object is initialized only with the bracket
1820 enclosed list if the types of the compound literal and the object match.
1821 The initializer list of the compound literal must be constant.
1822 If the object being initialized has array type of unknown size, the size is
1823 determined by compound literal size.
1826 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1827 static int y[] = (int []) @{1, 2, 3@};
1828 static int z[] = (int [3]) @{1@};
1832 The above lines are equivalent to the following:
1834 static struct foo x = @{1, 'a', 'b'@};
1835 static int y[] = @{1, 2, 3@};
1836 static int z[] = @{1, 0, 0@};
1839 In C, a compound literal designates an unnamed object with static or
1840 automatic storage duration. In C++, a compound literal designates a
1841 temporary object, which only lives until the end of its
1842 full-expression. As a result, well-defined C code that takes the
1843 address of a subobject of a compound literal can be undefined in C++.
1844 For instance, if the array compound literal example above appeared
1845 inside a function, any subsequent use of @samp{foo} in C++ has
1846 undefined behavior because the lifetime of the array ends after the
1847 declaration of @samp{foo}. As a result, the C++ compiler now rejects
1848 the conversion of a temporary array to a pointer.
1850 As an optimization, the C++ compiler sometimes gives array compound
1851 literals longer lifetimes: when the array either appears outside a
1852 function or has const-qualified type. If @samp{foo} and its
1853 initializer had elements of @samp{char *const} type rather than
1854 @samp{char *}, or if @samp{foo} were a global variable, the array
1855 would have static storage duration. But it is probably safest just to
1856 avoid the use of array compound literals in code compiled as C++.
1858 @node Designated Inits
1859 @section Designated Initializers
1860 @cindex initializers with labeled elements
1861 @cindex labeled elements in initializers
1862 @cindex case labels in initializers
1863 @cindex designated initializers
1865 Standard C90 requires the elements of an initializer to appear in a fixed
1866 order, the same as the order of the elements in the array or structure
1869 In ISO C99 you can give the elements in any order, specifying the array
1870 indices or structure field names they apply to, and GNU C allows this as
1871 an extension in C90 mode as well. This extension is not
1872 implemented in GNU C++.
1874 To specify an array index, write
1875 @samp{[@var{index}] =} before the element value. For example,
1878 int a[6] = @{ [4] = 29, [2] = 15 @};
1885 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1889 The index values must be constant expressions, even if the array being
1890 initialized is automatic.
1892 An alternative syntax for this that has been obsolete since GCC 2.5 but
1893 GCC still accepts is to write @samp{[@var{index}]} before the element
1894 value, with no @samp{=}.
1896 To initialize a range of elements to the same value, write
1897 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1898 extension. For example,
1901 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1905 If the value in it has side-effects, the side-effects happen only once,
1906 not for each initialized field by the range initializer.
1909 Note that the length of the array is the highest value specified
1912 In a structure initializer, specify the name of a field to initialize
1913 with @samp{.@var{fieldname} =} before the element value. For example,
1914 given the following structure,
1917 struct point @{ int x, y; @};
1921 the following initialization
1924 struct point p = @{ .y = yvalue, .x = xvalue @};
1931 struct point p = @{ xvalue, yvalue @};
1934 Another syntax that has the same meaning, obsolete since GCC 2.5, is
1935 @samp{@var{fieldname}:}, as shown here:
1938 struct point p = @{ y: yvalue, x: xvalue @};
1942 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1943 @dfn{designator}. You can also use a designator (or the obsolete colon
1944 syntax) when initializing a union, to specify which element of the union
1945 should be used. For example,
1948 union foo @{ int i; double d; @};
1950 union foo f = @{ .d = 4 @};
1954 converts 4 to a @code{double} to store it in the union using
1955 the second element. By contrast, casting 4 to type @code{union foo}
1956 stores it into the union as the integer @code{i}, since it is
1957 an integer. (@xref{Cast to Union}.)
1959 You can combine this technique of naming elements with ordinary C
1960 initialization of successive elements. Each initializer element that
1961 does not have a designator applies to the next consecutive element of the
1962 array or structure. For example,
1965 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1972 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1975 Labeling the elements of an array initializer is especially useful
1976 when the indices are characters or belong to an @code{enum} type.
1981 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1982 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1985 @cindex designator lists
1986 You can also write a series of @samp{.@var{fieldname}} and
1987 @samp{[@var{index}]} designators before an @samp{=} to specify a
1988 nested subobject to initialize; the list is taken relative to the
1989 subobject corresponding to the closest surrounding brace pair. For
1990 example, with the @samp{struct point} declaration above:
1993 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1997 If the same field is initialized multiple times, it has the value from
1998 the last initialization. If any such overridden initialization has
1999 side-effect, it is unspecified whether the side-effect happens or not.
2000 Currently, GCC discards them and issues a warning.
2003 @section Case Ranges
2005 @cindex ranges in case statements
2007 You can specify a range of consecutive values in a single @code{case} label,
2011 case @var{low} ... @var{high}:
2015 This has the same effect as the proper number of individual @code{case}
2016 labels, one for each integer value from @var{low} to @var{high}, inclusive.
2018 This feature is especially useful for ranges of ASCII character codes:
2024 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
2025 it may be parsed wrong when you use it with integer values. For example,
2040 @section Cast to a Union Type
2041 @cindex cast to a union
2042 @cindex union, casting to a
2044 A cast to union type is similar to other casts, except that the type
2045 specified is a union type. You can specify the type either with
2046 @code{union @var{tag}} or with a typedef name. A cast to union is actually
2047 a constructor, not a cast, and hence does not yield an lvalue like
2048 normal casts. (@xref{Compound Literals}.)
2050 The types that may be cast to the union type are those of the members
2051 of the union. Thus, given the following union and variables:
2054 union foo @{ int i; double d; @};
2060 both @code{x} and @code{y} can be cast to type @code{union foo}.
2062 Using the cast as the right-hand side of an assignment to a variable of
2063 union type is equivalent to storing in a member of the union:
2068 u = (union foo) x @equiv{} u.i = x
2069 u = (union foo) y @equiv{} u.d = y
2072 You can also use the union cast as a function argument:
2075 void hack (union foo);
2077 hack ((union foo) x);
2080 @node Mixed Declarations
2081 @section Mixed Declarations and Code
2082 @cindex mixed declarations and code
2083 @cindex declarations, mixed with code
2084 @cindex code, mixed with declarations
2086 ISO C99 and ISO C++ allow declarations and code to be freely mixed
2087 within compound statements. As an extension, GNU C also allows this in
2088 C90 mode. For example, you could do:
2097 Each identifier is visible from where it is declared until the end of
2098 the enclosing block.
2100 @node Function Attributes
2101 @section Declaring Attributes of Functions
2102 @cindex function attributes
2103 @cindex declaring attributes of functions
2104 @cindex functions that never return
2105 @cindex functions that return more than once
2106 @cindex functions that have no side effects
2107 @cindex functions in arbitrary sections
2108 @cindex functions that behave like malloc
2109 @cindex @code{volatile} applied to function
2110 @cindex @code{const} applied to function
2111 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
2112 @cindex functions with non-null pointer arguments
2113 @cindex functions that are passed arguments in registers on the 386
2114 @cindex functions that pop the argument stack on the 386
2115 @cindex functions that do not pop the argument stack on the 386
2116 @cindex functions that have different compilation options on the 386
2117 @cindex functions that have different optimization options
2118 @cindex functions that are dynamically resolved
2120 In GNU C, you declare certain things about functions called in your program
2121 which help the compiler optimize function calls and check your code more
2124 The keyword @code{__attribute__} allows you to specify special
2125 attributes when making a declaration. This keyword is followed by an
2126 attribute specification inside double parentheses. The following
2127 attributes are currently defined for functions on all targets:
2128 @code{aligned}, @code{alloc_size}, @code{noreturn},
2129 @code{returns_twice}, @code{noinline}, @code{noclone},
2130 @code{always_inline}, @code{flatten}, @code{pure}, @code{const},
2131 @code{nothrow}, @code{sentinel}, @code{format}, @code{format_arg},
2132 @code{no_instrument_function}, @code{no_split_stack},
2133 @code{section}, @code{constructor},
2134 @code{destructor}, @code{used}, @code{unused}, @code{deprecated},
2135 @code{weak}, @code{malloc}, @code{alias}, @code{ifunc},
2136 @code{warn_unused_result}, @code{nonnull}, @code{gnu_inline},
2137 @code{externally_visible}, @code{hot}, @code{cold}, @code{artificial},
2138 @code{no_sanitize_address}, @code{no_address_safety_analysis},
2139 @code{no_sanitize_undefined},
2140 @code{error} and @code{warning}.
2141 Several other attributes are defined for functions on particular
2142 target systems. Other attributes, including @code{section} are
2143 supported for variables declarations (@pxref{Variable Attributes})
2144 and for types (@pxref{Type Attributes}).
2146 GCC plugins may provide their own attributes.
2148 You may also specify attributes with @samp{__} preceding and following
2149 each keyword. This allows you to use them in header files without
2150 being concerned about a possible macro of the same name. For example,
2151 you may use @code{__noreturn__} instead of @code{noreturn}.
2153 @xref{Attribute Syntax}, for details of the exact syntax for using
2157 @c Keep this table alphabetized by attribute name. Treat _ as space.
2159 @item alias ("@var{target}")
2160 @cindex @code{alias} attribute
2161 The @code{alias} attribute causes the declaration to be emitted as an
2162 alias for another symbol, which must be specified. For instance,
2165 void __f () @{ /* @r{Do something.} */; @}
2166 void f () __attribute__ ((weak, alias ("__f")));
2170 defines @samp{f} to be a weak alias for @samp{__f}. In C++, the
2171 mangled name for the target must be used. It is an error if @samp{__f}
2172 is not defined in the same translation unit.
2174 Not all target machines support this attribute.
2176 @item aligned (@var{alignment})
2177 @cindex @code{aligned} attribute
2178 This attribute specifies a minimum alignment for the function,
2181 You cannot use this attribute to decrease the alignment of a function,
2182 only to increase it. However, when you explicitly specify a function
2183 alignment this overrides the effect of the
2184 @option{-falign-functions} (@pxref{Optimize Options}) option for this
2187 Note that the effectiveness of @code{aligned} attributes may be
2188 limited by inherent limitations in your linker. On many systems, the
2189 linker is only able to arrange for functions to be aligned up to a
2190 certain maximum alignment. (For some linkers, the maximum supported
2191 alignment may be very very small.) See your linker documentation for
2192 further information.
2194 The @code{aligned} attribute can also be used for variables and fields
2195 (@pxref{Variable Attributes}.)
2198 @cindex @code{alloc_size} attribute
2199 The @code{alloc_size} attribute is used to tell the compiler that the
2200 function return value points to memory, where the size is given by
2201 one or two of the functions parameters. GCC uses this
2202 information to improve the correctness of @code{__builtin_object_size}.
2204 The function parameter(s) denoting the allocated size are specified by
2205 one or two integer arguments supplied to the attribute. The allocated size
2206 is either the value of the single function argument specified or the product
2207 of the two function arguments specified. Argument numbering starts at
2213 void* my_calloc(size_t, size_t) __attribute__((alloc_size(1,2)))
2214 void my_realloc(void*, size_t) __attribute__((alloc_size(2)))
2218 declares that @code{my_calloc} returns memory of the size given by
2219 the product of parameter 1 and 2 and that @code{my_realloc} returns memory
2220 of the size given by parameter 2.
2223 @cindex @code{always_inline} function attribute
2224 Generally, functions are not inlined unless optimization is specified.
2225 For functions declared inline, this attribute inlines the function even
2226 if no optimization level is specified.
2229 @cindex @code{gnu_inline} function attribute
2230 This attribute should be used with a function that is also declared
2231 with the @code{inline} keyword. It directs GCC to treat the function
2232 as if it were defined in gnu90 mode even when compiling in C99 or
2235 If the function is declared @code{extern}, then this definition of the
2236 function is used only for inlining. In no case is the function
2237 compiled as a standalone function, not even if you take its address
2238 explicitly. Such an address becomes an external reference, as if you
2239 had only declared the function, and had not defined it. This has
2240 almost the effect of a macro. The way to use this is to put a
2241 function definition in a header file with this attribute, and put
2242 another copy of the function, without @code{extern}, in a library
2243 file. The definition in the header file causes most calls to the
2244 function to be inlined. If any uses of the function remain, they
2245 refer to the single copy in the library. Note that the two
2246 definitions of the functions need not be precisely the same, although
2247 if they do not have the same effect your program may behave oddly.
2249 In C, if the function is neither @code{extern} nor @code{static}, then
2250 the function is compiled as a standalone function, as well as being
2251 inlined where possible.
2253 This is how GCC traditionally handled functions declared
2254 @code{inline}. Since ISO C99 specifies a different semantics for
2255 @code{inline}, this function attribute is provided as a transition
2256 measure and as a useful feature in its own right. This attribute is
2257 available in GCC 4.1.3 and later. It is available if either of the
2258 preprocessor macros @code{__GNUC_GNU_INLINE__} or
2259 @code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline
2260 Function is As Fast As a Macro}.
2262 In C++, this attribute does not depend on @code{extern} in any way,
2263 but it still requires the @code{inline} keyword to enable its special
2267 @cindex @code{artificial} function attribute
2268 This attribute is useful for small inline wrappers that if possible
2269 should appear during debugging as a unit. Depending on the debug
2270 info format it either means marking the function as artificial
2271 or using the caller location for all instructions within the inlined
2275 @cindex interrupt handler functions
2276 When added to an interrupt handler with the M32C port, causes the
2277 prologue and epilogue to use bank switching to preserve the registers
2278 rather than saving them on the stack.
2281 @cindex @code{flatten} function attribute
2282 Generally, inlining into a function is limited. For a function marked with
2283 this attribute, every call inside this function is inlined, if possible.
2284 Whether the function itself is considered for inlining depends on its size and
2285 the current inlining parameters.
2287 @item error ("@var{message}")
2288 @cindex @code{error} function attribute
2289 If this attribute is used on a function declaration and a call to such a function
2290 is not eliminated through dead code elimination or other optimizations, an error
2291 that includes @var{message} is diagnosed. This is useful
2292 for compile-time checking, especially together with @code{__builtin_constant_p}
2293 and inline functions where checking the inline function arguments is not
2294 possible through @code{extern char [(condition) ? 1 : -1];} tricks.
2295 While it is possible to leave the function undefined and thus invoke
2296 a link failure, when using this attribute the problem is diagnosed
2297 earlier and with exact location of the call even in presence of inline
2298 functions or when not emitting debugging information.
2300 @item warning ("@var{message}")
2301 @cindex @code{warning} function attribute
2302 If this attribute is used on a function declaration and a call to such a function
2303 is not eliminated through dead code elimination or other optimizations, a warning
2304 that includes @var{message} is diagnosed. This is useful
2305 for compile-time checking, especially together with @code{__builtin_constant_p}
2306 and inline functions. While it is possible to define the function with
2307 a message in @code{.gnu.warning*} section, when using this attribute the problem
2308 is diagnosed earlier and with exact location of the call even in presence
2309 of inline functions or when not emitting debugging information.
2312 @cindex functions that do pop the argument stack on the 386
2314 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2315 assume that the calling function pops off the stack space used to
2316 pass arguments. This is
2317 useful to override the effects of the @option{-mrtd} switch.
2320 @cindex @code{const} function attribute
2321 Many functions do not examine any values except their arguments, and
2322 have no effects except the return value. Basically this is just slightly
2323 more strict class than the @code{pure} attribute below, since function is not
2324 allowed to read global memory.
2326 @cindex pointer arguments
2327 Note that a function that has pointer arguments and examines the data
2328 pointed to must @emph{not} be declared @code{const}. Likewise, a
2329 function that calls a non-@code{const} function usually must not be
2330 @code{const}. It does not make sense for a @code{const} function to
2333 The attribute @code{const} is not implemented in GCC versions earlier
2334 than 2.5. An alternative way to declare that a function has no side
2335 effects, which works in the current version and in some older versions,
2339 typedef int intfn ();
2341 extern const intfn square;
2345 This approach does not work in GNU C++ from 2.6.0 on, since the language
2346 specifies that the @samp{const} must be attached to the return value.
2350 @itemx constructor (@var{priority})
2351 @itemx destructor (@var{priority})
2352 @cindex @code{constructor} function attribute
2353 @cindex @code{destructor} function attribute
2354 The @code{constructor} attribute causes the function to be called
2355 automatically before execution enters @code{main ()}. Similarly, the
2356 @code{destructor} attribute causes the function to be called
2357 automatically after @code{main ()} completes or @code{exit ()} is
2358 called. Functions with these attributes are useful for
2359 initializing data that is used implicitly during the execution of
2362 You may provide an optional integer priority to control the order in
2363 which constructor and destructor functions are run. A constructor
2364 with a smaller priority number runs before a constructor with a larger
2365 priority number; the opposite relationship holds for destructors. So,
2366 if you have a constructor that allocates a resource and a destructor
2367 that deallocates the same resource, both functions typically have the
2368 same priority. The priorities for constructor and destructor
2369 functions are the same as those specified for namespace-scope C++
2370 objects (@pxref{C++ Attributes}).
2372 These attributes are not currently implemented for Objective-C@.
2375 @itemx deprecated (@var{msg})
2376 @cindex @code{deprecated} attribute.
2377 The @code{deprecated} attribute results in a warning if the function
2378 is used anywhere in the source file. This is useful when identifying
2379 functions that are expected to be removed in a future version of a
2380 program. The warning also includes the location of the declaration
2381 of the deprecated function, to enable users to easily find further
2382 information about why the function is deprecated, or what they should
2383 do instead. Note that the warnings only occurs for uses:
2386 int old_fn () __attribute__ ((deprecated));
2388 int (*fn_ptr)() = old_fn;
2392 results in a warning on line 3 but not line 2. The optional @var{msg}
2393 argument, which must be a string, is printed in the warning if
2396 The @code{deprecated} attribute can also be used for variables and
2397 types (@pxref{Variable Attributes}, @pxref{Type Attributes}.)
2400 @cindex @code{disinterrupt} attribute
2401 On Epiphany and MeP targets, this attribute causes the compiler to emit
2402 instructions to disable interrupts for the duration of the given
2406 @cindex @code{__declspec(dllexport)}
2407 On Microsoft Windows targets and Symbian OS targets the
2408 @code{dllexport} attribute causes the compiler to provide a global
2409 pointer to a pointer in a DLL, so that it can be referenced with the
2410 @code{dllimport} attribute. On Microsoft Windows targets, the pointer
2411 name is formed by combining @code{_imp__} and the function or variable
2414 You can use @code{__declspec(dllexport)} as a synonym for
2415 @code{__attribute__ ((dllexport))} for compatibility with other
2418 On systems that support the @code{visibility} attribute, this
2419 attribute also implies ``default'' visibility. It is an error to
2420 explicitly specify any other visibility.
2422 In previous versions of GCC, the @code{dllexport} attribute was ignored
2423 for inlined functions, unless the @option{-fkeep-inline-functions} flag
2424 had been used. The default behavior now is to emit all dllexported
2425 inline functions; however, this can cause object file-size bloat, in
2426 which case the old behavior can be restored by using
2427 @option{-fno-keep-inline-dllexport}.
2429 The attribute is also ignored for undefined symbols.
2431 When applied to C++ classes, the attribute marks defined non-inlined
2432 member functions and static data members as exports. Static consts
2433 initialized in-class are not marked unless they are also defined
2436 For Microsoft Windows targets there are alternative methods for
2437 including the symbol in the DLL's export table such as using a
2438 @file{.def} file with an @code{EXPORTS} section or, with GNU ld, using
2439 the @option{--export-all} linker flag.
2442 @cindex @code{__declspec(dllimport)}
2443 On Microsoft Windows and Symbian OS targets, the @code{dllimport}
2444 attribute causes the compiler to reference a function or variable via
2445 a global pointer to a pointer that is set up by the DLL exporting the
2446 symbol. The attribute implies @code{extern}. On Microsoft Windows
2447 targets, the pointer name is formed by combining @code{_imp__} and the
2448 function or variable name.
2450 You can use @code{__declspec(dllimport)} as a synonym for
2451 @code{__attribute__ ((dllimport))} for compatibility with other
2454 On systems that support the @code{visibility} attribute, this
2455 attribute also implies ``default'' visibility. It is an error to
2456 explicitly specify any other visibility.
2458 Currently, the attribute is ignored for inlined functions. If the
2459 attribute is applied to a symbol @emph{definition}, an error is reported.
2460 If a symbol previously declared @code{dllimport} is later defined, the
2461 attribute is ignored in subsequent references, and a warning is emitted.
2462 The attribute is also overridden by a subsequent declaration as
2465 When applied to C++ classes, the attribute marks non-inlined
2466 member functions and static data members as imports. However, the
2467 attribute is ignored for virtual methods to allow creation of vtables
2470 On the SH Symbian OS target the @code{dllimport} attribute also has
2471 another affect---it can cause the vtable and run-time type information
2472 for a class to be exported. This happens when the class has a
2473 dllimported constructor or a non-inline, non-pure virtual function
2474 and, for either of those two conditions, the class also has an inline
2475 constructor or destructor and has a key function that is defined in
2476 the current translation unit.
2478 For Microsoft Windows targets the use of the @code{dllimport}
2479 attribute on functions is not necessary, but provides a small
2480 performance benefit by eliminating a thunk in the DLL@. The use of the
2481 @code{dllimport} attribute on imported variables was required on older
2482 versions of the GNU linker, but can now be avoided by passing the
2483 @option{--enable-auto-import} switch to the GNU linker. As with
2484 functions, using the attribute for a variable eliminates a thunk in
2487 One drawback to using this attribute is that a pointer to a
2488 @emph{variable} marked as @code{dllimport} cannot be used as a constant
2489 address. However, a pointer to a @emph{function} with the
2490 @code{dllimport} attribute can be used as a constant initializer; in
2491 this case, the address of a stub function in the import lib is
2492 referenced. On Microsoft Windows targets, the attribute can be disabled
2493 for functions by setting the @option{-mnop-fun-dllimport} flag.
2496 @cindex eight-bit data on the H8/300, H8/300H, and H8S
2497 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2498 variable should be placed into the eight-bit data section.
2499 The compiler generates more efficient code for certain operations
2500 on data in the eight-bit data area. Note the eight-bit data area is limited to
2503 You must use GAS and GLD from GNU binutils version 2.7 or later for
2504 this attribute to work correctly.
2506 @item exception_handler
2507 @cindex exception handler functions on the Blackfin processor
2508 Use this attribute on the Blackfin to indicate that the specified function
2509 is an exception handler. The compiler generates function entry and
2510 exit sequences suitable for use in an exception handler when this
2511 attribute is present.
2513 @item externally_visible
2514 @cindex @code{externally_visible} attribute.
2515 This attribute, attached to a global variable or function, nullifies
2516 the effect of the @option{-fwhole-program} command-line option, so the
2517 object remains visible outside the current compilation unit.
2519 If @option{-fwhole-program} is used together with @option{-flto} and
2520 @command{gold} is used as the linker plugin,
2521 @code{externally_visible} attributes are automatically added to functions
2522 (not variable yet due to a current @command{gold} issue)
2523 that are accessed outside of LTO objects according to resolution file
2524 produced by @command{gold}.
2525 For other linkers that cannot generate resolution file,
2526 explicit @code{externally_visible} attributes are still necessary.
2529 @cindex functions that handle memory bank switching
2530 On 68HC11 and 68HC12 the @code{far} attribute causes the compiler to
2531 use a calling convention that takes care of switching memory banks when
2532 entering and leaving a function. This calling convention is also the
2533 default when using the @option{-mlong-calls} option.
2535 On 68HC12 the compiler uses the @code{call} and @code{rtc} instructions
2536 to call and return from a function.
2538 On 68HC11 the compiler generates a sequence of instructions
2539 to invoke a board-specific routine to switch the memory bank and call the
2540 real function. The board-specific routine simulates a @code{call}.
2541 At the end of a function, it jumps to a board-specific routine
2542 instead of using @code{rts}. The board-specific return routine simulates
2545 On MeP targets this causes the compiler to use a calling convention
2546 that assumes the called function is too far away for the built-in
2549 @item fast_interrupt
2550 @cindex interrupt handler functions
2551 Use this attribute on the M32C and RX ports to indicate that the specified
2552 function is a fast interrupt handler. This is just like the
2553 @code{interrupt} attribute, except that @code{freit} is used to return
2554 instead of @code{reit}.
2557 @cindex functions that pop the argument stack on the 386
2558 On the Intel 386, the @code{fastcall} attribute causes the compiler to
2559 pass the first argument (if of integral type) in the register ECX and
2560 the second argument (if of integral type) in the register EDX@. Subsequent
2561 and other typed arguments are passed on the stack. The called function
2562 pops the arguments off the stack. If the number of arguments is variable all
2563 arguments are pushed on the stack.
2566 @cindex functions that pop the argument stack on the 386
2567 On the Intel 386, the @code{thiscall} attribute causes the compiler to
2568 pass the first argument (if of integral type) in the register ECX.
2569 Subsequent and other typed arguments are passed on the stack. The called
2570 function pops the arguments off the stack.
2571 If the number of arguments is variable all arguments are pushed on the
2573 The @code{thiscall} attribute is intended for C++ non-static member functions.
2574 As a GCC extension, this calling convention can be used for C functions
2575 and for static member methods.
2577 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2578 @cindex @code{format} function attribute
2580 The @code{format} attribute specifies that a function takes @code{printf},
2581 @code{scanf}, @code{strftime} or @code{strfmon} style arguments that
2582 should be type-checked against a format string. For example, the
2587 my_printf (void *my_object, const char *my_format, ...)
2588 __attribute__ ((format (printf, 2, 3)));
2592 causes the compiler to check the arguments in calls to @code{my_printf}
2593 for consistency with the @code{printf} style format string argument
2596 The parameter @var{archetype} determines how the format string is
2597 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime},
2598 @code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or
2599 @code{strfmon}. (You can also use @code{__printf__},
2600 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On
2601 MinGW targets, @code{ms_printf}, @code{ms_scanf}, and
2602 @code{ms_strftime} are also present.
2603 @var{archetype} values such as @code{printf} refer to the formats accepted
2604 by the system's C runtime library,
2605 while values prefixed with @samp{gnu_} always refer
2606 to the formats accepted by the GNU C Library. On Microsoft Windows
2607 targets, values prefixed with @samp{ms_} refer to the formats accepted by the
2608 @file{msvcrt.dll} library.
2609 The parameter @var{string-index}
2610 specifies which argument is the format string argument (starting
2611 from 1), while @var{first-to-check} is the number of the first
2612 argument to check against the format string. For functions
2613 where the arguments are not available to be checked (such as
2614 @code{vprintf}), specify the third parameter as zero. In this case the
2615 compiler only checks the format string for consistency. For
2616 @code{strftime} formats, the third parameter is required to be zero.
2617 Since non-static C++ methods have an implicit @code{this} argument, the
2618 arguments of such methods should be counted from two, not one, when
2619 giving values for @var{string-index} and @var{first-to-check}.
2621 In the example above, the format string (@code{my_format}) is the second
2622 argument of the function @code{my_print}, and the arguments to check
2623 start with the third argument, so the correct parameters for the format
2624 attribute are 2 and 3.
2626 @opindex ffreestanding
2627 @opindex fno-builtin
2628 The @code{format} attribute allows you to identify your own functions
2629 that take format strings as arguments, so that GCC can check the
2630 calls to these functions for errors. The compiler always (unless
2631 @option{-ffreestanding} or @option{-fno-builtin} is used) checks formats
2632 for the standard library functions @code{printf}, @code{fprintf},
2633 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2634 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2635 warnings are requested (using @option{-Wformat}), so there is no need to
2636 modify the header file @file{stdio.h}. In C99 mode, the functions
2637 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2638 @code{vsscanf} are also checked. Except in strictly conforming C
2639 standard modes, the X/Open function @code{strfmon} is also checked as
2640 are @code{printf_unlocked} and @code{fprintf_unlocked}.
2641 @xref{C Dialect Options,,Options Controlling C Dialect}.
2643 For Objective-C dialects, @code{NSString} (or @code{__NSString__}) is
2644 recognized in the same context. Declarations including these format attributes
2645 are parsed for correct syntax, however the result of checking of such format
2646 strings is not yet defined, and is not carried out by this version of the
2649 The target may also provide additional types of format checks.
2650 @xref{Target Format Checks,,Format Checks Specific to Particular
2653 @item format_arg (@var{string-index})
2654 @cindex @code{format_arg} function attribute
2655 @opindex Wformat-nonliteral
2656 The @code{format_arg} attribute specifies that a function takes a format
2657 string for a @code{printf}, @code{scanf}, @code{strftime} or
2658 @code{strfmon} style function and modifies it (for example, to translate
2659 it into another language), so the result can be passed to a
2660 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2661 function (with the remaining arguments to the format function the same
2662 as they would have been for the unmodified string). For example, the
2667 my_dgettext (char *my_domain, const char *my_format)
2668 __attribute__ ((format_arg (2)));
2672 causes the compiler to check the arguments in calls to a @code{printf},
2673 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2674 format string argument is a call to the @code{my_dgettext} function, for
2675 consistency with the format string argument @code{my_format}. If the
2676 @code{format_arg} attribute had not been specified, all the compiler
2677 could tell in such calls to format functions would be that the format
2678 string argument is not constant; this would generate a warning when
2679 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2680 without the attribute.
2682 The parameter @var{string-index} specifies which argument is the format
2683 string argument (starting from one). Since non-static C++ methods have
2684 an implicit @code{this} argument, the arguments of such methods should
2685 be counted from two.
2687 The @code{format_arg} attribute allows you to identify your own
2688 functions that modify format strings, so that GCC can check the
2689 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2690 type function whose operands are a call to one of your own function.
2691 The compiler always treats @code{gettext}, @code{dgettext}, and
2692 @code{dcgettext} in this manner except when strict ISO C support is
2693 requested by @option{-ansi} or an appropriate @option{-std} option, or
2694 @option{-ffreestanding} or @option{-fno-builtin}
2695 is used. @xref{C Dialect Options,,Options
2696 Controlling C Dialect}.
2698 For Objective-C dialects, the @code{format-arg} attribute may refer to an
2699 @code{NSString} reference for compatibility with the @code{format} attribute
2702 The target may also allow additional types in @code{format-arg} attributes.
2703 @xref{Target Format Checks,,Format Checks Specific to Particular
2706 @item function_vector
2707 @cindex calling functions through the function vector on H8/300, M16C, M32C and SH2A processors
2708 Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified
2709 function should be called through the function vector. Calling a
2710 function through the function vector reduces code size, however;
2711 the function vector has a limited size (maximum 128 entries on the H8/300
2712 and 64 entries on the H8/300H and H8S) and shares space with the interrupt vector.
2714 On SH2A targets, this attribute declares a function to be called using the
2715 TBR relative addressing mode. The argument to this attribute is the entry
2716 number of the same function in a vector table containing all the TBR
2717 relative addressable functions. For correct operation the TBR must be setup
2718 accordingly to point to the start of the vector table before any functions with
2719 this attribute are invoked. Usually a good place to do the initialization is
2720 the startup routine. The TBR relative vector table can have at max 256 function
2721 entries. The jumps to these functions are generated using a SH2A specific,
2722 non delayed branch instruction JSR/N @@(disp8,TBR). You must use GAS and GLD
2723 from GNU binutils version 2.7 or later for this attribute to work correctly.
2725 Please refer the example of M16C target, to see the use of this
2726 attribute while declaring a function,
2728 In an application, for a function being called once, this attribute
2729 saves at least 8 bytes of code; and if other successive calls are being
2730 made to the same function, it saves 2 bytes of code per each of these
2733 On M16C/M32C targets, the @code{function_vector} attribute declares a
2734 special page subroutine call function. Use of this attribute reduces
2735 the code size by 2 bytes for each call generated to the
2736 subroutine. The argument to the attribute is the vector number entry
2737 from the special page vector table which contains the 16 low-order
2738 bits of the subroutine's entry address. Each vector table has special
2739 page number (18 to 255) that is used in @code{jsrs} instructions.
2740 Jump addresses of the routines are generated by adding 0x0F0000 (in
2741 case of M16C targets) or 0xFF0000 (in case of M32C targets), to the
2742 2-byte addresses set in the vector table. Therefore you need to ensure
2743 that all the special page vector routines should get mapped within the
2744 address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF
2747 In the following example 2 bytes are saved for each call to
2748 function @code{foo}.
2751 void foo (void) __attribute__((function_vector(0x18)));
2762 If functions are defined in one file and are called in another file,
2763 then be sure to write this declaration in both files.
2765 This attribute is ignored for R8C target.
2767 @item ifunc ("@var{resolver}")
2768 @cindex @code{ifunc} attribute
2769 The @code{ifunc} attribute is used to mark a function as an indirect
2770 function using the STT_GNU_IFUNC symbol type extension to the ELF
2771 standard. This allows the resolution of the symbol value to be
2772 determined dynamically at load time, and an optimized version of the
2773 routine can be selected for the particular processor or other system
2774 characteristics determined then. To use this attribute, first define
2775 the implementation functions available, and a resolver function that
2776 returns a pointer to the selected implementation function. The
2777 implementation functions' declarations must match the API of the
2778 function being implemented, the resolver's declaration is be a
2779 function returning pointer to void function returning void:
2782 void *my_memcpy (void *dst, const void *src, size_t len)
2787 static void (*resolve_memcpy (void)) (void)
2789 return my_memcpy; // we'll just always select this routine
2794 The exported header file declaring the function the user calls would
2798 extern void *memcpy (void *, const void *, size_t);
2802 allowing the user to call this as a regular function, unaware of the
2803 implementation. Finally, the indirect function needs to be defined in
2804 the same translation unit as the resolver function:
2807 void *memcpy (void *, const void *, size_t)
2808 __attribute__ ((ifunc ("resolve_memcpy")));
2811 Indirect functions cannot be weak, and require a recent binutils (at
2812 least version 2.20.1), and GNU C library (at least version 2.11.1).
2815 @cindex interrupt handler functions
2816 Use this attribute on the ARM, AVR, CR16, Epiphany, M32C, M32R/D, m68k, MeP, MIPS,
2817 MSP430, RL78, RX and Xstormy16 ports to indicate that the specified function is an
2818 interrupt handler. The compiler generates function entry and exit
2819 sequences suitable for use in an interrupt handler when this attribute
2820 is present. With Epiphany targets it may also generate a special section with
2821 code to initialize the interrupt vector table.
2823 Note, interrupt handlers for the Blackfin, H8/300, H8/300H, H8S, MicroBlaze,
2824 and SH processors can be specified via the @code{interrupt_handler} attribute.
2826 Note, on the AVR, the hardware globally disables interrupts when an
2827 interrupt is executed. The first instruction of an interrupt handler
2828 declared with this attribute is a @code{SEI} instruction to
2829 re-enable interrupts. See also the @code{signal} function attribute
2830 that does not insert a @code{SEI} instruction. If both @code{signal} and
2831 @code{interrupt} are specified for the same function, @code{signal}
2832 is silently ignored.
2834 Note, for the ARM, you can specify the kind of interrupt to be handled by
2835 adding an optional parameter to the interrupt attribute like this:
2838 void f () __attribute__ ((interrupt ("IRQ")));
2842 Permissible values for this parameter are: @code{IRQ}, @code{FIQ},
2843 @code{SWI}, @code{ABORT} and @code{UNDEF}.
2845 On ARMv7-M the interrupt type is ignored, and the attribute means the function
2846 may be called with a word-aligned stack pointer.
2848 Note, for the MSP430 you can provide an argument to the interrupt
2849 attribute which specifies a name or number. If the argument is a
2850 number it indicates the slot in the interrupt vector table (0 - 31) to
2851 which this handler should be assigned. If the argument is a name it
2852 is treated as a symbolic name for the vector slot. These names should
2853 match up with appropriate entries in the linker script. By default
2854 the names @code{watchdog} for vector 26, @code{nmi} for vector 30 and
2855 @code{reset} for vector 31 are recognised.
2857 You can also use the following function attributes to modify how
2858 normal functions interact with interrupt functions:
2862 @cindex @code{critical} attribute
2863 Critical functions disable interrupts upon entry and restore the
2864 previous interrupt state upon exit. Critical functions cannot also
2865 have the @code{naked} or @code{reentrant} attributes. They can have
2866 the @code{interrupt} attribute.
2869 @cindex @code{reentrant} attribute
2870 Reentrant functions disable interrupts upon entry and enable them
2871 upon exit. Reentrant functions cannot also have the @code{naked}
2872 or @code{critical} attributes. They can have the @code{interrupt}
2877 On Epiphany targets one or more optional parameters can be added like this:
2880 void __attribute__ ((interrupt ("dma0, dma1"))) universal_dma_handler ();
2883 Permissible values for these parameters are: @w{@code{reset}},
2884 @w{@code{software_exception}}, @w{@code{page_miss}},
2885 @w{@code{timer0}}, @w{@code{timer1}}, @w{@code{message}},
2886 @w{@code{dma0}}, @w{@code{dma1}}, @w{@code{wand}} and @w{@code{swi}}.
2887 Multiple parameters indicate that multiple entries in the interrupt
2888 vector table should be initialized for this function, i.e.@: for each
2889 parameter @w{@var{name}}, a jump to the function is emitted in
2890 the section @w{ivt_entry_@var{name}}. The parameter(s) may be omitted
2891 entirely, in which case no interrupt vector table entry is provided.
2893 Note, on Epiphany targets, interrupts are enabled inside the function
2894 unless the @code{disinterrupt} attribute is also specified.
2896 On Epiphany targets, you can also use the following attribute to
2897 modify the behavior of an interrupt handler:
2899 @item forwarder_section
2900 @cindex @code{forwarder_section} attribute
2901 The interrupt handler may be in external memory which cannot be
2902 reached by a branch instruction, so generate a local memory trampoline
2903 to transfer control. The single parameter identifies the section where
2904 the trampoline is placed.
2907 The following examples are all valid uses of these attributes on
2910 void __attribute__ ((interrupt)) universal_handler ();
2911 void __attribute__ ((interrupt ("dma1"))) dma1_handler ();
2912 void __attribute__ ((interrupt ("dma0, dma1"))) universal_dma_handler ();
2913 void __attribute__ ((interrupt ("timer0"), disinterrupt))
2914 fast_timer_handler ();
2915 void __attribute__ ((interrupt ("dma0, dma1"), forwarder_section ("tramp")))
2916 external_dma_handler ();
2919 On MIPS targets, you can use the following attributes to modify the behavior
2920 of an interrupt handler:
2922 @item use_shadow_register_set
2923 @cindex @code{use_shadow_register_set} attribute
2924 Assume that the handler uses a shadow register set, instead of
2925 the main general-purpose registers.
2927 @item keep_interrupts_masked
2928 @cindex @code{keep_interrupts_masked} attribute
2929 Keep interrupts masked for the whole function. Without this attribute,
2930 GCC tries to reenable interrupts for as much of the function as it can.
2932 @item use_debug_exception_return
2933 @cindex @code{use_debug_exception_return} attribute
2934 Return using the @code{deret} instruction. Interrupt handlers that don't
2935 have this attribute return using @code{eret} instead.
2938 You can use any combination of these attributes, as shown below:
2940 void __attribute__ ((interrupt)) v0 ();
2941 void __attribute__ ((interrupt, use_shadow_register_set)) v1 ();
2942 void __attribute__ ((interrupt, keep_interrupts_masked)) v2 ();
2943 void __attribute__ ((interrupt, use_debug_exception_return)) v3 ();
2944 void __attribute__ ((interrupt, use_shadow_register_set,
2945 keep_interrupts_masked)) v4 ();
2946 void __attribute__ ((interrupt, use_shadow_register_set,
2947 use_debug_exception_return)) v5 ();
2948 void __attribute__ ((interrupt, keep_interrupts_masked,
2949 use_debug_exception_return)) v6 ();
2950 void __attribute__ ((interrupt, use_shadow_register_set,
2951 keep_interrupts_masked,
2952 use_debug_exception_return)) v7 ();
2955 On RL78, use @code{brk_interrupt} instead of @code{interrupt} for
2956 handlers intended to be used with the @code{BRK} opcode (i.e.@: those
2957 that must end with @code{RETB} instead of @code{RETI}).
2959 @item interrupt_handler
2960 @cindex interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors
2961 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S, and SH to
2962 indicate that the specified function is an interrupt handler. The compiler
2963 generates function entry and exit sequences suitable for use in an
2964 interrupt handler when this attribute is present.
2966 @item interrupt_thread
2967 @cindex interrupt thread functions on fido
2968 Use this attribute on fido, a subarchitecture of the m68k, to indicate
2969 that the specified function is an interrupt handler that is designed
2970 to run as a thread. The compiler omits generate prologue/epilogue
2971 sequences and replaces the return instruction with a @code{sleep}
2972 instruction. This attribute is available only on fido.
2975 @cindex interrupt service routines on ARM
2976 Use this attribute on ARM to write Interrupt Service Routines. This is an
2977 alias to the @code{interrupt} attribute above.
2980 @cindex User stack pointer in interrupts on the Blackfin
2981 When used together with @code{interrupt_handler}, @code{exception_handler}
2982 or @code{nmi_handler}, code is generated to load the stack pointer
2983 from the USP register in the function prologue.
2986 @cindex @code{l1_text} function attribute
2987 This attribute specifies a function to be placed into L1 Instruction
2988 SRAM@. The function is put into a specific section named @code{.l1.text}.
2989 With @option{-mfdpic}, function calls with a such function as the callee
2990 or caller uses inlined PLT.
2993 @cindex @code{l2} function attribute
2994 On the Blackfin, this attribute specifies a function to be placed into L2
2995 SRAM. The function is put into a specific section named
2996 @code{.l1.text}. With @option{-mfdpic}, callers of such functions use
3000 @cindex @code{leaf} function attribute
3001 Calls to external functions with this attribute must return to the current
3002 compilation unit only by return or by exception handling. In particular, leaf
3003 functions are not allowed to call callback function passed to it from the current
3004 compilation unit or directly call functions exported by the unit or longjmp
3005 into the unit. Leaf function might still call functions from other compilation
3006 units and thus they are not necessarily leaf in the sense that they contain no
3007 function calls at all.
3009 The attribute is intended for library functions to improve dataflow analysis.
3010 The compiler takes the hint that any data not escaping the current compilation unit can
3011 not be used or modified by the leaf function. For example, the @code{sin} function
3012 is a leaf function, but @code{qsort} is not.
3014 Note that leaf functions might invoke signals and signal handlers might be
3015 defined in the current compilation unit and use static variables. The only
3016 compliant way to write such a signal handler is to declare such variables
3019 The attribute has no effect on functions defined within the current compilation
3020 unit. This is to allow easy merging of multiple compilation units into one,
3021 for example, by using the link-time optimization. For this reason the
3022 attribute is not allowed on types to annotate indirect calls.
3024 @item long_call/short_call
3025 @cindex indirect calls on ARM
3026 This attribute specifies how a particular function is called on
3027 ARM and Epiphany. Both attributes override the
3028 @option{-mlong-calls} (@pxref{ARM Options})
3029 command-line switch and @code{#pragma long_calls} settings. The
3030 @code{long_call} attribute indicates that the function might be far
3031 away from the call site and require a different (more expensive)
3032 calling sequence. The @code{short_call} attribute always places
3033 the offset to the function from the call site into the @samp{BL}
3034 instruction directly.
3036 @item longcall/shortcall
3037 @cindex functions called via pointer on the RS/6000 and PowerPC
3038 On the Blackfin, RS/6000 and PowerPC, the @code{longcall} attribute
3039 indicates that the function might be far away from the call site and
3040 require a different (more expensive) calling sequence. The
3041 @code{shortcall} attribute indicates that the function is always close
3042 enough for the shorter calling sequence to be used. These attributes
3043 override both the @option{-mlongcall} switch and, on the RS/6000 and
3044 PowerPC, the @code{#pragma longcall} setting.
3046 @xref{RS/6000 and PowerPC Options}, for more information on whether long
3047 calls are necessary.
3049 @item long_call/near/far
3050 @cindex indirect calls on MIPS
3051 These attributes specify how a particular function is called on MIPS@.
3052 The attributes override the @option{-mlong-calls} (@pxref{MIPS Options})
3053 command-line switch. The @code{long_call} and @code{far} attributes are
3054 synonyms, and cause the compiler to always call
3055 the function by first loading its address into a register, and then using
3056 the contents of that register. The @code{near} attribute has the opposite
3057 effect; it specifies that non-PIC calls should be made using the more
3058 efficient @code{jal} instruction.
3061 @cindex @code{malloc} attribute
3062 The @code{malloc} attribute is used to tell the compiler that a function
3063 may be treated as if any non-@code{NULL} pointer it returns cannot
3064 alias any other pointer valid when the function returns and that the memory
3065 has undefined content.
3066 This often improves optimization.
3067 Standard functions with this property include @code{malloc} and
3068 @code{calloc}. @code{realloc}-like functions do not have this
3069 property as the memory pointed to does not have undefined content.
3071 @item mips16/nomips16
3072 @cindex @code{mips16} attribute
3073 @cindex @code{nomips16} attribute
3075 On MIPS targets, you can use the @code{mips16} and @code{nomips16}
3076 function attributes to locally select or turn off MIPS16 code generation.
3077 A function with the @code{mips16} attribute is emitted as MIPS16 code,
3078 while MIPS16 code generation is disabled for functions with the
3079 @code{nomips16} attribute. These attributes override the
3080 @option{-mips16} and @option{-mno-mips16} options on the command line
3081 (@pxref{MIPS Options}).
3083 When compiling files containing mixed MIPS16 and non-MIPS16 code, the
3084 preprocessor symbol @code{__mips16} reflects the setting on the command line,
3085 not that within individual functions. Mixed MIPS16 and non-MIPS16 code
3086 may interact badly with some GCC extensions such as @code{__builtin_apply}
3087 (@pxref{Constructing Calls}).
3089 @item micromips/nomicromips
3090 @cindex @code{micromips} attribute
3091 @cindex @code{nomicromips} attribute
3093 On MIPS targets, you can use the @code{micromips} and @code{nomicromips}
3094 function attributes to locally select or turn off microMIPS code generation.
3095 A function with the @code{micromips} attribute is emitted as microMIPS code,
3096 while microMIPS code generation is disabled for functions with the
3097 @code{nomicromips} attribute. These attributes override the
3098 @option{-mmicromips} and @option{-mno-micromips} options on the command line
3099 (@pxref{MIPS Options}).
3101 When compiling files containing mixed microMIPS and non-microMIPS code, the
3102 preprocessor symbol @code{__mips_micromips} reflects the setting on the
3104 not that within individual functions. Mixed microMIPS and non-microMIPS code
3105 may interact badly with some GCC extensions such as @code{__builtin_apply}
3106 (@pxref{Constructing Calls}).
3108 @item model (@var{model-name})
3109 @cindex function addressability on the M32R/D
3110 @cindex variable addressability on the IA-64
3112 On the M32R/D, use this attribute to set the addressability of an
3113 object, and of the code generated for a function. The identifier
3114 @var{model-name} is one of @code{small}, @code{medium}, or
3115 @code{large}, representing each of the code models.
3117 Small model objects live in the lower 16MB of memory (so that their
3118 addresses can be loaded with the @code{ld24} instruction), and are
3119 callable with the @code{bl} instruction.
3121 Medium model objects may live anywhere in the 32-bit address space (the
3122 compiler generates @code{seth/add3} instructions to load their addresses),
3123 and are callable with the @code{bl} instruction.
3125 Large model objects may live anywhere in the 32-bit address space (the
3126 compiler generates @code{seth/add3} instructions to load their addresses),
3127 and may not be reachable with the @code{bl} instruction (the compiler
3128 generates the much slower @code{seth/add3/jl} instruction sequence).
3130 On IA-64, use this attribute to set the addressability of an object.
3131 At present, the only supported identifier for @var{model-name} is
3132 @code{small}, indicating addressability via ``small'' (22-bit)
3133 addresses (so that their addresses can be loaded with the @code{addl}
3134 instruction). Caveat: such addressing is by definition not position
3135 independent and hence this attribute must not be used for objects
3136 defined by shared libraries.
3138 @item ms_abi/sysv_abi
3139 @cindex @code{ms_abi} attribute
3140 @cindex @code{sysv_abi} attribute
3142 On 32-bit and 64-bit (i?86|x86_64)-*-* targets, you can use an ABI attribute
3143 to indicate which calling convention should be used for a function. The
3144 @code{ms_abi} attribute tells the compiler to use the Microsoft ABI,
3145 while the @code{sysv_abi} attribute tells the compiler to use the ABI
3146 used on GNU/Linux and other systems. The default is to use the Microsoft ABI
3147 when targeting Windows. On all other systems, the default is the x86/AMD ABI.
3149 Note, the @code{ms_abi} attribute for Microsoft Windows 64-bit targets currently
3150 requires the @option{-maccumulate-outgoing-args} option.
3152 @item callee_pop_aggregate_return (@var{number})
3153 @cindex @code{callee_pop_aggregate_return} attribute
3155 On 32-bit i?86-*-* targets, you can use this attribute to control how
3156 aggregates are returned in memory. If the caller is responsible for
3157 popping the hidden pointer together with the rest of the arguments, specify
3158 @var{number} equal to zero. If callee is responsible for popping the
3159 hidden pointer, specify @var{number} equal to one.
3161 The default i386 ABI assumes that the callee pops the
3162 stack for hidden pointer. However, on 32-bit i386 Microsoft Windows targets,
3163 the compiler assumes that the
3164 caller pops the stack for hidden pointer.
3166 @item ms_hook_prologue
3167 @cindex @code{ms_hook_prologue} attribute
3169 On 32-bit i[34567]86-*-* targets and 64-bit x86_64-*-* targets, you can use
3170 this function attribute to make GCC generate the ``hot-patching'' function
3171 prologue used in Win32 API functions in Microsoft Windows XP Service Pack 2
3175 @cindex function without a prologue/epilogue code
3176 Use this attribute on the ARM, AVR, MCORE, MSP430, RL78, RX and SPU ports to indicate that
3177 the specified function does not need prologue/epilogue sequences generated by
3178 the compiler. It is up to the programmer to provide these sequences. The
3179 only statements that can be safely included in naked functions are
3180 @code{asm} statements that do not have operands. All other statements,
3181 including declarations of local variables, @code{if} statements, and so
3182 forth, should be avoided. Naked functions should be used to implement the
3183 body of an assembly function, while allowing the compiler to construct
3184 the requisite function declaration for the assembler.
3187 @cindex functions that do not handle memory bank switching on 68HC11/68HC12
3188 On 68HC11 and 68HC12 the @code{near} attribute causes the compiler to
3189 use the normal calling convention based on @code{jsr} and @code{rts}.
3190 This attribute can be used to cancel the effect of the @option{-mlong-calls}
3193 On MeP targets this attribute causes the compiler to assume the called
3194 function is close enough to use the normal calling convention,
3195 overriding the @option{-mtf} command-line option.
3198 @cindex Allow nesting in an interrupt handler on the Blackfin processor.
3199 Use this attribute together with @code{interrupt_handler},
3200 @code{exception_handler} or @code{nmi_handler} to indicate that the function
3201 entry code should enable nested interrupts or exceptions.
3204 @cindex NMI handler functions on the Blackfin processor
3205 Use this attribute on the Blackfin to indicate that the specified function
3206 is an NMI handler. The compiler generates function entry and
3207 exit sequences suitable for use in an NMI handler when this
3208 attribute is present.
3211 @cindex @code{nocompression} attribute
3212 On MIPS targets, you can use the @code{nocompression} function attribute
3213 to locally turn off MIPS16 and microMIPS code generation. This attribute
3214 overrides the @option{-mips16} and @option{-mmicromips} options on the
3215 command line (@pxref{MIPS Options}).
3217 @item no_instrument_function
3218 @cindex @code{no_instrument_function} function attribute
3219 @opindex finstrument-functions
3220 If @option{-finstrument-functions} is given, profiling function calls are
3221 generated at entry and exit of most user-compiled functions.
3222 Functions with this attribute are not so instrumented.
3224 @item no_split_stack
3225 @cindex @code{no_split_stack} function attribute
3226 @opindex fsplit-stack
3227 If @option{-fsplit-stack} is given, functions have a small
3228 prologue which decides whether to split the stack. Functions with the
3229 @code{no_split_stack} attribute do not have that prologue, and thus
3230 may run with only a small amount of stack space available.
3233 @cindex @code{noinline} function attribute
3234 This function attribute prevents a function from being considered for
3236 @c Don't enumerate the optimizations by name here; we try to be
3237 @c future-compatible with this mechanism.
3238 If the function does not have side-effects, there are optimizations
3239 other than inlining that cause function calls to be optimized away,
3240 although the function call is live. To keep such calls from being
3247 (@pxref{Extended Asm}) in the called function, to serve as a special
3251 @cindex @code{noclone} function attribute
3252 This function attribute prevents a function from being considered for
3253 cloning---a mechanism that produces specialized copies of functions
3254 and which is (currently) performed by interprocedural constant
3257 @item nonnull (@var{arg-index}, @dots{})
3258 @cindex @code{nonnull} function attribute
3259 The @code{nonnull} attribute specifies that some function parameters should
3260 be non-null pointers. For instance, the declaration:
3264 my_memcpy (void *dest, const void *src, size_t len)
3265 __attribute__((nonnull (1, 2)));
3269 causes the compiler to check that, in calls to @code{my_memcpy},
3270 arguments @var{dest} and @var{src} are non-null. If the compiler
3271 determines that a null pointer is passed in an argument slot marked
3272 as non-null, and the @option{-Wnonnull} option is enabled, a warning
3273 is issued. The compiler may also choose to make optimizations based
3274 on the knowledge that certain function arguments will never be null.
3276 If no argument index list is given to the @code{nonnull} attribute,
3277 all pointer arguments are marked as non-null. To illustrate, the
3278 following declaration is equivalent to the previous example:
3282 my_memcpy (void *dest, const void *src, size_t len)
3283 __attribute__((nonnull));
3287 @cindex @code{noreturn} function attribute
3288 A few standard library functions, such as @code{abort} and @code{exit},
3289 cannot return. GCC knows this automatically. Some programs define
3290 their own functions that never return. You can declare them
3291 @code{noreturn} to tell the compiler this fact. For example,
3295 void fatal () __attribute__ ((noreturn));
3298 fatal (/* @r{@dots{}} */)
3300 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */
3306 The @code{noreturn} keyword tells the compiler to assume that
3307 @code{fatal} cannot return. It can then optimize without regard to what
3308 would happen if @code{fatal} ever did return. This makes slightly
3309 better code. More importantly, it helps avoid spurious warnings of
3310 uninitialized variables.
3312 The @code{noreturn} keyword does not affect the exceptional path when that
3313 applies: a @code{noreturn}-marked function may still return to the caller
3314 by throwing an exception or calling @code{longjmp}.
3316 Do not assume that registers saved by the calling function are
3317 restored before calling the @code{noreturn} function.
3319 It does not make sense for a @code{noreturn} function to have a return
3320 type other than @code{void}.
3322 The attribute @code{noreturn} is not implemented in GCC versions
3323 earlier than 2.5. An alternative way to declare that a function does
3324 not return, which works in the current version and in some older
3325 versions, is as follows:
3328 typedef void voidfn ();
3330 volatile voidfn fatal;
3334 This approach does not work in GNU C++.
3337 @cindex @code{nothrow} function attribute
3338 The @code{nothrow} attribute is used to inform the compiler that a
3339 function cannot throw an exception. For example, most functions in
3340 the standard C library can be guaranteed not to throw an exception
3341 with the notable exceptions of @code{qsort} and @code{bsearch} that
3342 take function pointer arguments. The @code{nothrow} attribute is not
3343 implemented in GCC versions earlier than 3.3.
3345 @item nosave_low_regs
3346 @cindex @code{nosave_low_regs} attribute
3347 Use this attribute on SH targets to indicate that an @code{interrupt_handler}
3348 function should not save and restore registers R0..R7. This can be used on SH3*
3349 and SH4* targets that have a second R0..R7 register bank for non-reentrant
3353 @cindex @code{optimize} function attribute
3354 The @code{optimize} attribute is used to specify that a function is to
3355 be compiled with different optimization options than specified on the
3356 command line. Arguments can either be numbers or strings. Numbers
3357 are assumed to be an optimization level. Strings that begin with
3358 @code{O} are assumed to be an optimization option, while other options
3359 are assumed to be used with a @code{-f} prefix. You can also use the
3360 @samp{#pragma GCC optimize} pragma to set the optimization options
3361 that affect more than one function.
3362 @xref{Function Specific Option Pragmas}, for details about the
3363 @samp{#pragma GCC optimize} pragma.
3365 This can be used for instance to have frequently-executed functions
3366 compiled with more aggressive optimization options that produce faster
3367 and larger code, while other functions can be compiled with less
3370 @item OS_main/OS_task
3371 @cindex @code{OS_main} AVR function attribute
3372 @cindex @code{OS_task} AVR function attribute
3373 On AVR, functions with the @code{OS_main} or @code{OS_task} attribute
3374 do not save/restore any call-saved register in their prologue/epilogue.
3376 The @code{OS_main} attribute can be used when there @emph{is
3377 guarantee} that interrupts are disabled at the time when the function
3378 is entered. This saves resources when the stack pointer has to be
3379 changed to set up a frame for local variables.
3381 The @code{OS_task} attribute can be used when there is @emph{no
3382 guarantee} that interrupts are disabled at that time when the function
3383 is entered like for, e@.g@. task functions in a multi-threading operating
3384 system. In that case, changing the stack pointer register is
3385 guarded by save/clear/restore of the global interrupt enable flag.
3387 The differences to the @code{naked} function attribute are:
3389 @item @code{naked} functions do not have a return instruction whereas
3390 @code{OS_main} and @code{OS_task} functions have a @code{RET} or
3391 @code{RETI} return instruction.
3392 @item @code{naked} functions do not set up a frame for local variables
3393 or a frame pointer whereas @code{OS_main} and @code{OS_task} do this
3398 @cindex @code{pcs} function attribute
3400 The @code{pcs} attribute can be used to control the calling convention
3401 used for a function on ARM. The attribute takes an argument that specifies
3402 the calling convention to use.
3404 When compiling using the AAPCS ABI (or a variant of it) then valid
3405 values for the argument are @code{"aapcs"} and @code{"aapcs-vfp"}. In
3406 order to use a variant other than @code{"aapcs"} then the compiler must
3407 be permitted to use the appropriate co-processor registers (i.e., the
3408 VFP registers must be available in order to use @code{"aapcs-vfp"}).
3412 /* Argument passed in r0, and result returned in r0+r1. */
3413 double f2d (float) __attribute__((pcs("aapcs")));
3416 Variadic functions always use the @code{"aapcs"} calling convention and
3417 the compiler rejects attempts to specify an alternative.
3420 @cindex @code{pure} function attribute
3421 Many functions have no effects except the return value and their
3422 return value depends only on the parameters and/or global variables.
3423 Such a function can be subject
3424 to common subexpression elimination and loop optimization just as an
3425 arithmetic operator would be. These functions should be declared
3426 with the attribute @code{pure}. For example,
3429 int square (int) __attribute__ ((pure));
3433 says that the hypothetical function @code{square} is safe to call
3434 fewer times than the program says.
3436 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
3437 Interesting non-pure functions are functions with infinite loops or those
3438 depending on volatile memory or other system resource, that may change between
3439 two consecutive calls (such as @code{feof} in a multithreading environment).
3441 The attribute @code{pure} is not implemented in GCC versions earlier
3445 @cindex @code{hot} function attribute
3446 The @code{hot} attribute on a function is used to inform the compiler that
3447 the function is a hot spot of the compiled program. The function is
3448 optimized more aggressively and on many target it is placed into special
3449 subsection of the text section so all hot functions appears close together
3452 When profile feedback is available, via @option{-fprofile-use}, hot functions
3453 are automatically detected and this attribute is ignored.
3455 The @code{hot} attribute on functions is not implemented in GCC versions
3458 @cindex @code{hot} label attribute
3459 The @code{hot} attribute on a label is used to inform the compiler that
3460 path following the label are more likely than paths that are not so
3461 annotated. This attribute is used in cases where @code{__builtin_expect}
3462 cannot be used, for instance with computed goto or @code{asm goto}.
3464 The @code{hot} attribute on labels is not implemented in GCC versions
3468 @cindex @code{cold} function attribute
3469 The @code{cold} attribute on functions is used to inform the compiler that
3470 the function is unlikely to be executed. The function is optimized for
3471 size rather than speed and on many targets it is placed into special
3472 subsection of the text section so all cold functions appears close together
3473 improving code locality of non-cold parts of program. The paths leading
3474 to call of cold functions within code are marked as unlikely by the branch
3475 prediction mechanism. It is thus useful to mark functions used to handle
3476 unlikely conditions, such as @code{perror}, as cold to improve optimization
3477 of hot functions that do call marked functions in rare occasions.
3479 When profile feedback is available, via @option{-fprofile-use}, cold functions
3480 are automatically detected and this attribute is ignored.
3482 The @code{cold} attribute on functions is not implemented in GCC versions
3485 @cindex @code{cold} label attribute
3486 The @code{cold} attribute on labels is used to inform the compiler that
3487 the path following the label is unlikely to be executed. This attribute
3488 is used in cases where @code{__builtin_expect} cannot be used, for instance
3489 with computed goto or @code{asm goto}.
3491 The @code{cold} attribute on labels is not implemented in GCC versions
3494 @item no_sanitize_address
3495 @itemx no_address_safety_analysis
3496 @cindex @code{no_sanitize_address} function attribute
3497 The @code{no_sanitize_address} attribute on functions is used
3498 to inform the compiler that it should not instrument memory accesses
3499 in the function when compiling with the @option{-fsanitize=address} option.
3500 The @code{no_address_safety_analysis} is a deprecated alias of the
3501 @code{no_sanitize_address} attribute, new code should use
3502 @code{no_sanitize_address}.
3504 @item no_sanitize_undefined
3505 @cindex @code{no_sanitize_undefined} function attribute
3506 The @code{no_sanitize_undefined} attribute on functions is used
3507 to inform the compiler that it should not check for undefined behavior
3508 in the function when compiling with the @option{-fsanitize=undefined} option.
3510 @item regparm (@var{number})
3511 @cindex @code{regparm} attribute
3512 @cindex functions that are passed arguments in registers on the 386
3513 On the Intel 386, the @code{regparm} attribute causes the compiler to
3514 pass arguments number one to @var{number} if they are of integral type
3515 in registers EAX, EDX, and ECX instead of on the stack. Functions that
3516 take a variable number of arguments continue to be passed all of their
3517 arguments on the stack.
3519 Beware that on some ELF systems this attribute is unsuitable for
3520 global functions in shared libraries with lazy binding (which is the
3521 default). Lazy binding sends the first call via resolving code in
3522 the loader, which might assume EAX, EDX and ECX can be clobbered, as
3523 per the standard calling conventions. Solaris 8 is affected by this.
3524 Systems with the GNU C Library version 2.1 or higher
3525 and FreeBSD are believed to be
3526 safe since the loaders there save EAX, EDX and ECX. (Lazy binding can be
3527 disabled with the linker or the loader if desired, to avoid the
3531 @cindex @code{sseregparm} attribute
3532 On the Intel 386 with SSE support, the @code{sseregparm} attribute
3533 causes the compiler to pass up to 3 floating-point arguments in
3534 SSE registers instead of on the stack. Functions that take a
3535 variable number of arguments continue to pass all of their
3536 floating-point arguments on the stack.
3538 @item force_align_arg_pointer
3539 @cindex @code{force_align_arg_pointer} attribute
3540 On the Intel x86, the @code{force_align_arg_pointer} attribute may be
3541 applied to individual function definitions, generating an alternate
3542 prologue and epilogue that realigns the run-time stack if necessary.
3543 This supports mixing legacy codes that run with a 4-byte aligned stack
3544 with modern codes that keep a 16-byte stack for SSE compatibility.
3547 @cindex @code{renesas} attribute
3548 On SH targets this attribute specifies that the function or struct follows the
3552 @cindex @code{resbank} attribute
3553 On the SH2A target, this attribute enables the high-speed register
3554 saving and restoration using a register bank for @code{interrupt_handler}
3555 routines. Saving to the bank is performed automatically after the CPU
3556 accepts an interrupt that uses a register bank.
3558 The nineteen 32-bit registers comprising general register R0 to R14,
3559 control register GBR, and system registers MACH, MACL, and PR and the
3560 vector table address offset are saved into a register bank. Register
3561 banks are stacked in first-in last-out (FILO) sequence. Restoration
3562 from the bank is executed by issuing a RESBANK instruction.
3565 @cindex @code{returns_twice} attribute
3566 The @code{returns_twice} attribute tells the compiler that a function may
3567 return more than one time. The compiler ensures that all registers
3568 are dead before calling such a function and emits a warning about
3569 the variables that may be clobbered after the second return from the
3570 function. Examples of such functions are @code{setjmp} and @code{vfork}.
3571 The @code{longjmp}-like counterpart of such function, if any, might need
3572 to be marked with the @code{noreturn} attribute.
3575 @cindex save all registers on the Blackfin, H8/300, H8/300H, and H8S
3576 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to indicate that
3577 all registers except the stack pointer should be saved in the prologue
3578 regardless of whether they are used or not.
3580 @item save_volatiles
3581 @cindex save volatile registers on the MicroBlaze
3582 Use this attribute on the MicroBlaze to indicate that the function is
3583 an interrupt handler. All volatile registers (in addition to non-volatile
3584 registers) are saved in the function prologue. If the function is a leaf
3585 function, only volatiles used by the function are saved. A normal function
3586 return is generated instead of a return from interrupt.
3588 @item section ("@var{section-name}")
3589 @cindex @code{section} function attribute
3590 Normally, the compiler places the code it generates in the @code{text} section.
3591 Sometimes, however, you need additional sections, or you need certain
3592 particular functions to appear in special sections. The @code{section}
3593 attribute specifies that a function lives in a particular section.
3594 For example, the declaration:
3597 extern void foobar (void) __attribute__ ((section ("bar")));
3601 puts the function @code{foobar} in the @code{bar} section.
3603 Some file formats do not support arbitrary sections so the @code{section}
3604 attribute is not available on all platforms.
3605 If you need to map the entire contents of a module to a particular
3606 section, consider using the facilities of the linker instead.
3609 @cindex @code{sentinel} function attribute
3610 This function attribute ensures that a parameter in a function call is
3611 an explicit @code{NULL}. The attribute is only valid on variadic
3612 functions. By default, the sentinel is located at position zero, the
3613 last parameter of the function call. If an optional integer position
3614 argument P is supplied to the attribute, the sentinel must be located at
3615 position P counting backwards from the end of the argument list.
3618 __attribute__ ((sentinel))
3620 __attribute__ ((sentinel(0)))
3623 The attribute is automatically set with a position of 0 for the built-in
3624 functions @code{execl} and @code{execlp}. The built-in function
3625 @code{execle} has the attribute set with a position of 1.
3627 A valid @code{NULL} in this context is defined as zero with any pointer
3628 type. If your system defines the @code{NULL} macro with an integer type
3629 then you need to add an explicit cast. GCC replaces @code{stddef.h}
3630 with a copy that redefines NULL appropriately.
3632 The warnings for missing or incorrect sentinels are enabled with
3636 See @code{long_call/short_call}.
3639 See @code{longcall/shortcall}.
3642 @cindex interrupt handler functions on the AVR processors
3643 Use this attribute on the AVR to indicate that the specified
3644 function is an interrupt handler. The compiler generates function
3645 entry and exit sequences suitable for use in an interrupt handler when this
3646 attribute is present.
3648 See also the @code{interrupt} function attribute.
3650 The AVR hardware globally disables interrupts when an interrupt is executed.
3651 Interrupt handler functions defined with the @code{signal} attribute
3652 do not re-enable interrupts. It is save to enable interrupts in a
3653 @code{signal} handler. This ``save'' only applies to the code
3654 generated by the compiler and not to the IRQ layout of the
3655 application which is responsibility of the application.
3657 If both @code{signal} and @code{interrupt} are specified for the same
3658 function, @code{signal} is silently ignored.
3661 @cindex @code{sp_switch} attribute
3662 Use this attribute on the SH to indicate an @code{interrupt_handler}
3663 function should switch to an alternate stack. It expects a string
3664 argument that names a global variable holding the address of the
3669 void f () __attribute__ ((interrupt_handler,
3670 sp_switch ("alt_stack")));
3674 @cindex functions that pop the argument stack on the 386
3675 On the Intel 386, the @code{stdcall} attribute causes the compiler to
3676 assume that the called function pops off the stack space used to
3677 pass arguments, unless it takes a variable number of arguments.
3679 @item syscall_linkage
3680 @cindex @code{syscall_linkage} attribute
3681 This attribute is used to modify the IA-64 calling convention by marking
3682 all input registers as live at all function exits. This makes it possible
3683 to restart a system call after an interrupt without having to save/restore
3684 the input registers. This also prevents kernel data from leaking into
3688 @cindex @code{target} function attribute
3689 The @code{target} attribute is used to specify that a function is to
3690 be compiled with different target options than specified on the
3691 command line. This can be used for instance to have functions
3692 compiled with a different ISA (instruction set architecture) than the
3693 default. You can also use the @samp{#pragma GCC target} pragma to set
3694 more than one function to be compiled with specific target options.
3695 @xref{Function Specific Option Pragmas}, for details about the
3696 @samp{#pragma GCC target} pragma.
3698 For instance on a 386, you could compile one function with
3699 @code{target("sse4.1,arch=core2")} and another with
3700 @code{target("sse4a,arch=amdfam10")}. This is equivalent to
3701 compiling the first function with @option{-msse4.1} and
3702 @option{-march=core2} options, and the second function with
3703 @option{-msse4a} and @option{-march=amdfam10} options. It is up to the
3704 user to make sure that a function is only invoked on a machine that
3705 supports the particular ISA it is compiled for (for example by using
3706 @code{cpuid} on 386 to determine what feature bits and architecture
3710 int core2_func (void) __attribute__ ((__target__ ("arch=core2")));
3711 int sse3_func (void) __attribute__ ((__target__ ("sse3")));
3714 On the 386, the following options are allowed:
3719 @cindex @code{target("abm")} attribute
3720 Enable/disable the generation of the advanced bit instructions.
3724 @cindex @code{target("aes")} attribute
3725 Enable/disable the generation of the AES instructions.
3728 @cindex @code{target("default")} attribute
3729 @xref{Function Multiversioning}, where it is used to specify the
3730 default function version.
3734 @cindex @code{target("mmx")} attribute
3735 Enable/disable the generation of the MMX instructions.
3739 @cindex @code{target("pclmul")} attribute
3740 Enable/disable the generation of the PCLMUL instructions.
3744 @cindex @code{target("popcnt")} attribute
3745 Enable/disable the generation of the POPCNT instruction.
3749 @cindex @code{target("sse")} attribute
3750 Enable/disable the generation of the SSE instructions.
3754 @cindex @code{target("sse2")} attribute
3755 Enable/disable the generation of the SSE2 instructions.
3759 @cindex @code{target("sse3")} attribute
3760 Enable/disable the generation of the SSE3 instructions.
3764 @cindex @code{target("sse4")} attribute
3765 Enable/disable the generation of the SSE4 instructions (both SSE4.1
3770 @cindex @code{target("sse4.1")} attribute
3771 Enable/disable the generation of the sse4.1 instructions.
3775 @cindex @code{target("sse4.2")} attribute
3776 Enable/disable the generation of the sse4.2 instructions.
3780 @cindex @code{target("sse4a")} attribute
3781 Enable/disable the generation of the SSE4A instructions.
3785 @cindex @code{target("fma4")} attribute
3786 Enable/disable the generation of the FMA4 instructions.
3790 @cindex @code{target("xop")} attribute
3791 Enable/disable the generation of the XOP instructions.
3795 @cindex @code{target("lwp")} attribute
3796 Enable/disable the generation of the LWP instructions.
3800 @cindex @code{target("ssse3")} attribute
3801 Enable/disable the generation of the SSSE3 instructions.
3805 @cindex @code{target("cld")} attribute
3806 Enable/disable the generation of the CLD before string moves.
3808 @item fancy-math-387
3809 @itemx no-fancy-math-387
3810 @cindex @code{target("fancy-math-387")} attribute
3811 Enable/disable the generation of the @code{sin}, @code{cos}, and
3812 @code{sqrt} instructions on the 387 floating-point unit.
3815 @itemx no-fused-madd
3816 @cindex @code{target("fused-madd")} attribute
3817 Enable/disable the generation of the fused multiply/add instructions.
3821 @cindex @code{target("ieee-fp")} attribute
3822 Enable/disable the generation of floating point that depends on IEEE arithmetic.
3824 @item inline-all-stringops
3825 @itemx no-inline-all-stringops
3826 @cindex @code{target("inline-all-stringops")} attribute
3827 Enable/disable inlining of string operations.
3829 @item inline-stringops-dynamically
3830 @itemx no-inline-stringops-dynamically
3831 @cindex @code{target("inline-stringops-dynamically")} attribute
3832 Enable/disable the generation of the inline code to do small string
3833 operations and calling the library routines for large operations.
3835 @item align-stringops
3836 @itemx no-align-stringops
3837 @cindex @code{target("align-stringops")} attribute
3838 Do/do not align destination of inlined string operations.
3842 @cindex @code{target("recip")} attribute
3843 Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS
3844 instructions followed an additional Newton-Raphson step instead of
3845 doing a floating-point division.
3847 @item arch=@var{ARCH}
3848 @cindex @code{target("arch=@var{ARCH}")} attribute
3849 Specify the architecture to generate code for in compiling the function.
3851 @item tune=@var{TUNE}
3852 @cindex @code{target("tune=@var{TUNE}")} attribute
3853 Specify the architecture to tune for in compiling the function.
3855 @item fpmath=@var{FPMATH}
3856 @cindex @code{target("fpmath=@var{FPMATH}")} attribute
3857 Specify which floating-point unit to use. The
3858 @code{target("fpmath=sse,387")} option must be specified as
3859 @code{target("fpmath=sse+387")} because the comma would separate
3863 On the PowerPC, the following options are allowed:
3868 @cindex @code{target("altivec")} attribute
3869 Generate code that uses (does not use) AltiVec instructions. In
3870 32-bit code, you cannot enable AltiVec instructions unless
3871 @option{-mabi=altivec} is used on the command line.
3875 @cindex @code{target("cmpb")} attribute
3876 Generate code that uses (does not use) the compare bytes instruction
3877 implemented on the POWER6 processor and other processors that support
3878 the PowerPC V2.05 architecture.
3882 @cindex @code{target("dlmzb")} attribute
3883 Generate code that uses (does not use) the string-search @samp{dlmzb}
3884 instruction on the IBM 405, 440, 464 and 476 processors. This instruction is
3885 generated by default when targeting those processors.
3889 @cindex @code{target("fprnd")} attribute
3890 Generate code that uses (does not use) the FP round to integer
3891 instructions implemented on the POWER5+ processor and other processors
3892 that support the PowerPC V2.03 architecture.
3896 @cindex @code{target("hard-dfp")} attribute
3897 Generate code that uses (does not use) the decimal floating-point
3898 instructions implemented on some POWER processors.
3902 @cindex @code{target("isel")} attribute
3903 Generate code that uses (does not use) ISEL instruction.
3907 @cindex @code{target("mfcrf")} attribute
3908 Generate code that uses (does not use) the move from condition
3909 register field instruction implemented on the POWER4 processor and
3910 other processors that support the PowerPC V2.01 architecture.
3914 @cindex @code{target("mfpgpr")} attribute
3915 Generate code that uses (does not use) the FP move to/from general
3916 purpose register instructions implemented on the POWER6X processor and
3917 other processors that support the extended PowerPC V2.05 architecture.
3921 @cindex @code{target("mulhw")} attribute
3922 Generate code that uses (does not use) the half-word multiply and
3923 multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors.
3924 These instructions are generated by default when targeting those
3929 @cindex @code{target("multiple")} attribute
3930 Generate code that uses (does not use) the load multiple word
3931 instructions and the store multiple word instructions.
3935 @cindex @code{target("update")} attribute
3936 Generate code that uses (does not use) the load or store instructions
3937 that update the base register to the address of the calculated memory
3942 @cindex @code{target("popcntb")} attribute
3943 Generate code that uses (does not use) the popcount and double-precision
3944 FP reciprocal estimate instruction implemented on the POWER5
3945 processor and other processors that support the PowerPC V2.02
3950 @cindex @code{target("popcntd")} attribute
3951 Generate code that uses (does not use) the popcount instruction
3952 implemented on the POWER7 processor and other processors that support
3953 the PowerPC V2.06 architecture.
3955 @item powerpc-gfxopt
3956 @itemx no-powerpc-gfxopt
3957 @cindex @code{target("powerpc-gfxopt")} attribute
3958 Generate code that uses (does not use) the optional PowerPC
3959 architecture instructions in the Graphics group, including
3960 floating-point select.
3963 @itemx no-powerpc-gpopt
3964 @cindex @code{target("powerpc-gpopt")} attribute
3965 Generate code that uses (does not use) the optional PowerPC
3966 architecture instructions in the General Purpose group, including
3967 floating-point square root.
3969 @item recip-precision
3970 @itemx no-recip-precision
3971 @cindex @code{target("recip-precision")} attribute
3972 Assume (do not assume) that the reciprocal estimate instructions
3973 provide higher-precision estimates than is mandated by the powerpc
3978 @cindex @code{target("string")} attribute
3979 Generate code that uses (does not use) the load string instructions
3980 and the store string word instructions to save multiple registers and
3981 do small block moves.
3985 @cindex @code{target("vsx")} attribute
3986 Generate code that uses (does not use) vector/scalar (VSX)
3987 instructions, and also enable the use of built-in functions that allow
3988 more direct access to the VSX instruction set. In 32-bit code, you
3989 cannot enable VSX or AltiVec instructions unless
3990 @option{-mabi=altivec} is used on the command line.
3994 @cindex @code{target("friz")} attribute
3995 Generate (do not generate) the @code{friz} instruction when the
3996 @option{-funsafe-math-optimizations} option is used to optimize
3997 rounding a floating-point value to 64-bit integer and back to floating
3998 point. The @code{friz} instruction does not return the same value if
3999 the floating-point number is too large to fit in an integer.
4001 @item avoid-indexed-addresses
4002 @itemx no-avoid-indexed-addresses
4003 @cindex @code{target("avoid-indexed-addresses")} attribute
4004 Generate code that tries to avoid (not avoid) the use of indexed load
4005 or store instructions.
4009 @cindex @code{target("paired")} attribute
4010 Generate code that uses (does not use) the generation of PAIRED simd
4015 @cindex @code{target("longcall")} attribute
4016 Generate code that assumes (does not assume) that all calls are far
4017 away so that a longer more expensive calling sequence is required.
4020 @cindex @code{target("cpu=@var{CPU}")} attribute
4021 Specify the architecture to generate code for when compiling the
4022 function. If you select the @code{target("cpu=power7")} attribute when
4023 generating 32-bit code, VSX and AltiVec instructions are not generated
4024 unless you use the @option{-mabi=altivec} option on the command line.
4026 @item tune=@var{TUNE}
4027 @cindex @code{target("tune=@var{TUNE}")} attribute
4028 Specify the architecture to tune for when compiling the function. If
4029 you do not specify the @code{target("tune=@var{TUNE}")} attribute and
4030 you do specify the @code{target("cpu=@var{CPU}")} attribute,
4031 compilation tunes for the @var{CPU} architecture, and not the
4032 default tuning specified on the command line.
4035 On the 386/x86_64 and PowerPC back ends, you can use either multiple
4036 strings to specify multiple options, or you can separate the option
4037 with a comma (@code{,}).
4039 On the 386/x86_64 and PowerPC back ends, the inliner does not inline a
4040 function that has different target options than the caller, unless the
4041 callee has a subset of the target options of the caller. For example
4042 a function declared with @code{target("sse3")} can inline a function
4043 with @code{target("sse2")}, since @code{-msse3} implies @code{-msse2}.
4045 The @code{target} attribute is not implemented in GCC versions earlier
4046 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC back ends. It is
4047 not currently implemented for other back ends.
4050 @cindex tiny data section on the H8/300H and H8S
4051 Use this attribute on the H8/300H and H8S to indicate that the specified
4052 variable should be placed into the tiny data section.
4053 The compiler generates more efficient code for loads and stores
4054 on data in the tiny data section. Note the tiny data area is limited to
4055 slightly under 32KB of data.
4058 @cindex @code{trap_exit} attribute
4059 Use this attribute on the SH for an @code{interrupt_handler} to return using
4060 @code{trapa} instead of @code{rte}. This attribute expects an integer
4061 argument specifying the trap number to be used.
4064 @cindex @code{trapa_handler} attribute
4065 On SH targets this function attribute is similar to @code{interrupt_handler}
4066 but it does not save and restore all registers.
4069 @cindex @code{unused} attribute.
4070 This attribute, attached to a function, means that the function is meant
4071 to be possibly unused. GCC does not produce a warning for this
4075 @cindex @code{used} attribute.
4076 This attribute, attached to a function, means that code must be emitted
4077 for the function even if it appears that the function is not referenced.
4078 This is useful, for example, when the function is referenced only in
4081 When applied to a member function of a C++ class template, the
4082 attribute also means that the function is instantiated if the
4083 class itself is instantiated.
4086 @cindex @code{version_id} attribute
4087 This IA-64 HP-UX attribute, attached to a global variable or function, renames a
4088 symbol to contain a version string, thus allowing for function level
4089 versioning. HP-UX system header files may use function level versioning
4090 for some system calls.
4093 extern int foo () __attribute__((version_id ("20040821")));
4097 Calls to @var{foo} are mapped to calls to @var{foo@{20040821@}}.
4099 @item visibility ("@var{visibility_type}")
4100 @cindex @code{visibility} attribute
4101 This attribute affects the linkage of the declaration to which it is attached.
4102 There are four supported @var{visibility_type} values: default,
4103 hidden, protected or internal visibility.
4106 void __attribute__ ((visibility ("protected")))
4107 f () @{ /* @r{Do something.} */; @}
4108 int i __attribute__ ((visibility ("hidden")));
4111 The possible values of @var{visibility_type} correspond to the
4112 visibility settings in the ELF gABI.
4115 @c keep this list of visibilities in alphabetical order.
4118 Default visibility is the normal case for the object file format.
4119 This value is available for the visibility attribute to override other
4120 options that may change the assumed visibility of entities.
4122 On ELF, default visibility means that the declaration is visible to other
4123 modules and, in shared libraries, means that the declared entity may be
4126 On Darwin, default visibility means that the declaration is visible to
4129 Default visibility corresponds to ``external linkage'' in the language.
4132 Hidden visibility indicates that the entity declared has a new
4133 form of linkage, which we call ``hidden linkage''. Two
4134 declarations of an object with hidden linkage refer to the same object
4135 if they are in the same shared object.
4138 Internal visibility is like hidden visibility, but with additional
4139 processor specific semantics. Unless otherwise specified by the
4140 psABI, GCC defines internal visibility to mean that a function is
4141 @emph{never} called from another module. Compare this with hidden
4142 functions which, while they cannot be referenced directly by other
4143 modules, can be referenced indirectly via function pointers. By
4144 indicating that a function cannot be called from outside the module,
4145 GCC may for instance omit the load of a PIC register since it is known
4146 that the calling function loaded the correct value.
4149 Protected visibility is like default visibility except that it
4150 indicates that references within the defining module bind to the
4151 definition in that module. That is, the declared entity cannot be
4152 overridden by another module.
4156 All visibilities are supported on many, but not all, ELF targets
4157 (supported when the assembler supports the @samp{.visibility}
4158 pseudo-op). Default visibility is supported everywhere. Hidden
4159 visibility is supported on Darwin targets.
4161 The visibility attribute should be applied only to declarations that
4162 would otherwise have external linkage. The attribute should be applied
4163 consistently, so that the same entity should not be declared with
4164 different settings of the attribute.
4166 In C++, the visibility attribute applies to types as well as functions
4167 and objects, because in C++ types have linkage. A class must not have
4168 greater visibility than its non-static data member types and bases,
4169 and class members default to the visibility of their class. Also, a
4170 declaration without explicit visibility is limited to the visibility
4173 In C++, you can mark member functions and static member variables of a
4174 class with the visibility attribute. This is useful if you know a
4175 particular method or static member variable should only be used from
4176 one shared object; then you can mark it hidden while the rest of the
4177 class has default visibility. Care must be taken to avoid breaking
4178 the One Definition Rule; for example, it is usually not useful to mark
4179 an inline method as hidden without marking the whole class as hidden.
4181 A C++ namespace declaration can also have the visibility attribute.
4182 This attribute applies only to the particular namespace body, not to
4183 other definitions of the same namespace; it is equivalent to using
4184 @samp{#pragma GCC visibility} before and after the namespace
4185 definition (@pxref{Visibility Pragmas}).
4187 In C++, if a template argument has limited visibility, this
4188 restriction is implicitly propagated to the template instantiation.
4189 Otherwise, template instantiations and specializations default to the
4190 visibility of their template.
4192 If both the template and enclosing class have explicit visibility, the
4193 visibility from the template is used.
4196 @cindex @code{vliw} attribute
4197 On MeP, the @code{vliw} attribute tells the compiler to emit
4198 instructions in VLIW mode instead of core mode. Note that this
4199 attribute is not allowed unless a VLIW coprocessor has been configured
4200 and enabled through command-line options.
4202 @item warn_unused_result
4203 @cindex @code{warn_unused_result} attribute
4204 The @code{warn_unused_result} attribute causes a warning to be emitted
4205 if a caller of the function with this attribute does not use its
4206 return value. This is useful for functions where not checking
4207 the result is either a security problem or always a bug, such as
4211 int fn () __attribute__ ((warn_unused_result));
4214 if (fn () < 0) return -1;
4221 results in warning on line 5.
4224 @cindex @code{weak} attribute
4225 The @code{weak} attribute causes the declaration to be emitted as a weak
4226 symbol rather than a global. This is primarily useful in defining
4227 library functions that can be overridden in user code, though it can
4228 also be used with non-function declarations. Weak symbols are supported
4229 for ELF targets, and also for a.out targets when using the GNU assembler
4233 @itemx weakref ("@var{target}")
4234 @cindex @code{weakref} attribute
4235 The @code{weakref} attribute marks a declaration as a weak reference.
4236 Without arguments, it should be accompanied by an @code{alias} attribute
4237 naming the target symbol. Optionally, the @var{target} may be given as
4238 an argument to @code{weakref} itself. In either case, @code{weakref}
4239 implicitly marks the declaration as @code{weak}. Without a
4240 @var{target}, given as an argument to @code{weakref} or to @code{alias},
4241 @code{weakref} is equivalent to @code{weak}.
4244 static int x() __attribute__ ((weakref ("y")));
4245 /* is equivalent to... */
4246 static int x() __attribute__ ((weak, weakref, alias ("y")));
4248 static int x() __attribute__ ((weakref));
4249 static int x() __attribute__ ((alias ("y")));
4252 A weak reference is an alias that does not by itself require a
4253 definition to be given for the target symbol. If the target symbol is
4254 only referenced through weak references, then it becomes a @code{weak}
4255 undefined symbol. If it is directly referenced, however, then such
4256 strong references prevail, and a definition is required for the
4257 symbol, not necessarily in the same translation unit.
4259 The effect is equivalent to moving all references to the alias to a
4260 separate translation unit, renaming the alias to the aliased symbol,
4261 declaring it as weak, compiling the two separate translation units and
4262 performing a reloadable link on them.
4264 At present, a declaration to which @code{weakref} is attached can
4265 only be @code{static}.
4269 You can specify multiple attributes in a declaration by separating them
4270 by commas within the double parentheses or by immediately following an
4271 attribute declaration with another attribute declaration.
4273 @cindex @code{#pragma}, reason for not using
4274 @cindex pragma, reason for not using
4275 Some people object to the @code{__attribute__} feature, suggesting that
4276 ISO C's @code{#pragma} should be used instead. At the time
4277 @code{__attribute__} was designed, there were two reasons for not doing
4282 It is impossible to generate @code{#pragma} commands from a macro.
4285 There is no telling what the same @code{#pragma} might mean in another
4289 These two reasons applied to almost any application that might have been
4290 proposed for @code{#pragma}. It was basically a mistake to use
4291 @code{#pragma} for @emph{anything}.
4293 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
4294 to be generated from macros. In addition, a @code{#pragma GCC}
4295 namespace is now in use for GCC-specific pragmas. However, it has been
4296 found convenient to use @code{__attribute__} to achieve a natural
4297 attachment of attributes to their corresponding declarations, whereas
4298 @code{#pragma GCC} is of use for constructs that do not naturally form
4299 part of the grammar. @xref{Pragmas,,Pragmas Accepted by GCC}.
4301 @node Attribute Syntax
4302 @section Attribute Syntax
4303 @cindex attribute syntax
4305 This section describes the syntax with which @code{__attribute__} may be
4306 used, and the constructs to which attribute specifiers bind, for the C
4307 language. Some details may vary for C++ and Objective-C@. Because of
4308 infelicities in the grammar for attributes, some forms described here
4309 may not be successfully parsed in all cases.
4311 There are some problems with the semantics of attributes in C++. For
4312 example, there are no manglings for attributes, although they may affect
4313 code generation, so problems may arise when attributed types are used in
4314 conjunction with templates or overloading. Similarly, @code{typeid}
4315 does not distinguish between types with different attributes. Support
4316 for attributes in C++ may be restricted in future to attributes on
4317 declarations only, but not on nested declarators.
4319 @xref{Function Attributes}, for details of the semantics of attributes
4320 applying to functions. @xref{Variable Attributes}, for details of the
4321 semantics of attributes applying to variables. @xref{Type Attributes},
4322 for details of the semantics of attributes applying to structure, union
4323 and enumerated types.
4325 An @dfn{attribute specifier} is of the form
4326 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
4327 is a possibly empty comma-separated sequence of @dfn{attributes}, where
4328 each attribute is one of the following:
4332 Empty. Empty attributes are ignored.
4335 A word (which may be an identifier such as @code{unused}, or a reserved
4336 word such as @code{const}).
4339 A word, followed by, in parentheses, parameters for the attribute.
4340 These parameters take one of the following forms:
4344 An identifier. For example, @code{mode} attributes use this form.
4347 An identifier followed by a comma and a non-empty comma-separated list
4348 of expressions. For example, @code{format} attributes use this form.
4351 A possibly empty comma-separated list of expressions. For example,
4352 @code{format_arg} attributes use this form with the list being a single
4353 integer constant expression, and @code{alias} attributes use this form
4354 with the list being a single string constant.
4358 An @dfn{attribute specifier list} is a sequence of one or more attribute
4359 specifiers, not separated by any other tokens.
4361 In GNU C, an attribute specifier list may appear after the colon following a
4362 label, other than a @code{case} or @code{default} label. The only
4363 attribute it makes sense to use after a label is @code{unused}. This
4364 feature is intended for program-generated code that may contain unused labels,
4365 but which is compiled with @option{-Wall}. It is
4366 not normally appropriate to use in it human-written code, though it
4367 could be useful in cases where the code that jumps to the label is
4368 contained within an @code{#ifdef} conditional. GNU C++ only permits
4369 attributes on labels if the attribute specifier is immediately
4370 followed by a semicolon (i.e., the label applies to an empty
4371 statement). If the semicolon is missing, C++ label attributes are
4372 ambiguous, as it is permissible for a declaration, which could begin
4373 with an attribute list, to be labelled in C++. Declarations cannot be
4374 labelled in C90 or C99, so the ambiguity does not arise there.
4376 An attribute specifier list may appear as part of a @code{struct},
4377 @code{union} or @code{enum} specifier. It may go either immediately
4378 after the @code{struct}, @code{union} or @code{enum} keyword, or after
4379 the closing brace. The former syntax is preferred.
4380 Where attribute specifiers follow the closing brace, they are considered
4381 to relate to the structure, union or enumerated type defined, not to any
4382 enclosing declaration the type specifier appears in, and the type
4383 defined is not complete until after the attribute specifiers.
4384 @c Otherwise, there would be the following problems: a shift/reduce
4385 @c conflict between attributes binding the struct/union/enum and
4386 @c binding to the list of specifiers/qualifiers; and "aligned"
4387 @c attributes could use sizeof for the structure, but the size could be
4388 @c changed later by "packed" attributes.
4390 Otherwise, an attribute specifier appears as part of a declaration,
4391 counting declarations of unnamed parameters and type names, and relates
4392 to that declaration (which may be nested in another declaration, for
4393 example in the case of a parameter declaration), or to a particular declarator
4394 within a declaration. Where an
4395 attribute specifier is applied to a parameter declared as a function or
4396 an array, it should apply to the function or array rather than the
4397 pointer to which the parameter is implicitly converted, but this is not
4398 yet correctly implemented.
4400 Any list of specifiers and qualifiers at the start of a declaration may
4401 contain attribute specifiers, whether or not such a list may in that
4402 context contain storage class specifiers. (Some attributes, however,
4403 are essentially in the nature of storage class specifiers, and only make
4404 sense where storage class specifiers may be used; for example,
4405 @code{section}.) There is one necessary limitation to this syntax: the
4406 first old-style parameter declaration in a function definition cannot
4407 begin with an attribute specifier, because such an attribute applies to
4408 the function instead by syntax described below (which, however, is not
4409 yet implemented in this case). In some other cases, attribute
4410 specifiers are permitted by this grammar but not yet supported by the
4411 compiler. All attribute specifiers in this place relate to the
4412 declaration as a whole. In the obsolescent usage where a type of
4413 @code{int} is implied by the absence of type specifiers, such a list of
4414 specifiers and qualifiers may be an attribute specifier list with no
4415 other specifiers or qualifiers.
4417 At present, the first parameter in a function prototype must have some
4418 type specifier that is not an attribute specifier; this resolves an
4419 ambiguity in the interpretation of @code{void f(int
4420 (__attribute__((foo)) x))}, but is subject to change. At present, if
4421 the parentheses of a function declarator contain only attributes then
4422 those attributes are ignored, rather than yielding an error or warning
4423 or implying a single parameter of type int, but this is subject to
4426 An attribute specifier list may appear immediately before a declarator
4427 (other than the first) in a comma-separated list of declarators in a
4428 declaration of more than one identifier using a single list of
4429 specifiers and qualifiers. Such attribute specifiers apply
4430 only to the identifier before whose declarator they appear. For
4434 __attribute__((noreturn)) void d0 (void),
4435 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
4440 the @code{noreturn} attribute applies to all the functions
4441 declared; the @code{format} attribute only applies to @code{d1}.
4443 An attribute specifier list may appear immediately before the comma,
4444 @code{=} or semicolon terminating the declaration of an identifier other
4445 than a function definition. Such attribute specifiers apply
4446 to the declared object or function. Where an
4447 assembler name for an object or function is specified (@pxref{Asm
4448 Labels}), the attribute must follow the @code{asm}
4451 An attribute specifier list may, in future, be permitted to appear after
4452 the declarator in a function definition (before any old-style parameter
4453 declarations or the function body).
4455 Attribute specifiers may be mixed with type qualifiers appearing inside
4456 the @code{[]} of a parameter array declarator, in the C99 construct by
4457 which such qualifiers are applied to the pointer to which the array is
4458 implicitly converted. Such attribute specifiers apply to the pointer,
4459 not to the array, but at present this is not implemented and they are
4462 An attribute specifier list may appear at the start of a nested
4463 declarator. At present, there are some limitations in this usage: the
4464 attributes correctly apply to the declarator, but for most individual
4465 attributes the semantics this implies are not implemented.
4466 When attribute specifiers follow the @code{*} of a pointer
4467 declarator, they may be mixed with any type qualifiers present.
4468 The following describes the formal semantics of this syntax. It makes the
4469 most sense if you are familiar with the formal specification of
4470 declarators in the ISO C standard.
4472 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
4473 D1}, where @code{T} contains declaration specifiers that specify a type
4474 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
4475 contains an identifier @var{ident}. The type specified for @var{ident}
4476 for derived declarators whose type does not include an attribute
4477 specifier is as in the ISO C standard.
4479 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
4480 and the declaration @code{T D} specifies the type
4481 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4482 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4483 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
4485 If @code{D1} has the form @code{*
4486 @var{type-qualifier-and-attribute-specifier-list} D}, and the
4487 declaration @code{T D} specifies the type
4488 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
4489 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
4490 @var{type-qualifier-and-attribute-specifier-list} pointer to @var{Type}'' for
4496 void (__attribute__((noreturn)) ****f) (void);
4500 specifies the type ``pointer to pointer to pointer to pointer to
4501 non-returning function returning @code{void}''. As another example,
4504 char *__attribute__((aligned(8))) *f;
4508 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
4509 Note again that this does not work with most attributes; for example,
4510 the usage of @samp{aligned} and @samp{noreturn} attributes given above
4511 is not yet supported.
4513 For compatibility with existing code written for compiler versions that
4514 did not implement attributes on nested declarators, some laxity is
4515 allowed in the placing of attributes. If an attribute that only applies
4516 to types is applied to a declaration, it is treated as applying to
4517 the type of that declaration. If an attribute that only applies to
4518 declarations is applied to the type of a declaration, it is treated
4519 as applying to that declaration; and, for compatibility with code
4520 placing the attributes immediately before the identifier declared, such
4521 an attribute applied to a function return type is treated as
4522 applying to the function type, and such an attribute applied to an array
4523 element type is treated as applying to the array type. If an
4524 attribute that only applies to function types is applied to a
4525 pointer-to-function type, it is treated as applying to the pointer
4526 target type; if such an attribute is applied to a function return type
4527 that is not a pointer-to-function type, it is treated as applying
4528 to the function type.
4530 @node Function Prototypes
4531 @section Prototypes and Old-Style Function Definitions
4532 @cindex function prototype declarations
4533 @cindex old-style function definitions
4534 @cindex promotion of formal parameters
4536 GNU C extends ISO C to allow a function prototype to override a later
4537 old-style non-prototype definition. Consider the following example:
4540 /* @r{Use prototypes unless the compiler is old-fashioned.} */
4547 /* @r{Prototype function declaration.} */
4548 int isroot P((uid_t));
4550 /* @r{Old-style function definition.} */
4552 isroot (x) /* @r{??? lossage here ???} */
4559 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
4560 not allow this example, because subword arguments in old-style
4561 non-prototype definitions are promoted. Therefore in this example the
4562 function definition's argument is really an @code{int}, which does not
4563 match the prototype argument type of @code{short}.
4565 This restriction of ISO C makes it hard to write code that is portable
4566 to traditional C compilers, because the programmer does not know
4567 whether the @code{uid_t} type is @code{short}, @code{int}, or
4568 @code{long}. Therefore, in cases like these GNU C allows a prototype
4569 to override a later old-style definition. More precisely, in GNU C, a
4570 function prototype argument type overrides the argument type specified
4571 by a later old-style definition if the former type is the same as the
4572 latter type before promotion. Thus in GNU C the above example is
4573 equivalent to the following:
4586 GNU C++ does not support old-style function definitions, so this
4587 extension is irrelevant.
4590 @section C++ Style Comments
4592 @cindex C++ comments
4593 @cindex comments, C++ style
4595 In GNU C, you may use C++ style comments, which start with @samp{//} and
4596 continue until the end of the line. Many other C implementations allow
4597 such comments, and they are included in the 1999 C standard. However,
4598 C++ style comments are not recognized if you specify an @option{-std}
4599 option specifying a version of ISO C before C99, or @option{-ansi}
4600 (equivalent to @option{-std=c90}).
4603 @section Dollar Signs in Identifier Names
4605 @cindex dollar signs in identifier names
4606 @cindex identifier names, dollar signs in
4608 In GNU C, you may normally use dollar signs in identifier names.
4609 This is because many traditional C implementations allow such identifiers.
4610 However, dollar signs in identifiers are not supported on a few target
4611 machines, typically because the target assembler does not allow them.
4613 @node Character Escapes
4614 @section The Character @key{ESC} in Constants
4616 You can use the sequence @samp{\e} in a string or character constant to
4617 stand for the ASCII character @key{ESC}.
4619 @node Variable Attributes
4620 @section Specifying Attributes of Variables
4621 @cindex attribute of variables
4622 @cindex variable attributes
4624 The keyword @code{__attribute__} allows you to specify special
4625 attributes of variables or structure fields. This keyword is followed
4626 by an attribute specification inside double parentheses. Some
4627 attributes are currently defined generically for variables.
4628 Other attributes are defined for variables on particular target
4629 systems. Other attributes are available for functions
4630 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
4631 Other front ends might define more attributes
4632 (@pxref{C++ Extensions,,Extensions to the C++ Language}).
4634 You may also specify attributes with @samp{__} preceding and following
4635 each keyword. This allows you to use them in header files without
4636 being concerned about a possible macro of the same name. For example,
4637 you may use @code{__aligned__} instead of @code{aligned}.
4639 @xref{Attribute Syntax}, for details of the exact syntax for using
4643 @cindex @code{aligned} attribute
4644 @item aligned (@var{alignment})
4645 This attribute specifies a minimum alignment for the variable or
4646 structure field, measured in bytes. For example, the declaration:
4649 int x __attribute__ ((aligned (16))) = 0;
4653 causes the compiler to allocate the global variable @code{x} on a
4654 16-byte boundary. On a 68040, this could be used in conjunction with
4655 an @code{asm} expression to access the @code{move16} instruction which
4656 requires 16-byte aligned operands.
4658 You can also specify the alignment of structure fields. For example, to
4659 create a double-word aligned @code{int} pair, you could write:
4662 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
4666 This is an alternative to creating a union with a @code{double} member,
4667 which forces the union to be double-word aligned.
4669 As in the preceding examples, you can explicitly specify the alignment
4670 (in bytes) that you wish the compiler to use for a given variable or
4671 structure field. Alternatively, you can leave out the alignment factor
4672 and just ask the compiler to align a variable or field to the
4673 default alignment for the target architecture you are compiling for.
4674 The default alignment is sufficient for all scalar types, but may not be
4675 enough for all vector types on a target that supports vector operations.
4676 The default alignment is fixed for a particular target ABI.
4678 GCC also provides a target specific macro @code{__BIGGEST_ALIGNMENT__},
4679 which is the largest alignment ever used for any data type on the
4680 target machine you are compiling for. For example, you could write:
4683 short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__)));
4686 The compiler automatically sets the alignment for the declared
4687 variable or field to @code{__BIGGEST_ALIGNMENT__}. Doing this can
4688 often make copy operations more efficient, because the compiler can
4689 use whatever instructions copy the biggest chunks of memory when
4690 performing copies to or from the variables or fields that you have
4691 aligned this way. Note that the value of @code{__BIGGEST_ALIGNMENT__}
4692 may change depending on command-line options.
4694 When used on a struct, or struct member, the @code{aligned} attribute can
4695 only increase the alignment; in order to decrease it, the @code{packed}
4696 attribute must be specified as well. When used as part of a typedef, the
4697 @code{aligned} attribute can both increase and decrease alignment, and
4698 specifying the @code{packed} attribute generates a warning.
4700 Note that the effectiveness of @code{aligned} attributes may be limited
4701 by inherent limitations in your linker. On many systems, the linker is
4702 only able to arrange for variables to be aligned up to a certain maximum
4703 alignment. (For some linkers, the maximum supported alignment may
4704 be very very small.) If your linker is only able to align variables
4705 up to a maximum of 8-byte alignment, then specifying @code{aligned(16)}
4706 in an @code{__attribute__} still only provides you with 8-byte
4707 alignment. See your linker documentation for further information.
4709 The @code{aligned} attribute can also be used for functions
4710 (@pxref{Function Attributes}.)
4712 @item cleanup (@var{cleanup_function})
4713 @cindex @code{cleanup} attribute
4714 The @code{cleanup} attribute runs a function when the variable goes
4715 out of scope. This attribute can only be applied to auto function
4716 scope variables; it may not be applied to parameters or variables
4717 with static storage duration. The function must take one parameter,
4718 a pointer to a type compatible with the variable. The return value
4719 of the function (if any) is ignored.
4721 If @option{-fexceptions} is enabled, then @var{cleanup_function}
4722 is run during the stack unwinding that happens during the
4723 processing of the exception. Note that the @code{cleanup} attribute
4724 does not allow the exception to be caught, only to perform an action.
4725 It is undefined what happens if @var{cleanup_function} does not
4730 @cindex @code{common} attribute
4731 @cindex @code{nocommon} attribute
4734 The @code{common} attribute requests GCC to place a variable in
4735 ``common'' storage. The @code{nocommon} attribute requests the
4736 opposite---to allocate space for it directly.
4738 These attributes override the default chosen by the
4739 @option{-fno-common} and @option{-fcommon} flags respectively.
4742 @itemx deprecated (@var{msg})
4743 @cindex @code{deprecated} attribute
4744 The @code{deprecated} attribute results in a warning if the variable
4745 is used anywhere in the source file. This is useful when identifying
4746 variables that are expected to be removed in a future version of a
4747 program. The warning also includes the location of the declaration
4748 of the deprecated variable, to enable users to easily find further
4749 information about why the variable is deprecated, or what they should
4750 do instead. Note that the warning only occurs for uses:
4753 extern int old_var __attribute__ ((deprecated));
4755 int new_fn () @{ return old_var; @}
4759 results in a warning on line 3 but not line 2. The optional @var{msg}
4760 argument, which must be a string, is printed in the warning if
4763 The @code{deprecated} attribute can also be used for functions and
4764 types (@pxref{Function Attributes}, @pxref{Type Attributes}.)
4766 @item mode (@var{mode})
4767 @cindex @code{mode} attribute
4768 This attribute specifies the data type for the declaration---whichever
4769 type corresponds to the mode @var{mode}. This in effect lets you
4770 request an integer or floating-point type according to its width.
4772 You may also specify a mode of @code{byte} or @code{__byte__} to
4773 indicate the mode corresponding to a one-byte integer, @code{word} or
4774 @code{__word__} for the mode of a one-word integer, and @code{pointer}
4775 or @code{__pointer__} for the mode used to represent pointers.
4778 @cindex @code{packed} attribute
4779 The @code{packed} attribute specifies that a variable or structure field
4780 should have the smallest possible alignment---one byte for a variable,
4781 and one bit for a field, unless you specify a larger value with the
4782 @code{aligned} attribute.
4784 Here is a structure in which the field @code{x} is packed, so that it
4785 immediately follows @code{a}:
4791 int x[2] __attribute__ ((packed));
4795 @emph{Note:} The 4.1, 4.2 and 4.3 series of GCC ignore the
4796 @code{packed} attribute on bit-fields of type @code{char}. This has
4797 been fixed in GCC 4.4 but the change can lead to differences in the
4798 structure layout. See the documentation of
4799 @option{-Wpacked-bitfield-compat} for more information.
4801 @item section ("@var{section-name}")
4802 @cindex @code{section} variable attribute
4803 Normally, the compiler places the objects it generates in sections like
4804 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
4805 or you need certain particular variables to appear in special sections,
4806 for example to map to special hardware. The @code{section}
4807 attribute specifies that a variable (or function) lives in a particular
4808 section. For example, this small program uses several specific section names:
4811 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
4812 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
4813 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
4814 int init_data __attribute__ ((section ("INITDATA")));
4818 /* @r{Initialize stack pointer} */
4819 init_sp (stack + sizeof (stack));
4821 /* @r{Initialize initialized data} */
4822 memcpy (&init_data, &data, &edata - &data);
4824 /* @r{Turn on the serial ports} */
4831 Use the @code{section} attribute with
4832 @emph{global} variables and not @emph{local} variables,
4833 as shown in the example.
4835 You may use the @code{section} attribute with initialized or
4836 uninitialized global variables but the linker requires
4837 each object be defined once, with the exception that uninitialized
4838 variables tentatively go in the @code{common} (or @code{bss}) section
4839 and can be multiply ``defined''. Using the @code{section} attribute
4840 changes what section the variable goes into and may cause the
4841 linker to issue an error if an uninitialized variable has multiple
4842 definitions. You can force a variable to be initialized with the
4843 @option{-fno-common} flag or the @code{nocommon} attribute.
4845 Some file formats do not support arbitrary sections so the @code{section}
4846 attribute is not available on all platforms.
4847 If you need to map the entire contents of a module to a particular
4848 section, consider using the facilities of the linker instead.
4851 @cindex @code{shared} variable attribute
4852 On Microsoft Windows, in addition to putting variable definitions in a named
4853 section, the section can also be shared among all running copies of an
4854 executable or DLL@. For example, this small program defines shared data
4855 by putting it in a named section @code{shared} and marking the section
4859 int foo __attribute__((section ("shared"), shared)) = 0;
4864 /* @r{Read and write foo. All running
4865 copies see the same value.} */
4871 You may only use the @code{shared} attribute along with @code{section}
4872 attribute with a fully-initialized global definition because of the way
4873 linkers work. See @code{section} attribute for more information.
4875 The @code{shared} attribute is only available on Microsoft Windows@.
4877 @item tls_model ("@var{tls_model}")
4878 @cindex @code{tls_model} attribute
4879 The @code{tls_model} attribute sets thread-local storage model
4880 (@pxref{Thread-Local}) of a particular @code{__thread} variable,
4881 overriding @option{-ftls-model=} command-line switch on a per-variable
4883 The @var{tls_model} argument should be one of @code{global-dynamic},
4884 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
4886 Not all targets support this attribute.
4889 This attribute, attached to a variable, means that the variable is meant
4890 to be possibly unused. GCC does not produce a warning for this
4894 This attribute, attached to a variable, means that the variable must be
4895 emitted even if it appears that the variable is not referenced.
4897 When applied to a static data member of a C++ class template, the
4898 attribute also means that the member is instantiated if the
4899 class itself is instantiated.
4901 @item vector_size (@var{bytes})
4902 This attribute specifies the vector size for the variable, measured in
4903 bytes. For example, the declaration:
4906 int foo __attribute__ ((vector_size (16)));
4910 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
4911 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
4912 4 units of 4 bytes), the corresponding mode of @code{foo} is V4SI@.
4914 This attribute is only applicable to integral and float scalars,
4915 although arrays, pointers, and function return values are allowed in
4916 conjunction with this construct.
4918 Aggregates with this attribute are invalid, even if they are of the same
4919 size as a corresponding scalar. For example, the declaration:
4922 struct S @{ int a; @};
4923 struct S __attribute__ ((vector_size (16))) foo;
4927 is invalid even if the size of the structure is the same as the size of
4931 The @code{selectany} attribute causes an initialized global variable to
4932 have link-once semantics. When multiple definitions of the variable are
4933 encountered by the linker, the first is selected and the remainder are
4934 discarded. Following usage by the Microsoft compiler, the linker is told
4935 @emph{not} to warn about size or content differences of the multiple
4938 Although the primary usage of this attribute is for POD types, the
4939 attribute can also be applied to global C++ objects that are initialized
4940 by a constructor. In this case, the static initialization and destruction
4941 code for the object is emitted in each translation defining the object,
4942 but the calls to the constructor and destructor are protected by a
4943 link-once guard variable.
4945 The @code{selectany} attribute is only available on Microsoft Windows
4946 targets. You can use @code{__declspec (selectany)} as a synonym for
4947 @code{__attribute__ ((selectany))} for compatibility with other
4951 The @code{weak} attribute is described in @ref{Function Attributes}.
4954 The @code{dllimport} attribute is described in @ref{Function Attributes}.
4957 The @code{dllexport} attribute is described in @ref{Function Attributes}.
4961 @anchor{AVR Variable Attributes}
4962 @subsection AVR Variable Attributes
4966 @cindex @code{progmem} AVR variable attribute
4967 The @code{progmem} attribute is used on the AVR to place read-only
4968 data in the non-volatile program memory (flash). The @code{progmem}
4969 attribute accomplishes this by putting respective variables into a
4970 section whose name starts with @code{.progmem}.
4972 This attribute works similar to the @code{section} attribute
4973 but adds additional checking. Notice that just like the
4974 @code{section} attribute, @code{progmem} affects the location
4975 of the data but not how this data is accessed.
4977 In order to read data located with the @code{progmem} attribute
4978 (inline) assembler must be used.
4980 /* Use custom macros from @w{@uref{http://nongnu.org/avr-libc/user-manual/,AVR-LibC}} */
4981 #include <avr/pgmspace.h>
4983 /* Locate var in flash memory */
4984 const int var[2] PROGMEM = @{ 1, 2 @};
4986 int read_var (int i)
4988 /* Access var[] by accessor macro from avr/pgmspace.h */
4989 return (int) pgm_read_word (& var[i]);
4993 AVR is a Harvard architecture processor and data and read-only data
4994 normally resides in the data memory (RAM).
4996 See also the @ref{AVR Named Address Spaces} section for
4997 an alternate way to locate and access data in flash memory.
5000 @subsection Blackfin Variable Attributes
5002 Three attributes are currently defined for the Blackfin.
5008 @cindex @code{l1_data} variable attribute
5009 @cindex @code{l1_data_A} variable attribute
5010 @cindex @code{l1_data_B} variable attribute
5011 Use these attributes on the Blackfin to place the variable into L1 Data SRAM.
5012 Variables with @code{l1_data} attribute are put into the specific section
5013 named @code{.l1.data}. Those with @code{l1_data_A} attribute are put into
5014 the specific section named @code{.l1.data.A}. Those with @code{l1_data_B}
5015 attribute are put into the specific section named @code{.l1.data.B}.
5018 @cindex @code{l2} variable attribute
5019 Use this attribute on the Blackfin to place the variable into L2 SRAM.
5020 Variables with @code{l2} attribute are put into the specific section
5021 named @code{.l2.data}.
5024 @subsection M32R/D Variable Attributes
5026 One attribute is currently defined for the M32R/D@.
5029 @item model (@var{model-name})
5030 @cindex variable addressability on the M32R/D
5031 Use this attribute on the M32R/D to set the addressability of an object.
5032 The identifier @var{model-name} is one of @code{small}, @code{medium},
5033 or @code{large}, representing each of the code models.
5035 Small model objects live in the lower 16MB of memory (so that their
5036 addresses can be loaded with the @code{ld24} instruction).
5038 Medium and large model objects may live anywhere in the 32-bit address space
5039 (the compiler generates @code{seth/add3} instructions to load their
5043 @anchor{MeP Variable Attributes}
5044 @subsection MeP Variable Attributes
5046 The MeP target has a number of addressing modes and busses. The
5047 @code{near} space spans the standard memory space's first 16 megabytes
5048 (24 bits). The @code{far} space spans the entire 32-bit memory space.
5049 The @code{based} space is a 128-byte region in the memory space that
5050 is addressed relative to the @code{$tp} register. The @code{tiny}
5051 space is a 65536-byte region relative to the @code{$gp} register. In
5052 addition to these memory regions, the MeP target has a separate 16-bit
5053 control bus which is specified with @code{cb} attributes.
5058 Any variable with the @code{based} attribute is assigned to the
5059 @code{.based} section, and is accessed with relative to the
5060 @code{$tp} register.
5063 Likewise, the @code{tiny} attribute assigned variables to the
5064 @code{.tiny} section, relative to the @code{$gp} register.
5067 Variables with the @code{near} attribute are assumed to have addresses
5068 that fit in a 24-bit addressing mode. This is the default for large
5069 variables (@code{-mtiny=4} is the default) but this attribute can
5070 override @code{-mtiny=} for small variables, or override @code{-ml}.
5073 Variables with the @code{far} attribute are addressed using a full
5074 32-bit address. Since this covers the entire memory space, this
5075 allows modules to make no assumptions about where variables might be
5079 @itemx io (@var{addr})
5080 Variables with the @code{io} attribute are used to address
5081 memory-mapped peripherals. If an address is specified, the variable
5082 is assigned that address, else it is not assigned an address (it is
5083 assumed some other module assigns an address). Example:
5086 int timer_count __attribute__((io(0x123)));
5090 @itemx cb (@var{addr})
5091 Variables with the @code{cb} attribute are used to access the control
5092 bus, using special instructions. @code{addr} indicates the control bus
5096 int cpu_clock __attribute__((cb(0x123)));
5101 @anchor{i386 Variable Attributes}
5102 @subsection i386 Variable Attributes
5104 Two attributes are currently defined for i386 configurations:
5105 @code{ms_struct} and @code{gcc_struct}
5110 @cindex @code{ms_struct} attribute
5111 @cindex @code{gcc_struct} attribute
5113 If @code{packed} is used on a structure, or if bit-fields are used,
5114 it may be that the Microsoft ABI lays out the structure differently
5115 than the way GCC normally does. Particularly when moving packed
5116 data between functions compiled with GCC and the native Microsoft compiler
5117 (either via function call or as data in a file), it may be necessary to access
5120 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
5121 compilers to match the native Microsoft compiler.
5123 The Microsoft structure layout algorithm is fairly simple with the exception
5124 of the bit-field packing.
5125 The padding and alignment of members of structures and whether a bit-field
5126 can straddle a storage-unit boundary are determine by these rules:
5129 @item Structure members are stored sequentially in the order in which they are
5130 declared: the first member has the lowest memory address and the last member
5133 @item Every data object has an alignment requirement. The alignment requirement
5134 for all data except structures, unions, and arrays is either the size of the
5135 object or the current packing size (specified with either the
5136 @code{aligned} attribute or the @code{pack} pragma),
5137 whichever is less. For structures, unions, and arrays,
5138 the alignment requirement is the largest alignment requirement of its members.
5139 Every object is allocated an offset so that:
5142 offset % alignment_requirement == 0
5145 @item Adjacent bit-fields are packed into the same 1-, 2-, or 4-byte allocation
5146 unit if the integral types are the same size and if the next bit-field fits
5147 into the current allocation unit without crossing the boundary imposed by the
5148 common alignment requirements of the bit-fields.
5151 MSVC interprets zero-length bit-fields in the following ways:
5154 @item If a zero-length bit-field is inserted between two bit-fields that
5155 are normally coalesced, the bit-fields are not coalesced.
5162 unsigned long bf_1 : 12;
5164 unsigned long bf_2 : 12;
5169 The size of @code{t1} is 8 bytes with the zero-length bit-field. If the
5170 zero-length bit-field were removed, @code{t1}'s size would be 4 bytes.
5172 @item If a zero-length bit-field is inserted after a bit-field, @code{foo}, and the
5173 alignment of the zero-length bit-field is greater than the member that follows it,
5174 @code{bar}, @code{bar} is aligned as the type of the zero-length bit-field.
5195 For @code{t2}, @code{bar} is placed at offset 2, rather than offset 1.
5196 Accordingly, the size of @code{t2} is 4. For @code{t3}, the zero-length
5197 bit-field does not affect the alignment of @code{bar} or, as a result, the size
5200 Taking this into account, it is important to note the following:
5203 @item If a zero-length bit-field follows a normal bit-field, the type of the
5204 zero-length bit-field may affect the alignment of the structure as whole. For
5205 example, @code{t2} has a size of 4 bytes, since the zero-length bit-field follows a
5206 normal bit-field, and is of type short.
5208 @item Even if a zero-length bit-field is not followed by a normal bit-field, it may
5209 still affect the alignment of the structure:
5220 Here, @code{t4} takes up 4 bytes.
5223 @item Zero-length bit-fields following non-bit-field members are ignored:
5235 Here, @code{t5} takes up 2 bytes.
5239 @subsection PowerPC Variable Attributes
5241 Three attributes currently are defined for PowerPC configurations:
5242 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
5244 For full documentation of the struct attributes please see the
5245 documentation in @ref{i386 Variable Attributes}.
5247 For documentation of @code{altivec} attribute please see the
5248 documentation in @ref{PowerPC Type Attributes}.
5250 @subsection SPU Variable Attributes
5252 The SPU supports the @code{spu_vector} attribute for variables. For
5253 documentation of this attribute please see the documentation in
5254 @ref{SPU Type Attributes}.
5256 @subsection Xstormy16 Variable Attributes
5258 One attribute is currently defined for xstormy16 configurations:
5263 @cindex @code{below100} attribute
5265 If a variable has the @code{below100} attribute (@code{BELOW100} is
5266 allowed also), GCC places the variable in the first 0x100 bytes of
5267 memory and use special opcodes to access it. Such variables are
5268 placed in either the @code{.bss_below100} section or the
5269 @code{.data_below100} section.
5273 @node Type Attributes
5274 @section Specifying Attributes of Types
5275 @cindex attribute of types
5276 @cindex type attributes
5278 The keyword @code{__attribute__} allows you to specify special
5279 attributes of @code{struct} and @code{union} types when you define
5280 such types. This keyword is followed by an attribute specification
5281 inside double parentheses. Seven attributes are currently defined for
5282 types: @code{aligned}, @code{packed}, @code{transparent_union},
5283 @code{unused}, @code{deprecated}, @code{visibility}, and
5284 @code{may_alias}. Other attributes are defined for functions
5285 (@pxref{Function Attributes}) and for variables (@pxref{Variable
5288 You may also specify any one of these attributes with @samp{__}
5289 preceding and following its keyword. This allows you to use these
5290 attributes in header files without being concerned about a possible
5291 macro of the same name. For example, you may use @code{__aligned__}
5292 instead of @code{aligned}.
5294 You may specify type attributes in an enum, struct or union type
5295 declaration or definition, or for other types in a @code{typedef}
5298 For an enum, struct or union type, you may specify attributes either
5299 between the enum, struct or union tag and the name of the type, or
5300 just past the closing curly brace of the @emph{definition}. The
5301 former syntax is preferred.
5303 @xref{Attribute Syntax}, for details of the exact syntax for using
5307 @cindex @code{aligned} attribute
5308 @item aligned (@var{alignment})
5309 This attribute specifies a minimum alignment (in bytes) for variables
5310 of the specified type. For example, the declarations:
5313 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
5314 typedef int more_aligned_int __attribute__ ((aligned (8)));
5318 force the compiler to ensure (as far as it can) that each variable whose
5319 type is @code{struct S} or @code{more_aligned_int} is allocated and
5320 aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all
5321 variables of type @code{struct S} aligned to 8-byte boundaries allows
5322 the compiler to use the @code{ldd} and @code{std} (doubleword load and
5323 store) instructions when copying one variable of type @code{struct S} to
5324 another, thus improving run-time efficiency.
5326 Note that the alignment of any given @code{struct} or @code{union} type
5327 is required by the ISO C standard to be at least a perfect multiple of
5328 the lowest common multiple of the alignments of all of the members of
5329 the @code{struct} or @code{union} in question. This means that you @emph{can}
5330 effectively adjust the alignment of a @code{struct} or @code{union}
5331 type by attaching an @code{aligned} attribute to any one of the members
5332 of such a type, but the notation illustrated in the example above is a
5333 more obvious, intuitive, and readable way to request the compiler to
5334 adjust the alignment of an entire @code{struct} or @code{union} type.
5336 As in the preceding example, you can explicitly specify the alignment
5337 (in bytes) that you wish the compiler to use for a given @code{struct}
5338 or @code{union} type. Alternatively, you can leave out the alignment factor
5339 and just ask the compiler to align a type to the maximum
5340 useful alignment for the target machine you are compiling for. For
5341 example, you could write:
5344 struct S @{ short f[3]; @} __attribute__ ((aligned));
5347 Whenever you leave out the alignment factor in an @code{aligned}
5348 attribute specification, the compiler automatically sets the alignment
5349 for the type to the largest alignment that is ever used for any data
5350 type on the target machine you are compiling for. Doing this can often
5351 make copy operations more efficient, because the compiler can use
5352 whatever instructions copy the biggest chunks of memory when performing
5353 copies to or from the variables that have types that you have aligned
5356 In the example above, if the size of each @code{short} is 2 bytes, then
5357 the size of the entire @code{struct S} type is 6 bytes. The smallest
5358 power of two that is greater than or equal to that is 8, so the
5359 compiler sets the alignment for the entire @code{struct S} type to 8
5362 Note that although you can ask the compiler to select a time-efficient
5363 alignment for a given type and then declare only individual stand-alone
5364 objects of that type, the compiler's ability to select a time-efficient
5365 alignment is primarily useful only when you plan to create arrays of
5366 variables having the relevant (efficiently aligned) type. If you
5367 declare or use arrays of variables of an efficiently-aligned type, then
5368 it is likely that your program also does pointer arithmetic (or
5369 subscripting, which amounts to the same thing) on pointers to the
5370 relevant type, and the code that the compiler generates for these
5371 pointer arithmetic operations is often more efficient for
5372 efficiently-aligned types than for other types.
5374 The @code{aligned} attribute can only increase the alignment; but you
5375 can decrease it by specifying @code{packed} as well. See below.
5377 Note that the effectiveness of @code{aligned} attributes may be limited
5378 by inherent limitations in your linker. On many systems, the linker is
5379 only able to arrange for variables to be aligned up to a certain maximum
5380 alignment. (For some linkers, the maximum supported alignment may
5381 be very very small.) If your linker is only able to align variables
5382 up to a maximum of 8-byte alignment, then specifying @code{aligned(16)}
5383 in an @code{__attribute__} still only provides you with 8-byte
5384 alignment. See your linker documentation for further information.
5387 This attribute, attached to @code{struct} or @code{union} type
5388 definition, specifies that each member (other than zero-width bit-fields)
5389 of the structure or union is placed to minimize the memory required. When
5390 attached to an @code{enum} definition, it indicates that the smallest
5391 integral type should be used.
5393 @opindex fshort-enums
5394 Specifying this attribute for @code{struct} and @code{union} types is
5395 equivalent to specifying the @code{packed} attribute on each of the
5396 structure or union members. Specifying the @option{-fshort-enums}
5397 flag on the line is equivalent to specifying the @code{packed}
5398 attribute on all @code{enum} definitions.
5400 In the following example @code{struct my_packed_struct}'s members are
5401 packed closely together, but the internal layout of its @code{s} member
5402 is not packed---to do that, @code{struct my_unpacked_struct} needs to
5406 struct my_unpacked_struct
5412 struct __attribute__ ((__packed__)) my_packed_struct
5416 struct my_unpacked_struct s;
5420 You may only specify this attribute on the definition of an @code{enum},
5421 @code{struct} or @code{union}, not on a @code{typedef} that does not
5422 also define the enumerated type, structure or union.
5424 @item transparent_union
5425 This attribute, attached to a @code{union} type definition, indicates
5426 that any function parameter having that union type causes calls to that
5427 function to be treated in a special way.
5429 First, the argument corresponding to a transparent union type can be of
5430 any type in the union; no cast is required. Also, if the union contains
5431 a pointer type, the corresponding argument can be a null pointer
5432 constant or a void pointer expression; and if the union contains a void
5433 pointer type, the corresponding argument can be any pointer expression.
5434 If the union member type is a pointer, qualifiers like @code{const} on
5435 the referenced type must be respected, just as with normal pointer
5438 Second, the argument is passed to the function using the calling
5439 conventions of the first member of the transparent union, not the calling
5440 conventions of the union itself. All members of the union must have the
5441 same machine representation; this is necessary for this argument passing
5444 Transparent unions are designed for library functions that have multiple
5445 interfaces for compatibility reasons. For example, suppose the
5446 @code{wait} function must accept either a value of type @code{int *} to
5447 comply with POSIX, or a value of type @code{union wait *} to comply with
5448 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
5449 @code{wait} would accept both kinds of arguments, but it would also
5450 accept any other pointer type and this would make argument type checking
5451 less useful. Instead, @code{<sys/wait.h>} might define the interface
5455 typedef union __attribute__ ((__transparent_union__))
5459 @} wait_status_ptr_t;
5461 pid_t wait (wait_status_ptr_t);
5465 This interface allows either @code{int *} or @code{union wait *}
5466 arguments to be passed, using the @code{int *} calling convention.
5467 The program can call @code{wait} with arguments of either type:
5470 int w1 () @{ int w; return wait (&w); @}
5471 int w2 () @{ union wait w; return wait (&w); @}
5475 With this interface, @code{wait}'s implementation might look like this:
5478 pid_t wait (wait_status_ptr_t p)
5480 return waitpid (-1, p.__ip, 0);
5485 When attached to a type (including a @code{union} or a @code{struct}),
5486 this attribute means that variables of that type are meant to appear
5487 possibly unused. GCC does not produce a warning for any variables of
5488 that type, even if the variable appears to do nothing. This is often
5489 the case with lock or thread classes, which are usually defined and then
5490 not referenced, but contain constructors and destructors that have
5491 nontrivial bookkeeping functions.
5494 @itemx deprecated (@var{msg})
5495 The @code{deprecated} attribute results in a warning if the type
5496 is used anywhere in the source file. This is useful when identifying
5497 types that are expected to be removed in a future version of a program.
5498 If possible, the warning also includes the location of the declaration
5499 of the deprecated type, to enable users to easily find further
5500 information about why the type is deprecated, or what they should do
5501 instead. Note that the warnings only occur for uses and then only
5502 if the type is being applied to an identifier that itself is not being
5503 declared as deprecated.
5506 typedef int T1 __attribute__ ((deprecated));
5510 typedef T1 T3 __attribute__ ((deprecated));
5511 T3 z __attribute__ ((deprecated));
5515 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
5516 warning is issued for line 4 because T2 is not explicitly
5517 deprecated. Line 5 has no warning because T3 is explicitly
5518 deprecated. Similarly for line 6. The optional @var{msg}
5519 argument, which must be a string, is printed in the warning if
5522 The @code{deprecated} attribute can also be used for functions and
5523 variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.)
5526 Accesses through pointers to types with this attribute are not subject
5527 to type-based alias analysis, but are instead assumed to be able to alias
5528 any other type of objects.
5529 In the context of section 6.5 paragraph 7 of the C99 standard,
5530 an lvalue expression
5531 dereferencing such a pointer is treated like having a character type.
5532 See @option{-fstrict-aliasing} for more information on aliasing issues.
5533 This extension exists to support some vector APIs, in which pointers to
5534 one vector type are permitted to alias pointers to a different vector type.
5536 Note that an object of a type with this attribute does not have any
5542 typedef short __attribute__((__may_alias__)) short_a;
5548 short_a *b = (short_a *) &a;
5552 if (a == 0x12345678)
5560 If you replaced @code{short_a} with @code{short} in the variable
5561 declaration, the above program would abort when compiled with
5562 @option{-fstrict-aliasing}, which is on by default at @option{-O2} or
5563 above in recent GCC versions.
5566 In C++, attribute visibility (@pxref{Function Attributes}) can also be
5567 applied to class, struct, union and enum types. Unlike other type
5568 attributes, the attribute must appear between the initial keyword and
5569 the name of the type; it cannot appear after the body of the type.
5571 Note that the type visibility is applied to vague linkage entities
5572 associated with the class (vtable, typeinfo node, etc.). In
5573 particular, if a class is thrown as an exception in one shared object
5574 and caught in another, the class must have default visibility.
5575 Otherwise the two shared objects are unable to use the same
5576 typeinfo node and exception handling will break.
5580 To specify multiple attributes, separate them by commas within the
5581 double parentheses: for example, @samp{__attribute__ ((aligned (16),
5584 @subsection ARM Type Attributes
5586 On those ARM targets that support @code{dllimport} (such as Symbian
5587 OS), you can use the @code{notshared} attribute to indicate that the
5588 virtual table and other similar data for a class should not be
5589 exported from a DLL@. For example:
5592 class __declspec(notshared) C @{
5594 __declspec(dllimport) C();
5598 __declspec(dllexport)
5603 In this code, @code{C::C} is exported from the current DLL, but the
5604 virtual table for @code{C} is not exported. (You can use
5605 @code{__attribute__} instead of @code{__declspec} if you prefer, but
5606 most Symbian OS code uses @code{__declspec}.)
5608 @anchor{MeP Type Attributes}
5609 @subsection MeP Type Attributes
5611 Many of the MeP variable attributes may be applied to types as well.
5612 Specifically, the @code{based}, @code{tiny}, @code{near}, and
5613 @code{far} attributes may be applied to either. The @code{io} and
5614 @code{cb} attributes may not be applied to types.
5616 @anchor{i386 Type Attributes}
5617 @subsection i386 Type Attributes
5619 Two attributes are currently defined for i386 configurations:
5620 @code{ms_struct} and @code{gcc_struct}.
5626 @cindex @code{ms_struct}
5627 @cindex @code{gcc_struct}
5629 If @code{packed} is used on a structure, or if bit-fields are used
5630 it may be that the Microsoft ABI packs them differently
5631 than GCC normally packs them. Particularly when moving packed
5632 data between functions compiled with GCC and the native Microsoft compiler
5633 (either via function call or as data in a file), it may be necessary to access
5636 Currently @option{-m[no-]ms-bitfields} is provided for the Microsoft Windows X86
5637 compilers to match the native Microsoft compiler.
5640 @anchor{PowerPC Type Attributes}
5641 @subsection PowerPC Type Attributes
5643 Three attributes currently are defined for PowerPC configurations:
5644 @code{altivec}, @code{ms_struct} and @code{gcc_struct}.
5646 For full documentation of the @code{ms_struct} and @code{gcc_struct}
5647 attributes please see the documentation in @ref{i386 Type Attributes}.
5649 The @code{altivec} attribute allows one to declare AltiVec vector data
5650 types supported by the AltiVec Programming Interface Manual. The
5651 attribute requires an argument to specify one of three vector types:
5652 @code{vector__}, @code{pixel__} (always followed by unsigned short),
5653 and @code{bool__} (always followed by unsigned).
5656 __attribute__((altivec(vector__)))
5657 __attribute__((altivec(pixel__))) unsigned short
5658 __attribute__((altivec(bool__))) unsigned
5661 These attributes mainly are intended to support the @code{__vector},
5662 @code{__pixel}, and @code{__bool} AltiVec keywords.
5664 @anchor{SPU Type Attributes}
5665 @subsection SPU Type Attributes
5667 The SPU supports the @code{spu_vector} attribute for types. This attribute
5668 allows one to declare vector data types supported by the Sony/Toshiba/IBM SPU
5669 Language Extensions Specification. It is intended to support the
5670 @code{__vector} keyword.
5673 @section Inquiring on Alignment of Types or Variables
5675 @cindex type alignment
5676 @cindex variable alignment
5678 The keyword @code{__alignof__} allows you to inquire about how an object
5679 is aligned, or the minimum alignment usually required by a type. Its
5680 syntax is just like @code{sizeof}.
5682 For example, if the target machine requires a @code{double} value to be
5683 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
5684 This is true on many RISC machines. On more traditional machine
5685 designs, @code{__alignof__ (double)} is 4 or even 2.
5687 Some machines never actually require alignment; they allow reference to any
5688 data type even at an odd address. For these machines, @code{__alignof__}
5689 reports the smallest alignment that GCC gives the data type, usually as
5690 mandated by the target ABI.
5692 If the operand of @code{__alignof__} is an lvalue rather than a type,
5693 its value is the required alignment for its type, taking into account
5694 any minimum alignment specified with GCC's @code{__attribute__}
5695 extension (@pxref{Variable Attributes}). For example, after this
5699 struct foo @{ int x; char y; @} foo1;
5703 the value of @code{__alignof__ (foo1.y)} is 1, even though its actual
5704 alignment is probably 2 or 4, the same as @code{__alignof__ (int)}.
5706 It is an error to ask for the alignment of an incomplete type.
5710 @section An Inline Function is As Fast As a Macro
5711 @cindex inline functions
5712 @cindex integrating function code
5714 @cindex macros, inline alternative
5716 By declaring a function inline, you can direct GCC to make
5717 calls to that function faster. One way GCC can achieve this is to
5718 integrate that function's code into the code for its callers. This
5719 makes execution faster by eliminating the function-call overhead; in
5720 addition, if any of the actual argument values are constant, their
5721 known values may permit simplifications at compile time so that not
5722 all of the inline function's code needs to be included. The effect on
5723 code size is less predictable; object code may be larger or smaller
5724 with function inlining, depending on the particular case. You can
5725 also direct GCC to try to integrate all ``simple enough'' functions
5726 into their callers with the option @option{-finline-functions}.
5728 GCC implements three different semantics of declaring a function
5729 inline. One is available with @option{-std=gnu89} or
5730 @option{-fgnu89-inline} or when @code{gnu_inline} attribute is present
5731 on all inline declarations, another when
5732 @option{-std=c99}, @option{-std=c11},
5733 @option{-std=gnu99} or @option{-std=gnu11}
5734 (without @option{-fgnu89-inline}), and the third
5735 is used when compiling C++.
5737 To declare a function inline, use the @code{inline} keyword in its
5738 declaration, like this:
5748 If you are writing a header file to be included in ISO C90 programs, write
5749 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.
5751 The three types of inlining behave similarly in two important cases:
5752 when the @code{inline} keyword is used on a @code{static} function,
5753 like the example above, and when a function is first declared without
5754 using the @code{inline} keyword and then is defined with
5755 @code{inline}, like this:
5758 extern int inc (int *a);
5766 In both of these common cases, the program behaves the same as if you
5767 had not used the @code{inline} keyword, except for its speed.
5769 @cindex inline functions, omission of
5770 @opindex fkeep-inline-functions
5771 When a function is both inline and @code{static}, if all calls to the
5772 function are integrated into the caller, and the function's address is
5773 never used, then the function's own assembler code is never referenced.
5774 In this case, GCC does not actually output assembler code for the
5775 function, unless you specify the option @option{-fkeep-inline-functions}.
5776 Some calls cannot be integrated for various reasons (in particular,
5777 calls that precede the function's definition cannot be integrated, and
5778 neither can recursive calls within the definition). If there is a
5779 nonintegrated call, then the function is compiled to assembler code as
5780 usual. The function must also be compiled as usual if the program
5781 refers to its address, because that can't be inlined.
5784 Note that certain usages in a function definition can make it unsuitable
5785 for inline substitution. Among these usages are: variadic functions, use of
5786 @code{alloca}, use of variable-length data types (@pxref{Variable Length}),
5787 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
5788 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
5789 warns when a function marked @code{inline} could not be substituted,
5790 and gives the reason for the failure.
5792 @cindex automatic @code{inline} for C++ member fns
5793 @cindex @code{inline} automatic for C++ member fns
5794 @cindex member fns, automatically @code{inline}
5795 @cindex C++ member fns, automatically @code{inline}
5796 @opindex fno-default-inline
5797 As required by ISO C++, GCC considers member functions defined within
5798 the body of a class to be marked inline even if they are
5799 not explicitly declared with the @code{inline} keyword. You can
5800 override this with @option{-fno-default-inline}; @pxref{C++ Dialect
5801 Options,,Options Controlling C++ Dialect}.
5803 GCC does not inline any functions when not optimizing unless you specify
5804 the @samp{always_inline} attribute for the function, like this:
5807 /* @r{Prototype.} */
5808 inline void foo (const char) __attribute__((always_inline));
5811 The remainder of this section is specific to GNU C90 inlining.
5813 @cindex non-static inline function
5814 When an inline function is not @code{static}, then the compiler must assume
5815 that there may be calls from other source files; since a global symbol can
5816 be defined only once in any program, the function must not be defined in
5817 the other source files, so the calls therein cannot be integrated.
5818 Therefore, a non-@code{static} inline function is always compiled on its
5819 own in the usual fashion.
5821 If you specify both @code{inline} and @code{extern} in the function
5822 definition, then the definition is used only for inlining. In no case
5823 is the function compiled on its own, not even if you refer to its
5824 address explicitly. Such an address becomes an external reference, as
5825 if you had only declared the function, and had not defined it.
5827 This combination of @code{inline} and @code{extern} has almost the
5828 effect of a macro. The way to use it is to put a function definition in
5829 a header file with these keywords, and put another copy of the
5830 definition (lacking @code{inline} and @code{extern}) in a library file.
5831 The definition in the header file causes most calls to the function
5832 to be inlined. If any uses of the function remain, they refer to
5833 the single copy in the library.
5836 @section When is a Volatile Object Accessed?
5837 @cindex accessing volatiles
5838 @cindex volatile read
5839 @cindex volatile write
5840 @cindex volatile access
5842 C has the concept of volatile objects. These are normally accessed by
5843 pointers and used for accessing hardware or inter-thread
5844 communication. The standard encourages compilers to refrain from
5845 optimizations concerning accesses to volatile objects, but leaves it
5846 implementation defined as to what constitutes a volatile access. The
5847 minimum requirement is that at a sequence point all previous accesses
5848 to volatile objects have stabilized and no subsequent accesses have
5849 occurred. Thus an implementation is free to reorder and combine
5850 volatile accesses that occur between sequence points, but cannot do
5851 so for accesses across a sequence point. The use of volatile does
5852 not allow you to violate the restriction on updating objects multiple
5853 times between two sequence points.
5855 Accesses to non-volatile objects are not ordered with respect to
5856 volatile accesses. You cannot use a volatile object as a memory
5857 barrier to order a sequence of writes to non-volatile memory. For
5861 int *ptr = @var{something};
5863 *ptr = @var{something};
5868 Unless @var{*ptr} and @var{vobj} can be aliased, it is not guaranteed
5869 that the write to @var{*ptr} occurs by the time the update
5870 of @var{vobj} happens. If you need this guarantee, you must use
5871 a stronger memory barrier such as:
5874 int *ptr = @var{something};
5876 *ptr = @var{something};
5877 asm volatile ("" : : : "memory");
5881 A scalar volatile object is read when it is accessed in a void context:
5884 volatile int *src = @var{somevalue};
5888 Such expressions are rvalues, and GCC implements this as a
5889 read of the volatile object being pointed to.
5891 Assignments are also expressions and have an rvalue. However when
5892 assigning to a scalar volatile, the volatile object is not reread,
5893 regardless of whether the assignment expression's rvalue is used or
5894 not. If the assignment's rvalue is used, the value is that assigned
5895 to the volatile object. For instance, there is no read of @var{vobj}
5896 in all the following cases:
5901 vobj = @var{something};
5902 obj = vobj = @var{something};
5903 obj ? vobj = @var{onething} : vobj = @var{anotherthing};
5904 obj = (@var{something}, vobj = @var{anotherthing});
5907 If you need to read the volatile object after an assignment has
5908 occurred, you must use a separate expression with an intervening
5911 As bit-fields are not individually addressable, volatile bit-fields may
5912 be implicitly read when written to, or when adjacent bit-fields are
5913 accessed. Bit-field operations may be optimized such that adjacent
5914 bit-fields are only partially accessed, if they straddle a storage unit
5915 boundary. For these reasons it is unwise to use volatile bit-fields to
5919 @section Assembler Instructions with C Expression Operands
5920 @cindex extended @code{asm}
5921 @cindex @code{asm} expressions
5922 @cindex assembler instructions
5925 In an assembler instruction using @code{asm}, you can specify the
5926 operands of the instruction using C expressions. This means you need not
5927 guess which registers or memory locations contain the data you want
5930 You must specify an assembler instruction template much like what
5931 appears in a machine description, plus an operand constraint string for
5934 For example, here is how to use the 68881's @code{fsinx} instruction:
5937 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
5941 Here @code{angle} is the C expression for the input operand while
5942 @code{result} is that of the output operand. Each has @samp{"f"} as its
5943 operand constraint, saying that a floating-point register is required.
5944 The @samp{=} in @samp{=f} indicates that the operand is an output; all
5945 output operands' constraints must use @samp{=}. The constraints use the
5946 same language used in the machine description (@pxref{Constraints}).
5948 Each operand is described by an operand-constraint string followed by
5949 the C expression in parentheses. A colon separates the assembler
5950 template from the first output operand and another separates the last
5951 output operand from the first input, if any. Commas separate the
5952 operands within each group. The total number of operands is currently
5953 limited to 30; this limitation may be lifted in some future version of
5956 If there are no output operands but there are input operands, you must
5957 place two consecutive colons surrounding the place where the output
5960 As of GCC version 3.1, it is also possible to specify input and output
5961 operands using symbolic names which can be referenced within the
5962 assembler code. These names are specified inside square brackets
5963 preceding the constraint string, and can be referenced inside the
5964 assembler code using @code{%[@var{name}]} instead of a percentage sign
5965 followed by the operand number. Using named operands the above example
5969 asm ("fsinx %[angle],%[output]"
5970 : [output] "=f" (result)
5971 : [angle] "f" (angle));
5975 Note that the symbolic operand names have no relation whatsoever to
5976 other C identifiers. You may use any name you like, even those of
5977 existing C symbols, but you must ensure that no two operands within the same
5978 assembler construct use the same symbolic name.
5980 Output operand expressions must be lvalues; the compiler can check this.
5981 The input operands need not be lvalues. The compiler cannot check
5982 whether the operands have data types that are reasonable for the
5983 instruction being executed. It does not parse the assembler instruction
5984 template and does not know what it means or even whether it is valid
5985 assembler input. The extended @code{asm} feature is most often used for
5986 machine instructions the compiler itself does not know exist. If
5987 the output expression cannot be directly addressed (for example, it is a
5988 bit-field), your constraint must allow a register. In that case, GCC
5989 uses the register as the output of the @code{asm}, and then stores
5990 that register into the output.
5992 The ordinary output operands must be write-only; GCC assumes that
5993 the values in these operands before the instruction are dead and need
5994 not be generated. Extended asm supports input-output or read-write
5995 operands. Use the constraint character @samp{+} to indicate such an
5996 operand and list it with the output operands.
5998 You may, as an alternative, logically split its function into two
5999 separate operands, one input operand and one write-only output
6000 operand. The connection between them is expressed by constraints
6001 that say they need to be in the same location when the instruction
6002 executes. You can use the same C expression for both operands, or
6003 different expressions. For example, here we write the (fictitious)
6004 @samp{combine} instruction with @code{bar} as its read-only source
6005 operand and @code{foo} as its read-write destination:
6008 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
6012 The constraint @samp{"0"} for operand 1 says that it must occupy the
6013 same location as operand 0. A number in constraint is allowed only in
6014 an input operand and it must refer to an output operand.
6016 Only a number in the constraint can guarantee that one operand is in
6017 the same place as another. The mere fact that @code{foo} is the value
6018 of both operands is not enough to guarantee that they are in the
6019 same place in the generated assembler code. The following does not
6023 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
6026 Various optimizations or reloading could cause operands 0 and 1 to be in
6027 different registers; GCC knows no reason not to do so. For example, the
6028 compiler might find a copy of the value of @code{foo} in one register and
6029 use it for operand 1, but generate the output operand 0 in a different
6030 register (copying it afterward to @code{foo}'s own address). Of course,
6031 since the register for operand 1 is not even mentioned in the assembler
6032 code, the result will not work, but GCC can't tell that.
6034 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
6035 the operand number for a matching constraint. For example:
6038 asm ("cmoveq %1,%2,%[result]"
6039 : [result] "=r"(result)
6040 : "r" (test), "r"(new), "[result]"(old));
6043 Sometimes you need to make an @code{asm} operand be a specific register,
6044 but there's no matching constraint letter for that register @emph{by
6045 itself}. To force the operand into that register, use a local variable
6046 for the operand and specify the register in the variable declaration.
6047 @xref{Explicit Reg Vars}. Then for the @code{asm} operand, use any
6048 register constraint letter that matches the register:
6051 register int *p1 asm ("r0") = @dots{};
6052 register int *p2 asm ("r1") = @dots{};
6053 register int *result asm ("r0");
6054 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
6057 @anchor{Example of asm with clobbered asm reg}
6058 In the above example, beware that a register that is call-clobbered by
6059 the target ABI will be overwritten by any function call in the
6060 assignment, including library calls for arithmetic operators.
6061 Also a register may be clobbered when generating some operations,
6062 like variable shift, memory copy or memory move on x86.
6063 Assuming it is a call-clobbered register, this may happen to @code{r0}
6064 above by the assignment to @code{p2}. If you have to use such a
6065 register, use temporary variables for expressions between the register
6070 register int *p1 asm ("r0") = @dots{};
6071 register int *p2 asm ("r1") = t1;
6072 register int *result asm ("r0");
6073 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
6076 Some instructions clobber specific hard registers. To describe this,
6077 write a third colon after the input operands, followed by the names of
6078 the clobbered hard registers (given as strings). Here is a realistic
6079 example for the VAX:
6082 asm volatile ("movc3 %0,%1,%2"
6083 : /* @r{no outputs} */
6084 : "g" (from), "g" (to), "g" (count)
6085 : "r0", "r1", "r2", "r3", "r4", "r5");
6088 You may not write a clobber description in a way that overlaps with an
6089 input or output operand. For example, you may not have an operand
6090 describing a register class with one member if you mention that register
6091 in the clobber list. Variables declared to live in specific registers
6092 (@pxref{Explicit Reg Vars}), and used as asm input or output operands must
6093 have no part mentioned in the clobber description.
6094 There is no way for you to specify that an input
6095 operand is modified without also specifying it as an output
6096 operand. Note that if all the output operands you specify are for this
6097 purpose (and hence unused), you then also need to specify
6098 @code{volatile} for the @code{asm} construct, as described below, to
6099 prevent GCC from deleting the @code{asm} statement as unused.
6101 If you refer to a particular hardware register from the assembler code,
6102 you probably have to list the register after the third colon to
6103 tell the compiler the register's value is modified. In some assemblers,
6104 the register names begin with @samp{%}; to produce one @samp{%} in the
6105 assembler code, you must write @samp{%%} in the input.
6107 If your assembler instruction can alter the condition code register, add
6108 @samp{cc} to the list of clobbered registers. GCC on some machines
6109 represents the condition codes as a specific hardware register;
6110 @samp{cc} serves to name this register. On other machines, the
6111 condition code is handled differently, and specifying @samp{cc} has no
6112 effect. But it is valid no matter what the machine.
6114 If your assembler instructions access memory in an unpredictable
6115 fashion, add @samp{memory} to the list of clobbered registers. This
6116 causes GCC to not keep memory values cached in registers across the
6117 assembler instruction and not optimize stores or loads to that memory.
6118 You also should add the @code{volatile} keyword if the memory
6119 affected is not listed in the inputs or outputs of the @code{asm}, as
6120 the @samp{memory} clobber does not count as a side-effect of the
6121 @code{asm}. If you know how large the accessed memory is, you can add
6122 it as input or output but if this is not known, you should add
6123 @samp{memory}. As an example, if you access ten bytes of a string, you
6124 can use a memory input like:
6127 @{"m"( (@{ struct @{ char x[10]; @} *p = (void *)ptr ; *p; @}) )@}.
6130 Note that in the following example the memory input is necessary,
6131 otherwise GCC might optimize the store to @code{x} away:
6138 asm ("magic stuff accessing an 'int' pointed to by '%1'"
6139 : "=&d" (r) : "a" (y), "m" (*y));
6144 You can put multiple assembler instructions together in a single
6145 @code{asm} template, separated by the characters normally used in assembly
6146 code for the system. A combination that works in most places is a newline
6147 to break the line, plus a tab character to move to the instruction field
6148 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
6149 assembler allows semicolons as a line-breaking character. Note that some
6150 assembler dialects use semicolons to start a comment.
6151 The input operands are guaranteed not to use any of the clobbered
6152 registers, and neither do the output operands' addresses, so you can
6153 read and write the clobbered registers as many times as you like. Here
6154 is an example of multiple instructions in a template; it assumes the
6155 subroutine @code{_foo} accepts arguments in registers 9 and 10:
6158 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
6160 : "g" (from), "g" (to)
6164 Unless an output operand has the @samp{&} constraint modifier, GCC
6165 may allocate it in the same register as an unrelated input operand, on
6166 the assumption the inputs are consumed before the outputs are produced.
6167 This assumption may be false if the assembler code actually consists of
6168 more than one instruction. In such a case, use @samp{&} for each output
6169 operand that may not overlap an input. @xref{Modifiers}.
6171 If you want to test the condition code produced by an assembler
6172 instruction, you must include a branch and a label in the @code{asm}
6173 construct, as follows:
6176 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
6182 This assumes your assembler supports local labels, as the GNU assembler
6183 and most Unix assemblers do.
6185 Speaking of labels, jumps from one @code{asm} to another are not
6186 supported. The compiler's optimizers do not know about these jumps, and
6187 therefore they cannot take account of them when deciding how to
6188 optimize. @xref{Extended asm with goto}.
6190 @cindex macros containing @code{asm}
6191 Usually the most convenient way to use these @code{asm} instructions is to
6192 encapsulate them in macros that look like functions. For example,
6196 (@{ double __value, __arg = (x); \
6197 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
6202 Here the variable @code{__arg} is used to make sure that the instruction
6203 operates on a proper @code{double} value, and to accept only those
6204 arguments @code{x} that can convert automatically to a @code{double}.
6206 Another way to make sure the instruction operates on the correct data
6207 type is to use a cast in the @code{asm}. This is different from using a
6208 variable @code{__arg} in that it converts more different types. For
6209 example, if the desired type is @code{int}, casting the argument to
6210 @code{int} accepts a pointer with no complaint, while assigning the
6211 argument to an @code{int} variable named @code{__arg} warns about
6212 using a pointer unless the caller explicitly casts it.
6214 If an @code{asm} has output operands, GCC assumes for optimization
6215 purposes the instruction has no side effects except to change the output
6216 operands. This does not mean instructions with a side effect cannot be
6217 used, but you must be careful, because the compiler may eliminate them
6218 if the output operands aren't used, or move them out of loops, or
6219 replace two with one if they constitute a common subexpression. Also,
6220 if your instruction does have a side effect on a variable that otherwise
6221 appears not to change, the old value of the variable may be reused later
6222 if it happens to be found in a register.
6224 You can prevent an @code{asm} instruction from being deleted
6225 by writing the keyword @code{volatile} after
6226 the @code{asm}. For example:
6229 #define get_and_set_priority(new) \
6231 asm volatile ("get_and_set_priority %0, %1" \
6232 : "=g" (__old) : "g" (new)); \
6237 The @code{volatile} keyword indicates that the instruction has
6238 important side-effects. GCC does not delete a volatile @code{asm} if
6239 it is reachable. (The instruction can still be deleted if GCC can
6240 prove that control flow never reaches the location of the
6241 instruction.) Note that even a volatile @code{asm} instruction
6242 can be moved relative to other code, including across jump
6243 instructions. For example, on many targets there is a system
6244 register that can be set to control the rounding mode of
6245 floating-point operations. You might try
6246 setting it with a volatile @code{asm}, like this PowerPC example:
6249 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
6254 This does not work reliably, as the compiler may move the addition back
6255 before the volatile @code{asm}. To make it work you need to add an
6256 artificial dependency to the @code{asm} referencing a variable in the code
6257 you don't want moved, for example:
6260 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
6264 Similarly, you can't expect a
6265 sequence of volatile @code{asm} instructions to remain perfectly
6266 consecutive. If you want consecutive output, use a single @code{asm}.
6267 Also, GCC performs some optimizations across a volatile @code{asm}
6268 instruction; GCC does not ``forget everything'' when it encounters
6269 a volatile @code{asm} instruction the way some other compilers do.
6271 An @code{asm} instruction without any output operands is treated
6272 identically to a volatile @code{asm} instruction.
6274 It is a natural idea to look for a way to give access to the condition
6275 code left by the assembler instruction. However, when we attempted to
6276 implement this, we found no way to make it work reliably. The problem
6277 is that output operands might need reloading, which result in
6278 additional following ``store'' instructions. On most machines, these
6279 instructions alter the condition code before there is time to
6280 test it. This problem doesn't arise for ordinary ``test'' and
6281 ``compare'' instructions because they don't have any output operands.
6283 For reasons similar to those described above, it is not possible to give
6284 an assembler instruction access to the condition code left by previous
6287 @anchor{Extended asm with goto}
6288 As of GCC version 4.5, @code{asm goto} may be used to have the assembly
6289 jump to one or more C labels. In this form, a fifth section after the
6290 clobber list contains a list of all C labels to which the assembly may jump.
6291 Each label operand is implicitly self-named. The @code{asm} is also assumed
6292 to fall through to the next statement.
6294 This form of @code{asm} is restricted to not have outputs. This is due
6295 to a internal restriction in the compiler that control transfer instructions
6296 cannot have outputs. This restriction on @code{asm goto} may be lifted
6297 in some future version of the compiler. In the meantime, @code{asm goto}
6298 may include a memory clobber, and so leave outputs in memory.
6304 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
6305 : : "r"(x), "r"(&y) : "r5", "memory" : error);
6313 In this (inefficient) example, the @code{frob} instruction sets the
6314 carry bit to indicate an error. The @code{jc} instruction detects
6315 this and branches to the @code{error} label. Finally, the output
6316 of the @code{frob} instruction (@code{%r5}) is stored into the memory
6317 for variable @code{y}, which is later read by the @code{return} statement.
6323 asm goto ("mfsr %%r1, 123; jmp %%r1;"
6324 ".pushsection doit_table;"
6325 ".long %l0, %l1, %l2, %l3;"
6327 : : : "r1" : label1, label2, label3, label4);
6328 __builtin_unreachable ();
6344 In this (also inefficient) example, the @code{mfsr} instruction reads
6345 an address from some out-of-band machine register, and the following
6346 @code{jmp} instruction branches to that address. The address read by
6347 the @code{mfsr} instruction is assumed to have been previously set via
6348 some application-specific mechanism to be one of the four values stored
6349 in the @code{doit_table} section. Finally, the @code{asm} is followed
6350 by a call to @code{__builtin_unreachable} to indicate that the @code{asm}
6351 does not in fact fall through.
6354 #define TRACE1(NUM) \
6356 asm goto ("0: nop;" \
6357 ".pushsection trace_table;" \
6360 : : : : trace#NUM); \
6361 if (0) @{ trace#NUM: trace(); @} \
6363 #define TRACE TRACE1(__COUNTER__)
6367 In this example (which in fact inspired the @code{asm goto} feature)
6368 we want on rare occasions to call the @code{trace} function; on other
6369 occasions we'd like to keep the overhead to the absolute minimum.
6370 The normal code path consists of a single @code{nop} instruction.
6371 However, we record the address of this @code{nop} together with the
6372 address of a label that calls the @code{trace} function. This allows
6373 the @code{nop} instruction to be patched at run time to be an
6374 unconditional branch to the stored label. It is assumed that an
6375 optimizing compiler moves the labeled block out of line, to
6376 optimize the fall through path from the @code{asm}.
6378 If you are writing a header file that should be includable in ISO C
6379 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
6382 @subsection Size of an @code{asm}
6384 Some targets require that GCC track the size of each instruction used in
6385 order to generate correct code. Because the final length of an
6386 @code{asm} is only known by the assembler, GCC must make an estimate as
6387 to how big it will be. The estimate is formed by counting the number of
6388 statements in the pattern of the @code{asm} and multiplying that by the
6389 length of the longest instruction on that processor. Statements in the
6390 @code{asm} are identified by newline characters and whatever statement
6391 separator characters are supported by the assembler; on most processors
6392 this is the @samp{;} character.
6394 Normally, GCC's estimate is perfectly adequate to ensure that correct
6395 code is generated, but it is possible to confuse the compiler if you use
6396 pseudo instructions or assembler macros that expand into multiple real
6397 instructions or if you use assembler directives that expand to more
6398 space in the object file than is needed for a single instruction.
6399 If this happens then the assembler produces a diagnostic saying that
6400 a label is unreachable.
6402 @subsection i386 floating-point asm operands
6404 On i386 targets, there are several rules on the usage of stack-like registers
6405 in the operands of an @code{asm}. These rules apply only to the operands
6406 that are stack-like registers:
6410 Given a set of input registers that die in an @code{asm}, it is
6411 necessary to know which are implicitly popped by the @code{asm}, and
6412 which must be explicitly popped by GCC@.
6414 An input register that is implicitly popped by the @code{asm} must be
6415 explicitly clobbered, unless it is constrained to match an
6419 For any input register that is implicitly popped by an @code{asm}, it is
6420 necessary to know how to adjust the stack to compensate for the pop.
6421 If any non-popped input is closer to the top of the reg-stack than
6422 the implicitly popped register, it would not be possible to know what the
6423 stack looked like---it's not clear how the rest of the stack ``slides
6426 All implicitly popped input registers must be closer to the top of
6427 the reg-stack than any input that is not implicitly popped.
6429 It is possible that if an input dies in an @code{asm}, the compiler might
6430 use the input register for an output reload. Consider this example:
6433 asm ("foo" : "=t" (a) : "f" (b));
6437 This code says that input @code{b} is not popped by the @code{asm}, and that
6438 the @code{asm} pushes a result onto the reg-stack, i.e., the stack is one
6439 deeper after the @code{asm} than it was before. But, it is possible that
6440 reload may think that it can use the same register for both the input and
6443 To prevent this from happening,
6444 if any input operand uses the @code{f} constraint, all output register
6445 constraints must use the @code{&} early-clobber modifier.
6447 The example above would be correctly written as:
6450 asm ("foo" : "=&t" (a) : "f" (b));
6454 Some operands need to be in particular places on the stack. All
6455 output operands fall in this category---GCC has no other way to
6456 know which registers the outputs appear in unless you indicate
6457 this in the constraints.
6459 Output operands must specifically indicate which register an output
6460 appears in after an @code{asm}. @code{=f} is not allowed: the operand
6461 constraints must select a class with a single register.
6464 Output operands may not be ``inserted'' between existing stack registers.
6465 Since no 387 opcode uses a read/write operand, all output operands
6466 are dead before the @code{asm}, and are pushed by the @code{asm}.
6467 It makes no sense to push anywhere but the top of the reg-stack.
6469 Output operands must start at the top of the reg-stack: output
6470 operands may not ``skip'' a register.
6473 Some @code{asm} statements may need extra stack space for internal
6474 calculations. This can be guaranteed by clobbering stack registers
6475 unrelated to the inputs and outputs.
6479 Here are a couple of reasonable @code{asm}s to want to write. This
6481 takes one input, which is internally popped, and produces two outputs.
6484 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
6488 This @code{asm} takes two inputs, which are popped by the @code{fyl2xp1} opcode,
6489 and replaces them with one output. The @code{st(1)} clobber is necessary
6490 for the compiler to know that @code{fyl2xp1} pops both inputs.
6493 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
6499 @section Controlling Names Used in Assembler Code
6500 @cindex assembler names for identifiers
6501 @cindex names used in assembler code
6502 @cindex identifiers, names in assembler code
6504 You can specify the name to be used in the assembler code for a C
6505 function or variable by writing the @code{asm} (or @code{__asm__})
6506 keyword after the declarator as follows:
6509 int foo asm ("myfoo") = 2;
6513 This specifies that the name to be used for the variable @code{foo} in
6514 the assembler code should be @samp{myfoo} rather than the usual
6517 On systems where an underscore is normally prepended to the name of a C
6518 function or variable, this feature allows you to define names for the
6519 linker that do not start with an underscore.
6521 It does not make sense to use this feature with a non-static local
6522 variable since such variables do not have assembler names. If you are
6523 trying to put the variable in a particular register, see @ref{Explicit
6524 Reg Vars}. GCC presently accepts such code with a warning, but will
6525 probably be changed to issue an error, rather than a warning, in the
6528 You cannot use @code{asm} in this way in a function @emph{definition}; but
6529 you can get the same effect by writing a declaration for the function
6530 before its definition and putting @code{asm} there, like this:
6533 extern func () asm ("FUNC");
6540 It is up to you to make sure that the assembler names you choose do not
6541 conflict with any other assembler symbols. Also, you must not use a
6542 register name; that would produce completely invalid assembler code. GCC
6543 does not as yet have the ability to store static variables in registers.
6544 Perhaps that will be added.
6546 @node Explicit Reg Vars
6547 @section Variables in Specified Registers
6548 @cindex explicit register variables
6549 @cindex variables in specified registers
6550 @cindex specified registers
6551 @cindex registers, global allocation
6553 GNU C allows you to put a few global variables into specified hardware
6554 registers. You can also specify the register in which an ordinary
6555 register variable should be allocated.
6559 Global register variables reserve registers throughout the program.
6560 This may be useful in programs such as programming language
6561 interpreters that have a couple of global variables that are accessed
6565 Local register variables in specific registers do not reserve the
6566 registers, except at the point where they are used as input or output
6567 operands in an @code{asm} statement and the @code{asm} statement itself is
6568 not deleted. The compiler's data flow analysis is capable of determining
6569 where the specified registers contain live values, and where they are
6570 available for other uses. Stores into local register variables may be deleted
6571 when they appear to be dead according to dataflow analysis. References
6572 to local register variables may be deleted or moved or simplified.
6574 These local variables are sometimes convenient for use with the extended
6575 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
6576 output of the assembler instruction directly into a particular register.
6577 (This works provided the register you specify fits the constraints
6578 specified for that operand in the @code{asm}.)
6586 @node Global Reg Vars
6587 @subsection Defining Global Register Variables
6588 @cindex global register variables
6589 @cindex registers, global variables in
6591 You can define a global register variable in GNU C like this:
6594 register int *foo asm ("a5");
6598 Here @code{a5} is the name of the register that should be used. Choose a
6599 register that is normally saved and restored by function calls on your
6600 machine, so that library routines will not clobber it.
6602 Naturally the register name is cpu-dependent, so you need to
6603 conditionalize your program according to cpu type. The register
6604 @code{a5} is a good choice on a 68000 for a variable of pointer
6605 type. On machines with register windows, be sure to choose a ``global''
6606 register that is not affected magically by the function call mechanism.
6608 In addition, different operating systems on the same CPU may differ in how they
6609 name the registers; then you need additional conditionals. For
6610 example, some 68000 operating systems call this register @code{%a5}.
6612 Eventually there may be a way of asking the compiler to choose a register
6613 automatically, but first we need to figure out how it should choose and
6614 how to enable you to guide the choice. No solution is evident.
6616 Defining a global register variable in a certain register reserves that
6617 register entirely for this use, at least within the current compilation.
6618 The register is not allocated for any other purpose in the functions
6619 in the current compilation, and is not saved and restored by
6620 these functions. Stores into this register are never deleted even if they
6621 appear to be dead, but references may be deleted or moved or
6624 It is not safe to access the global register variables from signal
6625 handlers, or from more than one thread of control, because the system
6626 library routines may temporarily use the register for other things (unless
6627 you recompile them specially for the task at hand).
6629 @cindex @code{qsort}, and global register variables
6630 It is not safe for one function that uses a global register variable to
6631 call another such function @code{foo} by way of a third function
6632 @code{lose} that is compiled without knowledge of this variable (i.e.@: in a
6633 different source file in which the variable isn't declared). This is
6634 because @code{lose} might save the register and put some other value there.
6635 For example, you can't expect a global register variable to be available in
6636 the comparison-function that you pass to @code{qsort}, since @code{qsort}
6637 might have put something else in that register. (If you are prepared to
6638 recompile @code{qsort} with the same global register variable, you can
6639 solve this problem.)
6641 If you want to recompile @code{qsort} or other source files that do not
6642 actually use your global register variable, so that they do not use that
6643 register for any other purpose, then it suffices to specify the compiler
6644 option @option{-ffixed-@var{reg}}. You need not actually add a global
6645 register declaration to their source code.
6647 A function that can alter the value of a global register variable cannot
6648 safely be called from a function compiled without this variable, because it
6649 could clobber the value the caller expects to find there on return.
6650 Therefore, the function that is the entry point into the part of the
6651 program that uses the global register variable must explicitly save and
6652 restore the value that belongs to its caller.
6654 @cindex register variable after @code{longjmp}
6655 @cindex global register after @code{longjmp}
6656 @cindex value after @code{longjmp}
6659 On most machines, @code{longjmp} restores to each global register
6660 variable the value it had at the time of the @code{setjmp}. On some
6661 machines, however, @code{longjmp} does not change the value of global
6662 register variables. To be portable, the function that called @code{setjmp}
6663 should make other arrangements to save the values of the global register
6664 variables, and to restore them in a @code{longjmp}. This way, the same
6665 thing happens regardless of what @code{longjmp} does.
6667 All global register variable declarations must precede all function
6668 definitions. If such a declaration could appear after function
6669 definitions, the declaration would be too late to prevent the register from
6670 being used for other purposes in the preceding functions.
6672 Global register variables may not have initial values, because an
6673 executable file has no means to supply initial contents for a register.
6675 On the SPARC, there are reports that g3 @dots{} g7 are suitable
6676 registers, but certain library functions, such as @code{getwd}, as well
6677 as the subroutines for division and remainder, modify g3 and g4. g1 and
6678 g2 are local temporaries.
6680 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
6681 Of course, it does not do to use more than a few of those.
6683 @node Local Reg Vars
6684 @subsection Specifying Registers for Local Variables
6685 @cindex local variables, specifying registers
6686 @cindex specifying registers for local variables
6687 @cindex registers for local variables
6689 You can define a local register variable with a specified register
6693 register int *foo asm ("a5");
6697 Here @code{a5} is the name of the register that should be used. Note
6698 that this is the same syntax used for defining global register
6699 variables, but for a local variable it appears within a function.
6701 Naturally the register name is cpu-dependent, but this is not a
6702 problem, since specific registers are most often useful with explicit
6703 assembler instructions (@pxref{Extended Asm}). Both of these things
6704 generally require that you conditionalize your program according to
6707 In addition, operating systems on one type of cpu may differ in how they
6708 name the registers; then you need additional conditionals. For
6709 example, some 68000 operating systems call this register @code{%a5}.
6711 Defining such a register variable does not reserve the register; it
6712 remains available for other uses in places where flow control determines
6713 the variable's value is not live.
6715 This option does not guarantee that GCC generates code that has
6716 this variable in the register you specify at all times. You may not
6717 code an explicit reference to this register in the @emph{assembler
6718 instruction template} part of an @code{asm} statement and assume it
6719 always refers to this variable. However, using the variable as an
6720 @code{asm} @emph{operand} guarantees that the specified register is used
6723 Stores into local register variables may be deleted when they appear to be dead
6724 according to dataflow analysis. References to local register variables may
6725 be deleted or moved or simplified.
6727 As for global register variables, it's recommended that you choose a
6728 register that is normally saved and restored by function calls on
6729 your machine, so that library routines will not clobber it. A common
6730 pitfall is to initialize multiple call-clobbered registers with
6731 arbitrary expressions, where a function call or library call for an
6732 arithmetic operator overwrites a register value from a previous
6733 assignment, for example @code{r0} below:
6735 register int *p1 asm ("r0") = @dots{};
6736 register int *p2 asm ("r1") = @dots{};
6740 In those cases, a solution is to use a temporary variable for
6741 each arbitrary expression. @xref{Example of asm with clobbered asm reg}.
6743 @node Alternate Keywords
6744 @section Alternate Keywords
6745 @cindex alternate keywords
6746 @cindex keywords, alternate
6748 @option{-ansi} and the various @option{-std} options disable certain
6749 keywords. This causes trouble when you want to use GNU C extensions, or
6750 a general-purpose header file that should be usable by all programs,
6751 including ISO C programs. The keywords @code{asm}, @code{typeof} and
6752 @code{inline} are not available in programs compiled with
6753 @option{-ansi} or @option{-std} (although @code{inline} can be used in a
6754 program compiled with @option{-std=c99} or @option{-std=c11}). The
6756 @code{restrict} is only available when @option{-std=gnu99} (which will
6757 eventually be the default) or @option{-std=c99} (or the equivalent
6758 @option{-std=iso9899:1999}), or an option for a later standard
6761 The way to solve these problems is to put @samp{__} at the beginning and
6762 end of each problematical keyword. For example, use @code{__asm__}
6763 instead of @code{asm}, and @code{__inline__} instead of @code{inline}.
6765 Other C compilers won't accept these alternative keywords; if you want to
6766 compile with another compiler, you can define the alternate keywords as
6767 macros to replace them with the customary keywords. It looks like this:
6775 @findex __extension__
6777 @option{-pedantic} and other options cause warnings for many GNU C extensions.
6779 prevent such warnings within one expression by writing
6780 @code{__extension__} before the expression. @code{__extension__} has no
6781 effect aside from this.
6783 @node Incomplete Enums
6784 @section Incomplete @code{enum} Types
6786 You can define an @code{enum} tag without specifying its possible values.
6787 This results in an incomplete type, much like what you get if you write
6788 @code{struct foo} without describing the elements. A later declaration
6789 that does specify the possible values completes the type.
6791 You can't allocate variables or storage using the type while it is
6792 incomplete. However, you can work with pointers to that type.
6794 This extension may not be very useful, but it makes the handling of
6795 @code{enum} more consistent with the way @code{struct} and @code{union}
6798 This extension is not supported by GNU C++.
6800 @node Function Names
6801 @section Function Names as Strings
6802 @cindex @code{__func__} identifier
6803 @cindex @code{__FUNCTION__} identifier
6804 @cindex @code{__PRETTY_FUNCTION__} identifier
6806 GCC provides three magic variables that hold the name of the current
6807 function, as a string. The first of these is @code{__func__}, which
6808 is part of the C99 standard:
6810 The identifier @code{__func__} is implicitly declared by the translator
6811 as if, immediately following the opening brace of each function
6812 definition, the declaration
6815 static const char __func__[] = "function-name";
6819 appeared, where function-name is the name of the lexically-enclosing
6820 function. This name is the unadorned name of the function.
6822 @code{__FUNCTION__} is another name for @code{__func__}. Older
6823 versions of GCC recognize only this name. However, it is not
6824 standardized. For maximum portability, we recommend you use
6825 @code{__func__}, but provide a fallback definition with the
6829 #if __STDC_VERSION__ < 199901L
6831 # define __func__ __FUNCTION__
6833 # define __func__ "<unknown>"
6838 In C, @code{__PRETTY_FUNCTION__} is yet another name for
6839 @code{__func__}. However, in C++, @code{__PRETTY_FUNCTION__} contains
6840 the type signature of the function as well as its bare name. For
6841 example, this program:
6845 extern int printf (char *, ...);
6852 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
6853 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
6871 __PRETTY_FUNCTION__ = void a::sub(int)
6874 These identifiers are not preprocessor macros. In GCC 3.3 and
6875 earlier, in C only, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__}
6876 were treated as string literals; they could be used to initialize
6877 @code{char} arrays, and they could be concatenated with other string
6878 literals. GCC 3.4 and later treat them as variables, like
6879 @code{__func__}. In C++, @code{__FUNCTION__} and
6880 @code{__PRETTY_FUNCTION__} have always been variables.
6882 @node Return Address
6883 @section Getting the Return or Frame Address of a Function
6885 These functions may be used to get information about the callers of a
6888 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
6889 This function returns the return address of the current function, or of
6890 one of its callers. The @var{level} argument is number of frames to
6891 scan up the call stack. A value of @code{0} yields the return address
6892 of the current function, a value of @code{1} yields the return address
6893 of the caller of the current function, and so forth. When inlining
6894 the expected behavior is that the function returns the address of
6895 the function that is returned to. To work around this behavior use
6896 the @code{noinline} function attribute.
6898 The @var{level} argument must be a constant integer.
6900 On some machines it may be impossible to determine the return address of
6901 any function other than the current one; in such cases, or when the top
6902 of the stack has been reached, this function returns @code{0} or a
6903 random value. In addition, @code{__builtin_frame_address} may be used
6904 to determine if the top of the stack has been reached.
6906 Additional post-processing of the returned value may be needed, see
6907 @code{__builtin_extract_return_addr}.
6909 This function should only be used with a nonzero argument for debugging
6913 @deftypefn {Built-in Function} {void *} __builtin_extract_return_addr (void *@var{addr})
6914 The address as returned by @code{__builtin_return_address} may have to be fed
6915 through this function to get the actual encoded address. For example, on the
6916 31-bit S/390 platform the highest bit has to be masked out, or on SPARC
6917 platforms an offset has to be added for the true next instruction to be
6920 If no fixup is needed, this function simply passes through @var{addr}.
6923 @deftypefn {Built-in Function} {void *} __builtin_frob_return_address (void *@var{addr})
6924 This function does the reverse of @code{__builtin_extract_return_addr}.
6927 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
6928 This function is similar to @code{__builtin_return_address}, but it
6929 returns the address of the function frame rather than the return address
6930 of the function. Calling @code{__builtin_frame_address} with a value of
6931 @code{0} yields the frame address of the current function, a value of
6932 @code{1} yields the frame address of the caller of the current function,
6935 The frame is the area on the stack that holds local variables and saved
6936 registers. The frame address is normally the address of the first word
6937 pushed on to the stack by the function. However, the exact definition
6938 depends upon the processor and the calling convention. If the processor
6939 has a dedicated frame pointer register, and the function has a frame,
6940 then @code{__builtin_frame_address} returns the value of the frame
6943 On some machines it may be impossible to determine the frame address of
6944 any function other than the current one; in such cases, or when the top
6945 of the stack has been reached, this function returns @code{0} if
6946 the first frame pointer is properly initialized by the startup code.
6948 This function should only be used with a nonzero argument for debugging
6952 @node Vector Extensions
6953 @section Using Vector Instructions through Built-in Functions
6955 On some targets, the instruction set contains SIMD vector instructions which
6956 operate on multiple values contained in one large register at the same time.
6957 For example, on the i386 the MMX, 3DNow!@: and SSE extensions can be used
6960 The first step in using these extensions is to provide the necessary data
6961 types. This should be done using an appropriate @code{typedef}:
6964 typedef int v4si __attribute__ ((vector_size (16)));
6968 The @code{int} type specifies the base type, while the attribute specifies
6969 the vector size for the variable, measured in bytes. For example, the
6970 declaration above causes the compiler to set the mode for the @code{v4si}
6971 type to be 16 bytes wide and divided into @code{int} sized units. For
6972 a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the
6973 corresponding mode of @code{foo} is @acronym{V4SI}.
6975 The @code{vector_size} attribute is only applicable to integral and
6976 float scalars, although arrays, pointers, and function return values
6977 are allowed in conjunction with this construct. Only sizes that are
6978 a power of two are currently allowed.
6980 All the basic integer types can be used as base types, both as signed
6981 and as unsigned: @code{char}, @code{short}, @code{int}, @code{long},
6982 @code{long long}. In addition, @code{float} and @code{double} can be
6983 used to build floating-point vector types.
6985 Specifying a combination that is not valid for the current architecture
6986 causes GCC to synthesize the instructions using a narrower mode.
6987 For example, if you specify a variable of type @code{V4SI} and your
6988 architecture does not allow for this specific SIMD type, GCC
6989 produces code that uses 4 @code{SIs}.
6991 The types defined in this manner can be used with a subset of normal C
6992 operations. Currently, GCC allows using the following operators
6993 on these types: @code{+, -, *, /, unary minus, ^, |, &, ~, %}@.
6995 The operations behave like C++ @code{valarrays}. Addition is defined as
6996 the addition of the corresponding elements of the operands. For
6997 example, in the code below, each of the 4 elements in @var{a} is
6998 added to the corresponding 4 elements in @var{b} and the resulting
6999 vector is stored in @var{c}.
7002 typedef int v4si __attribute__ ((vector_size (16)));
7009 Subtraction, multiplication, division, and the logical operations
7010 operate in a similar manner. Likewise, the result of using the unary
7011 minus or complement operators on a vector type is a vector whose
7012 elements are the negative or complemented values of the corresponding
7013 elements in the operand.
7015 It is possible to use shifting operators @code{<<}, @code{>>} on
7016 integer-type vectors. The operation is defined as following: @code{@{a0,
7017 a1, @dots{}, an@} >> @{b0, b1, @dots{}, bn@} == @{a0 >> b0, a1 >> b1,
7018 @dots{}, an >> bn@}}@. Vector operands must have the same number of
7021 For convenience, it is allowed to use a binary vector operation
7022 where one operand is a scalar. In that case the compiler transforms
7023 the scalar operand into a vector where each element is the scalar from
7024 the operation. The transformation happens only if the scalar could be
7025 safely converted to the vector-element type.
7026 Consider the following code.
7029 typedef int v4si __attribute__ ((vector_size (16)));
7034 a = b + 1; /* a = b + @{1,1,1,1@}; */
7035 a = 2 * b; /* a = @{2,2,2,2@} * b; */
7037 a = l + a; /* Error, cannot convert long to int. */
7040 Vectors can be subscripted as if the vector were an array with
7041 the same number of elements and base type. Out of bound accesses
7042 invoke undefined behavior at run time. Warnings for out of bound
7043 accesses for vector subscription can be enabled with
7044 @option{-Warray-bounds}.
7046 Vector comparison is supported with standard comparison
7047 operators: @code{==, !=, <, <=, >, >=}. Comparison operands can be
7048 vector expressions of integer-type or real-type. Comparison between
7049 integer-type vectors and real-type vectors are not supported. The
7050 result of the comparison is a vector of the same width and number of
7051 elements as the comparison operands with a signed integral element
7054 Vectors are compared element-wise producing 0 when comparison is false
7055 and -1 (constant of the appropriate type where all bits are set)
7056 otherwise. Consider the following example.
7059 typedef int v4si __attribute__ ((vector_size (16)));
7061 v4si a = @{1,2,3,4@};
7062 v4si b = @{3,2,1,4@};
7065 c = a > b; /* The result would be @{0, 0,-1, 0@} */
7066 c = a == b; /* The result would be @{0,-1, 0,-1@} */
7069 Vector shuffling is available using functions
7070 @code{__builtin_shuffle (vec, mask)} and
7071 @code{__builtin_shuffle (vec0, vec1, mask)}.
7072 Both functions construct a permutation of elements from one or two
7073 vectors and return a vector of the same type as the input vector(s).
7074 The @var{mask} is an integral vector with the same width (@var{W})
7075 and element count (@var{N}) as the output vector.
7077 The elements of the input vectors are numbered in memory ordering of
7078 @var{vec0} beginning at 0 and @var{vec1} beginning at @var{N}. The
7079 elements of @var{mask} are considered modulo @var{N} in the single-operand
7080 case and modulo @math{2*@var{N}} in the two-operand case.
7082 Consider the following example,
7085 typedef int v4si __attribute__ ((vector_size (16)));
7087 v4si a = @{1,2,3,4@};
7088 v4si b = @{5,6,7,8@};
7089 v4si mask1 = @{0,1,1,3@};
7090 v4si mask2 = @{0,4,2,5@};
7093 res = __builtin_shuffle (a, mask1); /* res is @{1,2,2,4@} */
7094 res = __builtin_shuffle (a, b, mask2); /* res is @{1,5,3,6@} */
7097 Note that @code{__builtin_shuffle} is intentionally semantically
7098 compatible with the OpenCL @code{shuffle} and @code{shuffle2} functions.
7100 You can declare variables and use them in function calls and returns, as
7101 well as in assignments and some casts. You can specify a vector type as
7102 a return type for a function. Vector types can also be used as function
7103 arguments. It is possible to cast from one vector type to another,
7104 provided they are of the same size (in fact, you can also cast vectors
7105 to and from other datatypes of the same size).
7107 You cannot operate between vectors of different lengths or different
7108 signedness without a cast.
7112 @findex __builtin_offsetof
7114 GCC implements for both C and C++ a syntactic extension to implement
7115 the @code{offsetof} macro.
7119 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")"
7121 offsetof_member_designator:
7123 | offsetof_member_designator "." @code{identifier}
7124 | offsetof_member_designator "[" @code{expr} "]"
7127 This extension is sufficient such that
7130 #define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member})
7134 is a suitable definition of the @code{offsetof} macro. In C++, @var{type}
7135 may be dependent. In either case, @var{member} may consist of a single
7136 identifier, or a sequence of member accesses and array references.
7138 @node __sync Builtins
7139 @section Legacy __sync Built-in Functions for Atomic Memory Access
7141 The following built-in functions
7142 are intended to be compatible with those described
7143 in the @cite{Intel Itanium Processor-specific Application Binary Interface},
7144 section 7.4. As such, they depart from the normal GCC practice of using
7145 the @samp{__builtin_} prefix, and further that they are overloaded such that
7146 they work on multiple types.
7148 The definition given in the Intel documentation allows only for the use of
7149 the types @code{int}, @code{long}, @code{long long} as well as their unsigned
7150 counterparts. GCC allows any integral scalar or pointer type that is
7151 1, 2, 4 or 8 bytes in length.
7153 Not all operations are supported by all target processors. If a particular
7154 operation cannot be implemented on the target processor, a warning is
7155 generated and a call an external function is generated. The external
7156 function carries the same name as the built-in version,
7157 with an additional suffix
7158 @samp{_@var{n}} where @var{n} is the size of the data type.
7160 @c ??? Should we have a mechanism to suppress this warning? This is almost
7161 @c useful for implementing the operation under the control of an external
7164 In most cases, these built-in functions are considered a @dfn{full barrier}.
7166 no memory operand is moved across the operation, either forward or
7167 backward. Further, instructions are issued as necessary to prevent the
7168 processor from speculating loads across the operation and from queuing stores
7169 after the operation.
7171 All of the routines are described in the Intel documentation to take
7172 ``an optional list of variables protected by the memory barrier''. It's
7173 not clear what is meant by that; it could mean that @emph{only} the
7174 following variables are protected, or it could mean that these variables
7175 should in addition be protected. At present GCC ignores this list and
7176 protects all variables that are globally accessible. If in the future
7177 we make some use of this list, an empty list will continue to mean all
7178 globally accessible variables.
7181 @item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...)
7182 @itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...)
7183 @itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...)
7184 @itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...)
7185 @itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...)
7186 @itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...)
7187 @findex __sync_fetch_and_add
7188 @findex __sync_fetch_and_sub
7189 @findex __sync_fetch_and_or
7190 @findex __sync_fetch_and_and
7191 @findex __sync_fetch_and_xor
7192 @findex __sync_fetch_and_nand
7193 These built-in functions perform the operation suggested by the name, and
7194 returns the value that had previously been in memory. That is,
7197 @{ tmp = *ptr; *ptr @var{op}= value; return tmp; @}
7198 @{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; @} // nand
7201 @emph{Note:} GCC 4.4 and later implement @code{__sync_fetch_and_nand}
7202 as @code{*ptr = ~(tmp & value)} instead of @code{*ptr = ~tmp & value}.
7204 @item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...)
7205 @itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...)
7206 @itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...)
7207 @itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...)
7208 @itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...)
7209 @itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...)
7210 @findex __sync_add_and_fetch
7211 @findex __sync_sub_and_fetch
7212 @findex __sync_or_and_fetch
7213 @findex __sync_and_and_fetch
7214 @findex __sync_xor_and_fetch
7215 @findex __sync_nand_and_fetch
7216 These built-in functions perform the operation suggested by the name, and
7217 return the new value. That is,
7220 @{ *ptr @var{op}= value; return *ptr; @}
7221 @{ *ptr = ~(*ptr & value); return *ptr; @} // nand
7224 @emph{Note:} GCC 4.4 and later implement @code{__sync_nand_and_fetch}
7225 as @code{*ptr = ~(*ptr & value)} instead of
7226 @code{*ptr = ~*ptr & value}.
7228 @item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
7229 @itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...)
7230 @findex __sync_bool_compare_and_swap
7231 @findex __sync_val_compare_and_swap
7232 These built-in functions perform an atomic compare and swap.
7233 That is, if the current
7234 value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into
7237 The ``bool'' version returns true if the comparison is successful and
7238 @var{newval} is written. The ``val'' version returns the contents
7239 of @code{*@var{ptr}} before the operation.
7241 @item __sync_synchronize (...)
7242 @findex __sync_synchronize
7243 This built-in function issues a full memory barrier.
7245 @item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...)
7246 @findex __sync_lock_test_and_set
7247 This built-in function, as described by Intel, is not a traditional test-and-set
7248 operation, but rather an atomic exchange operation. It writes @var{value}
7249 into @code{*@var{ptr}}, and returns the previous contents of
7252 Many targets have only minimal support for such locks, and do not support
7253 a full exchange operation. In this case, a target may support reduced
7254 functionality here by which the @emph{only} valid value to store is the
7255 immediate constant 1. The exact value actually stored in @code{*@var{ptr}}
7256 is implementation defined.
7258 This built-in function is not a full barrier,
7259 but rather an @dfn{acquire barrier}.
7260 This means that references after the operation cannot move to (or be
7261 speculated to) before the operation, but previous memory stores may not
7262 be globally visible yet, and previous memory loads may not yet be
7265 @item void __sync_lock_release (@var{type} *ptr, ...)
7266 @findex __sync_lock_release
7267 This built-in function releases the lock acquired by
7268 @code{__sync_lock_test_and_set}.
7269 Normally this means writing the constant 0 to @code{*@var{ptr}}.
7271 This built-in function is not a full barrier,
7272 but rather a @dfn{release barrier}.
7273 This means that all previous memory stores are globally visible, and all
7274 previous memory loads have been satisfied, but following memory reads
7275 are not prevented from being speculated to before the barrier.
7278 @node __atomic Builtins
7279 @section Built-in functions for memory model aware atomic operations
7281 The following built-in functions approximately match the requirements for
7282 C++11 memory model. Many are similar to the @samp{__sync} prefixed built-in
7283 functions, but all also have a memory model parameter. These are all
7284 identified by being prefixed with @samp{__atomic}, and most are overloaded
7285 such that they work with multiple types.
7287 GCC allows any integral scalar or pointer type that is 1, 2, 4, or 8
7288 bytes in length. 16-byte integral types are also allowed if
7289 @samp{__int128} (@pxref{__int128}) is supported by the architecture.
7291 Target architectures are encouraged to provide their own patterns for
7292 each of these built-in functions. If no target is provided, the original
7293 non-memory model set of @samp{__sync} atomic built-in functions are
7294 utilized, along with any required synchronization fences surrounding it in
7295 order to achieve the proper behavior. Execution in this case is subject
7296 to the same restrictions as those built-in functions.
7298 If there is no pattern or mechanism to provide a lock free instruction
7299 sequence, a call is made to an external routine with the same parameters
7300 to be resolved at run time.
7302 The four non-arithmetic functions (load, store, exchange, and
7303 compare_exchange) all have a generic version as well. This generic
7304 version works on any data type. If the data type size maps to one
7305 of the integral sizes that may have lock free support, the generic
7306 version utilizes the lock free built-in function. Otherwise an
7307 external call is left to be resolved at run time. This external call is
7308 the same format with the addition of a @samp{size_t} parameter inserted
7309 as the first parameter indicating the size of the object being pointed to.
7310 All objects must be the same size.
7312 There are 6 different memory models that can be specified. These map
7313 to the same names in the C++11 standard. Refer there or to the
7314 @uref{http://gcc.gnu.org/wiki/Atomic/GCCMM/AtomicSync,GCC wiki on
7315 atomic synchronization} for more detailed definitions. These memory
7316 models integrate both barriers to code motion as well as synchronization
7317 requirements with other threads. These are listed in approximately
7318 ascending order of strength. It is also possible to use target specific
7319 flags for memory model flags, like Hardware Lock Elision.
7322 @item __ATOMIC_RELAXED
7323 No barriers or synchronization.
7324 @item __ATOMIC_CONSUME
7325 Data dependency only for both barrier and synchronization with another
7327 @item __ATOMIC_ACQUIRE
7328 Barrier to hoisting of code and synchronizes with release (or stronger)
7329 semantic stores from another thread.
7330 @item __ATOMIC_RELEASE
7331 Barrier to sinking of code and synchronizes with acquire (or stronger)
7332 semantic loads from another thread.
7333 @item __ATOMIC_ACQ_REL
7334 Full barrier in both directions and synchronizes with acquire loads and
7335 release stores in another thread.
7336 @item __ATOMIC_SEQ_CST
7337 Full barrier in both directions and synchronizes with acquire loads and
7338 release stores in all threads.
7341 When implementing patterns for these built-in functions, the memory model
7342 parameter can be ignored as long as the pattern implements the most
7343 restrictive @code{__ATOMIC_SEQ_CST} model. Any of the other memory models
7344 execute correctly with this memory model but they may not execute as
7345 efficiently as they could with a more appropriate implementation of the
7346 relaxed requirements.
7348 Note that the C++11 standard allows for the memory model parameter to be
7349 determined at run time rather than at compile time. These built-in
7350 functions map any run-time value to @code{__ATOMIC_SEQ_CST} rather
7351 than invoke a runtime library call or inline a switch statement. This is
7352 standard compliant, safe, and the simplest approach for now.
7354 The memory model parameter is a signed int, but only the lower 8 bits are
7355 reserved for the memory model. The remainder of the signed int is reserved
7356 for future use and should be 0. Use of the predefined atomic values
7357 ensures proper usage.
7359 @deftypefn {Built-in Function} @var{type} __atomic_load_n (@var{type} *ptr, int memmodel)
7360 This built-in function implements an atomic load operation. It returns the
7361 contents of @code{*@var{ptr}}.
7363 The valid memory model variants are
7364 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
7365 and @code{__ATOMIC_CONSUME}.
7369 @deftypefn {Built-in Function} void __atomic_load (@var{type} *ptr, @var{type} *ret, int memmodel)
7370 This is the generic version of an atomic load. It returns the
7371 contents of @code{*@var{ptr}} in @code{*@var{ret}}.
7375 @deftypefn {Built-in Function} void __atomic_store_n (@var{type} *ptr, @var{type} val, int memmodel)
7376 This built-in function implements an atomic store operation. It writes
7377 @code{@var{val}} into @code{*@var{ptr}}.
7379 The valid memory model variants are
7380 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and @code{__ATOMIC_RELEASE}.
7384 @deftypefn {Built-in Function} void __atomic_store (@var{type} *ptr, @var{type} *val, int memmodel)
7385 This is the generic version of an atomic store. It stores the value
7386 of @code{*@var{val}} into @code{*@var{ptr}}.
7390 @deftypefn {Built-in Function} @var{type} __atomic_exchange_n (@var{type} *ptr, @var{type} val, int memmodel)
7391 This built-in function implements an atomic exchange operation. It writes
7392 @var{val} into @code{*@var{ptr}}, and returns the previous contents of
7395 The valid memory model variants are
7396 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE},
7397 @code{__ATOMIC_RELEASE}, and @code{__ATOMIC_ACQ_REL}.
7401 @deftypefn {Built-in Function} void __atomic_exchange (@var{type} *ptr, @var{type} *val, @var{type} *ret, int memmodel)
7402 This is the generic version of an atomic exchange. It stores the
7403 contents of @code{*@var{val}} into @code{*@var{ptr}}. The original value
7404 of @code{*@var{ptr}} is copied into @code{*@var{ret}}.
7408 @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)
7409 This built-in function implements an atomic compare and exchange operation.
7410 This compares the contents of @code{*@var{ptr}} with the contents of
7411 @code{*@var{expected}} and if equal, writes @var{desired} into
7412 @code{*@var{ptr}}. If they are not equal, the current contents of
7413 @code{*@var{ptr}} is written into @code{*@var{expected}}. @var{weak} is true
7414 for weak compare_exchange, and false for the strong variation. Many targets
7415 only offer the strong variation and ignore the parameter. When in doubt, use
7416 the strong variation.
7418 True is returned if @var{desired} is written into
7419 @code{*@var{ptr}} and the execution is considered to conform to the
7420 memory model specified by @var{success_memmodel}. There are no
7421 restrictions on what memory model can be used here.
7423 False is returned otherwise, and the execution is considered to conform
7424 to @var{failure_memmodel}. This memory model cannot be
7425 @code{__ATOMIC_RELEASE} nor @code{__ATOMIC_ACQ_REL}. It also cannot be a
7426 stronger model than that specified by @var{success_memmodel}.
7430 @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)
7431 This built-in function implements the generic version of
7432 @code{__atomic_compare_exchange}. The function is virtually identical to
7433 @code{__atomic_compare_exchange_n}, except the desired value is also a
7438 @deftypefn {Built-in Function} @var{type} __atomic_add_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7439 @deftypefnx {Built-in Function} @var{type} __atomic_sub_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7440 @deftypefnx {Built-in Function} @var{type} __atomic_and_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7441 @deftypefnx {Built-in Function} @var{type} __atomic_xor_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7442 @deftypefnx {Built-in Function} @var{type} __atomic_or_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7443 @deftypefnx {Built-in Function} @var{type} __atomic_nand_fetch (@var{type} *ptr, @var{type} val, int memmodel)
7444 These built-in functions perform the operation suggested by the name, and
7445 return the result of the operation. That is,
7448 @{ *ptr @var{op}= val; return *ptr; @}
7451 All memory models are valid.
7455 @deftypefn {Built-in Function} @var{type} __atomic_fetch_add (@var{type} *ptr, @var{type} val, int memmodel)
7456 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_sub (@var{type} *ptr, @var{type} val, int memmodel)
7457 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_and (@var{type} *ptr, @var{type} val, int memmodel)
7458 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_xor (@var{type} *ptr, @var{type} val, int memmodel)
7459 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_or (@var{type} *ptr, @var{type} val, int memmodel)
7460 @deftypefnx {Built-in Function} @var{type} __atomic_fetch_nand (@var{type} *ptr, @var{type} val, int memmodel)
7461 These built-in functions perform the operation suggested by the name, and
7462 return the value that had previously been in @code{*@var{ptr}}. That is,
7465 @{ tmp = *ptr; *ptr @var{op}= val; return tmp; @}
7468 All memory models are valid.
7472 @deftypefn {Built-in Function} bool __atomic_test_and_set (void *ptr, int memmodel)
7474 This built-in function performs an atomic test-and-set operation on
7475 the byte at @code{*@var{ptr}}. The byte is set to some implementation
7476 defined nonzero ``set'' value and the return value is @code{true} if and only
7477 if the previous contents were ``set''.
7478 It should be only used for operands of type @code{bool} or @code{char}. For
7479 other types only part of the value may be set.
7481 All memory models are valid.
7485 @deftypefn {Built-in Function} void __atomic_clear (bool *ptr, int memmodel)
7487 This built-in function performs an atomic clear operation on
7488 @code{*@var{ptr}}. After the operation, @code{*@var{ptr}} contains 0.
7489 It should be only used for operands of type @code{bool} or @code{char} and
7490 in conjunction with @code{__atomic_test_and_set}.
7491 For other types it may only clear partially. If the type is not @code{bool}
7492 prefer using @code{__atomic_store}.
7494 The valid memory model variants are
7495 @code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and
7496 @code{__ATOMIC_RELEASE}.
7500 @deftypefn {Built-in Function} void __atomic_thread_fence (int memmodel)
7502 This built-in function acts as a synchronization fence between threads
7503 based on the specified memory model.
7505 All memory orders are valid.
7509 @deftypefn {Built-in Function} void __atomic_signal_fence (int memmodel)
7511 This built-in function acts as a synchronization fence between a thread
7512 and signal handlers based in the same thread.
7514 All memory orders are valid.
7518 @deftypefn {Built-in Function} bool __atomic_always_lock_free (size_t size, void *ptr)
7520 This built-in function returns true if objects of @var{size} bytes always
7521 generate lock free atomic instructions for the target architecture.
7522 @var{size} must resolve to a compile-time constant and the result also
7523 resolves to a compile-time constant.
7525 @var{ptr} is an optional pointer to the object that may be used to determine
7526 alignment. A value of 0 indicates typical alignment should be used. The
7527 compiler may also ignore this parameter.
7530 if (_atomic_always_lock_free (sizeof (long long), 0))
7535 @deftypefn {Built-in Function} bool __atomic_is_lock_free (size_t size, void *ptr)
7537 This built-in function returns true if objects of @var{size} bytes always
7538 generate lock free atomic instructions for the target architecture. If
7539 it is not known to be lock free a call is made to a runtime routine named
7540 @code{__atomic_is_lock_free}.
7542 @var{ptr} is an optional pointer to the object that may be used to determine
7543 alignment. A value of 0 indicates typical alignment should be used. The
7544 compiler may also ignore this parameter.
7547 @node x86 specific memory model extensions for transactional memory
7548 @section x86 specific memory model extensions for transactional memory
7550 The i386 architecture supports additional memory ordering flags
7551 to mark lock critical sections for hardware lock elision.
7552 These must be specified in addition to an existing memory model to
7556 @item __ATOMIC_HLE_ACQUIRE
7557 Start lock elision on a lock variable.
7558 Memory model must be @code{__ATOMIC_ACQUIRE} or stronger.
7559 @item __ATOMIC_HLE_RELEASE
7560 End lock elision on a lock variable.
7561 Memory model must be @code{__ATOMIC_RELEASE} or stronger.
7564 When a lock acquire fails it is required for good performance to abort
7565 the transaction quickly. This can be done with a @code{_mm_pause}
7568 #include <immintrin.h> // For _mm_pause
7572 /* Acquire lock with lock elision */
7573 while (__atomic_exchange_n(&lockvar, 1, __ATOMIC_ACQUIRE|__ATOMIC_HLE_ACQUIRE))
7574 _mm_pause(); /* Abort failed transaction */
7576 /* Free lock with lock elision */
7577 __atomic_store_n(&lockvar, 0, __ATOMIC_RELEASE|__ATOMIC_HLE_RELEASE);
7580 @node Object Size Checking
7581 @section Object Size Checking Built-in Functions
7582 @findex __builtin_object_size
7583 @findex __builtin___memcpy_chk
7584 @findex __builtin___mempcpy_chk
7585 @findex __builtin___memmove_chk
7586 @findex __builtin___memset_chk
7587 @findex __builtin___strcpy_chk
7588 @findex __builtin___stpcpy_chk
7589 @findex __builtin___strncpy_chk
7590 @findex __builtin___strcat_chk
7591 @findex __builtin___strncat_chk
7592 @findex __builtin___sprintf_chk
7593 @findex __builtin___snprintf_chk
7594 @findex __builtin___vsprintf_chk
7595 @findex __builtin___vsnprintf_chk
7596 @findex __builtin___printf_chk
7597 @findex __builtin___vprintf_chk
7598 @findex __builtin___fprintf_chk
7599 @findex __builtin___vfprintf_chk
7601 GCC implements a limited buffer overflow protection mechanism
7602 that can prevent some buffer overflow attacks.
7604 @deftypefn {Built-in Function} {size_t} __builtin_object_size (void * @var{ptr}, int @var{type})
7605 is a built-in construct that returns a constant number of bytes from
7606 @var{ptr} to the end of the object @var{ptr} pointer points to
7607 (if known at compile time). @code{__builtin_object_size} never evaluates
7608 its arguments for side-effects. If there are any side-effects in them, it
7609 returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
7610 for @var{type} 2 or 3. If there are multiple objects @var{ptr} can
7611 point to and all of them are known at compile time, the returned number
7612 is the maximum of remaining byte counts in those objects if @var{type} & 2 is
7613 0 and minimum if nonzero. If it is not possible to determine which objects
7614 @var{ptr} points to at compile time, @code{__builtin_object_size} should
7615 return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0}
7616 for @var{type} 2 or 3.
7618 @var{type} is an integer constant from 0 to 3. If the least significant
7619 bit is clear, objects are whole variables, if it is set, a closest
7620 surrounding subobject is considered the object a pointer points to.
7621 The second bit determines if maximum or minimum of remaining bytes
7625 struct V @{ char buf1[10]; int b; char buf2[10]; @} var;
7626 char *p = &var.buf1[1], *q = &var.b;
7628 /* Here the object p points to is var. */
7629 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
7630 /* The subobject p points to is var.buf1. */
7631 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
7632 /* The object q points to is var. */
7633 assert (__builtin_object_size (q, 0)
7634 == (char *) (&var + 1) - (char *) &var.b);
7635 /* The subobject q points to is var.b. */
7636 assert (__builtin_object_size (q, 1) == sizeof (var.b));
7640 There are built-in functions added for many common string operation
7641 functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk}
7642 built-in is provided. This built-in has an additional last argument,
7643 which is the number of bytes remaining in object the @var{dest}
7644 argument points to or @code{(size_t) -1} if the size is not known.
7646 The built-in functions are optimized into the normal string functions
7647 like @code{memcpy} if the last argument is @code{(size_t) -1} or if
7648 it is known at compile time that the destination object will not
7649 be overflown. If the compiler can determine at compile time the
7650 object will be always overflown, it issues a warning.
7652 The intended use can be e.g.@:
7656 #define bos0(dest) __builtin_object_size (dest, 0)
7657 #define memcpy(dest, src, n) \
7658 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
7662 /* It is unknown what object p points to, so this is optimized
7663 into plain memcpy - no checking is possible. */
7664 memcpy (p, "abcde", n);
7665 /* Destination is known and length too. It is known at compile
7666 time there will be no overflow. */
7667 memcpy (&buf[5], "abcde", 5);
7668 /* Destination is known, but the length is not known at compile time.
7669 This will result in __memcpy_chk call that can check for overflow
7671 memcpy (&buf[5], "abcde", n);
7672 /* Destination is known and it is known at compile time there will
7673 be overflow. There will be a warning and __memcpy_chk call that
7674 will abort the program at run time. */
7675 memcpy (&buf[6], "abcde", 5);
7678 Such built-in functions are provided for @code{memcpy}, @code{mempcpy},
7679 @code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy},
7680 @code{strcat} and @code{strncat}.
7682 There are also checking built-in functions for formatted output functions.
7684 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
7685 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
7686 const char *fmt, ...);
7687 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
7689 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
7690 const char *fmt, va_list ap);
7693 The added @var{flag} argument is passed unchanged to @code{__sprintf_chk}
7694 etc.@: functions and can contain implementation specific flags on what
7695 additional security measures the checking function might take, such as
7696 handling @code{%n} differently.
7698 The @var{os} argument is the object size @var{s} points to, like in the
7699 other built-in functions. There is a small difference in the behavior
7700 though, if @var{os} is @code{(size_t) -1}, the built-in functions are
7701 optimized into the non-checking functions only if @var{flag} is 0, otherwise
7702 the checking function is called with @var{os} argument set to
7705 In addition to this, there are checking built-in functions
7706 @code{__builtin___printf_chk}, @code{__builtin___vprintf_chk},
7707 @code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}.
7708 These have just one additional argument, @var{flag}, right before
7709 format string @var{fmt}. If the compiler is able to optimize them to
7710 @code{fputc} etc.@: functions, it does, otherwise the checking function
7711 is called and the @var{flag} argument passed to it.
7713 @node Other Builtins
7714 @section Other Built-in Functions Provided by GCC
7715 @cindex built-in functions
7716 @findex __builtin_fpclassify
7717 @findex __builtin_isfinite
7718 @findex __builtin_isnormal
7719 @findex __builtin_isgreater
7720 @findex __builtin_isgreaterequal
7721 @findex __builtin_isinf_sign
7722 @findex __builtin_isless
7723 @findex __builtin_islessequal
7724 @findex __builtin_islessgreater
7725 @findex __builtin_isunordered
7726 @findex __builtin_powi
7727 @findex __builtin_powif
7728 @findex __builtin_powil
7886 @findex fprintf_unlocked
7888 @findex fputs_unlocked
8005 @findex printf_unlocked
8037 @findex significandf
8038 @findex significandl
8109 GCC provides a large number of built-in functions other than the ones
8110 mentioned above. Some of these are for internal use in the processing
8111 of exceptions or variable-length argument lists and are not
8112 documented here because they may change from time to time; we do not
8113 recommend general use of these functions.
8115 The remaining functions are provided for optimization purposes.
8117 @opindex fno-builtin
8118 GCC includes built-in versions of many of the functions in the standard
8119 C library. The versions prefixed with @code{__builtin_} are always
8120 treated as having the same meaning as the C library function even if you
8121 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
8122 Many of these functions are only optimized in certain cases; if they are
8123 not optimized in a particular case, a call to the library function is
8128 Outside strict ISO C mode (@option{-ansi}, @option{-std=c90},
8129 @option{-std=c99} or @option{-std=c11}), the functions
8130 @code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero},
8131 @code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml},
8132 @code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll},
8133 @code{ffsl}, @code{ffs}, @code{fprintf_unlocked},
8134 @code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma},
8135 @code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext},
8136 @code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0},
8137 @code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn},
8138 @code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy},
8139 @code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked},
8140 @code{rindex}, @code{scalbf}, @code{scalbl}, @code{scalb},
8141 @code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32},
8142 @code{signbitd64}, @code{signbitd128}, @code{significandf},
8143 @code{significandl}, @code{significand}, @code{sincosf},
8144 @code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy},
8145 @code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp},
8146 @code{strndup}, @code{toascii}, @code{y0f}, @code{y0l}, @code{y0},
8147 @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and
8149 may be handled as built-in functions.
8150 All these functions have corresponding versions
8151 prefixed with @code{__builtin_}, which may be used even in strict C90
8154 The ISO C99 functions
8155 @code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf},
8156 @code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh},
8157 @code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf},
8158 @code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos},
8159 @code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf},
8160 @code{casinhl}, @code{casinh}, @code{casinl}, @code{casin},
8161 @code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh},
8162 @code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt},
8163 @code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl},
8164 @code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf},
8165 @code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog},
8166 @code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl},
8167 @code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf},
8168 @code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal},
8169 @code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl},
8170 @code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf},
8171 @code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan},
8172 @code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl},
8173 @code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f},
8174 @code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim},
8175 @code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax},
8176 @code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf},
8177 @code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb},
8178 @code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf},
8179 @code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl},
8180 @code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround},
8181 @code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l},
8182 @code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf},
8183 @code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl},
8184 @code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint},
8185 @code{nextafterf}, @code{nextafterl}, @code{nextafter},
8186 @code{nexttowardf}, @code{nexttowardl}, @code{nexttoward},
8187 @code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof},
8188 @code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint},
8189 @code{roundf}, @code{roundl}, @code{round}, @code{scalblnf},
8190 @code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl},
8191 @code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal},
8192 @code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc},
8193 @code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf}
8194 are handled as built-in functions
8195 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
8197 There are also built-in versions of the ISO C99 functions
8198 @code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f},
8199 @code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill},
8200 @code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf},
8201 @code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl},
8202 @code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf},
8203 @code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl},
8204 @code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf},
8205 @code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl},
8206 @code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl}
8207 that are recognized in any mode since ISO C90 reserves these names for
8208 the purpose to which ISO C99 puts them. All these functions have
8209 corresponding versions prefixed with @code{__builtin_}.
8211 The ISO C94 functions
8212 @code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit},
8213 @code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct},
8214 @code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and
8216 are handled as built-in functions
8217 except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}).
8219 The ISO C90 functions
8220 @code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2},
8221 @code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos},
8222 @code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod},
8223 @code{fprintf}, @code{fputs}, @code{frexp}, @code{fscanf},
8224 @code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit},
8225 @code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct},
8226 @code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower},
8227 @code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log},
8228 @code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy},
8229 @code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar},
8230 @code{puts}, @code{scanf}, @code{sinh}, @code{sin}, @code{snprintf},
8231 @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat},
8232 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
8233 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
8234 @code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr},
8235 @code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf}
8236 are all recognized as built-in functions unless
8237 @option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}}
8238 is specified for an individual function). All of these functions have
8239 corresponding versions prefixed with @code{__builtin_}.
8241 GCC provides built-in versions of the ISO C99 floating-point comparison
8242 macros that avoid raising exceptions for unordered operands. They have
8243 the same names as the standard macros ( @code{isgreater},
8244 @code{isgreaterequal}, @code{isless}, @code{islessequal},
8245 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
8246 prefixed. We intend for a library implementor to be able to simply
8247 @code{#define} each standard macro to its built-in equivalent.
8248 In the same fashion, GCC provides @code{fpclassify}, @code{isfinite},
8249 @code{isinf_sign} and @code{isnormal} built-ins used with
8250 @code{__builtin_} prefixed. The @code{isinf} and @code{isnan}
8251 built-in functions appear both with and without the @code{__builtin_} prefix.
8253 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
8255 You can use the built-in function @code{__builtin_types_compatible_p} to
8256 determine whether two types are the same.
8258 This built-in function returns 1 if the unqualified versions of the
8259 types @var{type1} and @var{type2} (which are types, not expressions) are
8260 compatible, 0 otherwise. The result of this built-in function can be
8261 used in integer constant expressions.
8263 This built-in function ignores top level qualifiers (e.g., @code{const},
8264 @code{volatile}). For example, @code{int} is equivalent to @code{const
8267 The type @code{int[]} and @code{int[5]} are compatible. On the other
8268 hand, @code{int} and @code{char *} are not compatible, even if the size
8269 of their types, on the particular architecture are the same. Also, the
8270 amount of pointer indirection is taken into account when determining
8271 similarity. Consequently, @code{short *} is not similar to
8272 @code{short **}. Furthermore, two types that are typedefed are
8273 considered compatible if their underlying types are compatible.
8275 An @code{enum} type is not considered to be compatible with another
8276 @code{enum} type even if both are compatible with the same integer
8277 type; this is what the C standard specifies.
8278 For example, @code{enum @{foo, bar@}} is not similar to
8279 @code{enum @{hot, dog@}}.
8281 You typically use this function in code whose execution varies
8282 depending on the arguments' types. For example:
8287 typeof (x) tmp = (x); \
8288 if (__builtin_types_compatible_p (typeof (x), long double)) \
8289 tmp = foo_long_double (tmp); \
8290 else if (__builtin_types_compatible_p (typeof (x), double)) \
8291 tmp = foo_double (tmp); \
8292 else if (__builtin_types_compatible_p (typeof (x), float)) \
8293 tmp = foo_float (tmp); \
8300 @emph{Note:} This construct is only available for C@.
8304 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
8306 You can use the built-in function @code{__builtin_choose_expr} to
8307 evaluate code depending on the value of a constant expression. This
8308 built-in function returns @var{exp1} if @var{const_exp}, which is an
8309 integer constant expression, is nonzero. Otherwise it returns @var{exp2}.
8311 This built-in function is analogous to the @samp{? :} operator in C,
8312 except that the expression returned has its type unaltered by promotion
8313 rules. Also, the built-in function does not evaluate the expression
8314 that is not chosen. For example, if @var{const_exp} evaluates to true,
8315 @var{exp2} is not evaluated even if it has side-effects.
8317 This built-in function can return an lvalue if the chosen argument is an
8320 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
8321 type. Similarly, if @var{exp2} is returned, its return type is the same
8328 __builtin_choose_expr ( \
8329 __builtin_types_compatible_p (typeof (x), double), \
8331 __builtin_choose_expr ( \
8332 __builtin_types_compatible_p (typeof (x), float), \
8334 /* @r{The void expression results in a compile-time error} \
8335 @r{when assigning the result to something.} */ \
8339 @emph{Note:} This construct is only available for C@. Furthermore, the
8340 unused expression (@var{exp1} or @var{exp2} depending on the value of
8341 @var{const_exp}) may still generate syntax errors. This may change in
8346 @deftypefn {Built-in Function} @var{type} __builtin_complex (@var{real}, @var{imag})
8348 The built-in function @code{__builtin_complex} is provided for use in
8349 implementing the ISO C11 macros @code{CMPLXF}, @code{CMPLX} and
8350 @code{CMPLXL}. @var{real} and @var{imag} must have the same type, a
8351 real binary floating-point type, and the result has the corresponding
8352 complex type with real and imaginary parts @var{real} and @var{imag}.
8353 Unlike @samp{@var{real} + I * @var{imag}}, this works even when
8354 infinities, NaNs and negative zeros are involved.
8358 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
8359 You can use the built-in function @code{__builtin_constant_p} to
8360 determine if a value is known to be constant at compile time and hence
8361 that GCC can perform constant-folding on expressions involving that
8362 value. The argument of the function is the value to test. The function
8363 returns the integer 1 if the argument is known to be a compile-time
8364 constant and 0 if it is not known to be a compile-time constant. A
8365 return of 0 does not indicate that the value is @emph{not} a constant,
8366 but merely that GCC cannot prove it is a constant with the specified
8367 value of the @option{-O} option.
8369 You typically use this function in an embedded application where
8370 memory is a critical resource. If you have some complex calculation,
8371 you may want it to be folded if it involves constants, but need to call
8372 a function if it does not. For example:
8375 #define Scale_Value(X) \
8376 (__builtin_constant_p (X) \
8377 ? ((X) * SCALE + OFFSET) : Scale (X))
8380 You may use this built-in function in either a macro or an inline
8381 function. However, if you use it in an inlined function and pass an
8382 argument of the function as the argument to the built-in, GCC
8383 never returns 1 when you call the inline function with a string constant
8384 or compound literal (@pxref{Compound Literals}) and does not return 1
8385 when you pass a constant numeric value to the inline function unless you
8386 specify the @option{-O} option.
8388 You may also use @code{__builtin_constant_p} in initializers for static
8389 data. For instance, you can write
8392 static const int table[] = @{
8393 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
8399 This is an acceptable initializer even if @var{EXPRESSION} is not a
8400 constant expression, including the case where
8401 @code{__builtin_constant_p} returns 1 because @var{EXPRESSION} can be
8402 folded to a constant but @var{EXPRESSION} contains operands that are
8403 not otherwise permitted in a static initializer (for example,
8404 @code{0 && foo ()}). GCC must be more conservative about evaluating the
8405 built-in in this case, because it has no opportunity to perform
8408 Previous versions of GCC did not accept this built-in in data
8409 initializers. The earliest version where it is completely safe is
8413 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
8414 @opindex fprofile-arcs
8415 You may use @code{__builtin_expect} to provide the compiler with
8416 branch prediction information. In general, you should prefer to
8417 use actual profile feedback for this (@option{-fprofile-arcs}), as
8418 programmers are notoriously bad at predicting how their programs
8419 actually perform. However, there are applications in which this
8420 data is hard to collect.
8422 The return value is the value of @var{exp}, which should be an integral
8423 expression. The semantics of the built-in are that it is expected that
8424 @var{exp} == @var{c}. For example:
8427 if (__builtin_expect (x, 0))
8432 indicates that we do not expect to call @code{foo}, since
8433 we expect @code{x} to be zero. Since you are limited to integral
8434 expressions for @var{exp}, you should use constructions such as
8437 if (__builtin_expect (ptr != NULL, 1))
8442 when testing pointer or floating-point values.
8445 @deftypefn {Built-in Function} void __builtin_trap (void)
8446 This function causes the program to exit abnormally. GCC implements
8447 this function by using a target-dependent mechanism (such as
8448 intentionally executing an illegal instruction) or by calling
8449 @code{abort}. The mechanism used may vary from release to release so
8450 you should not rely on any particular implementation.
8453 @deftypefn {Built-in Function} void __builtin_unreachable (void)
8454 If control flow reaches the point of the @code{__builtin_unreachable},
8455 the program is undefined. It is useful in situations where the
8456 compiler cannot deduce the unreachability of the code.
8458 One such case is immediately following an @code{asm} statement that
8459 either never terminates, or one that transfers control elsewhere
8460 and never returns. In this example, without the
8461 @code{__builtin_unreachable}, GCC issues a warning that control
8462 reaches the end of a non-void function. It also generates code
8463 to return after the @code{asm}.
8466 int f (int c, int v)
8474 asm("jmp error_handler");
8475 __builtin_unreachable ();
8481 Because the @code{asm} statement unconditionally transfers control out
8482 of the function, control never reaches the end of the function
8483 body. The @code{__builtin_unreachable} is in fact unreachable and
8484 communicates this fact to the compiler.
8486 Another use for @code{__builtin_unreachable} is following a call a
8487 function that never returns but that is not declared
8488 @code{__attribute__((noreturn))}, as in this example:
8491 void function_that_never_returns (void);
8501 function_that_never_returns ();
8502 __builtin_unreachable ();
8509 @deftypefn {Built-in Function} void *__builtin_assume_aligned (const void *@var{exp}, size_t @var{align}, ...)
8510 This function returns its first argument, and allows the compiler
8511 to assume that the returned pointer is at least @var{align} bytes
8512 aligned. This built-in can have either two or three arguments,
8513 if it has three, the third argument should have integer type, and
8514 if it is nonzero means misalignment offset. For example:
8517 void *x = __builtin_assume_aligned (arg, 16);
8521 means that the compiler can assume @code{x}, set to @code{arg}, is at least
8522 16-byte aligned, while:
8525 void *x = __builtin_assume_aligned (arg, 32, 8);
8529 means that the compiler can assume for @code{x}, set to @code{arg}, that
8530 @code{(char *) x - 8} is 32-byte aligned.
8533 @deftypefn {Built-in Function} int __builtin_LINE ()
8534 This function is the equivalent to the preprocessor @code{__LINE__}
8535 macro and returns the line number of the invocation of the built-in.
8538 @deftypefn {Built-in Function} int __builtin_FUNCTION ()
8539 This function is the equivalent to the preprocessor @code{__FUNCTION__}
8540 macro and returns the function name the invocation of the built-in is in.
8543 @deftypefn {Built-in Function} int __builtin_FILE ()
8544 This function is the equivalent to the preprocessor @code{__FILE__}
8545 macro and returns the file name the invocation of the built-in is in.
8548 @deftypefn {Built-in Function} void __builtin___clear_cache (char *@var{begin}, char *@var{end})
8549 This function is used to flush the processor's instruction cache for
8550 the region of memory between @var{begin} inclusive and @var{end}
8551 exclusive. Some targets require that the instruction cache be
8552 flushed, after modifying memory containing code, in order to obtain
8553 deterministic behavior.
8555 If the target does not require instruction cache flushes,
8556 @code{__builtin___clear_cache} has no effect. Otherwise either
8557 instructions are emitted in-line to clear the instruction cache or a
8558 call to the @code{__clear_cache} function in libgcc is made.
8561 @deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...)
8562 This function is used to minimize cache-miss latency by moving data into
8563 a cache before it is accessed.
8564 You can insert calls to @code{__builtin_prefetch} into code for which
8565 you know addresses of data in memory that is likely to be accessed soon.
8566 If the target supports them, data prefetch instructions are generated.
8567 If the prefetch is done early enough before the access then the data will
8568 be in the cache by the time it is accessed.
8570 The value of @var{addr} is the address of the memory to prefetch.
8571 There are two optional arguments, @var{rw} and @var{locality}.
8572 The value of @var{rw} is a compile-time constant one or zero; one
8573 means that the prefetch is preparing for a write to the memory address
8574 and zero, the default, means that the prefetch is preparing for a read.
8575 The value @var{locality} must be a compile-time constant integer between
8576 zero and three. A value of zero means that the data has no temporal
8577 locality, so it need not be left in the cache after the access. A value
8578 of three means that the data has a high degree of temporal locality and
8579 should be left in all levels of cache possible. Values of one and two
8580 mean, respectively, a low or moderate degree of temporal locality. The
8584 for (i = 0; i < n; i++)
8587 __builtin_prefetch (&a[i+j], 1, 1);
8588 __builtin_prefetch (&b[i+j], 0, 1);
8593 Data prefetch does not generate faults if @var{addr} is invalid, but
8594 the address expression itself must be valid. For example, a prefetch
8595 of @code{p->next} does not fault if @code{p->next} is not a valid
8596 address, but evaluation faults if @code{p} is not a valid address.
8598 If the target does not support data prefetch, the address expression
8599 is evaluated if it includes side effects but no other code is generated
8600 and GCC does not issue a warning.
8603 @deftypefn {Built-in Function} double __builtin_huge_val (void)
8604 Returns a positive infinity, if supported by the floating-point format,
8605 else @code{DBL_MAX}. This function is suitable for implementing the
8606 ISO C macro @code{HUGE_VAL}.
8609 @deftypefn {Built-in Function} float __builtin_huge_valf (void)
8610 Similar to @code{__builtin_huge_val}, except the return type is @code{float}.
8613 @deftypefn {Built-in Function} {long double} __builtin_huge_vall (void)
8614 Similar to @code{__builtin_huge_val}, except the return
8615 type is @code{long double}.
8618 @deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...)
8619 This built-in implements the C99 fpclassify functionality. The first
8620 five int arguments should be the target library's notion of the
8621 possible FP classes and are used for return values. They must be
8622 constant values and they must appear in this order: @code{FP_NAN},
8623 @code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and
8624 @code{FP_ZERO}. The ellipsis is for exactly one floating-point value
8625 to classify. GCC treats the last argument as type-generic, which
8626 means it does not do default promotion from float to double.
8629 @deftypefn {Built-in Function} double __builtin_inf (void)
8630 Similar to @code{__builtin_huge_val}, except a warning is generated
8631 if the target floating-point format does not support infinities.
8634 @deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void)
8635 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}.
8638 @deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void)
8639 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}.
8642 @deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void)
8643 Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}.
8646 @deftypefn {Built-in Function} float __builtin_inff (void)
8647 Similar to @code{__builtin_inf}, except the return type is @code{float}.
8648 This function is suitable for implementing the ISO C99 macro @code{INFINITY}.
8651 @deftypefn {Built-in Function} {long double} __builtin_infl (void)
8652 Similar to @code{__builtin_inf}, except the return
8653 type is @code{long double}.
8656 @deftypefn {Built-in Function} int __builtin_isinf_sign (...)
8657 Similar to @code{isinf}, except the return value is -1 for
8658 an argument of @code{-Inf} and 1 for an argument of @code{+Inf}.
8659 Note while the parameter list is an
8660 ellipsis, this function only accepts exactly one floating-point
8661 argument. GCC treats this parameter as type-generic, which means it
8662 does not do default promotion from float to double.
8665 @deftypefn {Built-in Function} double __builtin_nan (const char *str)
8666 This is an implementation of the ISO C99 function @code{nan}.
8668 Since ISO C99 defines this function in terms of @code{strtod}, which we
8669 do not implement, a description of the parsing is in order. The string
8670 is parsed as by @code{strtol}; that is, the base is recognized by
8671 leading @samp{0} or @samp{0x} prefixes. The number parsed is placed
8672 in the significand such that the least significant bit of the number
8673 is at the least significant bit of the significand. The number is
8674 truncated to fit the significand field provided. The significand is
8675 forced to be a quiet NaN@.
8677 This function, if given a string literal all of which would have been
8678 consumed by @code{strtol}, is evaluated early enough that it is considered a
8679 compile-time constant.
8682 @deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str)
8683 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}.
8686 @deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str)
8687 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}.
8690 @deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str)
8691 Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}.
8694 @deftypefn {Built-in Function} float __builtin_nanf (const char *str)
8695 Similar to @code{__builtin_nan}, except the return type is @code{float}.
8698 @deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str)
8699 Similar to @code{__builtin_nan}, except the return type is @code{long double}.
8702 @deftypefn {Built-in Function} double __builtin_nans (const char *str)
8703 Similar to @code{__builtin_nan}, except the significand is forced
8704 to be a signaling NaN@. The @code{nans} function is proposed by
8705 @uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}.
8708 @deftypefn {Built-in Function} float __builtin_nansf (const char *str)
8709 Similar to @code{__builtin_nans}, except the return type is @code{float}.
8712 @deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str)
8713 Similar to @code{__builtin_nans}, except the return type is @code{long double}.
8716 @deftypefn {Built-in Function} int __builtin_ffs (unsigned int x)
8717 Returns one plus the index of the least significant 1-bit of @var{x}, or
8718 if @var{x} is zero, returns zero.
8721 @deftypefn {Built-in Function} int __builtin_clz (unsigned int x)
8722 Returns the number of leading 0-bits in @var{x}, starting at the most
8723 significant bit position. If @var{x} is 0, the result is undefined.
8726 @deftypefn {Built-in Function} int __builtin_ctz (unsigned int x)
8727 Returns the number of trailing 0-bits in @var{x}, starting at the least
8728 significant bit position. If @var{x} is 0, the result is undefined.
8731 @deftypefn {Built-in Function} int __builtin_clrsb (int x)
8732 Returns the number of leading redundant sign bits in @var{x}, i.e.@: the
8733 number of bits following the most significant bit that are identical
8734 to it. There are no special cases for 0 or other values.
8737 @deftypefn {Built-in Function} int __builtin_popcount (unsigned int x)
8738 Returns the number of 1-bits in @var{x}.
8741 @deftypefn {Built-in Function} int __builtin_parity (unsigned int x)
8742 Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x}
8746 @deftypefn {Built-in Function} int __builtin_ffsl (unsigned long)
8747 Similar to @code{__builtin_ffs}, except the argument type is
8748 @code{unsigned long}.
8751 @deftypefn {Built-in Function} int __builtin_clzl (unsigned long)
8752 Similar to @code{__builtin_clz}, except the argument type is
8753 @code{unsigned long}.
8756 @deftypefn {Built-in Function} int __builtin_ctzl (unsigned long)
8757 Similar to @code{__builtin_ctz}, except the argument type is
8758 @code{unsigned long}.
8761 @deftypefn {Built-in Function} int __builtin_clrsbl (long)
8762 Similar to @code{__builtin_clrsb}, except the argument type is
8766 @deftypefn {Built-in Function} int __builtin_popcountl (unsigned long)
8767 Similar to @code{__builtin_popcount}, except the argument type is
8768 @code{unsigned long}.
8771 @deftypefn {Built-in Function} int __builtin_parityl (unsigned long)
8772 Similar to @code{__builtin_parity}, except the argument type is
8773 @code{unsigned long}.
8776 @deftypefn {Built-in Function} int __builtin_ffsll (unsigned long long)
8777 Similar to @code{__builtin_ffs}, except the argument type is
8778 @code{unsigned long long}.
8781 @deftypefn {Built-in Function} int __builtin_clzll (unsigned long long)
8782 Similar to @code{__builtin_clz}, except the argument type is
8783 @code{unsigned long long}.
8786 @deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long)
8787 Similar to @code{__builtin_ctz}, except the argument type is
8788 @code{unsigned long long}.
8791 @deftypefn {Built-in Function} int __builtin_clrsbll (long long)
8792 Similar to @code{__builtin_clrsb}, except the argument type is
8796 @deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long)
8797 Similar to @code{__builtin_popcount}, except the argument type is
8798 @code{unsigned long long}.
8801 @deftypefn {Built-in Function} int __builtin_parityll (unsigned long long)
8802 Similar to @code{__builtin_parity}, except the argument type is
8803 @code{unsigned long long}.
8806 @deftypefn {Built-in Function} double __builtin_powi (double, int)
8807 Returns the first argument raised to the power of the second. Unlike the
8808 @code{pow} function no guarantees about precision and rounding are made.
8811 @deftypefn {Built-in Function} float __builtin_powif (float, int)
8812 Similar to @code{__builtin_powi}, except the argument and return types
8816 @deftypefn {Built-in Function} {long double} __builtin_powil (long double, int)
8817 Similar to @code{__builtin_powi}, except the argument and return types
8818 are @code{long double}.
8821 @deftypefn {Built-in Function} uint16_t __builtin_bswap16 (uint16_t x)
8822 Returns @var{x} with the order of the bytes reversed; for example,
8823 @code{0xaabb} becomes @code{0xbbaa}. Byte here always means
8827 @deftypefn {Built-in Function} uint32_t __builtin_bswap32 (uint32_t x)
8828 Similar to @code{__builtin_bswap16}, except the argument and return types
8832 @deftypefn {Built-in Function} uint64_t __builtin_bswap64 (uint64_t x)
8833 Similar to @code{__builtin_bswap32}, except the argument and return types
8837 @node Cilk Plus Builtins
8838 @section Cilk Plus C/C++ language extension Built-in Functions.
8840 GCC provides support for the following built-in reduction funtions if Cilk Plus
8841 is enabled. Cilk Plus can be enabled using the @option{-fcilkplus} flag.
8844 @item __sec_implicit_index
8846 @item __sec_reduce_add
8847 @item __sec_reduce_all_nonzero
8848 @item __sec_reduce_all_zero
8849 @item __sec_reduce_any_nonzero
8850 @item __sec_reduce_any_zero
8851 @item __sec_reduce_max
8852 @item __sec_reduce_min
8853 @item __sec_reduce_max_ind
8854 @item __sec_reduce_min_ind
8855 @item __sec_reduce_mul
8856 @item __sec_reduce_mutating
8859 Further details and examples about these built-in functions are described
8860 in the Cilk Plus language manual which can be found at
8861 @uref{http://www.cilkplus.org}.
8863 @node Target Builtins
8864 @section Built-in Functions Specific to Particular Target Machines
8866 On some target machines, GCC supports many built-in functions specific
8867 to those machines. Generally these generate calls to specific machine
8868 instructions, but allow the compiler to schedule those calls.
8871 * Alpha Built-in Functions::
8872 * ARM iWMMXt Built-in Functions::
8873 * ARM NEON Intrinsics::
8874 * AVR Built-in Functions::
8875 * Blackfin Built-in Functions::
8876 * FR-V Built-in Functions::
8877 * X86 Built-in Functions::
8878 * X86 transactional memory intrinsics::
8879 * MIPS DSP Built-in Functions::
8880 * MIPS Paired-Single Support::
8881 * MIPS Loongson Built-in Functions::
8882 * Other MIPS Built-in Functions::
8883 * MSP430 Built-in Functions::
8884 * picoChip Built-in Functions::
8885 * PowerPC Built-in Functions::
8886 * PowerPC AltiVec/VSX Built-in Functions::
8887 * RX Built-in Functions::
8888 * S/390 System z Built-in Functions::
8889 * SH Built-in Functions::
8890 * SPARC VIS Built-in Functions::
8891 * SPU Built-in Functions::
8892 * TI C6X Built-in Functions::
8893 * TILE-Gx Built-in Functions::
8894 * TILEPro Built-in Functions::
8897 @node Alpha Built-in Functions
8898 @subsection Alpha Built-in Functions
8900 These built-in functions are available for the Alpha family of
8901 processors, depending on the command-line switches used.
8903 The following built-in functions are always available. They
8904 all generate the machine instruction that is part of the name.
8907 long __builtin_alpha_implver (void)
8908 long __builtin_alpha_rpcc (void)
8909 long __builtin_alpha_amask (long)
8910 long __builtin_alpha_cmpbge (long, long)
8911 long __builtin_alpha_extbl (long, long)
8912 long __builtin_alpha_extwl (long, long)
8913 long __builtin_alpha_extll (long, long)
8914 long __builtin_alpha_extql (long, long)
8915 long __builtin_alpha_extwh (long, long)
8916 long __builtin_alpha_extlh (long, long)
8917 long __builtin_alpha_extqh (long, long)
8918 long __builtin_alpha_insbl (long, long)
8919 long __builtin_alpha_inswl (long, long)
8920 long __builtin_alpha_insll (long, long)
8921 long __builtin_alpha_insql (long, long)
8922 long __builtin_alpha_inswh (long, long)
8923 long __builtin_alpha_inslh (long, long)
8924 long __builtin_alpha_insqh (long, long)
8925 long __builtin_alpha_mskbl (long, long)
8926 long __builtin_alpha_mskwl (long, long)
8927 long __builtin_alpha_mskll (long, long)
8928 long __builtin_alpha_mskql (long, long)
8929 long __builtin_alpha_mskwh (long, long)
8930 long __builtin_alpha_msklh (long, long)
8931 long __builtin_alpha_mskqh (long, long)
8932 long __builtin_alpha_umulh (long, long)
8933 long __builtin_alpha_zap (long, long)
8934 long __builtin_alpha_zapnot (long, long)
8937 The following built-in functions are always with @option{-mmax}
8938 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or
8939 later. They all generate the machine instruction that is part
8943 long __builtin_alpha_pklb (long)
8944 long __builtin_alpha_pkwb (long)
8945 long __builtin_alpha_unpkbl (long)
8946 long __builtin_alpha_unpkbw (long)
8947 long __builtin_alpha_minub8 (long, long)
8948 long __builtin_alpha_minsb8 (long, long)
8949 long __builtin_alpha_minuw4 (long, long)
8950 long __builtin_alpha_minsw4 (long, long)
8951 long __builtin_alpha_maxub8 (long, long)
8952 long __builtin_alpha_maxsb8 (long, long)
8953 long __builtin_alpha_maxuw4 (long, long)
8954 long __builtin_alpha_maxsw4 (long, long)
8955 long __builtin_alpha_perr (long, long)
8958 The following built-in functions are always with @option{-mcix}
8959 or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or
8960 later. They all generate the machine instruction that is part
8964 long __builtin_alpha_cttz (long)
8965 long __builtin_alpha_ctlz (long)
8966 long __builtin_alpha_ctpop (long)
8969 The following built-in functions are available on systems that use the OSF/1
8970 PALcode. Normally they invoke the @code{rduniq} and @code{wruniq}
8971 PAL calls, but when invoked with @option{-mtls-kernel}, they invoke
8972 @code{rdval} and @code{wrval}.
8975 void *__builtin_thread_pointer (void)
8976 void __builtin_set_thread_pointer (void *)
8979 @node ARM iWMMXt Built-in Functions
8980 @subsection ARM iWMMXt Built-in Functions
8982 These built-in functions are available for the ARM family of
8983 processors when the @option{-mcpu=iwmmxt} switch is used:
8986 typedef int v2si __attribute__ ((vector_size (8)));
8987 typedef short v4hi __attribute__ ((vector_size (8)));
8988 typedef char v8qi __attribute__ ((vector_size (8)));
8990 int __builtin_arm_getwcgr0 (void)
8991 void __builtin_arm_setwcgr0 (int)
8992 int __builtin_arm_getwcgr1 (void)
8993 void __builtin_arm_setwcgr1 (int)
8994 int __builtin_arm_getwcgr2 (void)
8995 void __builtin_arm_setwcgr2 (int)
8996 int __builtin_arm_getwcgr3 (void)
8997 void __builtin_arm_setwcgr3 (int)
8998 int __builtin_arm_textrmsb (v8qi, int)
8999 int __builtin_arm_textrmsh (v4hi, int)
9000 int __builtin_arm_textrmsw (v2si, int)
9001 int __builtin_arm_textrmub (v8qi, int)
9002 int __builtin_arm_textrmuh (v4hi, int)
9003 int __builtin_arm_textrmuw (v2si, int)
9004 v8qi __builtin_arm_tinsrb (v8qi, int, int)
9005 v4hi __builtin_arm_tinsrh (v4hi, int, int)
9006 v2si __builtin_arm_tinsrw (v2si, int, int)
9007 long long __builtin_arm_tmia (long long, int, int)
9008 long long __builtin_arm_tmiabb (long long, int, int)
9009 long long __builtin_arm_tmiabt (long long, int, int)
9010 long long __builtin_arm_tmiaph (long long, int, int)
9011 long long __builtin_arm_tmiatb (long long, int, int)
9012 long long __builtin_arm_tmiatt (long long, int, int)
9013 int __builtin_arm_tmovmskb (v8qi)
9014 int __builtin_arm_tmovmskh (v4hi)
9015 int __builtin_arm_tmovmskw (v2si)
9016 long long __builtin_arm_waccb (v8qi)
9017 long long __builtin_arm_wacch (v4hi)
9018 long long __builtin_arm_waccw (v2si)
9019 v8qi __builtin_arm_waddb (v8qi, v8qi)
9020 v8qi __builtin_arm_waddbss (v8qi, v8qi)
9021 v8qi __builtin_arm_waddbus (v8qi, v8qi)
9022 v4hi __builtin_arm_waddh (v4hi, v4hi)
9023 v4hi __builtin_arm_waddhss (v4hi, v4hi)
9024 v4hi __builtin_arm_waddhus (v4hi, v4hi)
9025 v2si __builtin_arm_waddw (v2si, v2si)
9026 v2si __builtin_arm_waddwss (v2si, v2si)
9027 v2si __builtin_arm_waddwus (v2si, v2si)
9028 v8qi __builtin_arm_walign (v8qi, v8qi, int)
9029 long long __builtin_arm_wand(long long, long long)
9030 long long __builtin_arm_wandn (long long, long long)
9031 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
9032 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
9033 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
9034 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
9035 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
9036 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
9037 v2si __builtin_arm_wcmpeqw (v2si, v2si)
9038 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
9039 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
9040 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
9041 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
9042 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
9043 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
9044 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
9045 long long __builtin_arm_wmacsz (v4hi, v4hi)
9046 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
9047 long long __builtin_arm_wmacuz (v4hi, v4hi)
9048 v4hi __builtin_arm_wmadds (v4hi, v4hi)
9049 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
9050 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
9051 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
9052 v2si __builtin_arm_wmaxsw (v2si, v2si)
9053 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
9054 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
9055 v2si __builtin_arm_wmaxuw (v2si, v2si)
9056 v8qi __builtin_arm_wminsb (v8qi, v8qi)
9057 v4hi __builtin_arm_wminsh (v4hi, v4hi)
9058 v2si __builtin_arm_wminsw (v2si, v2si)
9059 v8qi __builtin_arm_wminub (v8qi, v8qi)
9060 v4hi __builtin_arm_wminuh (v4hi, v4hi)
9061 v2si __builtin_arm_wminuw (v2si, v2si)
9062 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
9063 v4hi __builtin_arm_wmulul (v4hi, v4hi)
9064 v4hi __builtin_arm_wmulum (v4hi, v4hi)
9065 long long __builtin_arm_wor (long long, long long)
9066 v2si __builtin_arm_wpackdss (long long, long long)
9067 v2si __builtin_arm_wpackdus (long long, long long)
9068 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
9069 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
9070 v4hi __builtin_arm_wpackwss (v2si, v2si)
9071 v4hi __builtin_arm_wpackwus (v2si, v2si)
9072 long long __builtin_arm_wrord (long long, long long)
9073 long long __builtin_arm_wrordi (long long, int)
9074 v4hi __builtin_arm_wrorh (v4hi, long long)
9075 v4hi __builtin_arm_wrorhi (v4hi, int)
9076 v2si __builtin_arm_wrorw (v2si, long long)
9077 v2si __builtin_arm_wrorwi (v2si, int)
9078 v2si __builtin_arm_wsadb (v2si, v8qi, v8qi)
9079 v2si __builtin_arm_wsadbz (v8qi, v8qi)
9080 v2si __builtin_arm_wsadh (v2si, v4hi, v4hi)
9081 v2si __builtin_arm_wsadhz (v4hi, v4hi)
9082 v4hi __builtin_arm_wshufh (v4hi, int)
9083 long long __builtin_arm_wslld (long long, long long)
9084 long long __builtin_arm_wslldi (long long, int)
9085 v4hi __builtin_arm_wsllh (v4hi, long long)
9086 v4hi __builtin_arm_wsllhi (v4hi, int)
9087 v2si __builtin_arm_wsllw (v2si, long long)
9088 v2si __builtin_arm_wsllwi (v2si, int)
9089 long long __builtin_arm_wsrad (long long, long long)
9090 long long __builtin_arm_wsradi (long long, int)
9091 v4hi __builtin_arm_wsrah (v4hi, long long)
9092 v4hi __builtin_arm_wsrahi (v4hi, int)
9093 v2si __builtin_arm_wsraw (v2si, long long)
9094 v2si __builtin_arm_wsrawi (v2si, int)
9095 long long __builtin_arm_wsrld (long long, long long)
9096 long long __builtin_arm_wsrldi (long long, int)
9097 v4hi __builtin_arm_wsrlh (v4hi, long long)
9098 v4hi __builtin_arm_wsrlhi (v4hi, int)
9099 v2si __builtin_arm_wsrlw (v2si, long long)
9100 v2si __builtin_arm_wsrlwi (v2si, int)
9101 v8qi __builtin_arm_wsubb (v8qi, v8qi)
9102 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
9103 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
9104 v4hi __builtin_arm_wsubh (v4hi, v4hi)
9105 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
9106 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
9107 v2si __builtin_arm_wsubw (v2si, v2si)
9108 v2si __builtin_arm_wsubwss (v2si, v2si)
9109 v2si __builtin_arm_wsubwus (v2si, v2si)
9110 v4hi __builtin_arm_wunpckehsb (v8qi)
9111 v2si __builtin_arm_wunpckehsh (v4hi)
9112 long long __builtin_arm_wunpckehsw (v2si)
9113 v4hi __builtin_arm_wunpckehub (v8qi)
9114 v2si __builtin_arm_wunpckehuh (v4hi)
9115 long long __builtin_arm_wunpckehuw (v2si)
9116 v4hi __builtin_arm_wunpckelsb (v8qi)
9117 v2si __builtin_arm_wunpckelsh (v4hi)
9118 long long __builtin_arm_wunpckelsw (v2si)
9119 v4hi __builtin_arm_wunpckelub (v8qi)
9120 v2si __builtin_arm_wunpckeluh (v4hi)
9121 long long __builtin_arm_wunpckeluw (v2si)
9122 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
9123 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
9124 v2si __builtin_arm_wunpckihw (v2si, v2si)
9125 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
9126 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
9127 v2si __builtin_arm_wunpckilw (v2si, v2si)
9128 long long __builtin_arm_wxor (long long, long long)
9129 long long __builtin_arm_wzero ()
9132 @node ARM NEON Intrinsics
9133 @subsection ARM NEON Intrinsics
9135 These built-in intrinsics for the ARM Advanced SIMD extension are available
9136 when the @option{-mfpu=neon} switch is used:
9138 @include arm-neon-intrinsics.texi
9140 @node AVR Built-in Functions
9141 @subsection AVR Built-in Functions
9143 For each built-in function for AVR, there is an equally named,
9144 uppercase built-in macro defined. That way users can easily query if
9145 or if not a specific built-in is implemented or not. For example, if
9146 @code{__builtin_avr_nop} is available the macro
9147 @code{__BUILTIN_AVR_NOP} is defined to @code{1} and undefined otherwise.
9149 The following built-in functions map to the respective machine
9150 instruction, i.e.@: @code{nop}, @code{sei}, @code{cli}, @code{sleep},
9151 @code{wdr}, @code{swap}, @code{fmul}, @code{fmuls}
9152 resp. @code{fmulsu}. The three @code{fmul*} built-ins are implemented
9153 as library call if no hardware multiplier is available.
9156 void __builtin_avr_nop (void)
9157 void __builtin_avr_sei (void)
9158 void __builtin_avr_cli (void)
9159 void __builtin_avr_sleep (void)
9160 void __builtin_avr_wdr (void)
9161 unsigned char __builtin_avr_swap (unsigned char)
9162 unsigned int __builtin_avr_fmul (unsigned char, unsigned char)
9163 int __builtin_avr_fmuls (char, char)
9164 int __builtin_avr_fmulsu (char, unsigned char)
9167 In order to delay execution for a specific number of cycles, GCC
9170 void __builtin_avr_delay_cycles (unsigned long ticks)
9174 @code{ticks} is the number of ticks to delay execution. Note that this
9175 built-in does not take into account the effect of interrupts that
9176 might increase delay time. @code{ticks} must be a compile-time
9177 integer constant; delays with a variable number of cycles are not supported.
9180 char __builtin_avr_flash_segment (const __memx void*)
9184 This built-in takes a byte address to the 24-bit
9185 @ref{AVR Named Address Spaces,address space} @code{__memx} and returns
9186 the number of the flash segment (the 64 KiB chunk) where the address
9187 points to. Counting starts at @code{0}.
9188 If the address does not point to flash memory, return @code{-1}.
9191 unsigned char __builtin_avr_insert_bits (unsigned long map, unsigned char bits, unsigned char val)
9195 Insert bits from @var{bits} into @var{val} and return the resulting
9196 value. The nibbles of @var{map} determine how the insertion is
9197 performed: Let @var{X} be the @var{n}-th nibble of @var{map}
9199 @item If @var{X} is @code{0xf},
9200 then the @var{n}-th bit of @var{val} is returned unaltered.
9202 @item If X is in the range 0@dots{}7,
9203 then the @var{n}-th result bit is set to the @var{X}-th bit of @var{bits}
9205 @item If X is in the range 8@dots{}@code{0xe},
9206 then the @var{n}-th result bit is undefined.
9210 One typical use case for this built-in is adjusting input and
9211 output values to non-contiguous port layouts. Some examples:
9214 // same as val, bits is unused
9215 __builtin_avr_insert_bits (0xffffffff, bits, val)
9219 // same as bits, val is unused
9220 __builtin_avr_insert_bits (0x76543210, bits, val)
9224 // same as rotating bits by 4
9225 __builtin_avr_insert_bits (0x32107654, bits, 0)
9229 // high nibble of result is the high nibble of val
9230 // low nibble of result is the low nibble of bits
9231 __builtin_avr_insert_bits (0xffff3210, bits, val)
9235 // reverse the bit order of bits
9236 __builtin_avr_insert_bits (0x01234567, bits, 0)
9239 @node Blackfin Built-in Functions
9240 @subsection Blackfin Built-in Functions
9242 Currently, there are two Blackfin-specific built-in functions. These are
9243 used for generating @code{CSYNC} and @code{SSYNC} machine insns without
9244 using inline assembly; by using these built-in functions the compiler can
9245 automatically add workarounds for hardware errata involving these
9246 instructions. These functions are named as follows:
9249 void __builtin_bfin_csync (void)
9250 void __builtin_bfin_ssync (void)
9253 @node FR-V Built-in Functions
9254 @subsection FR-V Built-in Functions
9256 GCC provides many FR-V-specific built-in functions. In general,
9257 these functions are intended to be compatible with those described
9258 by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu
9259 Semiconductor}. The two exceptions are @code{__MDUNPACKH} and
9260 @code{__MBTOHE}, the GCC forms of which pass 128-bit values by
9261 pointer rather than by value.
9263 Most of the functions are named after specific FR-V instructions.
9264 Such functions are said to be ``directly mapped'' and are summarized
9265 here in tabular form.
9269 * Directly-mapped Integer Functions::
9270 * Directly-mapped Media Functions::
9271 * Raw read/write Functions::
9272 * Other Built-in Functions::
9275 @node Argument Types
9276 @subsubsection Argument Types
9278 The arguments to the built-in functions can be divided into three groups:
9279 register numbers, compile-time constants and run-time values. In order
9280 to make this classification clear at a glance, the arguments and return
9281 values are given the following pseudo types:
9283 @multitable @columnfractions .20 .30 .15 .35
9284 @item Pseudo type @tab Real C type @tab Constant? @tab Description
9285 @item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword
9286 @item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word
9287 @item @code{sw1} @tab @code{int} @tab No @tab a signed word
9288 @item @code{uw2} @tab @code{unsigned long long} @tab No
9289 @tab an unsigned doubleword
9290 @item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword
9291 @item @code{const} @tab @code{int} @tab Yes @tab an integer constant
9292 @item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number
9293 @item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number
9296 These pseudo types are not defined by GCC, they are simply a notational
9297 convenience used in this manual.
9299 Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2}
9300 and @code{sw2} are evaluated at run time. They correspond to
9301 register operands in the underlying FR-V instructions.
9303 @code{const} arguments represent immediate operands in the underlying
9304 FR-V instructions. They must be compile-time constants.
9306 @code{acc} arguments are evaluated at compile time and specify the number
9307 of an accumulator register. For example, an @code{acc} argument of 2
9308 selects the ACC2 register.
9310 @code{iacc} arguments are similar to @code{acc} arguments but specify the
9311 number of an IACC register. See @pxref{Other Built-in Functions}
9314 @node Directly-mapped Integer Functions
9315 @subsubsection Directly-mapped Integer Functions
9317 The functions listed below map directly to FR-V I-type instructions.
9319 @multitable @columnfractions .45 .32 .23
9320 @item Function prototype @tab Example usage @tab Assembly output
9321 @item @code{sw1 __ADDSS (sw1, sw1)}
9322 @tab @code{@var{c} = __ADDSS (@var{a}, @var{b})}
9323 @tab @code{ADDSS @var{a},@var{b},@var{c}}
9324 @item @code{sw1 __SCAN (sw1, sw1)}
9325 @tab @code{@var{c} = __SCAN (@var{a}, @var{b})}
9326 @tab @code{SCAN @var{a},@var{b},@var{c}}
9327 @item @code{sw1 __SCUTSS (sw1)}
9328 @tab @code{@var{b} = __SCUTSS (@var{a})}
9329 @tab @code{SCUTSS @var{a},@var{b}}
9330 @item @code{sw1 __SLASS (sw1, sw1)}
9331 @tab @code{@var{c} = __SLASS (@var{a}, @var{b})}
9332 @tab @code{SLASS @var{a},@var{b},@var{c}}
9333 @item @code{void __SMASS (sw1, sw1)}
9334 @tab @code{__SMASS (@var{a}, @var{b})}
9335 @tab @code{SMASS @var{a},@var{b}}
9336 @item @code{void __SMSSS (sw1, sw1)}
9337 @tab @code{__SMSSS (@var{a}, @var{b})}
9338 @tab @code{SMSSS @var{a},@var{b}}
9339 @item @code{void __SMU (sw1, sw1)}
9340 @tab @code{__SMU (@var{a}, @var{b})}
9341 @tab @code{SMU @var{a},@var{b}}
9342 @item @code{sw2 __SMUL (sw1, sw1)}
9343 @tab @code{@var{c} = __SMUL (@var{a}, @var{b})}
9344 @tab @code{SMUL @var{a},@var{b},@var{c}}
9345 @item @code{sw1 __SUBSS (sw1, sw1)}
9346 @tab @code{@var{c} = __SUBSS (@var{a}, @var{b})}
9347 @tab @code{SUBSS @var{a},@var{b},@var{c}}
9348 @item @code{uw2 __UMUL (uw1, uw1)}
9349 @tab @code{@var{c} = __UMUL (@var{a}, @var{b})}
9350 @tab @code{UMUL @var{a},@var{b},@var{c}}
9353 @node Directly-mapped Media Functions
9354 @subsubsection Directly-mapped Media Functions
9356 The functions listed below map directly to FR-V M-type instructions.
9358 @multitable @columnfractions .45 .32 .23
9359 @item Function prototype @tab Example usage @tab Assembly output
9360 @item @code{uw1 __MABSHS (sw1)}
9361 @tab @code{@var{b} = __MABSHS (@var{a})}
9362 @tab @code{MABSHS @var{a},@var{b}}
9363 @item @code{void __MADDACCS (acc, acc)}
9364 @tab @code{__MADDACCS (@var{b}, @var{a})}
9365 @tab @code{MADDACCS @var{a},@var{b}}
9366 @item @code{sw1 __MADDHSS (sw1, sw1)}
9367 @tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})}
9368 @tab @code{MADDHSS @var{a},@var{b},@var{c}}
9369 @item @code{uw1 __MADDHUS (uw1, uw1)}
9370 @tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})}
9371 @tab @code{MADDHUS @var{a},@var{b},@var{c}}
9372 @item @code{uw1 __MAND (uw1, uw1)}
9373 @tab @code{@var{c} = __MAND (@var{a}, @var{b})}
9374 @tab @code{MAND @var{a},@var{b},@var{c}}
9375 @item @code{void __MASACCS (acc, acc)}
9376 @tab @code{__MASACCS (@var{b}, @var{a})}
9377 @tab @code{MASACCS @var{a},@var{b}}
9378 @item @code{uw1 __MAVEH (uw1, uw1)}
9379 @tab @code{@var{c} = __MAVEH (@var{a}, @var{b})}
9380 @tab @code{MAVEH @var{a},@var{b},@var{c}}
9381 @item @code{uw2 __MBTOH (uw1)}
9382 @tab @code{@var{b} = __MBTOH (@var{a})}
9383 @tab @code{MBTOH @var{a},@var{b}}
9384 @item @code{void __MBTOHE (uw1 *, uw1)}
9385 @tab @code{__MBTOHE (&@var{b}, @var{a})}
9386 @tab @code{MBTOHE @var{a},@var{b}}
9387 @item @code{void __MCLRACC (acc)}
9388 @tab @code{__MCLRACC (@var{a})}
9389 @tab @code{MCLRACC @var{a}}
9390 @item @code{void __MCLRACCA (void)}
9391 @tab @code{__MCLRACCA ()}
9392 @tab @code{MCLRACCA}
9393 @item @code{uw1 __Mcop1 (uw1, uw1)}
9394 @tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})}
9395 @tab @code{Mcop1 @var{a},@var{b},@var{c}}
9396 @item @code{uw1 __Mcop2 (uw1, uw1)}
9397 @tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})}
9398 @tab @code{Mcop2 @var{a},@var{b},@var{c}}
9399 @item @code{uw1 __MCPLHI (uw2, const)}
9400 @tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})}
9401 @tab @code{MCPLHI @var{a},#@var{b},@var{c}}
9402 @item @code{uw1 __MCPLI (uw2, const)}
9403 @tab @code{@var{c} = __MCPLI (@var{a}, @var{b})}
9404 @tab @code{MCPLI @var{a},#@var{b},@var{c}}
9405 @item @code{void __MCPXIS (acc, sw1, sw1)}
9406 @tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})}
9407 @tab @code{MCPXIS @var{a},@var{b},@var{c}}
9408 @item @code{void __MCPXIU (acc, uw1, uw1)}
9409 @tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})}
9410 @tab @code{MCPXIU @var{a},@var{b},@var{c}}
9411 @item @code{void __MCPXRS (acc, sw1, sw1)}
9412 @tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})}
9413 @tab @code{MCPXRS @var{a},@var{b},@var{c}}
9414 @item @code{void __MCPXRU (acc, uw1, uw1)}
9415 @tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})}
9416 @tab @code{MCPXRU @var{a},@var{b},@var{c}}
9417 @item @code{uw1 __MCUT (acc, uw1)}
9418 @tab @code{@var{c} = __MCUT (@var{a}, @var{b})}
9419 @tab @code{MCUT @var{a},@var{b},@var{c}}
9420 @item @code{uw1 __MCUTSS (acc, sw1)}
9421 @tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})}
9422 @tab @code{MCUTSS @var{a},@var{b},@var{c}}
9423 @item @code{void __MDADDACCS (acc, acc)}
9424 @tab @code{__MDADDACCS (@var{b}, @var{a})}
9425 @tab @code{MDADDACCS @var{a},@var{b}}
9426 @item @code{void __MDASACCS (acc, acc)}
9427 @tab @code{__MDASACCS (@var{b}, @var{a})}
9428 @tab @code{MDASACCS @var{a},@var{b}}
9429 @item @code{uw2 __MDCUTSSI (acc, const)}
9430 @tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})}
9431 @tab @code{MDCUTSSI @var{a},#@var{b},@var{c}}
9432 @item @code{uw2 __MDPACKH (uw2, uw2)}
9433 @tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})}
9434 @tab @code{MDPACKH @var{a},@var{b},@var{c}}
9435 @item @code{uw2 __MDROTLI (uw2, const)}
9436 @tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})}
9437 @tab @code{MDROTLI @var{a},#@var{b},@var{c}}
9438 @item @code{void __MDSUBACCS (acc, acc)}
9439 @tab @code{__MDSUBACCS (@var{b}, @var{a})}
9440 @tab @code{MDSUBACCS @var{a},@var{b}}
9441 @item @code{void __MDUNPACKH (uw1 *, uw2)}
9442 @tab @code{__MDUNPACKH (&@var{b}, @var{a})}
9443 @tab @code{MDUNPACKH @var{a},@var{b}}
9444 @item @code{uw2 __MEXPDHD (uw1, const)}
9445 @tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})}
9446 @tab @code{MEXPDHD @var{a},#@var{b},@var{c}}
9447 @item @code{uw1 __MEXPDHW (uw1, const)}
9448 @tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})}
9449 @tab @code{MEXPDHW @var{a},#@var{b},@var{c}}
9450 @item @code{uw1 __MHDSETH (uw1, const)}
9451 @tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})}
9452 @tab @code{MHDSETH @var{a},#@var{b},@var{c}}
9453 @item @code{sw1 __MHDSETS (const)}
9454 @tab @code{@var{b} = __MHDSETS (@var{a})}
9455 @tab @code{MHDSETS #@var{a},@var{b}}
9456 @item @code{uw1 __MHSETHIH (uw1, const)}
9457 @tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})}
9458 @tab @code{MHSETHIH #@var{a},@var{b}}
9459 @item @code{sw1 __MHSETHIS (sw1, const)}
9460 @tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})}
9461 @tab @code{MHSETHIS #@var{a},@var{b}}
9462 @item @code{uw1 __MHSETLOH (uw1, const)}
9463 @tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})}
9464 @tab @code{MHSETLOH #@var{a},@var{b}}
9465 @item @code{sw1 __MHSETLOS (sw1, const)}
9466 @tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})}
9467 @tab @code{MHSETLOS #@var{a},@var{b}}
9468 @item @code{uw1 __MHTOB (uw2)}
9469 @tab @code{@var{b} = __MHTOB (@var{a})}
9470 @tab @code{MHTOB @var{a},@var{b}}
9471 @item @code{void __MMACHS (acc, sw1, sw1)}
9472 @tab @code{__MMACHS (@var{c}, @var{a}, @var{b})}
9473 @tab @code{MMACHS @var{a},@var{b},@var{c}}
9474 @item @code{void __MMACHU (acc, uw1, uw1)}
9475 @tab @code{__MMACHU (@var{c}, @var{a}, @var{b})}
9476 @tab @code{MMACHU @var{a},@var{b},@var{c}}
9477 @item @code{void __MMRDHS (acc, sw1, sw1)}
9478 @tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})}
9479 @tab @code{MMRDHS @var{a},@var{b},@var{c}}
9480 @item @code{void __MMRDHU (acc, uw1, uw1)}
9481 @tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})}
9482 @tab @code{MMRDHU @var{a},@var{b},@var{c}}
9483 @item @code{void __MMULHS (acc, sw1, sw1)}
9484 @tab @code{__MMULHS (@var{c}, @var{a}, @var{b})}
9485 @tab @code{MMULHS @var{a},@var{b},@var{c}}
9486 @item @code{void __MMULHU (acc, uw1, uw1)}
9487 @tab @code{__MMULHU (@var{c}, @var{a}, @var{b})}
9488 @tab @code{MMULHU @var{a},@var{b},@var{c}}
9489 @item @code{void __MMULXHS (acc, sw1, sw1)}
9490 @tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})}
9491 @tab @code{MMULXHS @var{a},@var{b},@var{c}}
9492 @item @code{void __MMULXHU (acc, uw1, uw1)}
9493 @tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})}
9494 @tab @code{MMULXHU @var{a},@var{b},@var{c}}
9495 @item @code{uw1 __MNOT (uw1)}
9496 @tab @code{@var{b} = __MNOT (@var{a})}
9497 @tab @code{MNOT @var{a},@var{b}}
9498 @item @code{uw1 __MOR (uw1, uw1)}
9499 @tab @code{@var{c} = __MOR (@var{a}, @var{b})}
9500 @tab @code{MOR @var{a},@var{b},@var{c}}
9501 @item @code{uw1 __MPACKH (uh, uh)}
9502 @tab @code{@var{c} = __MPACKH (@var{a}, @var{b})}
9503 @tab @code{MPACKH @var{a},@var{b},@var{c}}
9504 @item @code{sw2 __MQADDHSS (sw2, sw2)}
9505 @tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})}
9506 @tab @code{MQADDHSS @var{a},@var{b},@var{c}}
9507 @item @code{uw2 __MQADDHUS (uw2, uw2)}
9508 @tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})}
9509 @tab @code{MQADDHUS @var{a},@var{b},@var{c}}
9510 @item @code{void __MQCPXIS (acc, sw2, sw2)}
9511 @tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})}
9512 @tab @code{MQCPXIS @var{a},@var{b},@var{c}}
9513 @item @code{void __MQCPXIU (acc, uw2, uw2)}
9514 @tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})}
9515 @tab @code{MQCPXIU @var{a},@var{b},@var{c}}
9516 @item @code{void __MQCPXRS (acc, sw2, sw2)}
9517 @tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})}
9518 @tab @code{MQCPXRS @var{a},@var{b},@var{c}}
9519 @item @code{void __MQCPXRU (acc, uw2, uw2)}
9520 @tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})}
9521 @tab @code{MQCPXRU @var{a},@var{b},@var{c}}
9522 @item @code{sw2 __MQLCLRHS (sw2, sw2)}
9523 @tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})}
9524 @tab @code{MQLCLRHS @var{a},@var{b},@var{c}}
9525 @item @code{sw2 __MQLMTHS (sw2, sw2)}
9526 @tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})}
9527 @tab @code{MQLMTHS @var{a},@var{b},@var{c}}
9528 @item @code{void __MQMACHS (acc, sw2, sw2)}
9529 @tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})}
9530 @tab @code{MQMACHS @var{a},@var{b},@var{c}}
9531 @item @code{void __MQMACHU (acc, uw2, uw2)}
9532 @tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})}
9533 @tab @code{MQMACHU @var{a},@var{b},@var{c}}
9534 @item @code{void __MQMACXHS (acc, sw2, sw2)}
9535 @tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})}
9536 @tab @code{MQMACXHS @var{a},@var{b},@var{c}}
9537 @item @code{void __MQMULHS (acc, sw2, sw2)}
9538 @tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})}
9539 @tab @code{MQMULHS @var{a},@var{b},@var{c}}
9540 @item @code{void __MQMULHU (acc, uw2, uw2)}
9541 @tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})}
9542 @tab @code{MQMULHU @var{a},@var{b},@var{c}}
9543 @item @code{void __MQMULXHS (acc, sw2, sw2)}
9544 @tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})}
9545 @tab @code{MQMULXHS @var{a},@var{b},@var{c}}
9546 @item @code{void __MQMULXHU (acc, uw2, uw2)}
9547 @tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})}
9548 @tab @code{MQMULXHU @var{a},@var{b},@var{c}}
9549 @item @code{sw2 __MQSATHS (sw2, sw2)}
9550 @tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})}
9551 @tab @code{MQSATHS @var{a},@var{b},@var{c}}
9552 @item @code{uw2 __MQSLLHI (uw2, int)}
9553 @tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})}
9554 @tab @code{MQSLLHI @var{a},@var{b},@var{c}}
9555 @item @code{sw2 __MQSRAHI (sw2, int)}
9556 @tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})}
9557 @tab @code{MQSRAHI @var{a},@var{b},@var{c}}
9558 @item @code{sw2 __MQSUBHSS (sw2, sw2)}
9559 @tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})}
9560 @tab @code{MQSUBHSS @var{a},@var{b},@var{c}}
9561 @item @code{uw2 __MQSUBHUS (uw2, uw2)}
9562 @tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})}
9563 @tab @code{MQSUBHUS @var{a},@var{b},@var{c}}
9564 @item @code{void __MQXMACHS (acc, sw2, sw2)}
9565 @tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})}
9566 @tab @code{MQXMACHS @var{a},@var{b},@var{c}}
9567 @item @code{void __MQXMACXHS (acc, sw2, sw2)}
9568 @tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})}
9569 @tab @code{MQXMACXHS @var{a},@var{b},@var{c}}
9570 @item @code{uw1 __MRDACC (acc)}
9571 @tab @code{@var{b} = __MRDACC (@var{a})}
9572 @tab @code{MRDACC @var{a},@var{b}}
9573 @item @code{uw1 __MRDACCG (acc)}
9574 @tab @code{@var{b} = __MRDACCG (@var{a})}
9575 @tab @code{MRDACCG @var{a},@var{b}}
9576 @item @code{uw1 __MROTLI (uw1, const)}
9577 @tab @code{@var{c} = __MROTLI (@var{a}, @var{b})}
9578 @tab @code{MROTLI @var{a},#@var{b},@var{c}}
9579 @item @code{uw1 __MROTRI (uw1, const)}
9580 @tab @code{@var{c} = __MROTRI (@var{a}, @var{b})}
9581 @tab @code{MROTRI @var{a},#@var{b},@var{c}}
9582 @item @code{sw1 __MSATHS (sw1, sw1)}
9583 @tab @code{@var{c} = __MSATHS (@var{a}, @var{b})}
9584 @tab @code{MSATHS @var{a},@var{b},@var{c}}
9585 @item @code{uw1 __MSATHU (uw1, uw1)}
9586 @tab @code{@var{c} = __MSATHU (@var{a}, @var{b})}
9587 @tab @code{MSATHU @var{a},@var{b},@var{c}}
9588 @item @code{uw1 __MSLLHI (uw1, const)}
9589 @tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})}
9590 @tab @code{MSLLHI @var{a},#@var{b},@var{c}}
9591 @item @code{sw1 __MSRAHI (sw1, const)}
9592 @tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})}
9593 @tab @code{MSRAHI @var{a},#@var{b},@var{c}}
9594 @item @code{uw1 __MSRLHI (uw1, const)}
9595 @tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})}
9596 @tab @code{MSRLHI @var{a},#@var{b},@var{c}}
9597 @item @code{void __MSUBACCS (acc, acc)}
9598 @tab @code{__MSUBACCS (@var{b}, @var{a})}
9599 @tab @code{MSUBACCS @var{a},@var{b}}
9600 @item @code{sw1 __MSUBHSS (sw1, sw1)}
9601 @tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})}
9602 @tab @code{MSUBHSS @var{a},@var{b},@var{c}}
9603 @item @code{uw1 __MSUBHUS (uw1, uw1)}
9604 @tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})}
9605 @tab @code{MSUBHUS @var{a},@var{b},@var{c}}
9606 @item @code{void __MTRAP (void)}
9607 @tab @code{__MTRAP ()}
9609 @item @code{uw2 __MUNPACKH (uw1)}
9610 @tab @code{@var{b} = __MUNPACKH (@var{a})}
9611 @tab @code{MUNPACKH @var{a},@var{b}}
9612 @item @code{uw1 __MWCUT (uw2, uw1)}
9613 @tab @code{@var{c} = __MWCUT (@var{a}, @var{b})}
9614 @tab @code{MWCUT @var{a},@var{b},@var{c}}
9615 @item @code{void __MWTACC (acc, uw1)}
9616 @tab @code{__MWTACC (@var{b}, @var{a})}
9617 @tab @code{MWTACC @var{a},@var{b}}
9618 @item @code{void __MWTACCG (acc, uw1)}
9619 @tab @code{__MWTACCG (@var{b}, @var{a})}
9620 @tab @code{MWTACCG @var{a},@var{b}}
9621 @item @code{uw1 __MXOR (uw1, uw1)}
9622 @tab @code{@var{c} = __MXOR (@var{a}, @var{b})}
9623 @tab @code{MXOR @var{a},@var{b},@var{c}}
9626 @node Raw read/write Functions
9627 @subsubsection Raw read/write Functions
9629 This sections describes built-in functions related to read and write
9630 instructions to access memory. These functions generate
9631 @code{membar} instructions to flush the I/O load and stores where
9632 appropriate, as described in Fujitsu's manual described above.
9636 @item unsigned char __builtin_read8 (void *@var{data})
9637 @item unsigned short __builtin_read16 (void *@var{data})
9638 @item unsigned long __builtin_read32 (void *@var{data})
9639 @item unsigned long long __builtin_read64 (void *@var{data})
9641 @item void __builtin_write8 (void *@var{data}, unsigned char @var{datum})
9642 @item void __builtin_write16 (void *@var{data}, unsigned short @var{datum})
9643 @item void __builtin_write32 (void *@var{data}, unsigned long @var{datum})
9644 @item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum})
9647 @node Other Built-in Functions
9648 @subsubsection Other Built-in Functions
9650 This section describes built-in functions that are not named after
9651 a specific FR-V instruction.
9654 @item sw2 __IACCreadll (iacc @var{reg})
9655 Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved
9656 for future expansion and must be 0.
9658 @item sw1 __IACCreadl (iacc @var{reg})
9659 Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1.
9660 Other values of @var{reg} are rejected as invalid.
9662 @item void __IACCsetll (iacc @var{reg}, sw2 @var{x})
9663 Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument
9664 is reserved for future expansion and must be 0.
9666 @item void __IACCsetl (iacc @var{reg}, sw1 @var{x})
9667 Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg}
9668 is 1. Other values of @var{reg} are rejected as invalid.
9670 @item void __data_prefetch0 (const void *@var{x})
9671 Use the @code{dcpl} instruction to load the contents of address @var{x}
9672 into the data cache.
9674 @item void __data_prefetch (const void *@var{x})
9675 Use the @code{nldub} instruction to load the contents of address @var{x}
9676 into the data cache. The instruction is issued in slot I1@.
9679 @node X86 Built-in Functions
9680 @subsection X86 Built-in Functions
9682 These built-in functions are available for the i386 and x86-64 family
9683 of computers, depending on the command-line switches used.
9685 If you specify command-line switches such as @option{-msse},
9686 the compiler could use the extended instruction sets even if the built-ins
9687 are not used explicitly in the program. For this reason, applications
9688 that perform run-time CPU detection must compile separate files for each
9689 supported architecture, using the appropriate flags. In particular,
9690 the file containing the CPU detection code should be compiled without
9693 The following machine modes are available for use with MMX built-in functions
9694 (@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers,
9695 @code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a
9696 vector of eight 8-bit integers. Some of the built-in functions operate on
9697 MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode.
9699 If 3DNow!@: extensions are enabled, @code{V2SF} is used as a mode for a vector
9700 of two 32-bit floating-point values.
9702 If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit
9703 floating-point values. Some instructions use a vector of four 32-bit
9704 integers, these use @code{V4SI}. Finally, some instructions operate on an
9705 entire vector register, interpreting it as a 128-bit integer, these use mode
9708 In 64-bit mode, the x86-64 family of processors uses additional built-in
9709 functions for efficient use of @code{TF} (@code{__float128}) 128-bit
9710 floating point and @code{TC} 128-bit complex floating-point values.
9712 The following floating-point built-in functions are available in 64-bit
9713 mode. All of them implement the function that is part of the name.
9716 __float128 __builtin_fabsq (__float128)
9717 __float128 __builtin_copysignq (__float128, __float128)
9720 The following built-in function is always available.
9723 @item void __builtin_ia32_pause (void)
9724 Generates the @code{pause} machine instruction with a compiler memory
9728 The following floating-point built-in functions are made available in the
9732 @item __float128 __builtin_infq (void)
9733 Similar to @code{__builtin_inf}, except the return type is @code{__float128}.
9734 @findex __builtin_infq
9736 @item __float128 __builtin_huge_valq (void)
9737 Similar to @code{__builtin_huge_val}, except the return type is @code{__float128}.
9738 @findex __builtin_huge_valq
9741 The following built-in functions are always available and can be used to
9742 check the target platform type.
9744 @deftypefn {Built-in Function} void __builtin_cpu_init (void)
9745 This function runs the CPU detection code to check the type of CPU and the
9746 features supported. This built-in function needs to be invoked along with the built-in functions
9747 to check CPU type and features, @code{__builtin_cpu_is} and
9748 @code{__builtin_cpu_supports}, only when used in a function that is
9749 executed before any constructors are called. The CPU detection code is
9750 automatically executed in a very high priority constructor.
9752 For example, this function has to be used in @code{ifunc} resolvers that
9753 check for CPU type using the built-in functions @code{__builtin_cpu_is}
9754 and @code{__builtin_cpu_supports}, or in constructors on targets that
9755 don't support constructor priority.
9758 static void (*resolve_memcpy (void)) (void)
9760 // ifunc resolvers fire before constructors, explicitly call the init
9762 __builtin_cpu_init ();
9763 if (__builtin_cpu_supports ("ssse3"))
9764 return ssse3_memcpy; // super fast memcpy with ssse3 instructions.
9766 return default_memcpy;
9769 void *memcpy (void *, const void *, size_t)
9770 __attribute__ ((ifunc ("resolve_memcpy")));
9775 @deftypefn {Built-in Function} int __builtin_cpu_is (const char *@var{cpuname})
9776 This function returns a positive integer if the run-time CPU
9777 is of type @var{cpuname}
9778 and returns @code{0} otherwise. The following CPU names can be detected:
9794 Intel Core i7 Nehalem CPU.
9797 Intel Core i7 Westmere CPU.
9800 Intel Core i7 Sandy Bridge CPU.
9809 AMD Family 10h Barcelona CPU.
9812 AMD Family 10h Shanghai CPU.
9815 AMD Family 10h Istanbul CPU.
9824 AMD Family 15h Bulldozer version 1.
9827 AMD Family 15h Bulldozer version 2.
9830 AMD Family 15h Bulldozer version 3.
9838 if (__builtin_cpu_is ("corei7"))
9840 do_corei7 (); // Core i7 specific implementation.
9844 do_generic (); // Generic implementation.
9849 @deftypefn {Built-in Function} int __builtin_cpu_supports (const char *@var{feature})
9850 This function returns a positive integer if the run-time CPU
9851 supports @var{feature}
9852 and returns @code{0} otherwise. The following features can be detected:
9870 SSE4.1 instructions.
9872 SSE4.2 instructions.
9881 if (__builtin_cpu_supports ("popcnt"))
9883 asm("popcnt %1,%0" : "=r"(count) : "rm"(n) : "cc");
9887 count = generic_countbits (n); //generic implementation.
9893 The following built-in functions are made available by @option{-mmmx}.
9894 All of them generate the machine instruction that is part of the name.
9897 v8qi __builtin_ia32_paddb (v8qi, v8qi)
9898 v4hi __builtin_ia32_paddw (v4hi, v4hi)
9899 v2si __builtin_ia32_paddd (v2si, v2si)
9900 v8qi __builtin_ia32_psubb (v8qi, v8qi)
9901 v4hi __builtin_ia32_psubw (v4hi, v4hi)
9902 v2si __builtin_ia32_psubd (v2si, v2si)
9903 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
9904 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
9905 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
9906 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
9907 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
9908 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
9909 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
9910 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
9911 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
9912 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
9913 di __builtin_ia32_pand (di, di)
9914 di __builtin_ia32_pandn (di,di)
9915 di __builtin_ia32_por (di, di)
9916 di __builtin_ia32_pxor (di, di)
9917 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
9918 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
9919 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
9920 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
9921 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
9922 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
9923 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
9924 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
9925 v2si __builtin_ia32_punpckhdq (v2si, v2si)
9926 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
9927 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
9928 v2si __builtin_ia32_punpckldq (v2si, v2si)
9929 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
9930 v4hi __builtin_ia32_packssdw (v2si, v2si)
9931 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
9933 v4hi __builtin_ia32_psllw (v4hi, v4hi)
9934 v2si __builtin_ia32_pslld (v2si, v2si)
9935 v1di __builtin_ia32_psllq (v1di, v1di)
9936 v4hi __builtin_ia32_psrlw (v4hi, v4hi)
9937 v2si __builtin_ia32_psrld (v2si, v2si)
9938 v1di __builtin_ia32_psrlq (v1di, v1di)
9939 v4hi __builtin_ia32_psraw (v4hi, v4hi)
9940 v2si __builtin_ia32_psrad (v2si, v2si)
9941 v4hi __builtin_ia32_psllwi (v4hi, int)
9942 v2si __builtin_ia32_pslldi (v2si, int)
9943 v1di __builtin_ia32_psllqi (v1di, int)
9944 v4hi __builtin_ia32_psrlwi (v4hi, int)
9945 v2si __builtin_ia32_psrldi (v2si, int)
9946 v1di __builtin_ia32_psrlqi (v1di, int)
9947 v4hi __builtin_ia32_psrawi (v4hi, int)
9948 v2si __builtin_ia32_psradi (v2si, int)
9952 The following built-in functions are made available either with
9953 @option{-msse}, or with a combination of @option{-m3dnow} and
9954 @option{-march=athlon}. All of them generate the machine
9955 instruction that is part of the name.
9958 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
9959 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
9960 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
9961 v1di __builtin_ia32_psadbw (v8qi, v8qi)
9962 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
9963 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
9964 v8qi __builtin_ia32_pminub (v8qi, v8qi)
9965 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
9966 int __builtin_ia32_pmovmskb (v8qi)
9967 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
9968 void __builtin_ia32_movntq (di *, di)
9969 void __builtin_ia32_sfence (void)
9972 The following built-in functions are available when @option{-msse} is used.
9973 All of them generate the machine instruction that is part of the name.
9976 int __builtin_ia32_comieq (v4sf, v4sf)
9977 int __builtin_ia32_comineq (v4sf, v4sf)
9978 int __builtin_ia32_comilt (v4sf, v4sf)
9979 int __builtin_ia32_comile (v4sf, v4sf)
9980 int __builtin_ia32_comigt (v4sf, v4sf)
9981 int __builtin_ia32_comige (v4sf, v4sf)
9982 int __builtin_ia32_ucomieq (v4sf, v4sf)
9983 int __builtin_ia32_ucomineq (v4sf, v4sf)
9984 int __builtin_ia32_ucomilt (v4sf, v4sf)
9985 int __builtin_ia32_ucomile (v4sf, v4sf)
9986 int __builtin_ia32_ucomigt (v4sf, v4sf)
9987 int __builtin_ia32_ucomige (v4sf, v4sf)
9988 v4sf __builtin_ia32_addps (v4sf, v4sf)
9989 v4sf __builtin_ia32_subps (v4sf, v4sf)
9990 v4sf __builtin_ia32_mulps (v4sf, v4sf)
9991 v4sf __builtin_ia32_divps (v4sf, v4sf)
9992 v4sf __builtin_ia32_addss (v4sf, v4sf)
9993 v4sf __builtin_ia32_subss (v4sf, v4sf)
9994 v4sf __builtin_ia32_mulss (v4sf, v4sf)
9995 v4sf __builtin_ia32_divss (v4sf, v4sf)
9996 v4sf __builtin_ia32_cmpeqps (v4sf, v4sf)
9997 v4sf __builtin_ia32_cmpltps (v4sf, v4sf)
9998 v4sf __builtin_ia32_cmpleps (v4sf, v4sf)
9999 v4sf __builtin_ia32_cmpgtps (v4sf, v4sf)
10000 v4sf __builtin_ia32_cmpgeps (v4sf, v4sf)
10001 v4sf __builtin_ia32_cmpunordps (v4sf, v4sf)
10002 v4sf __builtin_ia32_cmpneqps (v4sf, v4sf)
10003 v4sf __builtin_ia32_cmpnltps (v4sf, v4sf)
10004 v4sf __builtin_ia32_cmpnleps (v4sf, v4sf)
10005 v4sf __builtin_ia32_cmpngtps (v4sf, v4sf)
10006 v4sf __builtin_ia32_cmpngeps (v4sf, v4sf)
10007 v4sf __builtin_ia32_cmpordps (v4sf, v4sf)
10008 v4sf __builtin_ia32_cmpeqss (v4sf, v4sf)
10009 v4sf __builtin_ia32_cmpltss (v4sf, v4sf)
10010 v4sf __builtin_ia32_cmpless (v4sf, v4sf)
10011 v4sf __builtin_ia32_cmpunordss (v4sf, v4sf)
10012 v4sf __builtin_ia32_cmpneqss (v4sf, v4sf)
10013 v4sf __builtin_ia32_cmpnltss (v4sf, v4sf)
10014 v4sf __builtin_ia32_cmpnless (v4sf, v4sf)
10015 v4sf __builtin_ia32_cmpordss (v4sf, v4sf)
10016 v4sf __builtin_ia32_maxps (v4sf, v4sf)
10017 v4sf __builtin_ia32_maxss (v4sf, v4sf)
10018 v4sf __builtin_ia32_minps (v4sf, v4sf)
10019 v4sf __builtin_ia32_minss (v4sf, v4sf)
10020 v4sf __builtin_ia32_andps (v4sf, v4sf)
10021 v4sf __builtin_ia32_andnps (v4sf, v4sf)
10022 v4sf __builtin_ia32_orps (v4sf, v4sf)
10023 v4sf __builtin_ia32_xorps (v4sf, v4sf)
10024 v4sf __builtin_ia32_movss (v4sf, v4sf)
10025 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
10026 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
10027 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
10028 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
10029 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
10030 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
10031 v2si __builtin_ia32_cvtps2pi (v4sf)
10032 int __builtin_ia32_cvtss2si (v4sf)
10033 v2si __builtin_ia32_cvttps2pi (v4sf)
10034 int __builtin_ia32_cvttss2si (v4sf)
10035 v4sf __builtin_ia32_rcpps (v4sf)
10036 v4sf __builtin_ia32_rsqrtps (v4sf)
10037 v4sf __builtin_ia32_sqrtps (v4sf)
10038 v4sf __builtin_ia32_rcpss (v4sf)
10039 v4sf __builtin_ia32_rsqrtss (v4sf)
10040 v4sf __builtin_ia32_sqrtss (v4sf)
10041 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
10042 void __builtin_ia32_movntps (float *, v4sf)
10043 int __builtin_ia32_movmskps (v4sf)
10046 The following built-in functions are available when @option{-msse} is used.
10049 @item v4sf __builtin_ia32_loadups (float *)
10050 Generates the @code{movups} machine instruction as a load from memory.
10051 @item void __builtin_ia32_storeups (float *, v4sf)
10052 Generates the @code{movups} machine instruction as a store to memory.
10053 @item v4sf __builtin_ia32_loadss (float *)
10054 Generates the @code{movss} machine instruction as a load from memory.
10055 @item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)
10056 Generates the @code{movhps} machine instruction as a load from memory.
10057 @item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)
10058 Generates the @code{movlps} machine instruction as a load from memory
10059 @item void __builtin_ia32_storehps (v2sf *, v4sf)
10060 Generates the @code{movhps} machine instruction as a store to memory.
10061 @item void __builtin_ia32_storelps (v2sf *, v4sf)
10062 Generates the @code{movlps} machine instruction as a store to memory.
10065 The following built-in functions are available when @option{-msse2} is used.
10066 All of them generate the machine instruction that is part of the name.
10069 int __builtin_ia32_comisdeq (v2df, v2df)
10070 int __builtin_ia32_comisdlt (v2df, v2df)
10071 int __builtin_ia32_comisdle (v2df, v2df)
10072 int __builtin_ia32_comisdgt (v2df, v2df)
10073 int __builtin_ia32_comisdge (v2df, v2df)
10074 int __builtin_ia32_comisdneq (v2df, v2df)
10075 int __builtin_ia32_ucomisdeq (v2df, v2df)
10076 int __builtin_ia32_ucomisdlt (v2df, v2df)
10077 int __builtin_ia32_ucomisdle (v2df, v2df)
10078 int __builtin_ia32_ucomisdgt (v2df, v2df)
10079 int __builtin_ia32_ucomisdge (v2df, v2df)
10080 int __builtin_ia32_ucomisdneq (v2df, v2df)
10081 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
10082 v2df __builtin_ia32_cmpltpd (v2df, v2df)
10083 v2df __builtin_ia32_cmplepd (v2df, v2df)
10084 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
10085 v2df __builtin_ia32_cmpgepd (v2df, v2df)
10086 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
10087 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
10088 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
10089 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
10090 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
10091 v2df __builtin_ia32_cmpngepd (v2df, v2df)
10092 v2df __builtin_ia32_cmpordpd (v2df, v2df)
10093 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
10094 v2df __builtin_ia32_cmpltsd (v2df, v2df)
10095 v2df __builtin_ia32_cmplesd (v2df, v2df)
10096 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
10097 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
10098 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
10099 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
10100 v2df __builtin_ia32_cmpordsd (v2df, v2df)
10101 v2di __builtin_ia32_paddq (v2di, v2di)
10102 v2di __builtin_ia32_psubq (v2di, v2di)
10103 v2df __builtin_ia32_addpd (v2df, v2df)
10104 v2df __builtin_ia32_subpd (v2df, v2df)
10105 v2df __builtin_ia32_mulpd (v2df, v2df)
10106 v2df __builtin_ia32_divpd (v2df, v2df)
10107 v2df __builtin_ia32_addsd (v2df, v2df)
10108 v2df __builtin_ia32_subsd (v2df, v2df)
10109 v2df __builtin_ia32_mulsd (v2df, v2df)
10110 v2df __builtin_ia32_divsd (v2df, v2df)
10111 v2df __builtin_ia32_minpd (v2df, v2df)
10112 v2df __builtin_ia32_maxpd (v2df, v2df)
10113 v2df __builtin_ia32_minsd (v2df, v2df)
10114 v2df __builtin_ia32_maxsd (v2df, v2df)
10115 v2df __builtin_ia32_andpd (v2df, v2df)
10116 v2df __builtin_ia32_andnpd (v2df, v2df)
10117 v2df __builtin_ia32_orpd (v2df, v2df)
10118 v2df __builtin_ia32_xorpd (v2df, v2df)
10119 v2df __builtin_ia32_movsd (v2df, v2df)
10120 v2df __builtin_ia32_unpckhpd (v2df, v2df)
10121 v2df __builtin_ia32_unpcklpd (v2df, v2df)
10122 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
10123 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
10124 v4si __builtin_ia32_paddd128 (v4si, v4si)
10125 v2di __builtin_ia32_paddq128 (v2di, v2di)
10126 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
10127 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
10128 v4si __builtin_ia32_psubd128 (v4si, v4si)
10129 v2di __builtin_ia32_psubq128 (v2di, v2di)
10130 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
10131 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
10132 v2di __builtin_ia32_pand128 (v2di, v2di)
10133 v2di __builtin_ia32_pandn128 (v2di, v2di)
10134 v2di __builtin_ia32_por128 (v2di, v2di)
10135 v2di __builtin_ia32_pxor128 (v2di, v2di)
10136 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
10137 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
10138 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
10139 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
10140 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
10141 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
10142 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
10143 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
10144 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
10145 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
10146 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
10147 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
10148 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
10149 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
10150 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
10151 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
10152 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
10153 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
10154 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
10155 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
10156 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi)
10157 v8hi __builtin_ia32_packssdw128 (v4si, v4si)
10158 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi)
10159 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
10160 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
10161 v2df __builtin_ia32_loadupd (double *)
10162 void __builtin_ia32_storeupd (double *, v2df)
10163 v2df __builtin_ia32_loadhpd (v2df, double const *)
10164 v2df __builtin_ia32_loadlpd (v2df, double const *)
10165 int __builtin_ia32_movmskpd (v2df)
10166 int __builtin_ia32_pmovmskb128 (v16qi)
10167 void __builtin_ia32_movnti (int *, int)
10168 void __builtin_ia32_movnti64 (long long int *, long long int)
10169 void __builtin_ia32_movntpd (double *, v2df)
10170 void __builtin_ia32_movntdq (v2df *, v2df)
10171 v4si __builtin_ia32_pshufd (v4si, int)
10172 v8hi __builtin_ia32_pshuflw (v8hi, int)
10173 v8hi __builtin_ia32_pshufhw (v8hi, int)
10174 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
10175 v2df __builtin_ia32_sqrtpd (v2df)
10176 v2df __builtin_ia32_sqrtsd (v2df)
10177 v2df __builtin_ia32_shufpd (v2df, v2df, int)
10178 v2df __builtin_ia32_cvtdq2pd (v4si)
10179 v4sf __builtin_ia32_cvtdq2ps (v4si)
10180 v4si __builtin_ia32_cvtpd2dq (v2df)
10181 v2si __builtin_ia32_cvtpd2pi (v2df)
10182 v4sf __builtin_ia32_cvtpd2ps (v2df)
10183 v4si __builtin_ia32_cvttpd2dq (v2df)
10184 v2si __builtin_ia32_cvttpd2pi (v2df)
10185 v2df __builtin_ia32_cvtpi2pd (v2si)
10186 int __builtin_ia32_cvtsd2si (v2df)
10187 int __builtin_ia32_cvttsd2si (v2df)
10188 long long __builtin_ia32_cvtsd2si64 (v2df)
10189 long long __builtin_ia32_cvttsd2si64 (v2df)
10190 v4si __builtin_ia32_cvtps2dq (v4sf)
10191 v2df __builtin_ia32_cvtps2pd (v4sf)
10192 v4si __builtin_ia32_cvttps2dq (v4sf)
10193 v2df __builtin_ia32_cvtsi2sd (v2df, int)
10194 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
10195 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
10196 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
10197 void __builtin_ia32_clflush (const void *)
10198 void __builtin_ia32_lfence (void)
10199 void __builtin_ia32_mfence (void)
10200 v16qi __builtin_ia32_loaddqu (const char *)
10201 void __builtin_ia32_storedqu (char *, v16qi)
10202 v1di __builtin_ia32_pmuludq (v2si, v2si)
10203 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
10204 v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
10205 v4si __builtin_ia32_pslld128 (v4si, v4si)
10206 v2di __builtin_ia32_psllq128 (v2di, v2di)
10207 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
10208 v4si __builtin_ia32_psrld128 (v4si, v4si)
10209 v2di __builtin_ia32_psrlq128 (v2di, v2di)
10210 v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
10211 v4si __builtin_ia32_psrad128 (v4si, v4si)
10212 v2di __builtin_ia32_pslldqi128 (v2di, int)
10213 v8hi __builtin_ia32_psllwi128 (v8hi, int)
10214 v4si __builtin_ia32_pslldi128 (v4si, int)
10215 v2di __builtin_ia32_psllqi128 (v2di, int)
10216 v2di __builtin_ia32_psrldqi128 (v2di, int)
10217 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
10218 v4si __builtin_ia32_psrldi128 (v4si, int)
10219 v2di __builtin_ia32_psrlqi128 (v2di, int)
10220 v8hi __builtin_ia32_psrawi128 (v8hi, int)
10221 v4si __builtin_ia32_psradi128 (v4si, int)
10222 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
10223 v2di __builtin_ia32_movq128 (v2di)
10226 The following built-in functions are available when @option{-msse3} is used.
10227 All of them generate the machine instruction that is part of the name.
10230 v2df __builtin_ia32_addsubpd (v2df, v2df)
10231 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
10232 v2df __builtin_ia32_haddpd (v2df, v2df)
10233 v4sf __builtin_ia32_haddps (v4sf, v4sf)
10234 v2df __builtin_ia32_hsubpd (v2df, v2df)
10235 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
10236 v16qi __builtin_ia32_lddqu (char const *)
10237 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
10238 v4sf __builtin_ia32_movshdup (v4sf)
10239 v4sf __builtin_ia32_movsldup (v4sf)
10240 void __builtin_ia32_mwait (unsigned int, unsigned int)
10243 The following built-in functions are available when @option{-mssse3} is used.
10244 All of them generate the machine instruction that is part of the name.
10247 v2si __builtin_ia32_phaddd (v2si, v2si)
10248 v4hi __builtin_ia32_phaddw (v4hi, v4hi)
10249 v4hi __builtin_ia32_phaddsw (v4hi, v4hi)
10250 v2si __builtin_ia32_phsubd (v2si, v2si)
10251 v4hi __builtin_ia32_phsubw (v4hi, v4hi)
10252 v4hi __builtin_ia32_phsubsw (v4hi, v4hi)
10253 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi)
10254 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi)
10255 v8qi __builtin_ia32_pshufb (v8qi, v8qi)
10256 v8qi __builtin_ia32_psignb (v8qi, v8qi)
10257 v2si __builtin_ia32_psignd (v2si, v2si)
10258 v4hi __builtin_ia32_psignw (v4hi, v4hi)
10259 v1di __builtin_ia32_palignr (v1di, v1di, int)
10260 v8qi __builtin_ia32_pabsb (v8qi)
10261 v2si __builtin_ia32_pabsd (v2si)
10262 v4hi __builtin_ia32_pabsw (v4hi)
10265 The following built-in functions are available when @option{-mssse3} is used.
10266 All of them generate the machine instruction that is part of the name.
10269 v4si __builtin_ia32_phaddd128 (v4si, v4si)
10270 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi)
10271 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi)
10272 v4si __builtin_ia32_phsubd128 (v4si, v4si)
10273 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi)
10274 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi)
10275 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi)
10276 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi)
10277 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi)
10278 v16qi __builtin_ia32_psignb128 (v16qi, v16qi)
10279 v4si __builtin_ia32_psignd128 (v4si, v4si)
10280 v8hi __builtin_ia32_psignw128 (v8hi, v8hi)
10281 v2di __builtin_ia32_palignr128 (v2di, v2di, int)
10282 v16qi __builtin_ia32_pabsb128 (v16qi)
10283 v4si __builtin_ia32_pabsd128 (v4si)
10284 v8hi __builtin_ia32_pabsw128 (v8hi)
10287 The following built-in functions are available when @option{-msse4.1} is
10288 used. All of them generate the machine instruction that is part of the
10292 v2df __builtin_ia32_blendpd (v2df, v2df, const int)
10293 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int)
10294 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df)
10295 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf)
10296 v2df __builtin_ia32_dppd (v2df, v2df, const int)
10297 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int)
10298 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int)
10299 v2di __builtin_ia32_movntdqa (v2di *);
10300 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int)
10301 v8hi __builtin_ia32_packusdw128 (v4si, v4si)
10302 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi)
10303 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int)
10304 v2di __builtin_ia32_pcmpeqq (v2di, v2di)
10305 v8hi __builtin_ia32_phminposuw128 (v8hi)
10306 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi)
10307 v4si __builtin_ia32_pmaxsd128 (v4si, v4si)
10308 v4si __builtin_ia32_pmaxud128 (v4si, v4si)
10309 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi)
10310 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi)
10311 v4si __builtin_ia32_pminsd128 (v4si, v4si)
10312 v4si __builtin_ia32_pminud128 (v4si, v4si)
10313 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi)
10314 v4si __builtin_ia32_pmovsxbd128 (v16qi)
10315 v2di __builtin_ia32_pmovsxbq128 (v16qi)
10316 v8hi __builtin_ia32_pmovsxbw128 (v16qi)
10317 v2di __builtin_ia32_pmovsxdq128 (v4si)
10318 v4si __builtin_ia32_pmovsxwd128 (v8hi)
10319 v2di __builtin_ia32_pmovsxwq128 (v8hi)
10320 v4si __builtin_ia32_pmovzxbd128 (v16qi)
10321 v2di __builtin_ia32_pmovzxbq128 (v16qi)
10322 v8hi __builtin_ia32_pmovzxbw128 (v16qi)
10323 v2di __builtin_ia32_pmovzxdq128 (v4si)
10324 v4si __builtin_ia32_pmovzxwd128 (v8hi)
10325 v2di __builtin_ia32_pmovzxwq128 (v8hi)
10326 v2di __builtin_ia32_pmuldq128 (v4si, v4si)
10327 v4si __builtin_ia32_pmulld128 (v4si, v4si)
10328 int __builtin_ia32_ptestc128 (v2di, v2di)
10329 int __builtin_ia32_ptestnzc128 (v2di, v2di)
10330 int __builtin_ia32_ptestz128 (v2di, v2di)
10331 v2df __builtin_ia32_roundpd (v2df, const int)
10332 v4sf __builtin_ia32_roundps (v4sf, const int)
10333 v2df __builtin_ia32_roundsd (v2df, v2df, const int)
10334 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int)
10337 The following built-in functions are available when @option{-msse4.1} is
10341 @item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)
10342 Generates the @code{insertps} machine instruction.
10343 @item int __builtin_ia32_vec_ext_v16qi (v16qi, const int)
10344 Generates the @code{pextrb} machine instruction.
10345 @item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)
10346 Generates the @code{pinsrb} machine instruction.
10347 @item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)
10348 Generates the @code{pinsrd} machine instruction.
10349 @item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)
10350 Generates the @code{pinsrq} machine instruction in 64bit mode.
10353 The following built-in functions are changed to generate new SSE4.1
10354 instructions when @option{-msse4.1} is used.
10357 @item float __builtin_ia32_vec_ext_v4sf (v4sf, const int)
10358 Generates the @code{extractps} machine instruction.
10359 @item int __builtin_ia32_vec_ext_v4si (v4si, const int)
10360 Generates the @code{pextrd} machine instruction.
10361 @item long long __builtin_ia32_vec_ext_v2di (v2di, const int)
10362 Generates the @code{pextrq} machine instruction in 64bit mode.
10365 The following built-in functions are available when @option{-msse4.2} is
10366 used. All of them generate the machine instruction that is part of the
10370 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int)
10371 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int)
10372 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int)
10373 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int)
10374 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int)
10375 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int)
10376 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int)
10377 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int)
10378 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int)
10379 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int)
10380 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int)
10381 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int)
10382 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int)
10383 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int)
10384 v2di __builtin_ia32_pcmpgtq (v2di, v2di)
10387 The following built-in functions are available when @option{-msse4.2} is
10391 @item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)
10392 Generates the @code{crc32b} machine instruction.
10393 @item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)
10394 Generates the @code{crc32w} machine instruction.
10395 @item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)
10396 Generates the @code{crc32l} machine instruction.
10397 @item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)
10398 Generates the @code{crc32q} machine instruction.
10401 The following built-in functions are changed to generate new SSE4.2
10402 instructions when @option{-msse4.2} is used.
10405 @item int __builtin_popcount (unsigned int)
10406 Generates the @code{popcntl} machine instruction.
10407 @item int __builtin_popcountl (unsigned long)
10408 Generates the @code{popcntl} or @code{popcntq} machine instruction,
10409 depending on the size of @code{unsigned long}.
10410 @item int __builtin_popcountll (unsigned long long)
10411 Generates the @code{popcntq} machine instruction.
10414 The following built-in functions are available when @option{-mavx} is
10415 used. All of them generate the machine instruction that is part of the
10419 v4df __builtin_ia32_addpd256 (v4df,v4df)
10420 v8sf __builtin_ia32_addps256 (v8sf,v8sf)
10421 v4df __builtin_ia32_addsubpd256 (v4df,v4df)
10422 v8sf __builtin_ia32_addsubps256 (v8sf,v8sf)
10423 v4df __builtin_ia32_andnpd256 (v4df,v4df)
10424 v8sf __builtin_ia32_andnps256 (v8sf,v8sf)
10425 v4df __builtin_ia32_andpd256 (v4df,v4df)
10426 v8sf __builtin_ia32_andps256 (v8sf,v8sf)
10427 v4df __builtin_ia32_blendpd256 (v4df,v4df,int)
10428 v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int)
10429 v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df)
10430 v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf)
10431 v2df __builtin_ia32_cmppd (v2df,v2df,int)
10432 v4df __builtin_ia32_cmppd256 (v4df,v4df,int)
10433 v4sf __builtin_ia32_cmpps (v4sf,v4sf,int)
10434 v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int)
10435 v2df __builtin_ia32_cmpsd (v2df,v2df,int)
10436 v4sf __builtin_ia32_cmpss (v4sf,v4sf,int)
10437 v4df __builtin_ia32_cvtdq2pd256 (v4si)
10438 v8sf __builtin_ia32_cvtdq2ps256 (v8si)
10439 v4si __builtin_ia32_cvtpd2dq256 (v4df)
10440 v4sf __builtin_ia32_cvtpd2ps256 (v4df)
10441 v8si __builtin_ia32_cvtps2dq256 (v8sf)
10442 v4df __builtin_ia32_cvtps2pd256 (v4sf)
10443 v4si __builtin_ia32_cvttpd2dq256 (v4df)
10444 v8si __builtin_ia32_cvttps2dq256 (v8sf)
10445 v4df __builtin_ia32_divpd256 (v4df,v4df)
10446 v8sf __builtin_ia32_divps256 (v8sf,v8sf)
10447 v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int)
10448 v4df __builtin_ia32_haddpd256 (v4df,v4df)
10449 v8sf __builtin_ia32_haddps256 (v8sf,v8sf)
10450 v4df __builtin_ia32_hsubpd256 (v4df,v4df)
10451 v8sf __builtin_ia32_hsubps256 (v8sf,v8sf)
10452 v32qi __builtin_ia32_lddqu256 (pcchar)
10453 v32qi __builtin_ia32_loaddqu256 (pcchar)
10454 v4df __builtin_ia32_loadupd256 (pcdouble)
10455 v8sf __builtin_ia32_loadups256 (pcfloat)
10456 v2df __builtin_ia32_maskloadpd (pcv2df,v2df)
10457 v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df)
10458 v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf)
10459 v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf)
10460 void __builtin_ia32_maskstorepd (pv2df,v2df,v2df)
10461 void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df)
10462 void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf)
10463 void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf)
10464 v4df __builtin_ia32_maxpd256 (v4df,v4df)
10465 v8sf __builtin_ia32_maxps256 (v8sf,v8sf)
10466 v4df __builtin_ia32_minpd256 (v4df,v4df)
10467 v8sf __builtin_ia32_minps256 (v8sf,v8sf)
10468 v4df __builtin_ia32_movddup256 (v4df)
10469 int __builtin_ia32_movmskpd256 (v4df)
10470 int __builtin_ia32_movmskps256 (v8sf)
10471 v8sf __builtin_ia32_movshdup256 (v8sf)
10472 v8sf __builtin_ia32_movsldup256 (v8sf)
10473 v4df __builtin_ia32_mulpd256 (v4df,v4df)
10474 v8sf __builtin_ia32_mulps256 (v8sf,v8sf)
10475 v4df __builtin_ia32_orpd256 (v4df,v4df)
10476 v8sf __builtin_ia32_orps256 (v8sf,v8sf)
10477 v2df __builtin_ia32_pd_pd256 (v4df)
10478 v4df __builtin_ia32_pd256_pd (v2df)
10479 v4sf __builtin_ia32_ps_ps256 (v8sf)
10480 v8sf __builtin_ia32_ps256_ps (v4sf)
10481 int __builtin_ia32_ptestc256 (v4di,v4di,ptest)
10482 int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest)
10483 int __builtin_ia32_ptestz256 (v4di,v4di,ptest)
10484 v8sf __builtin_ia32_rcpps256 (v8sf)
10485 v4df __builtin_ia32_roundpd256 (v4df,int)
10486 v8sf __builtin_ia32_roundps256 (v8sf,int)
10487 v8sf __builtin_ia32_rsqrtps_nr256 (v8sf)
10488 v8sf __builtin_ia32_rsqrtps256 (v8sf)
10489 v4df __builtin_ia32_shufpd256 (v4df,v4df,int)
10490 v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int)
10491 v4si __builtin_ia32_si_si256 (v8si)
10492 v8si __builtin_ia32_si256_si (v4si)
10493 v4df __builtin_ia32_sqrtpd256 (v4df)
10494 v8sf __builtin_ia32_sqrtps_nr256 (v8sf)
10495 v8sf __builtin_ia32_sqrtps256 (v8sf)
10496 void __builtin_ia32_storedqu256 (pchar,v32qi)
10497 void __builtin_ia32_storeupd256 (pdouble,v4df)
10498 void __builtin_ia32_storeups256 (pfloat,v8sf)
10499 v4df __builtin_ia32_subpd256 (v4df,v4df)
10500 v8sf __builtin_ia32_subps256 (v8sf,v8sf)
10501 v4df __builtin_ia32_unpckhpd256 (v4df,v4df)
10502 v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf)
10503 v4df __builtin_ia32_unpcklpd256 (v4df,v4df)
10504 v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf)
10505 v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df)
10506 v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf)
10507 v4df __builtin_ia32_vbroadcastsd256 (pcdouble)
10508 v4sf __builtin_ia32_vbroadcastss (pcfloat)
10509 v8sf __builtin_ia32_vbroadcastss256 (pcfloat)
10510 v2df __builtin_ia32_vextractf128_pd256 (v4df,int)
10511 v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int)
10512 v4si __builtin_ia32_vextractf128_si256 (v8si,int)
10513 v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int)
10514 v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int)
10515 v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int)
10516 v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int)
10517 v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int)
10518 v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int)
10519 v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int)
10520 v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int)
10521 v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int)
10522 v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int)
10523 v2df __builtin_ia32_vpermilpd (v2df,int)
10524 v4df __builtin_ia32_vpermilpd256 (v4df,int)
10525 v4sf __builtin_ia32_vpermilps (v4sf,int)
10526 v8sf __builtin_ia32_vpermilps256 (v8sf,int)
10527 v2df __builtin_ia32_vpermilvarpd (v2df,v2di)
10528 v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di)
10529 v4sf __builtin_ia32_vpermilvarps (v4sf,v4si)
10530 v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si)
10531 int __builtin_ia32_vtestcpd (v2df,v2df,ptest)
10532 int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest)
10533 int __builtin_ia32_vtestcps (v4sf,v4sf,ptest)
10534 int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest)
10535 int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest)
10536 int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest)
10537 int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest)
10538 int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest)
10539 int __builtin_ia32_vtestzpd (v2df,v2df,ptest)
10540 int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest)
10541 int __builtin_ia32_vtestzps (v4sf,v4sf,ptest)
10542 int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest)
10543 void __builtin_ia32_vzeroall (void)
10544 void __builtin_ia32_vzeroupper (void)
10545 v4df __builtin_ia32_xorpd256 (v4df,v4df)
10546 v8sf __builtin_ia32_xorps256 (v8sf,v8sf)
10549 The following built-in functions are available when @option{-mavx2} is
10550 used. All of them generate the machine instruction that is part of the
10554 v32qi __builtin_ia32_mpsadbw256 (v32qi,v32qi,v32qi,int)
10555 v32qi __builtin_ia32_pabsb256 (v32qi)
10556 v16hi __builtin_ia32_pabsw256 (v16hi)
10557 v8si __builtin_ia32_pabsd256 (v8si)
10558 v16hi __builtin_ia32_packssdw256 (v8si,v8si)
10559 v32qi __builtin_ia32_packsswb256 (v16hi,v16hi)
10560 v16hi __builtin_ia32_packusdw256 (v8si,v8si)
10561 v32qi __builtin_ia32_packuswb256 (v16hi,v16hi)
10562 v32qi __builtin_ia32_paddb256 (v32qi,v32qi)
10563 v16hi __builtin_ia32_paddw256 (v16hi,v16hi)
10564 v8si __builtin_ia32_paddd256 (v8si,v8si)
10565 v4di __builtin_ia32_paddq256 (v4di,v4di)
10566 v32qi __builtin_ia32_paddsb256 (v32qi,v32qi)
10567 v16hi __builtin_ia32_paddsw256 (v16hi,v16hi)
10568 v32qi __builtin_ia32_paddusb256 (v32qi,v32qi)
10569 v16hi __builtin_ia32_paddusw256 (v16hi,v16hi)
10570 v4di __builtin_ia32_palignr256 (v4di,v4di,int)
10571 v4di __builtin_ia32_andsi256 (v4di,v4di)
10572 v4di __builtin_ia32_andnotsi256 (v4di,v4di)
10573 v32qi __builtin_ia32_pavgb256 (v32qi,v32qi)
10574 v16hi __builtin_ia32_pavgw256 (v16hi,v16hi)
10575 v32qi __builtin_ia32_pblendvb256 (v32qi,v32qi,v32qi)
10576 v16hi __builtin_ia32_pblendw256 (v16hi,v16hi,int)
10577 v32qi __builtin_ia32_pcmpeqb256 (v32qi,v32qi)
10578 v16hi __builtin_ia32_pcmpeqw256 (v16hi,v16hi)
10579 v8si __builtin_ia32_pcmpeqd256 (c8si,v8si)
10580 v4di __builtin_ia32_pcmpeqq256 (v4di,v4di)
10581 v32qi __builtin_ia32_pcmpgtb256 (v32qi,v32qi)
10582 v16hi __builtin_ia32_pcmpgtw256 (16hi,v16hi)
10583 v8si __builtin_ia32_pcmpgtd256 (v8si,v8si)
10584 v4di __builtin_ia32_pcmpgtq256 (v4di,v4di)
10585 v16hi __builtin_ia32_phaddw256 (v16hi,v16hi)
10586 v8si __builtin_ia32_phaddd256 (v8si,v8si)
10587 v16hi __builtin_ia32_phaddsw256 (v16hi,v16hi)
10588 v16hi __builtin_ia32_phsubw256 (v16hi,v16hi)
10589 v8si __builtin_ia32_phsubd256 (v8si,v8si)
10590 v16hi __builtin_ia32_phsubsw256 (v16hi,v16hi)
10591 v32qi __builtin_ia32_pmaddubsw256 (v32qi,v32qi)
10592 v16hi __builtin_ia32_pmaddwd256 (v16hi,v16hi)
10593 v32qi __builtin_ia32_pmaxsb256 (v32qi,v32qi)
10594 v16hi __builtin_ia32_pmaxsw256 (v16hi,v16hi)
10595 v8si __builtin_ia32_pmaxsd256 (v8si,v8si)
10596 v32qi __builtin_ia32_pmaxub256 (v32qi,v32qi)
10597 v16hi __builtin_ia32_pmaxuw256 (v16hi,v16hi)
10598 v8si __builtin_ia32_pmaxud256 (v8si,v8si)
10599 v32qi __builtin_ia32_pminsb256 (v32qi,v32qi)
10600 v16hi __builtin_ia32_pminsw256 (v16hi,v16hi)
10601 v8si __builtin_ia32_pminsd256 (v8si,v8si)
10602 v32qi __builtin_ia32_pminub256 (v32qi,v32qi)
10603 v16hi __builtin_ia32_pminuw256 (v16hi,v16hi)
10604 v8si __builtin_ia32_pminud256 (v8si,v8si)
10605 int __builtin_ia32_pmovmskb256 (v32qi)
10606 v16hi __builtin_ia32_pmovsxbw256 (v16qi)
10607 v8si __builtin_ia32_pmovsxbd256 (v16qi)
10608 v4di __builtin_ia32_pmovsxbq256 (v16qi)
10609 v8si __builtin_ia32_pmovsxwd256 (v8hi)
10610 v4di __builtin_ia32_pmovsxwq256 (v8hi)
10611 v4di __builtin_ia32_pmovsxdq256 (v4si)
10612 v16hi __builtin_ia32_pmovzxbw256 (v16qi)
10613 v8si __builtin_ia32_pmovzxbd256 (v16qi)
10614 v4di __builtin_ia32_pmovzxbq256 (v16qi)
10615 v8si __builtin_ia32_pmovzxwd256 (v8hi)
10616 v4di __builtin_ia32_pmovzxwq256 (v8hi)
10617 v4di __builtin_ia32_pmovzxdq256 (v4si)
10618 v4di __builtin_ia32_pmuldq256 (v8si,v8si)
10619 v16hi __builtin_ia32_pmulhrsw256 (v16hi, v16hi)
10620 v16hi __builtin_ia32_pmulhuw256 (v16hi,v16hi)
10621 v16hi __builtin_ia32_pmulhw256 (v16hi,v16hi)
10622 v16hi __builtin_ia32_pmullw256 (v16hi,v16hi)
10623 v8si __builtin_ia32_pmulld256 (v8si,v8si)
10624 v4di __builtin_ia32_pmuludq256 (v8si,v8si)
10625 v4di __builtin_ia32_por256 (v4di,v4di)
10626 v16hi __builtin_ia32_psadbw256 (v32qi,v32qi)
10627 v32qi __builtin_ia32_pshufb256 (v32qi,v32qi)
10628 v8si __builtin_ia32_pshufd256 (v8si,int)
10629 v16hi __builtin_ia32_pshufhw256 (v16hi,int)
10630 v16hi __builtin_ia32_pshuflw256 (v16hi,int)
10631 v32qi __builtin_ia32_psignb256 (v32qi,v32qi)
10632 v16hi __builtin_ia32_psignw256 (v16hi,v16hi)
10633 v8si __builtin_ia32_psignd256 (v8si,v8si)
10634 v4di __builtin_ia32_pslldqi256 (v4di,int)
10635 v16hi __builtin_ia32_psllwi256 (16hi,int)
10636 v16hi __builtin_ia32_psllw256(v16hi,v8hi)
10637 v8si __builtin_ia32_pslldi256 (v8si,int)
10638 v8si __builtin_ia32_pslld256(v8si,v4si)
10639 v4di __builtin_ia32_psllqi256 (v4di,int)
10640 v4di __builtin_ia32_psllq256(v4di,v2di)
10641 v16hi __builtin_ia32_psrawi256 (v16hi,int)
10642 v16hi __builtin_ia32_psraw256 (v16hi,v8hi)
10643 v8si __builtin_ia32_psradi256 (v8si,int)
10644 v8si __builtin_ia32_psrad256 (v8si,v4si)
10645 v4di __builtin_ia32_psrldqi256 (v4di, int)
10646 v16hi __builtin_ia32_psrlwi256 (v16hi,int)
10647 v16hi __builtin_ia32_psrlw256 (v16hi,v8hi)
10648 v8si __builtin_ia32_psrldi256 (v8si,int)
10649 v8si __builtin_ia32_psrld256 (v8si,v4si)
10650 v4di __builtin_ia32_psrlqi256 (v4di,int)
10651 v4di __builtin_ia32_psrlq256(v4di,v2di)
10652 v32qi __builtin_ia32_psubb256 (v32qi,v32qi)
10653 v32hi __builtin_ia32_psubw256 (v16hi,v16hi)
10654 v8si __builtin_ia32_psubd256 (v8si,v8si)
10655 v4di __builtin_ia32_psubq256 (v4di,v4di)
10656 v32qi __builtin_ia32_psubsb256 (v32qi,v32qi)
10657 v16hi __builtin_ia32_psubsw256 (v16hi,v16hi)
10658 v32qi __builtin_ia32_psubusb256 (v32qi,v32qi)
10659 v16hi __builtin_ia32_psubusw256 (v16hi,v16hi)
10660 v32qi __builtin_ia32_punpckhbw256 (v32qi,v32qi)
10661 v16hi __builtin_ia32_punpckhwd256 (v16hi,v16hi)
10662 v8si __builtin_ia32_punpckhdq256 (v8si,v8si)
10663 v4di __builtin_ia32_punpckhqdq256 (v4di,v4di)
10664 v32qi __builtin_ia32_punpcklbw256 (v32qi,v32qi)
10665 v16hi __builtin_ia32_punpcklwd256 (v16hi,v16hi)
10666 v8si __builtin_ia32_punpckldq256 (v8si,v8si)
10667 v4di __builtin_ia32_punpcklqdq256 (v4di,v4di)
10668 v4di __builtin_ia32_pxor256 (v4di,v4di)
10669 v4di __builtin_ia32_movntdqa256 (pv4di)
10670 v4sf __builtin_ia32_vbroadcastss_ps (v4sf)
10671 v8sf __builtin_ia32_vbroadcastss_ps256 (v4sf)
10672 v4df __builtin_ia32_vbroadcastsd_pd256 (v2df)
10673 v4di __builtin_ia32_vbroadcastsi256 (v2di)
10674 v4si __builtin_ia32_pblendd128 (v4si,v4si)
10675 v8si __builtin_ia32_pblendd256 (v8si,v8si)
10676 v32qi __builtin_ia32_pbroadcastb256 (v16qi)
10677 v16hi __builtin_ia32_pbroadcastw256 (v8hi)
10678 v8si __builtin_ia32_pbroadcastd256 (v4si)
10679 v4di __builtin_ia32_pbroadcastq256 (v2di)
10680 v16qi __builtin_ia32_pbroadcastb128 (v16qi)
10681 v8hi __builtin_ia32_pbroadcastw128 (v8hi)
10682 v4si __builtin_ia32_pbroadcastd128 (v4si)
10683 v2di __builtin_ia32_pbroadcastq128 (v2di)
10684 v8si __builtin_ia32_permvarsi256 (v8si,v8si)
10685 v4df __builtin_ia32_permdf256 (v4df,int)
10686 v8sf __builtin_ia32_permvarsf256 (v8sf,v8sf)
10687 v4di __builtin_ia32_permdi256 (v4di,int)
10688 v4di __builtin_ia32_permti256 (v4di,v4di,int)
10689 v4di __builtin_ia32_extract128i256 (v4di,int)
10690 v4di __builtin_ia32_insert128i256 (v4di,v2di,int)
10691 v8si __builtin_ia32_maskloadd256 (pcv8si,v8si)
10692 v4di __builtin_ia32_maskloadq256 (pcv4di,v4di)
10693 v4si __builtin_ia32_maskloadd (pcv4si,v4si)
10694 v2di __builtin_ia32_maskloadq (pcv2di,v2di)
10695 void __builtin_ia32_maskstored256 (pv8si,v8si,v8si)
10696 void __builtin_ia32_maskstoreq256 (pv4di,v4di,v4di)
10697 void __builtin_ia32_maskstored (pv4si,v4si,v4si)
10698 void __builtin_ia32_maskstoreq (pv2di,v2di,v2di)
10699 v8si __builtin_ia32_psllv8si (v8si,v8si)
10700 v4si __builtin_ia32_psllv4si (v4si,v4si)
10701 v4di __builtin_ia32_psllv4di (v4di,v4di)
10702 v2di __builtin_ia32_psllv2di (v2di,v2di)
10703 v8si __builtin_ia32_psrav8si (v8si,v8si)
10704 v4si __builtin_ia32_psrav4si (v4si,v4si)
10705 v8si __builtin_ia32_psrlv8si (v8si,v8si)
10706 v4si __builtin_ia32_psrlv4si (v4si,v4si)
10707 v4di __builtin_ia32_psrlv4di (v4di,v4di)
10708 v2di __builtin_ia32_psrlv2di (v2di,v2di)
10709 v2df __builtin_ia32_gathersiv2df (v2df, pcdouble,v4si,v2df,int)
10710 v4df __builtin_ia32_gathersiv4df (v4df, pcdouble,v4si,v4df,int)
10711 v2df __builtin_ia32_gatherdiv2df (v2df, pcdouble,v2di,v2df,int)
10712 v4df __builtin_ia32_gatherdiv4df (v4df, pcdouble,v4di,v4df,int)
10713 v4sf __builtin_ia32_gathersiv4sf (v4sf, pcfloat,v4si,v4sf,int)
10714 v8sf __builtin_ia32_gathersiv8sf (v8sf, pcfloat,v8si,v8sf,int)
10715 v4sf __builtin_ia32_gatherdiv4sf (v4sf, pcfloat,v2di,v4sf,int)
10716 v4sf __builtin_ia32_gatherdiv4sf256 (v4sf, pcfloat,v4di,v4sf,int)
10717 v2di __builtin_ia32_gathersiv2di (v2di, pcint64,v4si,v2di,int)
10718 v4di __builtin_ia32_gathersiv4di (v4di, pcint64,v4si,v4di,int)
10719 v2di __builtin_ia32_gatherdiv2di (v2di, pcint64,v2di,v2di,int)
10720 v4di __builtin_ia32_gatherdiv4di (v4di, pcint64,v4di,v4di,int)
10721 v4si __builtin_ia32_gathersiv4si (v4si, pcint,v4si,v4si,int)
10722 v8si __builtin_ia32_gathersiv8si (v8si, pcint,v8si,v8si,int)
10723 v4si __builtin_ia32_gatherdiv4si (v4si, pcint,v2di,v4si,int)
10724 v4si __builtin_ia32_gatherdiv4si256 (v4si, pcint,v4di,v4si,int)
10727 The following built-in functions are available when @option{-maes} is
10728 used. All of them generate the machine instruction that is part of the
10732 v2di __builtin_ia32_aesenc128 (v2di, v2di)
10733 v2di __builtin_ia32_aesenclast128 (v2di, v2di)
10734 v2di __builtin_ia32_aesdec128 (v2di, v2di)
10735 v2di __builtin_ia32_aesdeclast128 (v2di, v2di)
10736 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int)
10737 v2di __builtin_ia32_aesimc128 (v2di)
10740 The following built-in function is available when @option{-mpclmul} is
10744 @item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)
10745 Generates the @code{pclmulqdq} machine instruction.
10748 The following built-in function is available when @option{-mfsgsbase} is
10749 used. All of them generate the machine instruction that is part of the
10753 unsigned int __builtin_ia32_rdfsbase32 (void)
10754 unsigned long long __builtin_ia32_rdfsbase64 (void)
10755 unsigned int __builtin_ia32_rdgsbase32 (void)
10756 unsigned long long __builtin_ia32_rdgsbase64 (void)
10757 void _writefsbase_u32 (unsigned int)
10758 void _writefsbase_u64 (unsigned long long)
10759 void _writegsbase_u32 (unsigned int)
10760 void _writegsbase_u64 (unsigned long long)
10763 The following built-in function is available when @option{-mrdrnd} is
10764 used. All of them generate the machine instruction that is part of the
10768 unsigned int __builtin_ia32_rdrand16_step (unsigned short *)
10769 unsigned int __builtin_ia32_rdrand32_step (unsigned int *)
10770 unsigned int __builtin_ia32_rdrand64_step (unsigned long long *)
10773 The following built-in functions are available when @option{-msse4a} is used.
10774 All of them generate the machine instruction that is part of the name.
10777 void __builtin_ia32_movntsd (double *, v2df)
10778 void __builtin_ia32_movntss (float *, v4sf)
10779 v2di __builtin_ia32_extrq (v2di, v16qi)
10780 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int)
10781 v2di __builtin_ia32_insertq (v2di, v2di)
10782 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int)
10785 The following built-in functions are available when @option{-mxop} is used.
10787 v2df __builtin_ia32_vfrczpd (v2df)
10788 v4sf __builtin_ia32_vfrczps (v4sf)
10789 v2df __builtin_ia32_vfrczsd (v2df, v2df)
10790 v4sf __builtin_ia32_vfrczss (v4sf, v4sf)
10791 v4df __builtin_ia32_vfrczpd256 (v4df)
10792 v8sf __builtin_ia32_vfrczps256 (v8sf)
10793 v2di __builtin_ia32_vpcmov (v2di, v2di, v2di)
10794 v2di __builtin_ia32_vpcmov_v2di (v2di, v2di, v2di)
10795 v4si __builtin_ia32_vpcmov_v4si (v4si, v4si, v4si)
10796 v8hi __builtin_ia32_vpcmov_v8hi (v8hi, v8hi, v8hi)
10797 v16qi __builtin_ia32_vpcmov_v16qi (v16qi, v16qi, v16qi)
10798 v2df __builtin_ia32_vpcmov_v2df (v2df, v2df, v2df)
10799 v4sf __builtin_ia32_vpcmov_v4sf (v4sf, v4sf, v4sf)
10800 v4di __builtin_ia32_vpcmov_v4di256 (v4di, v4di, v4di)
10801 v8si __builtin_ia32_vpcmov_v8si256 (v8si, v8si, v8si)
10802 v16hi __builtin_ia32_vpcmov_v16hi256 (v16hi, v16hi, v16hi)
10803 v32qi __builtin_ia32_vpcmov_v32qi256 (v32qi, v32qi, v32qi)
10804 v4df __builtin_ia32_vpcmov_v4df256 (v4df, v4df, v4df)
10805 v8sf __builtin_ia32_vpcmov_v8sf256 (v8sf, v8sf, v8sf)
10806 v16qi __builtin_ia32_vpcomeqb (v16qi, v16qi)
10807 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
10808 v4si __builtin_ia32_vpcomeqd (v4si, v4si)
10809 v2di __builtin_ia32_vpcomeqq (v2di, v2di)
10810 v16qi __builtin_ia32_vpcomequb (v16qi, v16qi)
10811 v4si __builtin_ia32_vpcomequd (v4si, v4si)
10812 v2di __builtin_ia32_vpcomequq (v2di, v2di)
10813 v8hi __builtin_ia32_vpcomequw (v8hi, v8hi)
10814 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi)
10815 v16qi __builtin_ia32_vpcomfalseb (v16qi, v16qi)
10816 v4si __builtin_ia32_vpcomfalsed (v4si, v4si)
10817 v2di __builtin_ia32_vpcomfalseq (v2di, v2di)
10818 v16qi __builtin_ia32_vpcomfalseub (v16qi, v16qi)
10819 v4si __builtin_ia32_vpcomfalseud (v4si, v4si)
10820 v2di __builtin_ia32_vpcomfalseuq (v2di, v2di)
10821 v8hi __builtin_ia32_vpcomfalseuw (v8hi, v8hi)
10822 v8hi __builtin_ia32_vpcomfalsew (v8hi, v8hi)
10823 v16qi __builtin_ia32_vpcomgeb (v16qi, v16qi)
10824 v4si __builtin_ia32_vpcomged (v4si, v4si)
10825 v2di __builtin_ia32_vpcomgeq (v2di, v2di)
10826 v16qi __builtin_ia32_vpcomgeub (v16qi, v16qi)
10827 v4si __builtin_ia32_vpcomgeud (v4si, v4si)
10828 v2di __builtin_ia32_vpcomgeuq (v2di, v2di)
10829 v8hi __builtin_ia32_vpcomgeuw (v8hi, v8hi)
10830 v8hi __builtin_ia32_vpcomgew (v8hi, v8hi)
10831 v16qi __builtin_ia32_vpcomgtb (v16qi, v16qi)
10832 v4si __builtin_ia32_vpcomgtd (v4si, v4si)
10833 v2di __builtin_ia32_vpcomgtq (v2di, v2di)
10834 v16qi __builtin_ia32_vpcomgtub (v16qi, v16qi)
10835 v4si __builtin_ia32_vpcomgtud (v4si, v4si)
10836 v2di __builtin_ia32_vpcomgtuq (v2di, v2di)
10837 v8hi __builtin_ia32_vpcomgtuw (v8hi, v8hi)
10838 v8hi __builtin_ia32_vpcomgtw (v8hi, v8hi)
10839 v16qi __builtin_ia32_vpcomleb (v16qi, v16qi)
10840 v4si __builtin_ia32_vpcomled (v4si, v4si)
10841 v2di __builtin_ia32_vpcomleq (v2di, v2di)
10842 v16qi __builtin_ia32_vpcomleub (v16qi, v16qi)
10843 v4si __builtin_ia32_vpcomleud (v4si, v4si)
10844 v2di __builtin_ia32_vpcomleuq (v2di, v2di)
10845 v8hi __builtin_ia32_vpcomleuw (v8hi, v8hi)
10846 v8hi __builtin_ia32_vpcomlew (v8hi, v8hi)
10847 v16qi __builtin_ia32_vpcomltb (v16qi, v16qi)
10848 v4si __builtin_ia32_vpcomltd (v4si, v4si)
10849 v2di __builtin_ia32_vpcomltq (v2di, v2di)
10850 v16qi __builtin_ia32_vpcomltub (v16qi, v16qi)
10851 v4si __builtin_ia32_vpcomltud (v4si, v4si)
10852 v2di __builtin_ia32_vpcomltuq (v2di, v2di)
10853 v8hi __builtin_ia32_vpcomltuw (v8hi, v8hi)
10854 v8hi __builtin_ia32_vpcomltw (v8hi, v8hi)
10855 v16qi __builtin_ia32_vpcomneb (v16qi, v16qi)
10856 v4si __builtin_ia32_vpcomned (v4si, v4si)
10857 v2di __builtin_ia32_vpcomneq (v2di, v2di)
10858 v16qi __builtin_ia32_vpcomneub (v16qi, v16qi)
10859 v4si __builtin_ia32_vpcomneud (v4si, v4si)
10860 v2di __builtin_ia32_vpcomneuq (v2di, v2di)
10861 v8hi __builtin_ia32_vpcomneuw (v8hi, v8hi)
10862 v8hi __builtin_ia32_vpcomnew (v8hi, v8hi)
10863 v16qi __builtin_ia32_vpcomtrueb (v16qi, v16qi)
10864 v4si __builtin_ia32_vpcomtrued (v4si, v4si)
10865 v2di __builtin_ia32_vpcomtrueq (v2di, v2di)
10866 v16qi __builtin_ia32_vpcomtrueub (v16qi, v16qi)
10867 v4si __builtin_ia32_vpcomtrueud (v4si, v4si)
10868 v2di __builtin_ia32_vpcomtrueuq (v2di, v2di)
10869 v8hi __builtin_ia32_vpcomtrueuw (v8hi, v8hi)
10870 v8hi __builtin_ia32_vpcomtruew (v8hi, v8hi)
10871 v4si __builtin_ia32_vphaddbd (v16qi)
10872 v2di __builtin_ia32_vphaddbq (v16qi)
10873 v8hi __builtin_ia32_vphaddbw (v16qi)
10874 v2di __builtin_ia32_vphadddq (v4si)
10875 v4si __builtin_ia32_vphaddubd (v16qi)
10876 v2di __builtin_ia32_vphaddubq (v16qi)
10877 v8hi __builtin_ia32_vphaddubw (v16qi)
10878 v2di __builtin_ia32_vphaddudq (v4si)
10879 v4si __builtin_ia32_vphadduwd (v8hi)
10880 v2di __builtin_ia32_vphadduwq (v8hi)
10881 v4si __builtin_ia32_vphaddwd (v8hi)
10882 v2di __builtin_ia32_vphaddwq (v8hi)
10883 v8hi __builtin_ia32_vphsubbw (v16qi)
10884 v2di __builtin_ia32_vphsubdq (v4si)
10885 v4si __builtin_ia32_vphsubwd (v8hi)
10886 v4si __builtin_ia32_vpmacsdd (v4si, v4si, v4si)
10887 v2di __builtin_ia32_vpmacsdqh (v4si, v4si, v2di)
10888 v2di __builtin_ia32_vpmacsdql (v4si, v4si, v2di)
10889 v4si __builtin_ia32_vpmacssdd (v4si, v4si, v4si)
10890 v2di __builtin_ia32_vpmacssdqh (v4si, v4si, v2di)
10891 v2di __builtin_ia32_vpmacssdql (v4si, v4si, v2di)
10892 v4si __builtin_ia32_vpmacsswd (v8hi, v8hi, v4si)
10893 v8hi __builtin_ia32_vpmacssww (v8hi, v8hi, v8hi)
10894 v4si __builtin_ia32_vpmacswd (v8hi, v8hi, v4si)
10895 v8hi __builtin_ia32_vpmacsww (v8hi, v8hi, v8hi)
10896 v4si __builtin_ia32_vpmadcsswd (v8hi, v8hi, v4si)
10897 v4si __builtin_ia32_vpmadcswd (v8hi, v8hi, v4si)
10898 v16qi __builtin_ia32_vpperm (v16qi, v16qi, v16qi)
10899 v16qi __builtin_ia32_vprotb (v16qi, v16qi)
10900 v4si __builtin_ia32_vprotd (v4si, v4si)
10901 v2di __builtin_ia32_vprotq (v2di, v2di)
10902 v8hi __builtin_ia32_vprotw (v8hi, v8hi)
10903 v16qi __builtin_ia32_vpshab (v16qi, v16qi)
10904 v4si __builtin_ia32_vpshad (v4si, v4si)
10905 v2di __builtin_ia32_vpshaq (v2di, v2di)
10906 v8hi __builtin_ia32_vpshaw (v8hi, v8hi)
10907 v16qi __builtin_ia32_vpshlb (v16qi, v16qi)
10908 v4si __builtin_ia32_vpshld (v4si, v4si)
10909 v2di __builtin_ia32_vpshlq (v2di, v2di)
10910 v8hi __builtin_ia32_vpshlw (v8hi, v8hi)
10913 The following built-in functions are available when @option{-mfma4} is used.
10914 All of them generate the machine instruction that is part of the name.
10917 v2df __builtin_ia32_vfmaddpd (v2df, v2df, v2df)
10918 v4sf __builtin_ia32_vfmaddps (v4sf, v4sf, v4sf)
10919 v2df __builtin_ia32_vfmaddsd (v2df, v2df, v2df)
10920 v4sf __builtin_ia32_vfmaddss (v4sf, v4sf, v4sf)
10921 v2df __builtin_ia32_vfmsubpd (v2df, v2df, v2df)
10922 v4sf __builtin_ia32_vfmsubps (v4sf, v4sf, v4sf)
10923 v2df __builtin_ia32_vfmsubsd (v2df, v2df, v2df)
10924 v4sf __builtin_ia32_vfmsubss (v4sf, v4sf, v4sf)
10925 v2df __builtin_ia32_vfnmaddpd (v2df, v2df, v2df)
10926 v4sf __builtin_ia32_vfnmaddps (v4sf, v4sf, v4sf)
10927 v2df __builtin_ia32_vfnmaddsd (v2df, v2df, v2df)
10928 v4sf __builtin_ia32_vfnmaddss (v4sf, v4sf, v4sf)
10929 v2df __builtin_ia32_vfnmsubpd (v2df, v2df, v2df)
10930 v4sf __builtin_ia32_vfnmsubps (v4sf, v4sf, v4sf)
10931 v2df __builtin_ia32_vfnmsubsd (v2df, v2df, v2df)
10932 v4sf __builtin_ia32_vfnmsubss (v4sf, v4sf, v4sf)
10933 v2df __builtin_ia32_vfmaddsubpd (v2df, v2df, v2df)
10934 v4sf __builtin_ia32_vfmaddsubps (v4sf, v4sf, v4sf)
10935 v2df __builtin_ia32_vfmsubaddpd (v2df, v2df, v2df)
10936 v4sf __builtin_ia32_vfmsubaddps (v4sf, v4sf, v4sf)
10937 v4df __builtin_ia32_vfmaddpd256 (v4df, v4df, v4df)
10938 v8sf __builtin_ia32_vfmaddps256 (v8sf, v8sf, v8sf)
10939 v4df __builtin_ia32_vfmsubpd256 (v4df, v4df, v4df)
10940 v8sf __builtin_ia32_vfmsubps256 (v8sf, v8sf, v8sf)
10941 v4df __builtin_ia32_vfnmaddpd256 (v4df, v4df, v4df)
10942 v8sf __builtin_ia32_vfnmaddps256 (v8sf, v8sf, v8sf)
10943 v4df __builtin_ia32_vfnmsubpd256 (v4df, v4df, v4df)
10944 v8sf __builtin_ia32_vfnmsubps256 (v8sf, v8sf, v8sf)
10945 v4df __builtin_ia32_vfmaddsubpd256 (v4df, v4df, v4df)
10946 v8sf __builtin_ia32_vfmaddsubps256 (v8sf, v8sf, v8sf)
10947 v4df __builtin_ia32_vfmsubaddpd256 (v4df, v4df, v4df)
10948 v8sf __builtin_ia32_vfmsubaddps256 (v8sf, v8sf, v8sf)
10952 The following built-in functions are available when @option{-mlwp} is used.
10955 void __builtin_ia32_llwpcb16 (void *);
10956 void __builtin_ia32_llwpcb32 (void *);
10957 void __builtin_ia32_llwpcb64 (void *);
10958 void * __builtin_ia32_llwpcb16 (void);
10959 void * __builtin_ia32_llwpcb32 (void);
10960 void * __builtin_ia32_llwpcb64 (void);
10961 void __builtin_ia32_lwpval16 (unsigned short, unsigned int, unsigned short)
10962 void __builtin_ia32_lwpval32 (unsigned int, unsigned int, unsigned int)
10963 void __builtin_ia32_lwpval64 (unsigned __int64, unsigned int, unsigned int)
10964 unsigned char __builtin_ia32_lwpins16 (unsigned short, unsigned int, unsigned short)
10965 unsigned char __builtin_ia32_lwpins32 (unsigned int, unsigned int, unsigned int)
10966 unsigned char __builtin_ia32_lwpins64 (unsigned __int64, unsigned int, unsigned int)
10969 The following built-in functions are available when @option{-mbmi} is used.
10970 All of them generate the machine instruction that is part of the name.
10972 unsigned int __builtin_ia32_bextr_u32(unsigned int, unsigned int);
10973 unsigned long long __builtin_ia32_bextr_u64 (unsigned long long, unsigned long long);
10976 The following built-in functions are available when @option{-mbmi2} is used.
10977 All of them generate the machine instruction that is part of the name.
10979 unsigned int _bzhi_u32 (unsigned int, unsigned int)
10980 unsigned int _pdep_u32 (unsigned int, unsigned int)
10981 unsigned int _pext_u32 (unsigned int, unsigned int)
10982 unsigned long long _bzhi_u64 (unsigned long long, unsigned long long)
10983 unsigned long long _pdep_u64 (unsigned long long, unsigned long long)
10984 unsigned long long _pext_u64 (unsigned long long, unsigned long long)
10987 The following built-in functions are available when @option{-mlzcnt} is used.
10988 All of them generate the machine instruction that is part of the name.
10990 unsigned short __builtin_ia32_lzcnt_16(unsigned short);
10991 unsigned int __builtin_ia32_lzcnt_u32(unsigned int);
10992 unsigned long long __builtin_ia32_lzcnt_u64 (unsigned long long);
10995 The following built-in functions are available when @option{-mfxsr} is used.
10996 All of them generate the machine instruction that is part of the name.
10998 void __builtin_ia32_fxsave (void *)
10999 void __builtin_ia32_fxrstor (void *)
11000 void __builtin_ia32_fxsave64 (void *)
11001 void __builtin_ia32_fxrstor64 (void *)
11004 The following built-in functions are available when @option{-mxsave} is used.
11005 All of them generate the machine instruction that is part of the name.
11007 void __builtin_ia32_xsave (void *, long long)
11008 void __builtin_ia32_xrstor (void *, long long)
11009 void __builtin_ia32_xsave64 (void *, long long)
11010 void __builtin_ia32_xrstor64 (void *, long long)
11013 The following built-in functions are available when @option{-mxsaveopt} is used.
11014 All of them generate the machine instruction that is part of the name.
11016 void __builtin_ia32_xsaveopt (void *, long long)
11017 void __builtin_ia32_xsaveopt64 (void *, long long)
11020 The following built-in functions are available when @option{-mtbm} is used.
11021 Both of them generate the immediate form of the bextr machine instruction.
11023 unsigned int __builtin_ia32_bextri_u32 (unsigned int, const unsigned int);
11024 unsigned long long __builtin_ia32_bextri_u64 (unsigned long long, const unsigned long long);
11028 The following built-in functions are available when @option{-m3dnow} is used.
11029 All of them generate the machine instruction that is part of the name.
11032 void __builtin_ia32_femms (void)
11033 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
11034 v2si __builtin_ia32_pf2id (v2sf)
11035 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
11036 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
11037 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
11038 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
11039 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
11040 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
11041 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
11042 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
11043 v2sf __builtin_ia32_pfrcp (v2sf)
11044 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
11045 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
11046 v2sf __builtin_ia32_pfrsqrt (v2sf)
11047 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
11048 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
11049 v2sf __builtin_ia32_pi2fd (v2si)
11050 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
11053 The following built-in functions are available when both @option{-m3dnow}
11054 and @option{-march=athlon} are used. All of them generate the machine
11055 instruction that is part of the name.
11058 v2si __builtin_ia32_pf2iw (v2sf)
11059 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
11060 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
11061 v2sf __builtin_ia32_pi2fw (v2si)
11062 v2sf __builtin_ia32_pswapdsf (v2sf)
11063 v2si __builtin_ia32_pswapdsi (v2si)
11066 The following built-in functions are available when @option{-mrtm} is used
11067 They are used for restricted transactional memory. These are the internal
11068 low level functions. Normally the functions in
11069 @ref{X86 transactional memory intrinsics} should be used instead.
11072 int __builtin_ia32_xbegin ()
11073 void __builtin_ia32_xend ()
11074 void __builtin_ia32_xabort (status)
11075 int __builtin_ia32_xtest ()
11078 @node X86 transactional memory intrinsics
11079 @subsection X86 transaction memory intrinsics
11081 Hardware transactional memory intrinsics for i386. These allow to use
11082 memory transactions with RTM (Restricted Transactional Memory).
11083 For using HLE (Hardware Lock Elision) see @ref{x86 specific memory model extensions for transactional memory} instead.
11084 This support is enabled with the @option{-mrtm} option.
11086 A memory transaction commits all changes to memory in an atomic way,
11087 as visible to other threads. If the transaction fails it is rolled back
11088 and all side effects discarded.
11090 Generally there is no guarantee that a memory transaction ever succeeds
11091 and suitable fallback code always needs to be supplied.
11093 @deftypefn {RTM Function} {unsigned} _xbegin ()
11094 Start a RTM (Restricted Transactional Memory) transaction.
11095 Returns _XBEGIN_STARTED when the transaction
11096 started successfully (note this is not 0, so the constant has to be
11097 explicitely tested). When the transaction aborts all side effects
11098 are undone and an abort code is returned. There is no guarantee
11099 any transaction ever succeeds, so there always needs to be a valid
11100 tested fallback path.
11104 #include <immintrin.h>
11106 if ((status = _xbegin ()) == _XBEGIN_STARTED) @{
11107 ... transaction code...
11110 ... non transactional fallback path...
11114 Valid abort status bits (when the value is not @code{_XBEGIN_STARTED}) are:
11117 @item _XABORT_EXPLICIT
11118 Transaction explicitely aborted with @code{_xabort}. The parameter passed
11119 to @code{_xabort} is available with @code{_XABORT_CODE(status)}
11120 @item _XABORT_RETRY
11121 Transaction retry is possible.
11122 @item _XABORT_CONFLICT
11123 Transaction abort due to a memory conflict with another thread
11124 @item _XABORT_CAPACITY
11125 Transaction abort due to the transaction using too much memory
11126 @item _XABORT_DEBUG
11127 Transaction abort due to a debug trap
11128 @item _XABORT_NESTED
11129 Transaction abort in a inner nested transaction
11132 @deftypefn {RTM Function} {void} _xend ()
11133 Commit the current transaction. When no transaction is active this will
11134 fault. All memory side effects of the transactions will become visible
11135 to other threads in an atomic matter.
11138 @deftypefn {RTM Function} {int} _xtest ()
11139 Return a value not zero when a transaction is currently active, otherwise 0.
11142 @deftypefn {RTM Function} {void} _xabort (status)
11143 Abort the current transaction. When no transaction is active this is a no-op.
11144 status must be a 8bit constant, that is included in the status code returned
11148 @node MIPS DSP Built-in Functions
11149 @subsection MIPS DSP Built-in Functions
11151 The MIPS DSP Application-Specific Extension (ASE) includes new
11152 instructions that are designed to improve the performance of DSP and
11153 media applications. It provides instructions that operate on packed
11154 8-bit/16-bit integer data, Q7, Q15 and Q31 fractional data.
11156 GCC supports MIPS DSP operations using both the generic
11157 vector extensions (@pxref{Vector Extensions}) and a collection of
11158 MIPS-specific built-in functions. Both kinds of support are
11159 enabled by the @option{-mdsp} command-line option.
11161 Revision 2 of the ASE was introduced in the second half of 2006.
11162 This revision adds extra instructions to the original ASE, but is
11163 otherwise backwards-compatible with it. You can select revision 2
11164 using the command-line option @option{-mdspr2}; this option implies
11167 The SCOUNT and POS bits of the DSP control register are global. The
11168 WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and
11169 POS bits. During optimization, the compiler does not delete these
11170 instructions and it does not delete calls to functions containing
11171 these instructions.
11173 At present, GCC only provides support for operations on 32-bit
11174 vectors. The vector type associated with 8-bit integer data is
11175 usually called @code{v4i8}, the vector type associated with Q7
11176 is usually called @code{v4q7}, the vector type associated with 16-bit
11177 integer data is usually called @code{v2i16}, and the vector type
11178 associated with Q15 is usually called @code{v2q15}. They can be
11179 defined in C as follows:
11182 typedef signed char v4i8 __attribute__ ((vector_size(4)));
11183 typedef signed char v4q7 __attribute__ ((vector_size(4)));
11184 typedef short v2i16 __attribute__ ((vector_size(4)));
11185 typedef short v2q15 __attribute__ ((vector_size(4)));
11188 @code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are
11189 initialized in the same way as aggregates. For example:
11192 v4i8 a = @{1, 2, 3, 4@};
11194 b = (v4i8) @{5, 6, 7, 8@};
11196 v2q15 c = @{0x0fcb, 0x3a75@};
11198 d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@};
11201 @emph{Note:} The CPU's endianness determines the order in which values
11202 are packed. On little-endian targets, the first value is the least
11203 significant and the last value is the most significant. The opposite
11204 order applies to big-endian targets. For example, the code above
11205 sets the lowest byte of @code{a} to @code{1} on little-endian targets
11206 and @code{4} on big-endian targets.
11208 @emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer
11209 representation. As shown in this example, the integer representation
11210 of a Q7 value can be obtained by multiplying the fractional value by
11211 @code{0x1.0p7}. The equivalent for Q15 values is to multiply by
11212 @code{0x1.0p15}. The equivalent for Q31 values is to multiply by
11215 The table below lists the @code{v4i8} and @code{v2q15} operations for which
11216 hardware support exists. @code{a} and @code{b} are @code{v4i8} values,
11217 and @code{c} and @code{d} are @code{v2q15} values.
11219 @multitable @columnfractions .50 .50
11220 @item C code @tab MIPS instruction
11221 @item @code{a + b} @tab @code{addu.qb}
11222 @item @code{c + d} @tab @code{addq.ph}
11223 @item @code{a - b} @tab @code{subu.qb}
11224 @item @code{c - d} @tab @code{subq.ph}
11227 The table below lists the @code{v2i16} operation for which
11228 hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are
11229 @code{v2i16} values.
11231 @multitable @columnfractions .50 .50
11232 @item C code @tab MIPS instruction
11233 @item @code{e * f} @tab @code{mul.ph}
11236 It is easier to describe the DSP built-in functions if we first define
11237 the following types:
11242 typedef unsigned int ui32;
11243 typedef long long a64;
11246 @code{q31} and @code{i32} are actually the same as @code{int}, but we
11247 use @code{q31} to indicate a Q31 fractional value and @code{i32} to
11248 indicate a 32-bit integer value. Similarly, @code{a64} is the same as
11249 @code{long long}, but we use @code{a64} to indicate values that are
11250 placed in one of the four DSP accumulators (@code{$ac0},
11251 @code{$ac1}, @code{$ac2} or @code{$ac3}).
11253 Also, some built-in functions prefer or require immediate numbers as
11254 parameters, because the corresponding DSP instructions accept both immediate
11255 numbers and register operands, or accept immediate numbers only. The
11256 immediate parameters are listed as follows.
11264 imm0_255: 0 to 255.
11265 imm_n32_31: -32 to 31.
11266 imm_n512_511: -512 to 511.
11269 The following built-in functions map directly to a particular MIPS DSP
11270 instruction. Please refer to the architecture specification
11271 for details on what each instruction does.
11274 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
11275 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
11276 q31 __builtin_mips_addq_s_w (q31, q31)
11277 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
11278 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
11279 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
11280 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
11281 q31 __builtin_mips_subq_s_w (q31, q31)
11282 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
11283 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
11284 i32 __builtin_mips_addsc (i32, i32)
11285 i32 __builtin_mips_addwc (i32, i32)
11286 i32 __builtin_mips_modsub (i32, i32)
11287 i32 __builtin_mips_raddu_w_qb (v4i8)
11288 v2q15 __builtin_mips_absq_s_ph (v2q15)
11289 q31 __builtin_mips_absq_s_w (q31)
11290 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
11291 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
11292 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
11293 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
11294 q31 __builtin_mips_preceq_w_phl (v2q15)
11295 q31 __builtin_mips_preceq_w_phr (v2q15)
11296 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
11297 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
11298 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
11299 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
11300 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
11301 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
11302 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
11303 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
11304 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
11305 v4i8 __builtin_mips_shll_qb (v4i8, i32)
11306 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
11307 v2q15 __builtin_mips_shll_ph (v2q15, i32)
11308 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
11309 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
11310 q31 __builtin_mips_shll_s_w (q31, imm0_31)
11311 q31 __builtin_mips_shll_s_w (q31, i32)
11312 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
11313 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
11314 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
11315 v2q15 __builtin_mips_shra_ph (v2q15, i32)
11316 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
11317 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
11318 q31 __builtin_mips_shra_r_w (q31, imm0_31)
11319 q31 __builtin_mips_shra_r_w (q31, i32)
11320 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
11321 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
11322 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
11323 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
11324 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
11325 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
11326 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
11327 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
11328 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
11329 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
11330 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
11331 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
11332 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
11333 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
11334 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
11335 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
11336 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
11337 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
11338 i32 __builtin_mips_bitrev (i32)
11339 i32 __builtin_mips_insv (i32, i32)
11340 v4i8 __builtin_mips_repl_qb (imm0_255)
11341 v4i8 __builtin_mips_repl_qb (i32)
11342 v2q15 __builtin_mips_repl_ph (imm_n512_511)
11343 v2q15 __builtin_mips_repl_ph (i32)
11344 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
11345 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
11346 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
11347 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
11348 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
11349 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
11350 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
11351 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
11352 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
11353 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
11354 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
11355 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
11356 i32 __builtin_mips_extr_w (a64, imm0_31)
11357 i32 __builtin_mips_extr_w (a64, i32)
11358 i32 __builtin_mips_extr_r_w (a64, imm0_31)
11359 i32 __builtin_mips_extr_s_h (a64, i32)
11360 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
11361 i32 __builtin_mips_extr_rs_w (a64, i32)
11362 i32 __builtin_mips_extr_s_h (a64, imm0_31)
11363 i32 __builtin_mips_extr_r_w (a64, i32)
11364 i32 __builtin_mips_extp (a64, imm0_31)
11365 i32 __builtin_mips_extp (a64, i32)
11366 i32 __builtin_mips_extpdp (a64, imm0_31)
11367 i32 __builtin_mips_extpdp (a64, i32)
11368 a64 __builtin_mips_shilo (a64, imm_n32_31)
11369 a64 __builtin_mips_shilo (a64, i32)
11370 a64 __builtin_mips_mthlip (a64, i32)
11371 void __builtin_mips_wrdsp (i32, imm0_63)
11372 i32 __builtin_mips_rddsp (imm0_63)
11373 i32 __builtin_mips_lbux (void *, i32)
11374 i32 __builtin_mips_lhx (void *, i32)
11375 i32 __builtin_mips_lwx (void *, i32)
11376 a64 __builtin_mips_ldx (void *, i32) [MIPS64 only]
11377 i32 __builtin_mips_bposge32 (void)
11378 a64 __builtin_mips_madd (a64, i32, i32);
11379 a64 __builtin_mips_maddu (a64, ui32, ui32);
11380 a64 __builtin_mips_msub (a64, i32, i32);
11381 a64 __builtin_mips_msubu (a64, ui32, ui32);
11382 a64 __builtin_mips_mult (i32, i32);
11383 a64 __builtin_mips_multu (ui32, ui32);
11386 The following built-in functions map directly to a particular MIPS DSP REV 2
11387 instruction. Please refer to the architecture specification
11388 for details on what each instruction does.
11391 v4q7 __builtin_mips_absq_s_qb (v4q7);
11392 v2i16 __builtin_mips_addu_ph (v2i16, v2i16);
11393 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16);
11394 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8);
11395 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8);
11396 i32 __builtin_mips_append (i32, i32, imm0_31);
11397 i32 __builtin_mips_balign (i32, i32, imm0_3);
11398 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8);
11399 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8);
11400 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8);
11401 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16);
11402 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16);
11403 v2i16 __builtin_mips_mul_ph (v2i16, v2i16);
11404 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16);
11405 q31 __builtin_mips_mulq_rs_w (q31, q31);
11406 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15);
11407 q31 __builtin_mips_mulq_s_w (q31, q31);
11408 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16);
11409 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16);
11410 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31);
11411 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31);
11412 i32 __builtin_mips_prepend (i32, i32, imm0_31);
11413 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7);
11414 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7);
11415 v4i8 __builtin_mips_shra_qb (v4i8, i32);
11416 v4i8 __builtin_mips_shra_r_qb (v4i8, i32);
11417 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15);
11418 v2i16 __builtin_mips_shrl_ph (v2i16, i32);
11419 v2i16 __builtin_mips_subu_ph (v2i16, v2i16);
11420 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16);
11421 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8);
11422 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8);
11423 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15);
11424 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15);
11425 q31 __builtin_mips_addqh_w (q31, q31);
11426 q31 __builtin_mips_addqh_r_w (q31, q31);
11427 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15);
11428 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15);
11429 q31 __builtin_mips_subqh_w (q31, q31);
11430 q31 __builtin_mips_subqh_r_w (q31, q31);
11431 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16);
11432 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16);
11433 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15);
11434 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15);
11435 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15);
11436 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15);
11440 @node MIPS Paired-Single Support
11441 @subsection MIPS Paired-Single Support
11443 The MIPS64 architecture includes a number of instructions that
11444 operate on pairs of single-precision floating-point values.
11445 Each pair is packed into a 64-bit floating-point register,
11446 with one element being designated the ``upper half'' and
11447 the other being designated the ``lower half''.
11449 GCC supports paired-single operations using both the generic
11450 vector extensions (@pxref{Vector Extensions}) and a collection of
11451 MIPS-specific built-in functions. Both kinds of support are
11452 enabled by the @option{-mpaired-single} command-line option.
11454 The vector type associated with paired-single values is usually
11455 called @code{v2sf}. It can be defined in C as follows:
11458 typedef float v2sf __attribute__ ((vector_size (8)));
11461 @code{v2sf} values are initialized in the same way as aggregates.
11465 v2sf a = @{1.5, 9.1@};
11468 b = (v2sf) @{e, f@};
11471 @emph{Note:} The CPU's endianness determines which value is stored in
11472 the upper half of a register and which value is stored in the lower half.
11473 On little-endian targets, the first value is the lower one and the second
11474 value is the upper one. The opposite order applies to big-endian targets.
11475 For example, the code above sets the lower half of @code{a} to
11476 @code{1.5} on little-endian targets and @code{9.1} on big-endian targets.
11478 @node MIPS Loongson Built-in Functions
11479 @subsection MIPS Loongson Built-in Functions
11481 GCC provides intrinsics to access the SIMD instructions provided by the
11482 ST Microelectronics Loongson-2E and -2F processors. These intrinsics,
11483 available after inclusion of the @code{loongson.h} header file,
11484 operate on the following 64-bit vector types:
11487 @item @code{uint8x8_t}, a vector of eight unsigned 8-bit integers;
11488 @item @code{uint16x4_t}, a vector of four unsigned 16-bit integers;
11489 @item @code{uint32x2_t}, a vector of two unsigned 32-bit integers;
11490 @item @code{int8x8_t}, a vector of eight signed 8-bit integers;
11491 @item @code{int16x4_t}, a vector of four signed 16-bit integers;
11492 @item @code{int32x2_t}, a vector of two signed 32-bit integers.
11495 The intrinsics provided are listed below; each is named after the
11496 machine instruction to which it corresponds, with suffixes added as
11497 appropriate to distinguish intrinsics that expand to the same machine
11498 instruction yet have different argument types. Refer to the architecture
11499 documentation for a description of the functionality of each
11503 int16x4_t packsswh (int32x2_t s, int32x2_t t);
11504 int8x8_t packsshb (int16x4_t s, int16x4_t t);
11505 uint8x8_t packushb (uint16x4_t s, uint16x4_t t);
11506 uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t);
11507 uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t);
11508 uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t);
11509 int32x2_t paddw_s (int32x2_t s, int32x2_t t);
11510 int16x4_t paddh_s (int16x4_t s, int16x4_t t);
11511 int8x8_t paddb_s (int8x8_t s, int8x8_t t);
11512 uint64_t paddd_u (uint64_t s, uint64_t t);
11513 int64_t paddd_s (int64_t s, int64_t t);
11514 int16x4_t paddsh (int16x4_t s, int16x4_t t);
11515 int8x8_t paddsb (int8x8_t s, int8x8_t t);
11516 uint16x4_t paddush (uint16x4_t s, uint16x4_t t);
11517 uint8x8_t paddusb (uint8x8_t s, uint8x8_t t);
11518 uint64_t pandn_ud (uint64_t s, uint64_t t);
11519 uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t);
11520 uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t);
11521 uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t);
11522 int64_t pandn_sd (int64_t s, int64_t t);
11523 int32x2_t pandn_sw (int32x2_t s, int32x2_t t);
11524 int16x4_t pandn_sh (int16x4_t s, int16x4_t t);
11525 int8x8_t pandn_sb (int8x8_t s, int8x8_t t);
11526 uint16x4_t pavgh (uint16x4_t s, uint16x4_t t);
11527 uint8x8_t pavgb (uint8x8_t s, uint8x8_t t);
11528 uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t);
11529 uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t);
11530 uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t);
11531 int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t);
11532 int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t);
11533 int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t);
11534 uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t);
11535 uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t);
11536 uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t);
11537 int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t);
11538 int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t);
11539 int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t);
11540 uint16x4_t pextrh_u (uint16x4_t s, int field);
11541 int16x4_t pextrh_s (int16x4_t s, int field);
11542 uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t);
11543 uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t);
11544 uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t);
11545 uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t);
11546 int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t);
11547 int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t);
11548 int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t);
11549 int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t);
11550 int32x2_t pmaddhw (int16x4_t s, int16x4_t t);
11551 int16x4_t pmaxsh (int16x4_t s, int16x4_t t);
11552 uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t);
11553 int16x4_t pminsh (int16x4_t s, int16x4_t t);
11554 uint8x8_t pminub (uint8x8_t s, uint8x8_t t);
11555 uint8x8_t pmovmskb_u (uint8x8_t s);
11556 int8x8_t pmovmskb_s (int8x8_t s);
11557 uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t);
11558 int16x4_t pmulhh (int16x4_t s, int16x4_t t);
11559 int16x4_t pmullh (int16x4_t s, int16x4_t t);
11560 int64_t pmuluw (uint32x2_t s, uint32x2_t t);
11561 uint8x8_t pasubub (uint8x8_t s, uint8x8_t t);
11562 uint16x4_t biadd (uint8x8_t s);
11563 uint16x4_t psadbh (uint8x8_t s, uint8x8_t t);
11564 uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order);
11565 int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order);
11566 uint16x4_t psllh_u (uint16x4_t s, uint8_t amount);
11567 int16x4_t psllh_s (int16x4_t s, uint8_t amount);
11568 uint32x2_t psllw_u (uint32x2_t s, uint8_t amount);
11569 int32x2_t psllw_s (int32x2_t s, uint8_t amount);
11570 uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount);
11571 int16x4_t psrlh_s (int16x4_t s, uint8_t amount);
11572 uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount);
11573 int32x2_t psrlw_s (int32x2_t s, uint8_t amount);
11574 uint16x4_t psrah_u (uint16x4_t s, uint8_t amount);
11575 int16x4_t psrah_s (int16x4_t s, uint8_t amount);
11576 uint32x2_t psraw_u (uint32x2_t s, uint8_t amount);
11577 int32x2_t psraw_s (int32x2_t s, uint8_t amount);
11578 uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t);
11579 uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t);
11580 uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t);
11581 int32x2_t psubw_s (int32x2_t s, int32x2_t t);
11582 int16x4_t psubh_s (int16x4_t s, int16x4_t t);
11583 int8x8_t psubb_s (int8x8_t s, int8x8_t t);
11584 uint64_t psubd_u (uint64_t s, uint64_t t);
11585 int64_t psubd_s (int64_t s, int64_t t);
11586 int16x4_t psubsh (int16x4_t s, int16x4_t t);
11587 int8x8_t psubsb (int8x8_t s, int8x8_t t);
11588 uint16x4_t psubush (uint16x4_t s, uint16x4_t t);
11589 uint8x8_t psubusb (uint8x8_t s, uint8x8_t t);
11590 uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t);
11591 uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t);
11592 uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t);
11593 int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t);
11594 int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t);
11595 int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t);
11596 uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t);
11597 uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t);
11598 uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t);
11599 int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t);
11600 int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t);
11601 int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t);
11605 * Paired-Single Arithmetic::
11606 * Paired-Single Built-in Functions::
11607 * MIPS-3D Built-in Functions::
11610 @node Paired-Single Arithmetic
11611 @subsubsection Paired-Single Arithmetic
11613 The table below lists the @code{v2sf} operations for which hardware
11614 support exists. @code{a}, @code{b} and @code{c} are @code{v2sf}
11615 values and @code{x} is an integral value.
11617 @multitable @columnfractions .50 .50
11618 @item C code @tab MIPS instruction
11619 @item @code{a + b} @tab @code{add.ps}
11620 @item @code{a - b} @tab @code{sub.ps}
11621 @item @code{-a} @tab @code{neg.ps}
11622 @item @code{a * b} @tab @code{mul.ps}
11623 @item @code{a * b + c} @tab @code{madd.ps}
11624 @item @code{a * b - c} @tab @code{msub.ps}
11625 @item @code{-(a * b + c)} @tab @code{nmadd.ps}
11626 @item @code{-(a * b - c)} @tab @code{nmsub.ps}
11627 @item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps}
11630 Note that the multiply-accumulate instructions can be disabled
11631 using the command-line option @code{-mno-fused-madd}.
11633 @node Paired-Single Built-in Functions
11634 @subsubsection Paired-Single Built-in Functions
11636 The following paired-single functions map directly to a particular
11637 MIPS instruction. Please refer to the architecture specification
11638 for details on what each instruction does.
11641 @item v2sf __builtin_mips_pll_ps (v2sf, v2sf)
11642 Pair lower lower (@code{pll.ps}).
11644 @item v2sf __builtin_mips_pul_ps (v2sf, v2sf)
11645 Pair upper lower (@code{pul.ps}).
11647 @item v2sf __builtin_mips_plu_ps (v2sf, v2sf)
11648 Pair lower upper (@code{plu.ps}).
11650 @item v2sf __builtin_mips_puu_ps (v2sf, v2sf)
11651 Pair upper upper (@code{puu.ps}).
11653 @item v2sf __builtin_mips_cvt_ps_s (float, float)
11654 Convert pair to paired single (@code{cvt.ps.s}).
11656 @item float __builtin_mips_cvt_s_pl (v2sf)
11657 Convert pair lower to single (@code{cvt.s.pl}).
11659 @item float __builtin_mips_cvt_s_pu (v2sf)
11660 Convert pair upper to single (@code{cvt.s.pu}).
11662 @item v2sf __builtin_mips_abs_ps (v2sf)
11663 Absolute value (@code{abs.ps}).
11665 @item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)
11666 Align variable (@code{alnv.ps}).
11668 @emph{Note:} The value of the third parameter must be 0 or 4
11669 modulo 8, otherwise the result is unpredictable. Please read the
11670 instruction description for details.
11673 The following multi-instruction functions are also available.
11674 In each case, @var{cond} can be any of the 16 floating-point conditions:
11675 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
11676 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl},
11677 @code{lt}, @code{nge}, @code{le} or @code{ngt}.
11680 @item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11681 @itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11682 Conditional move based on floating-point comparison (@code{c.@var{cond}.ps},
11683 @code{movt.ps}/@code{movf.ps}).
11685 The @code{movt} functions return the value @var{x} computed by:
11688 c.@var{cond}.ps @var{cc},@var{a},@var{b}
11689 mov.ps @var{x},@var{c}
11690 movt.ps @var{x},@var{d},@var{cc}
11693 The @code{movf} functions are similar but use @code{movf.ps} instead
11696 @item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11697 @itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11698 Comparison of two paired-single values (@code{c.@var{cond}.ps},
11699 @code{bc1t}/@code{bc1f}).
11701 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
11702 and return either the upper or lower half of the result. For example:
11706 if (__builtin_mips_upper_c_eq_ps (a, b))
11707 upper_halves_are_equal ();
11709 upper_halves_are_unequal ();
11711 if (__builtin_mips_lower_c_eq_ps (a, b))
11712 lower_halves_are_equal ();
11714 lower_halves_are_unequal ();
11718 @node MIPS-3D Built-in Functions
11719 @subsubsection MIPS-3D Built-in Functions
11721 The MIPS-3D Application-Specific Extension (ASE) includes additional
11722 paired-single instructions that are designed to improve the performance
11723 of 3D graphics operations. Support for these instructions is controlled
11724 by the @option{-mips3d} command-line option.
11726 The functions listed below map directly to a particular MIPS-3D
11727 instruction. Please refer to the architecture specification for
11728 more details on what each instruction does.
11731 @item v2sf __builtin_mips_addr_ps (v2sf, v2sf)
11732 Reduction add (@code{addr.ps}).
11734 @item v2sf __builtin_mips_mulr_ps (v2sf, v2sf)
11735 Reduction multiply (@code{mulr.ps}).
11737 @item v2sf __builtin_mips_cvt_pw_ps (v2sf)
11738 Convert paired single to paired word (@code{cvt.pw.ps}).
11740 @item v2sf __builtin_mips_cvt_ps_pw (v2sf)
11741 Convert paired word to paired single (@code{cvt.ps.pw}).
11743 @item float __builtin_mips_recip1_s (float)
11744 @itemx double __builtin_mips_recip1_d (double)
11745 @itemx v2sf __builtin_mips_recip1_ps (v2sf)
11746 Reduced-precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}).
11748 @item float __builtin_mips_recip2_s (float, float)
11749 @itemx double __builtin_mips_recip2_d (double, double)
11750 @itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf)
11751 Reduced-precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}).
11753 @item float __builtin_mips_rsqrt1_s (float)
11754 @itemx double __builtin_mips_rsqrt1_d (double)
11755 @itemx v2sf __builtin_mips_rsqrt1_ps (v2sf)
11756 Reduced-precision reciprocal square root (sequence step 1)
11757 (@code{rsqrt1.@var{fmt}}).
11759 @item float __builtin_mips_rsqrt2_s (float, float)
11760 @itemx double __builtin_mips_rsqrt2_d (double, double)
11761 @itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)
11762 Reduced-precision reciprocal square root (sequence step 2)
11763 (@code{rsqrt2.@var{fmt}}).
11766 The following multi-instruction functions are also available.
11767 In each case, @var{cond} can be any of the 16 floating-point conditions:
11768 @code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult},
11769 @code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq},
11770 @code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}.
11773 @item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b})
11774 @itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b})
11775 Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}},
11776 @code{bc1t}/@code{bc1f}).
11778 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s}
11779 or @code{cabs.@var{cond}.d} and return the result as a boolean value.
11784 if (__builtin_mips_cabs_eq_s (a, b))
11790 @item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11791 @itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11792 Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps},
11793 @code{bc1t}/@code{bc1f}).
11795 These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps}
11796 and return either the upper or lower half of the result. For example:
11800 if (__builtin_mips_upper_cabs_eq_ps (a, b))
11801 upper_halves_are_equal ();
11803 upper_halves_are_unequal ();
11805 if (__builtin_mips_lower_cabs_eq_ps (a, b))
11806 lower_halves_are_equal ();
11808 lower_halves_are_unequal ();
11811 @item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11812 @itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11813 Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps},
11814 @code{movt.ps}/@code{movf.ps}).
11816 The @code{movt} functions return the value @var{x} computed by:
11819 cabs.@var{cond}.ps @var{cc},@var{a},@var{b}
11820 mov.ps @var{x},@var{c}
11821 movt.ps @var{x},@var{d},@var{cc}
11824 The @code{movf} functions are similar but use @code{movf.ps} instead
11827 @item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11828 @itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11829 @itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11830 @itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b})
11831 Comparison of two paired-single values
11832 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
11833 @code{bc1any2t}/@code{bc1any2f}).
11835 These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps}
11836 or @code{cabs.@var{cond}.ps}. The @code{any} forms return true if either
11837 result is true and the @code{all} forms return true if both results are true.
11842 if (__builtin_mips_any_c_eq_ps (a, b))
11847 if (__builtin_mips_all_c_eq_ps (a, b))
11853 @item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11854 @itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11855 @itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11856 @itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d})
11857 Comparison of four paired-single values
11858 (@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps},
11859 @code{bc1any4t}/@code{bc1any4f}).
11861 These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps}
11862 to compare @var{a} with @var{b} and to compare @var{c} with @var{d}.
11863 The @code{any} forms return true if any of the four results are true
11864 and the @code{all} forms return true if all four results are true.
11869 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
11874 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
11881 @node Other MIPS Built-in Functions
11882 @subsection Other MIPS Built-in Functions
11884 GCC provides other MIPS-specific built-in functions:
11887 @item void __builtin_mips_cache (int @var{op}, const volatile void *@var{addr})
11888 Insert a @samp{cache} instruction with operands @var{op} and @var{addr}.
11889 GCC defines the preprocessor macro @code{___GCC_HAVE_BUILTIN_MIPS_CACHE}
11890 when this function is available.
11893 @node MSP430 Built-in Functions
11894 @subsection MSP430 Built-in Functions
11896 GCC provides a couple of special builtin functions to aid in the
11897 writing of interrupt handlers in C.
11900 @item __bic_SR_register_on_exit (int @var{mask})
11901 This clears the indicated bits in the saved copy of the status register
11902 currently residing on the stack. This only works inside interrupt
11903 handlers and the changes to the status register will only take affect
11904 once the handler returns.
11906 @item __bis_SR_register_on_exit (int @var{mask})
11907 This sets the indicated bits in the saved copy of the status register
11908 currently residing on the stack. This only works inside interrupt
11909 handlers and the changes to the status register will only take affect
11910 once the handler returns.
11913 @node picoChip Built-in Functions
11914 @subsection picoChip Built-in Functions
11916 GCC provides an interface to selected machine instructions from the
11917 picoChip instruction set.
11920 @item int __builtin_sbc (int @var{value})
11921 Sign bit count. Return the number of consecutive bits in @var{value}
11922 that have the same value as the sign bit. The result is the number of
11923 leading sign bits minus one, giving the number of redundant sign bits in
11926 @item int __builtin_byteswap (int @var{value})
11927 Byte swap. Return the result of swapping the upper and lower bytes of
11930 @item int __builtin_brev (int @var{value})
11931 Bit reversal. Return the result of reversing the bits in
11932 @var{value}. Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1,
11935 @item int __builtin_adds (int @var{x}, int @var{y})
11936 Saturating addition. Return the result of adding @var{x} and @var{y},
11937 storing the value 32767 if the result overflows.
11939 @item int __builtin_subs (int @var{x}, int @var{y})
11940 Saturating subtraction. Return the result of subtracting @var{y} from
11941 @var{x}, storing the value @minus{}32768 if the result overflows.
11943 @item void __builtin_halt (void)
11944 Halt. The processor stops execution. This built-in is useful for
11945 implementing assertions.
11949 @node PowerPC Built-in Functions
11950 @subsection PowerPC Built-in Functions
11952 These built-in functions are available for the PowerPC family of
11955 float __builtin_recipdivf (float, float);
11956 float __builtin_rsqrtf (float);
11957 double __builtin_recipdiv (double, double);
11958 double __builtin_rsqrt (double);
11959 long __builtin_bpermd (long, long);
11960 uint64_t __builtin_ppc_get_timebase ();
11961 unsigned long __builtin_ppc_mftb ();
11964 The @code{vec_rsqrt}, @code{__builtin_rsqrt}, and
11965 @code{__builtin_rsqrtf} functions generate multiple instructions to
11966 implement the reciprocal sqrt functionality using reciprocal sqrt
11967 estimate instructions.
11969 The @code{__builtin_recipdiv}, and @code{__builtin_recipdivf}
11970 functions generate multiple instructions to implement division using
11971 the reciprocal estimate instructions.
11973 The @code{__builtin_ppc_get_timebase} and @code{__builtin_ppc_mftb}
11974 functions generate instructions to read the Time Base Register. The
11975 @code{__builtin_ppc_get_timebase} function may generate multiple
11976 instructions and always returns the 64 bits of the Time Base Register.
11977 The @code{__builtin_ppc_mftb} function always generates one instruction and
11978 returns the Time Base Register value as an unsigned long, throwing away
11979 the most significant word on 32-bit environments.
11981 @node PowerPC AltiVec/VSX Built-in Functions
11982 @subsection PowerPC AltiVec Built-in Functions
11984 GCC provides an interface for the PowerPC family of processors to access
11985 the AltiVec operations described in Motorola's AltiVec Programming
11986 Interface Manual. The interface is made available by including
11987 @code{<altivec.h>} and using @option{-maltivec} and
11988 @option{-mabi=altivec}. The interface supports the following vector
11992 vector unsigned char
11996 vector unsigned short
11997 vector signed short
12001 vector unsigned int
12007 If @option{-mvsx} is used the following additional vector types are
12011 vector unsigned long
12016 The long types are only implemented for 64-bit code generation, and
12017 the long type is only used in the floating point/integer conversion
12020 GCC's implementation of the high-level language interface available from
12021 C and C++ code differs from Motorola's documentation in several ways.
12026 A vector constant is a list of constant expressions within curly braces.
12029 A vector initializer requires no cast if the vector constant is of the
12030 same type as the variable it is initializing.
12033 If @code{signed} or @code{unsigned} is omitted, the signedness of the
12034 vector type is the default signedness of the base type. The default
12035 varies depending on the operating system, so a portable program should
12036 always specify the signedness.
12039 Compiling with @option{-maltivec} adds keywords @code{__vector},
12040 @code{vector}, @code{__pixel}, @code{pixel}, @code{__bool} and
12041 @code{bool}. When compiling ISO C, the context-sensitive substitution
12042 of the keywords @code{vector}, @code{pixel} and @code{bool} is
12043 disabled. To use them, you must include @code{<altivec.h>} instead.
12046 GCC allows using a @code{typedef} name as the type specifier for a
12050 For C, overloaded functions are implemented with macros so the following
12054 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo);
12058 Since @code{vec_add} is a macro, the vector constant in the example
12059 is treated as four separate arguments. Wrap the entire argument in
12060 parentheses for this to work.
12063 @emph{Note:} Only the @code{<altivec.h>} interface is supported.
12064 Internally, GCC uses built-in functions to achieve the functionality in
12065 the aforementioned header file, but they are not supported and are
12066 subject to change without notice.
12068 The following interfaces are supported for the generic and specific
12069 AltiVec operations and the AltiVec predicates. In cases where there
12070 is a direct mapping between generic and specific operations, only the
12071 generic names are shown here, although the specific operations can also
12074 Arguments that are documented as @code{const int} require literal
12075 integral values within the range required for that operation.
12078 vector signed char vec_abs (vector signed char);
12079 vector signed short vec_abs (vector signed short);
12080 vector signed int vec_abs (vector signed int);
12081 vector float vec_abs (vector float);
12083 vector signed char vec_abss (vector signed char);
12084 vector signed short vec_abss (vector signed short);
12085 vector signed int vec_abss (vector signed int);
12087 vector signed char vec_add (vector bool char, vector signed char);
12088 vector signed char vec_add (vector signed char, vector bool char);
12089 vector signed char vec_add (vector signed char, vector signed char);
12090 vector unsigned char vec_add (vector bool char, vector unsigned char);
12091 vector unsigned char vec_add (vector unsigned char, vector bool char);
12092 vector unsigned char vec_add (vector unsigned char,
12093 vector unsigned char);
12094 vector signed short vec_add (vector bool short, vector signed short);
12095 vector signed short vec_add (vector signed short, vector bool short);
12096 vector signed short vec_add (vector signed short, vector signed short);
12097 vector unsigned short vec_add (vector bool short,
12098 vector unsigned short);
12099 vector unsigned short vec_add (vector unsigned short,
12100 vector bool short);
12101 vector unsigned short vec_add (vector unsigned short,
12102 vector unsigned short);
12103 vector signed int vec_add (vector bool int, vector signed int);
12104 vector signed int vec_add (vector signed int, vector bool int);
12105 vector signed int vec_add (vector signed int, vector signed int);
12106 vector unsigned int vec_add (vector bool int, vector unsigned int);
12107 vector unsigned int vec_add (vector unsigned int, vector bool int);
12108 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
12109 vector float vec_add (vector float, vector float);
12111 vector float vec_vaddfp (vector float, vector float);
12113 vector signed int vec_vadduwm (vector bool int, vector signed int);
12114 vector signed int vec_vadduwm (vector signed int, vector bool int);
12115 vector signed int vec_vadduwm (vector signed int, vector signed int);
12116 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
12117 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
12118 vector unsigned int vec_vadduwm (vector unsigned int,
12119 vector unsigned int);
12121 vector signed short vec_vadduhm (vector bool short,
12122 vector signed short);
12123 vector signed short vec_vadduhm (vector signed short,
12124 vector bool short);
12125 vector signed short vec_vadduhm (vector signed short,
12126 vector signed short);
12127 vector unsigned short vec_vadduhm (vector bool short,
12128 vector unsigned short);
12129 vector unsigned short vec_vadduhm (vector unsigned short,
12130 vector bool short);
12131 vector unsigned short vec_vadduhm (vector unsigned short,
12132 vector unsigned short);
12134 vector signed char vec_vaddubm (vector bool char, vector signed char);
12135 vector signed char vec_vaddubm (vector signed char, vector bool char);
12136 vector signed char vec_vaddubm (vector signed char, vector signed char);
12137 vector unsigned char vec_vaddubm (vector bool char,
12138 vector unsigned char);
12139 vector unsigned char vec_vaddubm (vector unsigned char,
12141 vector unsigned char vec_vaddubm (vector unsigned char,
12142 vector unsigned char);
12144 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
12146 vector unsigned char vec_adds (vector bool char, vector unsigned char);
12147 vector unsigned char vec_adds (vector unsigned char, vector bool char);
12148 vector unsigned char vec_adds (vector unsigned char,
12149 vector unsigned char);
12150 vector signed char vec_adds (vector bool char, vector signed char);
12151 vector signed char vec_adds (vector signed char, vector bool char);
12152 vector signed char vec_adds (vector signed char, vector signed char);
12153 vector unsigned short vec_adds (vector bool short,
12154 vector unsigned short);
12155 vector unsigned short vec_adds (vector unsigned short,
12156 vector bool short);
12157 vector unsigned short vec_adds (vector unsigned short,
12158 vector unsigned short);
12159 vector signed short vec_adds (vector bool short, vector signed short);
12160 vector signed short vec_adds (vector signed short, vector bool short);
12161 vector signed short vec_adds (vector signed short, vector signed short);
12162 vector unsigned int vec_adds (vector bool int, vector unsigned int);
12163 vector unsigned int vec_adds (vector unsigned int, vector bool int);
12164 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
12165 vector signed int vec_adds (vector bool int, vector signed int);
12166 vector signed int vec_adds (vector signed int, vector bool int);
12167 vector signed int vec_adds (vector signed int, vector signed int);
12169 vector signed int vec_vaddsws (vector bool int, vector signed int);
12170 vector signed int vec_vaddsws (vector signed int, vector bool int);
12171 vector signed int vec_vaddsws (vector signed int, vector signed int);
12173 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
12174 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
12175 vector unsigned int vec_vadduws (vector unsigned int,
12176 vector unsigned int);
12178 vector signed short vec_vaddshs (vector bool short,
12179 vector signed short);
12180 vector signed short vec_vaddshs (vector signed short,
12181 vector bool short);
12182 vector signed short vec_vaddshs (vector signed short,
12183 vector signed short);
12185 vector unsigned short vec_vadduhs (vector bool short,
12186 vector unsigned short);
12187 vector unsigned short vec_vadduhs (vector unsigned short,
12188 vector bool short);
12189 vector unsigned short vec_vadduhs (vector unsigned short,
12190 vector unsigned short);
12192 vector signed char vec_vaddsbs (vector bool char, vector signed char);
12193 vector signed char vec_vaddsbs (vector signed char, vector bool char);
12194 vector signed char vec_vaddsbs (vector signed char, vector signed char);
12196 vector unsigned char vec_vaddubs (vector bool char,
12197 vector unsigned char);
12198 vector unsigned char vec_vaddubs (vector unsigned char,
12200 vector unsigned char vec_vaddubs (vector unsigned char,
12201 vector unsigned char);
12203 vector float vec_and (vector float, vector float);
12204 vector float vec_and (vector float, vector bool int);
12205 vector float vec_and (vector bool int, vector float);
12206 vector bool int vec_and (vector bool int, vector bool int);
12207 vector signed int vec_and (vector bool int, vector signed int);
12208 vector signed int vec_and (vector signed int, vector bool int);
12209 vector signed int vec_and (vector signed int, vector signed int);
12210 vector unsigned int vec_and (vector bool int, vector unsigned int);
12211 vector unsigned int vec_and (vector unsigned int, vector bool int);
12212 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
12213 vector bool short vec_and (vector bool short, vector bool short);
12214 vector signed short vec_and (vector bool short, vector signed short);
12215 vector signed short vec_and (vector signed short, vector bool short);
12216 vector signed short vec_and (vector signed short, vector signed short);
12217 vector unsigned short vec_and (vector bool short,
12218 vector unsigned short);
12219 vector unsigned short vec_and (vector unsigned short,
12220 vector bool short);
12221 vector unsigned short vec_and (vector unsigned short,
12222 vector unsigned short);
12223 vector signed char vec_and (vector bool char, vector signed char);
12224 vector bool char vec_and (vector bool char, vector bool char);
12225 vector signed char vec_and (vector signed char, vector bool char);
12226 vector signed char vec_and (vector signed char, vector signed char);
12227 vector unsigned char vec_and (vector bool char, vector unsigned char);
12228 vector unsigned char vec_and (vector unsigned char, vector bool char);
12229 vector unsigned char vec_and (vector unsigned char,
12230 vector unsigned char);
12232 vector float vec_andc (vector float, vector float);
12233 vector float vec_andc (vector float, vector bool int);
12234 vector float vec_andc (vector bool int, vector float);
12235 vector bool int vec_andc (vector bool int, vector bool int);
12236 vector signed int vec_andc (vector bool int, vector signed int);
12237 vector signed int vec_andc (vector signed int, vector bool int);
12238 vector signed int vec_andc (vector signed int, vector signed int);
12239 vector unsigned int vec_andc (vector bool int, vector unsigned int);
12240 vector unsigned int vec_andc (vector unsigned int, vector bool int);
12241 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
12242 vector bool short vec_andc (vector bool short, vector bool short);
12243 vector signed short vec_andc (vector bool short, vector signed short);
12244 vector signed short vec_andc (vector signed short, vector bool short);
12245 vector signed short vec_andc (vector signed short, vector signed short);
12246 vector unsigned short vec_andc (vector bool short,
12247 vector unsigned short);
12248 vector unsigned short vec_andc (vector unsigned short,
12249 vector bool short);
12250 vector unsigned short vec_andc (vector unsigned short,
12251 vector unsigned short);
12252 vector signed char vec_andc (vector bool char, vector signed char);
12253 vector bool char vec_andc (vector bool char, vector bool char);
12254 vector signed char vec_andc (vector signed char, vector bool char);
12255 vector signed char vec_andc (vector signed char, vector signed char);
12256 vector unsigned char vec_andc (vector bool char, vector unsigned char);
12257 vector unsigned char vec_andc (vector unsigned char, vector bool char);
12258 vector unsigned char vec_andc (vector unsigned char,
12259 vector unsigned char);
12261 vector unsigned char vec_avg (vector unsigned char,
12262 vector unsigned char);
12263 vector signed char vec_avg (vector signed char, vector signed char);
12264 vector unsigned short vec_avg (vector unsigned short,
12265 vector unsigned short);
12266 vector signed short vec_avg (vector signed short, vector signed short);
12267 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
12268 vector signed int vec_avg (vector signed int, vector signed int);
12270 vector signed int vec_vavgsw (vector signed int, vector signed int);
12272 vector unsigned int vec_vavguw (vector unsigned int,
12273 vector unsigned int);
12275 vector signed short vec_vavgsh (vector signed short,
12276 vector signed short);
12278 vector unsigned short vec_vavguh (vector unsigned short,
12279 vector unsigned short);
12281 vector signed char vec_vavgsb (vector signed char, vector signed char);
12283 vector unsigned char vec_vavgub (vector unsigned char,
12284 vector unsigned char);
12286 vector float vec_copysign (vector float);
12288 vector float vec_ceil (vector float);
12290 vector signed int vec_cmpb (vector float, vector float);
12292 vector bool char vec_cmpeq (vector signed char, vector signed char);
12293 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
12294 vector bool short vec_cmpeq (vector signed short, vector signed short);
12295 vector bool short vec_cmpeq (vector unsigned short,
12296 vector unsigned short);
12297 vector bool int vec_cmpeq (vector signed int, vector signed int);
12298 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
12299 vector bool int vec_cmpeq (vector float, vector float);
12301 vector bool int vec_vcmpeqfp (vector float, vector float);
12303 vector bool int vec_vcmpequw (vector signed int, vector signed int);
12304 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
12306 vector bool short vec_vcmpequh (vector signed short,
12307 vector signed short);
12308 vector bool short vec_vcmpequh (vector unsigned short,
12309 vector unsigned short);
12311 vector bool char vec_vcmpequb (vector signed char, vector signed char);
12312 vector bool char vec_vcmpequb (vector unsigned char,
12313 vector unsigned char);
12315 vector bool int vec_cmpge (vector float, vector float);
12317 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
12318 vector bool char vec_cmpgt (vector signed char, vector signed char);
12319 vector bool short vec_cmpgt (vector unsigned short,
12320 vector unsigned short);
12321 vector bool short vec_cmpgt (vector signed short, vector signed short);
12322 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
12323 vector bool int vec_cmpgt (vector signed int, vector signed int);
12324 vector bool int vec_cmpgt (vector float, vector float);
12326 vector bool int vec_vcmpgtfp (vector float, vector float);
12328 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
12330 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
12332 vector bool short vec_vcmpgtsh (vector signed short,
12333 vector signed short);
12335 vector bool short vec_vcmpgtuh (vector unsigned short,
12336 vector unsigned short);
12338 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
12340 vector bool char vec_vcmpgtub (vector unsigned char,
12341 vector unsigned char);
12343 vector bool int vec_cmple (vector float, vector float);
12345 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
12346 vector bool char vec_cmplt (vector signed char, vector signed char);
12347 vector bool short vec_cmplt (vector unsigned short,
12348 vector unsigned short);
12349 vector bool short vec_cmplt (vector signed short, vector signed short);
12350 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
12351 vector bool int vec_cmplt (vector signed int, vector signed int);
12352 vector bool int vec_cmplt (vector float, vector float);
12354 vector float vec_ctf (vector unsigned int, const int);
12355 vector float vec_ctf (vector signed int, const int);
12357 vector float vec_vcfsx (vector signed int, const int);
12359 vector float vec_vcfux (vector unsigned int, const int);
12361 vector signed int vec_cts (vector float, const int);
12363 vector unsigned int vec_ctu (vector float, const int);
12365 void vec_dss (const int);
12367 void vec_dssall (void);
12369 void vec_dst (const vector unsigned char *, int, const int);
12370 void vec_dst (const vector signed char *, int, const int);
12371 void vec_dst (const vector bool char *, int, const int);
12372 void vec_dst (const vector unsigned short *, int, const int);
12373 void vec_dst (const vector signed short *, int, const int);
12374 void vec_dst (const vector bool short *, int, const int);
12375 void vec_dst (const vector pixel *, int, const int);
12376 void vec_dst (const vector unsigned int *, int, const int);
12377 void vec_dst (const vector signed int *, int, const int);
12378 void vec_dst (const vector bool int *, int, const int);
12379 void vec_dst (const vector float *, int, const int);
12380 void vec_dst (const unsigned char *, int, const int);
12381 void vec_dst (const signed char *, int, const int);
12382 void vec_dst (const unsigned short *, int, const int);
12383 void vec_dst (const short *, int, const int);
12384 void vec_dst (const unsigned int *, int, const int);
12385 void vec_dst (const int *, int, const int);
12386 void vec_dst (const unsigned long *, int, const int);
12387 void vec_dst (const long *, int, const int);
12388 void vec_dst (const float *, int, const int);
12390 void vec_dstst (const vector unsigned char *, int, const int);
12391 void vec_dstst (const vector signed char *, int, const int);
12392 void vec_dstst (const vector bool char *, int, const int);
12393 void vec_dstst (const vector unsigned short *, int, const int);
12394 void vec_dstst (const vector signed short *, int, const int);
12395 void vec_dstst (const vector bool short *, int, const int);
12396 void vec_dstst (const vector pixel *, int, const int);
12397 void vec_dstst (const vector unsigned int *, int, const int);
12398 void vec_dstst (const vector signed int *, int, const int);
12399 void vec_dstst (const vector bool int *, int, const int);
12400 void vec_dstst (const vector float *, int, const int);
12401 void vec_dstst (const unsigned char *, int, const int);
12402 void vec_dstst (const signed char *, int, const int);
12403 void vec_dstst (const unsigned short *, int, const int);
12404 void vec_dstst (const short *, int, const int);
12405 void vec_dstst (const unsigned int *, int, const int);
12406 void vec_dstst (const int *, int, const int);
12407 void vec_dstst (const unsigned long *, int, const int);
12408 void vec_dstst (const long *, int, const int);
12409 void vec_dstst (const float *, int, const int);
12411 void vec_dststt (const vector unsigned char *, int, const int);
12412 void vec_dststt (const vector signed char *, int, const int);
12413 void vec_dststt (const vector bool char *, int, const int);
12414 void vec_dststt (const vector unsigned short *, int, const int);
12415 void vec_dststt (const vector signed short *, int, const int);
12416 void vec_dststt (const vector bool short *, int, const int);
12417 void vec_dststt (const vector pixel *, int, const int);
12418 void vec_dststt (const vector unsigned int *, int, const int);
12419 void vec_dststt (const vector signed int *, int, const int);
12420 void vec_dststt (const vector bool int *, int, const int);
12421 void vec_dststt (const vector float *, int, const int);
12422 void vec_dststt (const unsigned char *, int, const int);
12423 void vec_dststt (const signed char *, int, const int);
12424 void vec_dststt (const unsigned short *, int, const int);
12425 void vec_dststt (const short *, int, const int);
12426 void vec_dststt (const unsigned int *, int, const int);
12427 void vec_dststt (const int *, int, const int);
12428 void vec_dststt (const unsigned long *, int, const int);
12429 void vec_dststt (const long *, int, const int);
12430 void vec_dststt (const float *, int, const int);
12432 void vec_dstt (const vector unsigned char *, int, const int);
12433 void vec_dstt (const vector signed char *, int, const int);
12434 void vec_dstt (const vector bool char *, int, const int);
12435 void vec_dstt (const vector unsigned short *, int, const int);
12436 void vec_dstt (const vector signed short *, int, const int);
12437 void vec_dstt (const vector bool short *, int, const int);
12438 void vec_dstt (const vector pixel *, int, const int);
12439 void vec_dstt (const vector unsigned int *, int, const int);
12440 void vec_dstt (const vector signed int *, int, const int);
12441 void vec_dstt (const vector bool int *, int, const int);
12442 void vec_dstt (const vector float *, int, const int);
12443 void vec_dstt (const unsigned char *, int, const int);
12444 void vec_dstt (const signed char *, int, const int);
12445 void vec_dstt (const unsigned short *, int, const int);
12446 void vec_dstt (const short *, int, const int);
12447 void vec_dstt (const unsigned int *, int, const int);
12448 void vec_dstt (const int *, int, const int);
12449 void vec_dstt (const unsigned long *, int, const int);
12450 void vec_dstt (const long *, int, const int);
12451 void vec_dstt (const float *, int, const int);
12453 vector float vec_expte (vector float);
12455 vector float vec_floor (vector float);
12457 vector float vec_ld (int, const vector float *);
12458 vector float vec_ld (int, const float *);
12459 vector bool int vec_ld (int, const vector bool int *);
12460 vector signed int vec_ld (int, const vector signed int *);
12461 vector signed int vec_ld (int, const int *);
12462 vector signed int vec_ld (int, const long *);
12463 vector unsigned int vec_ld (int, const vector unsigned int *);
12464 vector unsigned int vec_ld (int, const unsigned int *);
12465 vector unsigned int vec_ld (int, const unsigned long *);
12466 vector bool short vec_ld (int, const vector bool short *);
12467 vector pixel vec_ld (int, const vector pixel *);
12468 vector signed short vec_ld (int, const vector signed short *);
12469 vector signed short vec_ld (int, const short *);
12470 vector unsigned short vec_ld (int, const vector unsigned short *);
12471 vector unsigned short vec_ld (int, const unsigned short *);
12472 vector bool char vec_ld (int, const vector bool char *);
12473 vector signed char vec_ld (int, const vector signed char *);
12474 vector signed char vec_ld (int, const signed char *);
12475 vector unsigned char vec_ld (int, const vector unsigned char *);
12476 vector unsigned char vec_ld (int, const unsigned char *);
12478 vector signed char vec_lde (int, const signed char *);
12479 vector unsigned char vec_lde (int, const unsigned char *);
12480 vector signed short vec_lde (int, const short *);
12481 vector unsigned short vec_lde (int, const unsigned short *);
12482 vector float vec_lde (int, const float *);
12483 vector signed int vec_lde (int, const int *);
12484 vector unsigned int vec_lde (int, const unsigned int *);
12485 vector signed int vec_lde (int, const long *);
12486 vector unsigned int vec_lde (int, const unsigned long *);
12488 vector float vec_lvewx (int, float *);
12489 vector signed int vec_lvewx (int, int *);
12490 vector unsigned int vec_lvewx (int, unsigned int *);
12491 vector signed int vec_lvewx (int, long *);
12492 vector unsigned int vec_lvewx (int, unsigned long *);
12494 vector signed short vec_lvehx (int, short *);
12495 vector unsigned short vec_lvehx (int, unsigned short *);
12497 vector signed char vec_lvebx (int, char *);
12498 vector unsigned char vec_lvebx (int, unsigned char *);
12500 vector float vec_ldl (int, const vector float *);
12501 vector float vec_ldl (int, const float *);
12502 vector bool int vec_ldl (int, const vector bool int *);
12503 vector signed int vec_ldl (int, const vector signed int *);
12504 vector signed int vec_ldl (int, const int *);
12505 vector signed int vec_ldl (int, const long *);
12506 vector unsigned int vec_ldl (int, const vector unsigned int *);
12507 vector unsigned int vec_ldl (int, const unsigned int *);
12508 vector unsigned int vec_ldl (int, const unsigned long *);
12509 vector bool short vec_ldl (int, const vector bool short *);
12510 vector pixel vec_ldl (int, const vector pixel *);
12511 vector signed short vec_ldl (int, const vector signed short *);
12512 vector signed short vec_ldl (int, const short *);
12513 vector unsigned short vec_ldl (int, const vector unsigned short *);
12514 vector unsigned short vec_ldl (int, const unsigned short *);
12515 vector bool char vec_ldl (int, const vector bool char *);
12516 vector signed char vec_ldl (int, const vector signed char *);
12517 vector signed char vec_ldl (int, const signed char *);
12518 vector unsigned char vec_ldl (int, const vector unsigned char *);
12519 vector unsigned char vec_ldl (int, const unsigned char *);
12521 vector float vec_loge (vector float);
12523 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
12524 vector unsigned char vec_lvsl (int, const volatile signed char *);
12525 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
12526 vector unsigned char vec_lvsl (int, const volatile short *);
12527 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
12528 vector unsigned char vec_lvsl (int, const volatile int *);
12529 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
12530 vector unsigned char vec_lvsl (int, const volatile long *);
12531 vector unsigned char vec_lvsl (int, const volatile float *);
12533 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
12534 vector unsigned char vec_lvsr (int, const volatile signed char *);
12535 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
12536 vector unsigned char vec_lvsr (int, const volatile short *);
12537 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
12538 vector unsigned char vec_lvsr (int, const volatile int *);
12539 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
12540 vector unsigned char vec_lvsr (int, const volatile long *);
12541 vector unsigned char vec_lvsr (int, const volatile float *);
12543 vector float vec_madd (vector float, vector float, vector float);
12545 vector signed short vec_madds (vector signed short,
12546 vector signed short,
12547 vector signed short);
12549 vector unsigned char vec_max (vector bool char, vector unsigned char);
12550 vector unsigned char vec_max (vector unsigned char, vector bool char);
12551 vector unsigned char vec_max (vector unsigned char,
12552 vector unsigned char);
12553 vector signed char vec_max (vector bool char, vector signed char);
12554 vector signed char vec_max (vector signed char, vector bool char);
12555 vector signed char vec_max (vector signed char, vector signed char);
12556 vector unsigned short vec_max (vector bool short,
12557 vector unsigned short);
12558 vector unsigned short vec_max (vector unsigned short,
12559 vector bool short);
12560 vector unsigned short vec_max (vector unsigned short,
12561 vector unsigned short);
12562 vector signed short vec_max (vector bool short, vector signed short);
12563 vector signed short vec_max (vector signed short, vector bool short);
12564 vector signed short vec_max (vector signed short, vector signed short);
12565 vector unsigned int vec_max (vector bool int, vector unsigned int);
12566 vector unsigned int vec_max (vector unsigned int, vector bool int);
12567 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
12568 vector signed int vec_max (vector bool int, vector signed int);
12569 vector signed int vec_max (vector signed int, vector bool int);
12570 vector signed int vec_max (vector signed int, vector signed int);
12571 vector float vec_max (vector float, vector float);
12573 vector float vec_vmaxfp (vector float, vector float);
12575 vector signed int vec_vmaxsw (vector bool int, vector signed int);
12576 vector signed int vec_vmaxsw (vector signed int, vector bool int);
12577 vector signed int vec_vmaxsw (vector signed int, vector signed int);
12579 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
12580 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
12581 vector unsigned int vec_vmaxuw (vector unsigned int,
12582 vector unsigned int);
12584 vector signed short vec_vmaxsh (vector bool short, vector signed short);
12585 vector signed short vec_vmaxsh (vector signed short, vector bool short);
12586 vector signed short vec_vmaxsh (vector signed short,
12587 vector signed short);
12589 vector unsigned short vec_vmaxuh (vector bool short,
12590 vector unsigned short);
12591 vector unsigned short vec_vmaxuh (vector unsigned short,
12592 vector bool short);
12593 vector unsigned short vec_vmaxuh (vector unsigned short,
12594 vector unsigned short);
12596 vector signed char vec_vmaxsb (vector bool char, vector signed char);
12597 vector signed char vec_vmaxsb (vector signed char, vector bool char);
12598 vector signed char vec_vmaxsb (vector signed char, vector signed char);
12600 vector unsigned char vec_vmaxub (vector bool char,
12601 vector unsigned char);
12602 vector unsigned char vec_vmaxub (vector unsigned char,
12604 vector unsigned char vec_vmaxub (vector unsigned char,
12605 vector unsigned char);
12607 vector bool char vec_mergeh (vector bool char, vector bool char);
12608 vector signed char vec_mergeh (vector signed char, vector signed char);
12609 vector unsigned char vec_mergeh (vector unsigned char,
12610 vector unsigned char);
12611 vector bool short vec_mergeh (vector bool short, vector bool short);
12612 vector pixel vec_mergeh (vector pixel, vector pixel);
12613 vector signed short vec_mergeh (vector signed short,
12614 vector signed short);
12615 vector unsigned short vec_mergeh (vector unsigned short,
12616 vector unsigned short);
12617 vector float vec_mergeh (vector float, vector float);
12618 vector bool int vec_mergeh (vector bool int, vector bool int);
12619 vector signed int vec_mergeh (vector signed int, vector signed int);
12620 vector unsigned int vec_mergeh (vector unsigned int,
12621 vector unsigned int);
12623 vector float vec_vmrghw (vector float, vector float);
12624 vector bool int vec_vmrghw (vector bool int, vector bool int);
12625 vector signed int vec_vmrghw (vector signed int, vector signed int);
12626 vector unsigned int vec_vmrghw (vector unsigned int,
12627 vector unsigned int);
12629 vector bool short vec_vmrghh (vector bool short, vector bool short);
12630 vector signed short vec_vmrghh (vector signed short,
12631 vector signed short);
12632 vector unsigned short vec_vmrghh (vector unsigned short,
12633 vector unsigned short);
12634 vector pixel vec_vmrghh (vector pixel, vector pixel);
12636 vector bool char vec_vmrghb (vector bool char, vector bool char);
12637 vector signed char vec_vmrghb (vector signed char, vector signed char);
12638 vector unsigned char vec_vmrghb (vector unsigned char,
12639 vector unsigned char);
12641 vector bool char vec_mergel (vector bool char, vector bool char);
12642 vector signed char vec_mergel (vector signed char, vector signed char);
12643 vector unsigned char vec_mergel (vector unsigned char,
12644 vector unsigned char);
12645 vector bool short vec_mergel (vector bool short, vector bool short);
12646 vector pixel vec_mergel (vector pixel, vector pixel);
12647 vector signed short vec_mergel (vector signed short,
12648 vector signed short);
12649 vector unsigned short vec_mergel (vector unsigned short,
12650 vector unsigned short);
12651 vector float vec_mergel (vector float, vector float);
12652 vector bool int vec_mergel (vector bool int, vector bool int);
12653 vector signed int vec_mergel (vector signed int, vector signed int);
12654 vector unsigned int vec_mergel (vector unsigned int,
12655 vector unsigned int);
12657 vector float vec_vmrglw (vector float, vector float);
12658 vector signed int vec_vmrglw (vector signed int, vector signed int);
12659 vector unsigned int vec_vmrglw (vector unsigned int,
12660 vector unsigned int);
12661 vector bool int vec_vmrglw (vector bool int, vector bool int);
12663 vector bool short vec_vmrglh (vector bool short, vector bool short);
12664 vector signed short vec_vmrglh (vector signed short,
12665 vector signed short);
12666 vector unsigned short vec_vmrglh (vector unsigned short,
12667 vector unsigned short);
12668 vector pixel vec_vmrglh (vector pixel, vector pixel);
12670 vector bool char vec_vmrglb (vector bool char, vector bool char);
12671 vector signed char vec_vmrglb (vector signed char, vector signed char);
12672 vector unsigned char vec_vmrglb (vector unsigned char,
12673 vector unsigned char);
12675 vector unsigned short vec_mfvscr (void);
12677 vector unsigned char vec_min (vector bool char, vector unsigned char);
12678 vector unsigned char vec_min (vector unsigned char, vector bool char);
12679 vector unsigned char vec_min (vector unsigned char,
12680 vector unsigned char);
12681 vector signed char vec_min (vector bool char, vector signed char);
12682 vector signed char vec_min (vector signed char, vector bool char);
12683 vector signed char vec_min (vector signed char, vector signed char);
12684 vector unsigned short vec_min (vector bool short,
12685 vector unsigned short);
12686 vector unsigned short vec_min (vector unsigned short,
12687 vector bool short);
12688 vector unsigned short vec_min (vector unsigned short,
12689 vector unsigned short);
12690 vector signed short vec_min (vector bool short, vector signed short);
12691 vector signed short vec_min (vector signed short, vector bool short);
12692 vector signed short vec_min (vector signed short, vector signed short);
12693 vector unsigned int vec_min (vector bool int, vector unsigned int);
12694 vector unsigned int vec_min (vector unsigned int, vector bool int);
12695 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
12696 vector signed int vec_min (vector bool int, vector signed int);
12697 vector signed int vec_min (vector signed int, vector bool int);
12698 vector signed int vec_min (vector signed int, vector signed int);
12699 vector float vec_min (vector float, vector float);
12701 vector float vec_vminfp (vector float, vector float);
12703 vector signed int vec_vminsw (vector bool int, vector signed int);
12704 vector signed int vec_vminsw (vector signed int, vector bool int);
12705 vector signed int vec_vminsw (vector signed int, vector signed int);
12707 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
12708 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
12709 vector unsigned int vec_vminuw (vector unsigned int,
12710 vector unsigned int);
12712 vector signed short vec_vminsh (vector bool short, vector signed short);
12713 vector signed short vec_vminsh (vector signed short, vector bool short);
12714 vector signed short vec_vminsh (vector signed short,
12715 vector signed short);
12717 vector unsigned short vec_vminuh (vector bool short,
12718 vector unsigned short);
12719 vector unsigned short vec_vminuh (vector unsigned short,
12720 vector bool short);
12721 vector unsigned short vec_vminuh (vector unsigned short,
12722 vector unsigned short);
12724 vector signed char vec_vminsb (vector bool char, vector signed char);
12725 vector signed char vec_vminsb (vector signed char, vector bool char);
12726 vector signed char vec_vminsb (vector signed char, vector signed char);
12728 vector unsigned char vec_vminub (vector bool char,
12729 vector unsigned char);
12730 vector unsigned char vec_vminub (vector unsigned char,
12732 vector unsigned char vec_vminub (vector unsigned char,
12733 vector unsigned char);
12735 vector signed short vec_mladd (vector signed short,
12736 vector signed short,
12737 vector signed short);
12738 vector signed short vec_mladd (vector signed short,
12739 vector unsigned short,
12740 vector unsigned short);
12741 vector signed short vec_mladd (vector unsigned short,
12742 vector signed short,
12743 vector signed short);
12744 vector unsigned short vec_mladd (vector unsigned short,
12745 vector unsigned short,
12746 vector unsigned short);
12748 vector signed short vec_mradds (vector signed short,
12749 vector signed short,
12750 vector signed short);
12752 vector unsigned int vec_msum (vector unsigned char,
12753 vector unsigned char,
12754 vector unsigned int);
12755 vector signed int vec_msum (vector signed char,
12756 vector unsigned char,
12757 vector signed int);
12758 vector unsigned int vec_msum (vector unsigned short,
12759 vector unsigned short,
12760 vector unsigned int);
12761 vector signed int vec_msum (vector signed short,
12762 vector signed short,
12763 vector signed int);
12765 vector signed int vec_vmsumshm (vector signed short,
12766 vector signed short,
12767 vector signed int);
12769 vector unsigned int vec_vmsumuhm (vector unsigned short,
12770 vector unsigned short,
12771 vector unsigned int);
12773 vector signed int vec_vmsummbm (vector signed char,
12774 vector unsigned char,
12775 vector signed int);
12777 vector unsigned int vec_vmsumubm (vector unsigned char,
12778 vector unsigned char,
12779 vector unsigned int);
12781 vector unsigned int vec_msums (vector unsigned short,
12782 vector unsigned short,
12783 vector unsigned int);
12784 vector signed int vec_msums (vector signed short,
12785 vector signed short,
12786 vector signed int);
12788 vector signed int vec_vmsumshs (vector signed short,
12789 vector signed short,
12790 vector signed int);
12792 vector unsigned int vec_vmsumuhs (vector unsigned short,
12793 vector unsigned short,
12794 vector unsigned int);
12796 void vec_mtvscr (vector signed int);
12797 void vec_mtvscr (vector unsigned int);
12798 void vec_mtvscr (vector bool int);
12799 void vec_mtvscr (vector signed short);
12800 void vec_mtvscr (vector unsigned short);
12801 void vec_mtvscr (vector bool short);
12802 void vec_mtvscr (vector pixel);
12803 void vec_mtvscr (vector signed char);
12804 void vec_mtvscr (vector unsigned char);
12805 void vec_mtvscr (vector bool char);
12807 vector unsigned short vec_mule (vector unsigned char,
12808 vector unsigned char);
12809 vector signed short vec_mule (vector signed char,
12810 vector signed char);
12811 vector unsigned int vec_mule (vector unsigned short,
12812 vector unsigned short);
12813 vector signed int vec_mule (vector signed short, vector signed short);
12815 vector signed int vec_vmulesh (vector signed short,
12816 vector signed short);
12818 vector unsigned int vec_vmuleuh (vector unsigned short,
12819 vector unsigned short);
12821 vector signed short vec_vmulesb (vector signed char,
12822 vector signed char);
12824 vector unsigned short vec_vmuleub (vector unsigned char,
12825 vector unsigned char);
12827 vector unsigned short vec_mulo (vector unsigned char,
12828 vector unsigned char);
12829 vector signed short vec_mulo (vector signed char, vector signed char);
12830 vector unsigned int vec_mulo (vector unsigned short,
12831 vector unsigned short);
12832 vector signed int vec_mulo (vector signed short, vector signed short);
12834 vector signed int vec_vmulosh (vector signed short,
12835 vector signed short);
12837 vector unsigned int vec_vmulouh (vector unsigned short,
12838 vector unsigned short);
12840 vector signed short vec_vmulosb (vector signed char,
12841 vector signed char);
12843 vector unsigned short vec_vmuloub (vector unsigned char,
12844 vector unsigned char);
12846 vector float vec_nmsub (vector float, vector float, vector float);
12848 vector float vec_nor (vector float, vector float);
12849 vector signed int vec_nor (vector signed int, vector signed int);
12850 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
12851 vector bool int vec_nor (vector bool int, vector bool int);
12852 vector signed short vec_nor (vector signed short, vector signed short);
12853 vector unsigned short vec_nor (vector unsigned short,
12854 vector unsigned short);
12855 vector bool short vec_nor (vector bool short, vector bool short);
12856 vector signed char vec_nor (vector signed char, vector signed char);
12857 vector unsigned char vec_nor (vector unsigned char,
12858 vector unsigned char);
12859 vector bool char vec_nor (vector bool char, vector bool char);
12861 vector float vec_or (vector float, vector float);
12862 vector float vec_or (vector float, vector bool int);
12863 vector float vec_or (vector bool int, vector float);
12864 vector bool int vec_or (vector bool int, vector bool int);
12865 vector signed int vec_or (vector bool int, vector signed int);
12866 vector signed int vec_or (vector signed int, vector bool int);
12867 vector signed int vec_or (vector signed int, vector signed int);
12868 vector unsigned int vec_or (vector bool int, vector unsigned int);
12869 vector unsigned int vec_or (vector unsigned int, vector bool int);
12870 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
12871 vector bool short vec_or (vector bool short, vector bool short);
12872 vector signed short vec_or (vector bool short, vector signed short);
12873 vector signed short vec_or (vector signed short, vector bool short);
12874 vector signed short vec_or (vector signed short, vector signed short);
12875 vector unsigned short vec_or (vector bool short, vector unsigned short);
12876 vector unsigned short vec_or (vector unsigned short, vector bool short);
12877 vector unsigned short vec_or (vector unsigned short,
12878 vector unsigned short);
12879 vector signed char vec_or (vector bool char, vector signed char);
12880 vector bool char vec_or (vector bool char, vector bool char);
12881 vector signed char vec_or (vector signed char, vector bool char);
12882 vector signed char vec_or (vector signed char, vector signed char);
12883 vector unsigned char vec_or (vector bool char, vector unsigned char);
12884 vector unsigned char vec_or (vector unsigned char, vector bool char);
12885 vector unsigned char vec_or (vector unsigned char,
12886 vector unsigned char);
12888 vector signed char vec_pack (vector signed short, vector signed short);
12889 vector unsigned char vec_pack (vector unsigned short,
12890 vector unsigned short);
12891 vector bool char vec_pack (vector bool short, vector bool short);
12892 vector signed short vec_pack (vector signed int, vector signed int);
12893 vector unsigned short vec_pack (vector unsigned int,
12894 vector unsigned int);
12895 vector bool short vec_pack (vector bool int, vector bool int);
12897 vector bool short vec_vpkuwum (vector bool int, vector bool int);
12898 vector signed short vec_vpkuwum (vector signed int, vector signed int);
12899 vector unsigned short vec_vpkuwum (vector unsigned int,
12900 vector unsigned int);
12902 vector bool char vec_vpkuhum (vector bool short, vector bool short);
12903 vector signed char vec_vpkuhum (vector signed short,
12904 vector signed short);
12905 vector unsigned char vec_vpkuhum (vector unsigned short,
12906 vector unsigned short);
12908 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
12910 vector unsigned char vec_packs (vector unsigned short,
12911 vector unsigned short);
12912 vector signed char vec_packs (vector signed short, vector signed short);
12913 vector unsigned short vec_packs (vector unsigned int,
12914 vector unsigned int);
12915 vector signed short vec_packs (vector signed int, vector signed int);
12917 vector signed short vec_vpkswss (vector signed int, vector signed int);
12919 vector unsigned short vec_vpkuwus (vector unsigned int,
12920 vector unsigned int);
12922 vector signed char vec_vpkshss (vector signed short,
12923 vector signed short);
12925 vector unsigned char vec_vpkuhus (vector unsigned short,
12926 vector unsigned short);
12928 vector unsigned char vec_packsu (vector unsigned short,
12929 vector unsigned short);
12930 vector unsigned char vec_packsu (vector signed short,
12931 vector signed short);
12932 vector unsigned short vec_packsu (vector unsigned int,
12933 vector unsigned int);
12934 vector unsigned short vec_packsu (vector signed int, vector signed int);
12936 vector unsigned short vec_vpkswus (vector signed int,
12937 vector signed int);
12939 vector unsigned char vec_vpkshus (vector signed short,
12940 vector signed short);
12942 vector float vec_perm (vector float,
12944 vector unsigned char);
12945 vector signed int vec_perm (vector signed int,
12947 vector unsigned char);
12948 vector unsigned int vec_perm (vector unsigned int,
12949 vector unsigned int,
12950 vector unsigned char);
12951 vector bool int vec_perm (vector bool int,
12953 vector unsigned char);
12954 vector signed short vec_perm (vector signed short,
12955 vector signed short,
12956 vector unsigned char);
12957 vector unsigned short vec_perm (vector unsigned short,
12958 vector unsigned short,
12959 vector unsigned char);
12960 vector bool short vec_perm (vector bool short,
12962 vector unsigned char);
12963 vector pixel vec_perm (vector pixel,
12965 vector unsigned char);
12966 vector signed char vec_perm (vector signed char,
12967 vector signed char,
12968 vector unsigned char);
12969 vector unsigned char vec_perm (vector unsigned char,
12970 vector unsigned char,
12971 vector unsigned char);
12972 vector bool char vec_perm (vector bool char,
12974 vector unsigned char);
12976 vector float vec_re (vector float);
12978 vector signed char vec_rl (vector signed char,
12979 vector unsigned char);
12980 vector unsigned char vec_rl (vector unsigned char,
12981 vector unsigned char);
12982 vector signed short vec_rl (vector signed short, vector unsigned short);
12983 vector unsigned short vec_rl (vector unsigned short,
12984 vector unsigned short);
12985 vector signed int vec_rl (vector signed int, vector unsigned int);
12986 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
12988 vector signed int vec_vrlw (vector signed int, vector unsigned int);
12989 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
12991 vector signed short vec_vrlh (vector signed short,
12992 vector unsigned short);
12993 vector unsigned short vec_vrlh (vector unsigned short,
12994 vector unsigned short);
12996 vector signed char vec_vrlb (vector signed char, vector unsigned char);
12997 vector unsigned char vec_vrlb (vector unsigned char,
12998 vector unsigned char);
13000 vector float vec_round (vector float);
13002 vector float vec_recip (vector float, vector float);
13004 vector float vec_rsqrt (vector float);
13006 vector float vec_rsqrte (vector float);
13008 vector float vec_sel (vector float, vector float, vector bool int);
13009 vector float vec_sel (vector float, vector float, vector unsigned int);
13010 vector signed int vec_sel (vector signed int,
13013 vector signed int vec_sel (vector signed int,
13015 vector unsigned int);
13016 vector unsigned int vec_sel (vector unsigned int,
13017 vector unsigned int,
13019 vector unsigned int vec_sel (vector unsigned int,
13020 vector unsigned int,
13021 vector unsigned int);
13022 vector bool int vec_sel (vector bool int,
13025 vector bool int vec_sel (vector bool int,
13027 vector unsigned int);
13028 vector signed short vec_sel (vector signed short,
13029 vector signed short,
13030 vector bool short);
13031 vector signed short vec_sel (vector signed short,
13032 vector signed short,
13033 vector unsigned short);
13034 vector unsigned short vec_sel (vector unsigned short,
13035 vector unsigned short,
13036 vector bool short);
13037 vector unsigned short vec_sel (vector unsigned short,
13038 vector unsigned short,
13039 vector unsigned short);
13040 vector bool short vec_sel (vector bool short,
13042 vector bool short);
13043 vector bool short vec_sel (vector bool short,
13045 vector unsigned short);
13046 vector signed char vec_sel (vector signed char,
13047 vector signed char,
13049 vector signed char vec_sel (vector signed char,
13050 vector signed char,
13051 vector unsigned char);
13052 vector unsigned char vec_sel (vector unsigned char,
13053 vector unsigned char,
13055 vector unsigned char vec_sel (vector unsigned char,
13056 vector unsigned char,
13057 vector unsigned char);
13058 vector bool char vec_sel (vector bool char,
13061 vector bool char vec_sel (vector bool char,
13063 vector unsigned char);
13065 vector signed char vec_sl (vector signed char,
13066 vector unsigned char);
13067 vector unsigned char vec_sl (vector unsigned char,
13068 vector unsigned char);
13069 vector signed short vec_sl (vector signed short, vector unsigned short);
13070 vector unsigned short vec_sl (vector unsigned short,
13071 vector unsigned short);
13072 vector signed int vec_sl (vector signed int, vector unsigned int);
13073 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
13075 vector signed int vec_vslw (vector signed int, vector unsigned int);
13076 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
13078 vector signed short vec_vslh (vector signed short,
13079 vector unsigned short);
13080 vector unsigned short vec_vslh (vector unsigned short,
13081 vector unsigned short);
13083 vector signed char vec_vslb (vector signed char, vector unsigned char);
13084 vector unsigned char vec_vslb (vector unsigned char,
13085 vector unsigned char);
13087 vector float vec_sld (vector float, vector float, const int);
13088 vector signed int vec_sld (vector signed int,
13091 vector unsigned int vec_sld (vector unsigned int,
13092 vector unsigned int,
13094 vector bool int vec_sld (vector bool int,
13097 vector signed short vec_sld (vector signed short,
13098 vector signed short,
13100 vector unsigned short vec_sld (vector unsigned short,
13101 vector unsigned short,
13103 vector bool short vec_sld (vector bool short,
13106 vector pixel vec_sld (vector pixel,
13109 vector signed char vec_sld (vector signed char,
13110 vector signed char,
13112 vector unsigned char vec_sld (vector unsigned char,
13113 vector unsigned char,
13115 vector bool char vec_sld (vector bool char,
13119 vector signed int vec_sll (vector signed int,
13120 vector unsigned int);
13121 vector signed int vec_sll (vector signed int,
13122 vector unsigned short);
13123 vector signed int vec_sll (vector signed int,
13124 vector unsigned char);
13125 vector unsigned int vec_sll (vector unsigned int,
13126 vector unsigned int);
13127 vector unsigned int vec_sll (vector unsigned int,
13128 vector unsigned short);
13129 vector unsigned int vec_sll (vector unsigned int,
13130 vector unsigned char);
13131 vector bool int vec_sll (vector bool int,
13132 vector unsigned int);
13133 vector bool int vec_sll (vector bool int,
13134 vector unsigned short);
13135 vector bool int vec_sll (vector bool int,
13136 vector unsigned char);
13137 vector signed short vec_sll (vector signed short,
13138 vector unsigned int);
13139 vector signed short vec_sll (vector signed short,
13140 vector unsigned short);
13141 vector signed short vec_sll (vector signed short,
13142 vector unsigned char);
13143 vector unsigned short vec_sll (vector unsigned short,
13144 vector unsigned int);
13145 vector unsigned short vec_sll (vector unsigned short,
13146 vector unsigned short);
13147 vector unsigned short vec_sll (vector unsigned short,
13148 vector unsigned char);
13149 vector bool short vec_sll (vector bool short, vector unsigned int);
13150 vector bool short vec_sll (vector bool short, vector unsigned short);
13151 vector bool short vec_sll (vector bool short, vector unsigned char);
13152 vector pixel vec_sll (vector pixel, vector unsigned int);
13153 vector pixel vec_sll (vector pixel, vector unsigned short);
13154 vector pixel vec_sll (vector pixel, vector unsigned char);
13155 vector signed char vec_sll (vector signed char, vector unsigned int);
13156 vector signed char vec_sll (vector signed char, vector unsigned short);
13157 vector signed char vec_sll (vector signed char, vector unsigned char);
13158 vector unsigned char vec_sll (vector unsigned char,
13159 vector unsigned int);
13160 vector unsigned char vec_sll (vector unsigned char,
13161 vector unsigned short);
13162 vector unsigned char vec_sll (vector unsigned char,
13163 vector unsigned char);
13164 vector bool char vec_sll (vector bool char, vector unsigned int);
13165 vector bool char vec_sll (vector bool char, vector unsigned short);
13166 vector bool char vec_sll (vector bool char, vector unsigned char);
13168 vector float vec_slo (vector float, vector signed char);
13169 vector float vec_slo (vector float, vector unsigned char);
13170 vector signed int vec_slo (vector signed int, vector signed char);
13171 vector signed int vec_slo (vector signed int, vector unsigned char);
13172 vector unsigned int vec_slo (vector unsigned int, vector signed char);
13173 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
13174 vector signed short vec_slo (vector signed short, vector signed char);
13175 vector signed short vec_slo (vector signed short, vector unsigned char);
13176 vector unsigned short vec_slo (vector unsigned short,
13177 vector signed char);
13178 vector unsigned short vec_slo (vector unsigned short,
13179 vector unsigned char);
13180 vector pixel vec_slo (vector pixel, vector signed char);
13181 vector pixel vec_slo (vector pixel, vector unsigned char);
13182 vector signed char vec_slo (vector signed char, vector signed char);
13183 vector signed char vec_slo (vector signed char, vector unsigned char);
13184 vector unsigned char vec_slo (vector unsigned char, vector signed char);
13185 vector unsigned char vec_slo (vector unsigned char,
13186 vector unsigned char);
13188 vector signed char vec_splat (vector signed char, const int);
13189 vector unsigned char vec_splat (vector unsigned char, const int);
13190 vector bool char vec_splat (vector bool char, const int);
13191 vector signed short vec_splat (vector signed short, const int);
13192 vector unsigned short vec_splat (vector unsigned short, const int);
13193 vector bool short vec_splat (vector bool short, const int);
13194 vector pixel vec_splat (vector pixel, const int);
13195 vector float vec_splat (vector float, const int);
13196 vector signed int vec_splat (vector signed int, const int);
13197 vector unsigned int vec_splat (vector unsigned int, const int);
13198 vector bool int vec_splat (vector bool int, const int);
13200 vector float vec_vspltw (vector float, const int);
13201 vector signed int vec_vspltw (vector signed int, const int);
13202 vector unsigned int vec_vspltw (vector unsigned int, const int);
13203 vector bool int vec_vspltw (vector bool int, const int);
13205 vector bool short vec_vsplth (vector bool short, const int);
13206 vector signed short vec_vsplth (vector signed short, const int);
13207 vector unsigned short vec_vsplth (vector unsigned short, const int);
13208 vector pixel vec_vsplth (vector pixel, const int);
13210 vector signed char vec_vspltb (vector signed char, const int);
13211 vector unsigned char vec_vspltb (vector unsigned char, const int);
13212 vector bool char vec_vspltb (vector bool char, const int);
13214 vector signed char vec_splat_s8 (const int);
13216 vector signed short vec_splat_s16 (const int);
13218 vector signed int vec_splat_s32 (const int);
13220 vector unsigned char vec_splat_u8 (const int);
13222 vector unsigned short vec_splat_u16 (const int);
13224 vector unsigned int vec_splat_u32 (const int);
13226 vector signed char vec_sr (vector signed char, vector unsigned char);
13227 vector unsigned char vec_sr (vector unsigned char,
13228 vector unsigned char);
13229 vector signed short vec_sr (vector signed short,
13230 vector unsigned short);
13231 vector unsigned short vec_sr (vector unsigned short,
13232 vector unsigned short);
13233 vector signed int vec_sr (vector signed int, vector unsigned int);
13234 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
13236 vector signed int vec_vsrw (vector signed int, vector unsigned int);
13237 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
13239 vector signed short vec_vsrh (vector signed short,
13240 vector unsigned short);
13241 vector unsigned short vec_vsrh (vector unsigned short,
13242 vector unsigned short);
13244 vector signed char vec_vsrb (vector signed char, vector unsigned char);
13245 vector unsigned char vec_vsrb (vector unsigned char,
13246 vector unsigned char);
13248 vector signed char vec_sra (vector signed char, vector unsigned char);
13249 vector unsigned char vec_sra (vector unsigned char,
13250 vector unsigned char);
13251 vector signed short vec_sra (vector signed short,
13252 vector unsigned short);
13253 vector unsigned short vec_sra (vector unsigned short,
13254 vector unsigned short);
13255 vector signed int vec_sra (vector signed int, vector unsigned int);
13256 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
13258 vector signed int vec_vsraw (vector signed int, vector unsigned int);
13259 vector unsigned int vec_vsraw (vector unsigned int,
13260 vector unsigned int);
13262 vector signed short vec_vsrah (vector signed short,
13263 vector unsigned short);
13264 vector unsigned short vec_vsrah (vector unsigned short,
13265 vector unsigned short);
13267 vector signed char vec_vsrab (vector signed char, vector unsigned char);
13268 vector unsigned char vec_vsrab (vector unsigned char,
13269 vector unsigned char);
13271 vector signed int vec_srl (vector signed int, vector unsigned int);
13272 vector signed int vec_srl (vector signed int, vector unsigned short);
13273 vector signed int vec_srl (vector signed int, vector unsigned char);
13274 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
13275 vector unsigned int vec_srl (vector unsigned int,
13276 vector unsigned short);
13277 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
13278 vector bool int vec_srl (vector bool int, vector unsigned int);
13279 vector bool int vec_srl (vector bool int, vector unsigned short);
13280 vector bool int vec_srl (vector bool int, vector unsigned char);
13281 vector signed short vec_srl (vector signed short, vector unsigned int);
13282 vector signed short vec_srl (vector signed short,
13283 vector unsigned short);
13284 vector signed short vec_srl (vector signed short, vector unsigned char);
13285 vector unsigned short vec_srl (vector unsigned short,
13286 vector unsigned int);
13287 vector unsigned short vec_srl (vector unsigned short,
13288 vector unsigned short);
13289 vector unsigned short vec_srl (vector unsigned short,
13290 vector unsigned char);
13291 vector bool short vec_srl (vector bool short, vector unsigned int);
13292 vector bool short vec_srl (vector bool short, vector unsigned short);
13293 vector bool short vec_srl (vector bool short, vector unsigned char);
13294 vector pixel vec_srl (vector pixel, vector unsigned int);
13295 vector pixel vec_srl (vector pixel, vector unsigned short);
13296 vector pixel vec_srl (vector pixel, vector unsigned char);
13297 vector signed char vec_srl (vector signed char, vector unsigned int);
13298 vector signed char vec_srl (vector signed char, vector unsigned short);
13299 vector signed char vec_srl (vector signed char, vector unsigned char);
13300 vector unsigned char vec_srl (vector unsigned char,
13301 vector unsigned int);
13302 vector unsigned char vec_srl (vector unsigned char,
13303 vector unsigned short);
13304 vector unsigned char vec_srl (vector unsigned char,
13305 vector unsigned char);
13306 vector bool char vec_srl (vector bool char, vector unsigned int);
13307 vector bool char vec_srl (vector bool char, vector unsigned short);
13308 vector bool char vec_srl (vector bool char, vector unsigned char);
13310 vector float vec_sro (vector float, vector signed char);
13311 vector float vec_sro (vector float, vector unsigned char);
13312 vector signed int vec_sro (vector signed int, vector signed char);
13313 vector signed int vec_sro (vector signed int, vector unsigned char);
13314 vector unsigned int vec_sro (vector unsigned int, vector signed char);
13315 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
13316 vector signed short vec_sro (vector signed short, vector signed char);
13317 vector signed short vec_sro (vector signed short, vector unsigned char);
13318 vector unsigned short vec_sro (vector unsigned short,
13319 vector signed char);
13320 vector unsigned short vec_sro (vector unsigned short,
13321 vector unsigned char);
13322 vector pixel vec_sro (vector pixel, vector signed char);
13323 vector pixel vec_sro (vector pixel, vector unsigned char);
13324 vector signed char vec_sro (vector signed char, vector signed char);
13325 vector signed char vec_sro (vector signed char, vector unsigned char);
13326 vector unsigned char vec_sro (vector unsigned char, vector signed char);
13327 vector unsigned char vec_sro (vector unsigned char,
13328 vector unsigned char);
13330 void vec_st (vector float, int, vector float *);
13331 void vec_st (vector float, int, float *);
13332 void vec_st (vector signed int, int, vector signed int *);
13333 void vec_st (vector signed int, int, int *);
13334 void vec_st (vector unsigned int, int, vector unsigned int *);
13335 void vec_st (vector unsigned int, int, unsigned int *);
13336 void vec_st (vector bool int, int, vector bool int *);
13337 void vec_st (vector bool int, int, unsigned int *);
13338 void vec_st (vector bool int, int, int *);
13339 void vec_st (vector signed short, int, vector signed short *);
13340 void vec_st (vector signed short, int, short *);
13341 void vec_st (vector unsigned short, int, vector unsigned short *);
13342 void vec_st (vector unsigned short, int, unsigned short *);
13343 void vec_st (vector bool short, int, vector bool short *);
13344 void vec_st (vector bool short, int, unsigned short *);
13345 void vec_st (vector pixel, int, vector pixel *);
13346 void vec_st (vector pixel, int, unsigned short *);
13347 void vec_st (vector pixel, int, short *);
13348 void vec_st (vector bool short, int, short *);
13349 void vec_st (vector signed char, int, vector signed char *);
13350 void vec_st (vector signed char, int, signed char *);
13351 void vec_st (vector unsigned char, int, vector unsigned char *);
13352 void vec_st (vector unsigned char, int, unsigned char *);
13353 void vec_st (vector bool char, int, vector bool char *);
13354 void vec_st (vector bool char, int, unsigned char *);
13355 void vec_st (vector bool char, int, signed char *);
13357 void vec_ste (vector signed char, int, signed char *);
13358 void vec_ste (vector unsigned char, int, unsigned char *);
13359 void vec_ste (vector bool char, int, signed char *);
13360 void vec_ste (vector bool char, int, unsigned char *);
13361 void vec_ste (vector signed short, int, short *);
13362 void vec_ste (vector unsigned short, int, unsigned short *);
13363 void vec_ste (vector bool short, int, short *);
13364 void vec_ste (vector bool short, int, unsigned short *);
13365 void vec_ste (vector pixel, int, short *);
13366 void vec_ste (vector pixel, int, unsigned short *);
13367 void vec_ste (vector float, int, float *);
13368 void vec_ste (vector signed int, int, int *);
13369 void vec_ste (vector unsigned int, int, unsigned int *);
13370 void vec_ste (vector bool int, int, int *);
13371 void vec_ste (vector bool int, int, unsigned int *);
13373 void vec_stvewx (vector float, int, float *);
13374 void vec_stvewx (vector signed int, int, int *);
13375 void vec_stvewx (vector unsigned int, int, unsigned int *);
13376 void vec_stvewx (vector bool int, int, int *);
13377 void vec_stvewx (vector bool int, int, unsigned int *);
13379 void vec_stvehx (vector signed short, int, short *);
13380 void vec_stvehx (vector unsigned short, int, unsigned short *);
13381 void vec_stvehx (vector bool short, int, short *);
13382 void vec_stvehx (vector bool short, int, unsigned short *);
13383 void vec_stvehx (vector pixel, int, short *);
13384 void vec_stvehx (vector pixel, int, unsigned short *);
13386 void vec_stvebx (vector signed char, int, signed char *);
13387 void vec_stvebx (vector unsigned char, int, unsigned char *);
13388 void vec_stvebx (vector bool char, int, signed char *);
13389 void vec_stvebx (vector bool char, int, unsigned char *);
13391 void vec_stl (vector float, int, vector float *);
13392 void vec_stl (vector float, int, float *);
13393 void vec_stl (vector signed int, int, vector signed int *);
13394 void vec_stl (vector signed int, int, int *);
13395 void vec_stl (vector unsigned int, int, vector unsigned int *);
13396 void vec_stl (vector unsigned int, int, unsigned int *);
13397 void vec_stl (vector bool int, int, vector bool int *);
13398 void vec_stl (vector bool int, int, unsigned int *);
13399 void vec_stl (vector bool int, int, int *);
13400 void vec_stl (vector signed short, int, vector signed short *);
13401 void vec_stl (vector signed short, int, short *);
13402 void vec_stl (vector unsigned short, int, vector unsigned short *);
13403 void vec_stl (vector unsigned short, int, unsigned short *);
13404 void vec_stl (vector bool short, int, vector bool short *);
13405 void vec_stl (vector bool short, int, unsigned short *);
13406 void vec_stl (vector bool short, int, short *);
13407 void vec_stl (vector pixel, int, vector pixel *);
13408 void vec_stl (vector pixel, int, unsigned short *);
13409 void vec_stl (vector pixel, int, short *);
13410 void vec_stl (vector signed char, int, vector signed char *);
13411 void vec_stl (vector signed char, int, signed char *);
13412 void vec_stl (vector unsigned char, int, vector unsigned char *);
13413 void vec_stl (vector unsigned char, int, unsigned char *);
13414 void vec_stl (vector bool char, int, vector bool char *);
13415 void vec_stl (vector bool char, int, unsigned char *);
13416 void vec_stl (vector bool char, int, signed char *);
13418 vector signed char vec_sub (vector bool char, vector signed char);
13419 vector signed char vec_sub (vector signed char, vector bool char);
13420 vector signed char vec_sub (vector signed char, vector signed char);
13421 vector unsigned char vec_sub (vector bool char, vector unsigned char);
13422 vector unsigned char vec_sub (vector unsigned char, vector bool char);
13423 vector unsigned char vec_sub (vector unsigned char,
13424 vector unsigned char);
13425 vector signed short vec_sub (vector bool short, vector signed short);
13426 vector signed short vec_sub (vector signed short, vector bool short);
13427 vector signed short vec_sub (vector signed short, vector signed short);
13428 vector unsigned short vec_sub (vector bool short,
13429 vector unsigned short);
13430 vector unsigned short vec_sub (vector unsigned short,
13431 vector bool short);
13432 vector unsigned short vec_sub (vector unsigned short,
13433 vector unsigned short);
13434 vector signed int vec_sub (vector bool int, vector signed int);
13435 vector signed int vec_sub (vector signed int, vector bool int);
13436 vector signed int vec_sub (vector signed int, vector signed int);
13437 vector unsigned int vec_sub (vector bool int, vector unsigned int);
13438 vector unsigned int vec_sub (vector unsigned int, vector bool int);
13439 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
13440 vector float vec_sub (vector float, vector float);
13442 vector float vec_vsubfp (vector float, vector float);
13444 vector signed int vec_vsubuwm (vector bool int, vector signed int);
13445 vector signed int vec_vsubuwm (vector signed int, vector bool int);
13446 vector signed int vec_vsubuwm (vector signed int, vector signed int);
13447 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
13448 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
13449 vector unsigned int vec_vsubuwm (vector unsigned int,
13450 vector unsigned int);
13452 vector signed short vec_vsubuhm (vector bool short,
13453 vector signed short);
13454 vector signed short vec_vsubuhm (vector signed short,
13455 vector bool short);
13456 vector signed short vec_vsubuhm (vector signed short,
13457 vector signed short);
13458 vector unsigned short vec_vsubuhm (vector bool short,
13459 vector unsigned short);
13460 vector unsigned short vec_vsubuhm (vector unsigned short,
13461 vector bool short);
13462 vector unsigned short vec_vsubuhm (vector unsigned short,
13463 vector unsigned short);
13465 vector signed char vec_vsububm (vector bool char, vector signed char);
13466 vector signed char vec_vsububm (vector signed char, vector bool char);
13467 vector signed char vec_vsububm (vector signed char, vector signed char);
13468 vector unsigned char vec_vsububm (vector bool char,
13469 vector unsigned char);
13470 vector unsigned char vec_vsububm (vector unsigned char,
13472 vector unsigned char vec_vsububm (vector unsigned char,
13473 vector unsigned char);
13475 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
13477 vector unsigned char vec_subs (vector bool char, vector unsigned char);
13478 vector unsigned char vec_subs (vector unsigned char, vector bool char);
13479 vector unsigned char vec_subs (vector unsigned char,
13480 vector unsigned char);
13481 vector signed char vec_subs (vector bool char, vector signed char);
13482 vector signed char vec_subs (vector signed char, vector bool char);
13483 vector signed char vec_subs (vector signed char, vector signed char);
13484 vector unsigned short vec_subs (vector bool short,
13485 vector unsigned short);
13486 vector unsigned short vec_subs (vector unsigned short,
13487 vector bool short);
13488 vector unsigned short vec_subs (vector unsigned short,
13489 vector unsigned short);
13490 vector signed short vec_subs (vector bool short, vector signed short);
13491 vector signed short vec_subs (vector signed short, vector bool short);
13492 vector signed short vec_subs (vector signed short, vector signed short);
13493 vector unsigned int vec_subs (vector bool int, vector unsigned int);
13494 vector unsigned int vec_subs (vector unsigned int, vector bool int);
13495 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
13496 vector signed int vec_subs (vector bool int, vector signed int);
13497 vector signed int vec_subs (vector signed int, vector bool int);
13498 vector signed int vec_subs (vector signed int, vector signed int);
13500 vector signed int vec_vsubsws (vector bool int, vector signed int);
13501 vector signed int vec_vsubsws (vector signed int, vector bool int);
13502 vector signed int vec_vsubsws (vector signed int, vector signed int);
13504 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
13505 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
13506 vector unsigned int vec_vsubuws (vector unsigned int,
13507 vector unsigned int);
13509 vector signed short vec_vsubshs (vector bool short,
13510 vector signed short);
13511 vector signed short vec_vsubshs (vector signed short,
13512 vector bool short);
13513 vector signed short vec_vsubshs (vector signed short,
13514 vector signed short);
13516 vector unsigned short vec_vsubuhs (vector bool short,
13517 vector unsigned short);
13518 vector unsigned short vec_vsubuhs (vector unsigned short,
13519 vector bool short);
13520 vector unsigned short vec_vsubuhs (vector unsigned short,
13521 vector unsigned short);
13523 vector signed char vec_vsubsbs (vector bool char, vector signed char);
13524 vector signed char vec_vsubsbs (vector signed char, vector bool char);
13525 vector signed char vec_vsubsbs (vector signed char, vector signed char);
13527 vector unsigned char vec_vsububs (vector bool char,
13528 vector unsigned char);
13529 vector unsigned char vec_vsububs (vector unsigned char,
13531 vector unsigned char vec_vsububs (vector unsigned char,
13532 vector unsigned char);
13534 vector unsigned int vec_sum4s (vector unsigned char,
13535 vector unsigned int);
13536 vector signed int vec_sum4s (vector signed char, vector signed int);
13537 vector signed int vec_sum4s (vector signed short, vector signed int);
13539 vector signed int vec_vsum4shs (vector signed short, vector signed int);
13541 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
13543 vector unsigned int vec_vsum4ubs (vector unsigned char,
13544 vector unsigned int);
13546 vector signed int vec_sum2s (vector signed int, vector signed int);
13548 vector signed int vec_sums (vector signed int, vector signed int);
13550 vector float vec_trunc (vector float);
13552 vector signed short vec_unpackh (vector signed char);
13553 vector bool short vec_unpackh (vector bool char);
13554 vector signed int vec_unpackh (vector signed short);
13555 vector bool int vec_unpackh (vector bool short);
13556 vector unsigned int vec_unpackh (vector pixel);
13558 vector bool int vec_vupkhsh (vector bool short);
13559 vector signed int vec_vupkhsh (vector signed short);
13561 vector unsigned int vec_vupkhpx (vector pixel);
13563 vector bool short vec_vupkhsb (vector bool char);
13564 vector signed short vec_vupkhsb (vector signed char);
13566 vector signed short vec_unpackl (vector signed char);
13567 vector bool short vec_unpackl (vector bool char);
13568 vector unsigned int vec_unpackl (vector pixel);
13569 vector signed int vec_unpackl (vector signed short);
13570 vector bool int vec_unpackl (vector bool short);
13572 vector unsigned int vec_vupklpx (vector pixel);
13574 vector bool int vec_vupklsh (vector bool short);
13575 vector signed int vec_vupklsh (vector signed short);
13577 vector bool short vec_vupklsb (vector bool char);
13578 vector signed short vec_vupklsb (vector signed char);
13580 vector float vec_xor (vector float, vector float);
13581 vector float vec_xor (vector float, vector bool int);
13582 vector float vec_xor (vector bool int, vector float);
13583 vector bool int vec_xor (vector bool int, vector bool int);
13584 vector signed int vec_xor (vector bool int, vector signed int);
13585 vector signed int vec_xor (vector signed int, vector bool int);
13586 vector signed int vec_xor (vector signed int, vector signed int);
13587 vector unsigned int vec_xor (vector bool int, vector unsigned int);
13588 vector unsigned int vec_xor (vector unsigned int, vector bool int);
13589 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
13590 vector bool short vec_xor (vector bool short, vector bool short);
13591 vector signed short vec_xor (vector bool short, vector signed short);
13592 vector signed short vec_xor (vector signed short, vector bool short);
13593 vector signed short vec_xor (vector signed short, vector signed short);
13594 vector unsigned short vec_xor (vector bool short,
13595 vector unsigned short);
13596 vector unsigned short vec_xor (vector unsigned short,
13597 vector bool short);
13598 vector unsigned short vec_xor (vector unsigned short,
13599 vector unsigned short);
13600 vector signed char vec_xor (vector bool char, vector signed char);
13601 vector bool char vec_xor (vector bool char, vector bool char);
13602 vector signed char vec_xor (vector signed char, vector bool char);
13603 vector signed char vec_xor (vector signed char, vector signed char);
13604 vector unsigned char vec_xor (vector bool char, vector unsigned char);
13605 vector unsigned char vec_xor (vector unsigned char, vector bool char);
13606 vector unsigned char vec_xor (vector unsigned char,
13607 vector unsigned char);
13609 int vec_all_eq (vector signed char, vector bool char);
13610 int vec_all_eq (vector signed char, vector signed char);
13611 int vec_all_eq (vector unsigned char, vector bool char);
13612 int vec_all_eq (vector unsigned char, vector unsigned char);
13613 int vec_all_eq (vector bool char, vector bool char);
13614 int vec_all_eq (vector bool char, vector unsigned char);
13615 int vec_all_eq (vector bool char, vector signed char);
13616 int vec_all_eq (vector signed short, vector bool short);
13617 int vec_all_eq (vector signed short, vector signed short);
13618 int vec_all_eq (vector unsigned short, vector bool short);
13619 int vec_all_eq (vector unsigned short, vector unsigned short);
13620 int vec_all_eq (vector bool short, vector bool short);
13621 int vec_all_eq (vector bool short, vector unsigned short);
13622 int vec_all_eq (vector bool short, vector signed short);
13623 int vec_all_eq (vector pixel, vector pixel);
13624 int vec_all_eq (vector signed int, vector bool int);
13625 int vec_all_eq (vector signed int, vector signed int);
13626 int vec_all_eq (vector unsigned int, vector bool int);
13627 int vec_all_eq (vector unsigned int, vector unsigned int);
13628 int vec_all_eq (vector bool int, vector bool int);
13629 int vec_all_eq (vector bool int, vector unsigned int);
13630 int vec_all_eq (vector bool int, vector signed int);
13631 int vec_all_eq (vector float, vector float);
13633 int vec_all_ge (vector bool char, vector unsigned char);
13634 int vec_all_ge (vector unsigned char, vector bool char);
13635 int vec_all_ge (vector unsigned char, vector unsigned char);
13636 int vec_all_ge (vector bool char, vector signed char);
13637 int vec_all_ge (vector signed char, vector bool char);
13638 int vec_all_ge (vector signed char, vector signed char);
13639 int vec_all_ge (vector bool short, vector unsigned short);
13640 int vec_all_ge (vector unsigned short, vector bool short);
13641 int vec_all_ge (vector unsigned short, vector unsigned short);
13642 int vec_all_ge (vector signed short, vector signed short);
13643 int vec_all_ge (vector bool short, vector signed short);
13644 int vec_all_ge (vector signed short, vector bool short);
13645 int vec_all_ge (vector bool int, vector unsigned int);
13646 int vec_all_ge (vector unsigned int, vector bool int);
13647 int vec_all_ge (vector unsigned int, vector unsigned int);
13648 int vec_all_ge (vector bool int, vector signed int);
13649 int vec_all_ge (vector signed int, vector bool int);
13650 int vec_all_ge (vector signed int, vector signed int);
13651 int vec_all_ge (vector float, vector float);
13653 int vec_all_gt (vector bool char, vector unsigned char);
13654 int vec_all_gt (vector unsigned char, vector bool char);
13655 int vec_all_gt (vector unsigned char, vector unsigned char);
13656 int vec_all_gt (vector bool char, vector signed char);
13657 int vec_all_gt (vector signed char, vector bool char);
13658 int vec_all_gt (vector signed char, vector signed char);
13659 int vec_all_gt (vector bool short, vector unsigned short);
13660 int vec_all_gt (vector unsigned short, vector bool short);
13661 int vec_all_gt (vector unsigned short, vector unsigned short);
13662 int vec_all_gt (vector bool short, vector signed short);
13663 int vec_all_gt (vector signed short, vector bool short);
13664 int vec_all_gt (vector signed short, vector signed short);
13665 int vec_all_gt (vector bool int, vector unsigned int);
13666 int vec_all_gt (vector unsigned int, vector bool int);
13667 int vec_all_gt (vector unsigned int, vector unsigned int);
13668 int vec_all_gt (vector bool int, vector signed int);
13669 int vec_all_gt (vector signed int, vector bool int);
13670 int vec_all_gt (vector signed int, vector signed int);
13671 int vec_all_gt (vector float, vector float);
13673 int vec_all_in (vector float, vector float);
13675 int vec_all_le (vector bool char, vector unsigned char);
13676 int vec_all_le (vector unsigned char, vector bool char);
13677 int vec_all_le (vector unsigned char, vector unsigned char);
13678 int vec_all_le (vector bool char, vector signed char);
13679 int vec_all_le (vector signed char, vector bool char);
13680 int vec_all_le (vector signed char, vector signed char);
13681 int vec_all_le (vector bool short, vector unsigned short);
13682 int vec_all_le (vector unsigned short, vector bool short);
13683 int vec_all_le (vector unsigned short, vector unsigned short);
13684 int vec_all_le (vector bool short, vector signed short);
13685 int vec_all_le (vector signed short, vector bool short);
13686 int vec_all_le (vector signed short, vector signed short);
13687 int vec_all_le (vector bool int, vector unsigned int);
13688 int vec_all_le (vector unsigned int, vector bool int);
13689 int vec_all_le (vector unsigned int, vector unsigned int);
13690 int vec_all_le (vector bool int, vector signed int);
13691 int vec_all_le (vector signed int, vector bool int);
13692 int vec_all_le (vector signed int, vector signed int);
13693 int vec_all_le (vector float, vector float);
13695 int vec_all_lt (vector bool char, vector unsigned char);
13696 int vec_all_lt (vector unsigned char, vector bool char);
13697 int vec_all_lt (vector unsigned char, vector unsigned char);
13698 int vec_all_lt (vector bool char, vector signed char);
13699 int vec_all_lt (vector signed char, vector bool char);
13700 int vec_all_lt (vector signed char, vector signed char);
13701 int vec_all_lt (vector bool short, vector unsigned short);
13702 int vec_all_lt (vector unsigned short, vector bool short);
13703 int vec_all_lt (vector unsigned short, vector unsigned short);
13704 int vec_all_lt (vector bool short, vector signed short);
13705 int vec_all_lt (vector signed short, vector bool short);
13706 int vec_all_lt (vector signed short, vector signed short);
13707 int vec_all_lt (vector bool int, vector unsigned int);
13708 int vec_all_lt (vector unsigned int, vector bool int);
13709 int vec_all_lt (vector unsigned int, vector unsigned int);
13710 int vec_all_lt (vector bool int, vector signed int);
13711 int vec_all_lt (vector signed int, vector bool int);
13712 int vec_all_lt (vector signed int, vector signed int);
13713 int vec_all_lt (vector float, vector float);
13715 int vec_all_nan (vector float);
13717 int vec_all_ne (vector signed char, vector bool char);
13718 int vec_all_ne (vector signed char, vector signed char);
13719 int vec_all_ne (vector unsigned char, vector bool char);
13720 int vec_all_ne (vector unsigned char, vector unsigned char);
13721 int vec_all_ne (vector bool char, vector bool char);
13722 int vec_all_ne (vector bool char, vector unsigned char);
13723 int vec_all_ne (vector bool char, vector signed char);
13724 int vec_all_ne (vector signed short, vector bool short);
13725 int vec_all_ne (vector signed short, vector signed short);
13726 int vec_all_ne (vector unsigned short, vector bool short);
13727 int vec_all_ne (vector unsigned short, vector unsigned short);
13728 int vec_all_ne (vector bool short, vector bool short);
13729 int vec_all_ne (vector bool short, vector unsigned short);
13730 int vec_all_ne (vector bool short, vector signed short);
13731 int vec_all_ne (vector pixel, vector pixel);
13732 int vec_all_ne (vector signed int, vector bool int);
13733 int vec_all_ne (vector signed int, vector signed int);
13734 int vec_all_ne (vector unsigned int, vector bool int);
13735 int vec_all_ne (vector unsigned int, vector unsigned int);
13736 int vec_all_ne (vector bool int, vector bool int);
13737 int vec_all_ne (vector bool int, vector unsigned int);
13738 int vec_all_ne (vector bool int, vector signed int);
13739 int vec_all_ne (vector float, vector float);
13741 int vec_all_nge (vector float, vector float);
13743 int vec_all_ngt (vector float, vector float);
13745 int vec_all_nle (vector float, vector float);
13747 int vec_all_nlt (vector float, vector float);
13749 int vec_all_numeric (vector float);
13751 int vec_any_eq (vector signed char, vector bool char);
13752 int vec_any_eq (vector signed char, vector signed char);
13753 int vec_any_eq (vector unsigned char, vector bool char);
13754 int vec_any_eq (vector unsigned char, vector unsigned char);
13755 int vec_any_eq (vector bool char, vector bool char);
13756 int vec_any_eq (vector bool char, vector unsigned char);
13757 int vec_any_eq (vector bool char, vector signed char);
13758 int vec_any_eq (vector signed short, vector bool short);
13759 int vec_any_eq (vector signed short, vector signed short);
13760 int vec_any_eq (vector unsigned short, vector bool short);
13761 int vec_any_eq (vector unsigned short, vector unsigned short);
13762 int vec_any_eq (vector bool short, vector bool short);
13763 int vec_any_eq (vector bool short, vector unsigned short);
13764 int vec_any_eq (vector bool short, vector signed short);
13765 int vec_any_eq (vector pixel, vector pixel);
13766 int vec_any_eq (vector signed int, vector bool int);
13767 int vec_any_eq (vector signed int, vector signed int);
13768 int vec_any_eq (vector unsigned int, vector bool int);
13769 int vec_any_eq (vector unsigned int, vector unsigned int);
13770 int vec_any_eq (vector bool int, vector bool int);
13771 int vec_any_eq (vector bool int, vector unsigned int);
13772 int vec_any_eq (vector bool int, vector signed int);
13773 int vec_any_eq (vector float, vector float);
13775 int vec_any_ge (vector signed char, vector bool char);
13776 int vec_any_ge (vector unsigned char, vector bool char);
13777 int vec_any_ge (vector unsigned char, vector unsigned char);
13778 int vec_any_ge (vector signed char, vector signed char);
13779 int vec_any_ge (vector bool char, vector unsigned char);
13780 int vec_any_ge (vector bool char, vector signed char);
13781 int vec_any_ge (vector unsigned short, vector bool short);
13782 int vec_any_ge (vector unsigned short, vector unsigned short);
13783 int vec_any_ge (vector signed short, vector signed short);
13784 int vec_any_ge (vector signed short, vector bool short);
13785 int vec_any_ge (vector bool short, vector unsigned short);
13786 int vec_any_ge (vector bool short, vector signed short);
13787 int vec_any_ge (vector signed int, vector bool int);
13788 int vec_any_ge (vector unsigned int, vector bool int);
13789 int vec_any_ge (vector unsigned int, vector unsigned int);
13790 int vec_any_ge (vector signed int, vector signed int);
13791 int vec_any_ge (vector bool int, vector unsigned int);
13792 int vec_any_ge (vector bool int, vector signed int);
13793 int vec_any_ge (vector float, vector float);
13795 int vec_any_gt (vector bool char, vector unsigned char);
13796 int vec_any_gt (vector unsigned char, vector bool char);
13797 int vec_any_gt (vector unsigned char, vector unsigned char);
13798 int vec_any_gt (vector bool char, vector signed char);
13799 int vec_any_gt (vector signed char, vector bool char);
13800 int vec_any_gt (vector signed char, vector signed char);
13801 int vec_any_gt (vector bool short, vector unsigned short);
13802 int vec_any_gt (vector unsigned short, vector bool short);
13803 int vec_any_gt (vector unsigned short, vector unsigned short);
13804 int vec_any_gt (vector bool short, vector signed short);
13805 int vec_any_gt (vector signed short, vector bool short);
13806 int vec_any_gt (vector signed short, vector signed short);
13807 int vec_any_gt (vector bool int, vector unsigned int);
13808 int vec_any_gt (vector unsigned int, vector bool int);
13809 int vec_any_gt (vector unsigned int, vector unsigned int);
13810 int vec_any_gt (vector bool int, vector signed int);
13811 int vec_any_gt (vector signed int, vector bool int);
13812 int vec_any_gt (vector signed int, vector signed int);
13813 int vec_any_gt (vector float, vector float);
13815 int vec_any_le (vector bool char, vector unsigned char);
13816 int vec_any_le (vector unsigned char, vector bool char);
13817 int vec_any_le (vector unsigned char, vector unsigned char);
13818 int vec_any_le (vector bool char, vector signed char);
13819 int vec_any_le (vector signed char, vector bool char);
13820 int vec_any_le (vector signed char, vector signed char);
13821 int vec_any_le (vector bool short, vector unsigned short);
13822 int vec_any_le (vector unsigned short, vector bool short);
13823 int vec_any_le (vector unsigned short, vector unsigned short);
13824 int vec_any_le (vector bool short, vector signed short);
13825 int vec_any_le (vector signed short, vector bool short);
13826 int vec_any_le (vector signed short, vector signed short);
13827 int vec_any_le (vector bool int, vector unsigned int);
13828 int vec_any_le (vector unsigned int, vector bool int);
13829 int vec_any_le (vector unsigned int, vector unsigned int);
13830 int vec_any_le (vector bool int, vector signed int);
13831 int vec_any_le (vector signed int, vector bool int);
13832 int vec_any_le (vector signed int, vector signed int);
13833 int vec_any_le (vector float, vector float);
13835 int vec_any_lt (vector bool char, vector unsigned char);
13836 int vec_any_lt (vector unsigned char, vector bool char);
13837 int vec_any_lt (vector unsigned char, vector unsigned char);
13838 int vec_any_lt (vector bool char, vector signed char);
13839 int vec_any_lt (vector signed char, vector bool char);
13840 int vec_any_lt (vector signed char, vector signed char);
13841 int vec_any_lt (vector bool short, vector unsigned short);
13842 int vec_any_lt (vector unsigned short, vector bool short);
13843 int vec_any_lt (vector unsigned short, vector unsigned short);
13844 int vec_any_lt (vector bool short, vector signed short);
13845 int vec_any_lt (vector signed short, vector bool short);
13846 int vec_any_lt (vector signed short, vector signed short);
13847 int vec_any_lt (vector bool int, vector unsigned int);
13848 int vec_any_lt (vector unsigned int, vector bool int);
13849 int vec_any_lt (vector unsigned int, vector unsigned int);
13850 int vec_any_lt (vector bool int, vector signed int);
13851 int vec_any_lt (vector signed int, vector bool int);
13852 int vec_any_lt (vector signed int, vector signed int);
13853 int vec_any_lt (vector float, vector float);
13855 int vec_any_nan (vector float);
13857 int vec_any_ne (vector signed char, vector bool char);
13858 int vec_any_ne (vector signed char, vector signed char);
13859 int vec_any_ne (vector unsigned char, vector bool char);
13860 int vec_any_ne (vector unsigned char, vector unsigned char);
13861 int vec_any_ne (vector bool char, vector bool char);
13862 int vec_any_ne (vector bool char, vector unsigned char);
13863 int vec_any_ne (vector bool char, vector signed char);
13864 int vec_any_ne (vector signed short, vector bool short);
13865 int vec_any_ne (vector signed short, vector signed short);
13866 int vec_any_ne (vector unsigned short, vector bool short);
13867 int vec_any_ne (vector unsigned short, vector unsigned short);
13868 int vec_any_ne (vector bool short, vector bool short);
13869 int vec_any_ne (vector bool short, vector unsigned short);
13870 int vec_any_ne (vector bool short, vector signed short);
13871 int vec_any_ne (vector pixel, vector pixel);
13872 int vec_any_ne (vector signed int, vector bool int);
13873 int vec_any_ne (vector signed int, vector signed int);
13874 int vec_any_ne (vector unsigned int, vector bool int);
13875 int vec_any_ne (vector unsigned int, vector unsigned int);
13876 int vec_any_ne (vector bool int, vector bool int);
13877 int vec_any_ne (vector bool int, vector unsigned int);
13878 int vec_any_ne (vector bool int, vector signed int);
13879 int vec_any_ne (vector float, vector float);
13881 int vec_any_nge (vector float, vector float);
13883 int vec_any_ngt (vector float, vector float);
13885 int vec_any_nle (vector float, vector float);
13887 int vec_any_nlt (vector float, vector float);
13889 int vec_any_numeric (vector float);
13891 int vec_any_out (vector float, vector float);
13894 If the vector/scalar (VSX) instruction set is available, the following
13895 additional functions are available:
13898 vector double vec_abs (vector double);
13899 vector double vec_add (vector double, vector double);
13900 vector double vec_and (vector double, vector double);
13901 vector double vec_and (vector double, vector bool long);
13902 vector double vec_and (vector bool long, vector double);
13903 vector double vec_andc (vector double, vector double);
13904 vector double vec_andc (vector double, vector bool long);
13905 vector double vec_andc (vector bool long, vector double);
13906 vector double vec_ceil (vector double);
13907 vector bool long vec_cmpeq (vector double, vector double);
13908 vector bool long vec_cmpge (vector double, vector double);
13909 vector bool long vec_cmpgt (vector double, vector double);
13910 vector bool long vec_cmple (vector double, vector double);
13911 vector bool long vec_cmplt (vector double, vector double);
13912 vector float vec_div (vector float, vector float);
13913 vector double vec_div (vector double, vector double);
13914 vector double vec_floor (vector double);
13915 vector double vec_ld (int, const vector double *);
13916 vector double vec_ld (int, const double *);
13917 vector double vec_ldl (int, const vector double *);
13918 vector double vec_ldl (int, const double *);
13919 vector unsigned char vec_lvsl (int, const volatile double *);
13920 vector unsigned char vec_lvsr (int, const volatile double *);
13921 vector double vec_madd (vector double, vector double, vector double);
13922 vector double vec_max (vector double, vector double);
13923 vector double vec_min (vector double, vector double);
13924 vector float vec_msub (vector float, vector float, vector float);
13925 vector double vec_msub (vector double, vector double, vector double);
13926 vector float vec_mul (vector float, vector float);
13927 vector double vec_mul (vector double, vector double);
13928 vector float vec_nearbyint (vector float);
13929 vector double vec_nearbyint (vector double);
13930 vector float vec_nmadd (vector float, vector float, vector float);
13931 vector double vec_nmadd (vector double, vector double, vector double);
13932 vector double vec_nmsub (vector double, vector double, vector double);
13933 vector double vec_nor (vector double, vector double);
13934 vector double vec_or (vector double, vector double);
13935 vector double vec_or (vector double, vector bool long);
13936 vector double vec_or (vector bool long, vector double);
13937 vector double vec_perm (vector double,
13939 vector unsigned char);
13940 vector double vec_rint (vector double);
13941 vector double vec_recip (vector double, vector double);
13942 vector double vec_rsqrt (vector double);
13943 vector double vec_rsqrte (vector double);
13944 vector double vec_sel (vector double, vector double, vector bool long);
13945 vector double vec_sel (vector double, vector double, vector unsigned long);
13946 vector double vec_sub (vector double, vector double);
13947 vector float vec_sqrt (vector float);
13948 vector double vec_sqrt (vector double);
13949 void vec_st (vector double, int, vector double *);
13950 void vec_st (vector double, int, double *);
13951 vector double vec_trunc (vector double);
13952 vector double vec_xor (vector double, vector double);
13953 vector double vec_xor (vector double, vector bool long);
13954 vector double vec_xor (vector bool long, vector double);
13955 int vec_all_eq (vector double, vector double);
13956 int vec_all_ge (vector double, vector double);
13957 int vec_all_gt (vector double, vector double);
13958 int vec_all_le (vector double, vector double);
13959 int vec_all_lt (vector double, vector double);
13960 int vec_all_nan (vector double);
13961 int vec_all_ne (vector double, vector double);
13962 int vec_all_nge (vector double, vector double);
13963 int vec_all_ngt (vector double, vector double);
13964 int vec_all_nle (vector double, vector double);
13965 int vec_all_nlt (vector double, vector double);
13966 int vec_all_numeric (vector double);
13967 int vec_any_eq (vector double, vector double);
13968 int vec_any_ge (vector double, vector double);
13969 int vec_any_gt (vector double, vector double);
13970 int vec_any_le (vector double, vector double);
13971 int vec_any_lt (vector double, vector double);
13972 int vec_any_nan (vector double);
13973 int vec_any_ne (vector double, vector double);
13974 int vec_any_nge (vector double, vector double);
13975 int vec_any_ngt (vector double, vector double);
13976 int vec_any_nle (vector double, vector double);
13977 int vec_any_nlt (vector double, vector double);
13978 int vec_any_numeric (vector double);
13980 vector double vec_vsx_ld (int, const vector double *);
13981 vector double vec_vsx_ld (int, const double *);
13982 vector float vec_vsx_ld (int, const vector float *);
13983 vector float vec_vsx_ld (int, const float *);
13984 vector bool int vec_vsx_ld (int, const vector bool int *);
13985 vector signed int vec_vsx_ld (int, const vector signed int *);
13986 vector signed int vec_vsx_ld (int, const int *);
13987 vector signed int vec_vsx_ld (int, const long *);
13988 vector unsigned int vec_vsx_ld (int, const vector unsigned int *);
13989 vector unsigned int vec_vsx_ld (int, const unsigned int *);
13990 vector unsigned int vec_vsx_ld (int, const unsigned long *);
13991 vector bool short vec_vsx_ld (int, const vector bool short *);
13992 vector pixel vec_vsx_ld (int, const vector pixel *);
13993 vector signed short vec_vsx_ld (int, const vector signed short *);
13994 vector signed short vec_vsx_ld (int, const short *);
13995 vector unsigned short vec_vsx_ld (int, const vector unsigned short *);
13996 vector unsigned short vec_vsx_ld (int, const unsigned short *);
13997 vector bool char vec_vsx_ld (int, const vector bool char *);
13998 vector signed char vec_vsx_ld (int, const vector signed char *);
13999 vector signed char vec_vsx_ld (int, const signed char *);
14000 vector unsigned char vec_vsx_ld (int, const vector unsigned char *);
14001 vector unsigned char vec_vsx_ld (int, const unsigned char *);
14003 void vec_vsx_st (vector double, int, vector double *);
14004 void vec_vsx_st (vector double, int, double *);
14005 void vec_vsx_st (vector float, int, vector float *);
14006 void vec_vsx_st (vector float, int, float *);
14007 void vec_vsx_st (vector signed int, int, vector signed int *);
14008 void vec_vsx_st (vector signed int, int, int *);
14009 void vec_vsx_st (vector unsigned int, int, vector unsigned int *);
14010 void vec_vsx_st (vector unsigned int, int, unsigned int *);
14011 void vec_vsx_st (vector bool int, int, vector bool int *);
14012 void vec_vsx_st (vector bool int, int, unsigned int *);
14013 void vec_vsx_st (vector bool int, int, int *);
14014 void vec_vsx_st (vector signed short, int, vector signed short *);
14015 void vec_vsx_st (vector signed short, int, short *);
14016 void vec_vsx_st (vector unsigned short, int, vector unsigned short *);
14017 void vec_vsx_st (vector unsigned short, int, unsigned short *);
14018 void vec_vsx_st (vector bool short, int, vector bool short *);
14019 void vec_vsx_st (vector bool short, int, unsigned short *);
14020 void vec_vsx_st (vector pixel, int, vector pixel *);
14021 void vec_vsx_st (vector pixel, int, unsigned short *);
14022 void vec_vsx_st (vector pixel, int, short *);
14023 void vec_vsx_st (vector bool short, int, short *);
14024 void vec_vsx_st (vector signed char, int, vector signed char *);
14025 void vec_vsx_st (vector signed char, int, signed char *);
14026 void vec_vsx_st (vector unsigned char, int, vector unsigned char *);
14027 void vec_vsx_st (vector unsigned char, int, unsigned char *);
14028 void vec_vsx_st (vector bool char, int, vector bool char *);
14029 void vec_vsx_st (vector bool char, int, unsigned char *);
14030 void vec_vsx_st (vector bool char, int, signed char *);
14033 Note that the @samp{vec_ld} and @samp{vec_st} built-in functions always
14034 generate the AltiVec @samp{LVX} and @samp{STVX} instructions even
14035 if the VSX instruction set is available. The @samp{vec_vsx_ld} and
14036 @samp{vec_vsx_st} built-in functions always generate the VSX @samp{LXVD2X},
14037 @samp{LXVW4X}, @samp{STXVD2X}, and @samp{STXVW4X} instructions.
14039 If the ISA 2.07 additions to the vector/scalar (power8-vector)
14040 instruction set is available, the following additional functions are
14041 available for both 32-bit and 64-bit targets. For 64-bit targets, you
14042 can use @var{vector long} instead of @var{vector long long},
14043 @var{vector bool long} instead of @var{vector bool long long}, and
14044 @var{vector unsigned long} instead of @var{vector unsigned long long}.
14047 vector long long vec_abs (vector long long);
14049 vector long long vec_add (vector long long, vector long long);
14050 vector unsigned long long vec_add (vector unsigned long long,
14051 vector unsigned long long);
14053 int vec_all_eq (vector long long, vector long long);
14054 int vec_all_ge (vector long long, vector long long);
14055 int vec_all_gt (vector long long, vector long long);
14056 int vec_all_le (vector long long, vector long long);
14057 int vec_all_lt (vector long long, vector long long);
14058 int vec_all_ne (vector long long, vector long long);
14059 int vec_any_eq (vector long long, vector long long);
14060 int vec_any_ge (vector long long, vector long long);
14061 int vec_any_gt (vector long long, vector long long);
14062 int vec_any_le (vector long long, vector long long);
14063 int vec_any_lt (vector long long, vector long long);
14064 int vec_any_ne (vector long long, vector long long);
14066 vector long long vec_eqv (vector long long, vector long long);
14067 vector long long vec_eqv (vector bool long long, vector long long);
14068 vector long long vec_eqv (vector long long, vector bool long long);
14069 vector unsigned long long vec_eqv (vector unsigned long long,
14070 vector unsigned long long);
14071 vector unsigned long long vec_eqv (vector bool long long,
14072 vector unsigned long long);
14073 vector unsigned long long vec_eqv (vector unsigned long long,
14074 vector bool long long);
14075 vector int vec_eqv (vector int, vector int);
14076 vector int vec_eqv (vector bool int, vector int);
14077 vector int vec_eqv (vector int, vector bool int);
14078 vector unsigned int vec_eqv (vector unsigned int, vector unsigned int);
14079 vector unsigned int vec_eqv (vector bool unsigned int,
14080 vector unsigned int);
14081 vector unsigned int vec_eqv (vector unsigned int,
14082 vector bool unsigned int);
14083 vector short vec_eqv (vector short, vector short);
14084 vector short vec_eqv (vector bool short, vector short);
14085 vector short vec_eqv (vector short, vector bool short);
14086 vector unsigned short vec_eqv (vector unsigned short, vector unsigned short);
14087 vector unsigned short vec_eqv (vector bool unsigned short,
14088 vector unsigned short);
14089 vector unsigned short vec_eqv (vector unsigned short,
14090 vector bool unsigned short);
14091 vector signed char vec_eqv (vector signed char, vector signed char);
14092 vector signed char vec_eqv (vector bool signed char, vector signed char);
14093 vector signed char vec_eqv (vector signed char, vector bool signed char);
14094 vector unsigned char vec_eqv (vector unsigned char, vector unsigned char);
14095 vector unsigned char vec_eqv (vector bool unsigned char, vector unsigned char);
14096 vector unsigned char vec_eqv (vector unsigned char, vector bool unsigned char);
14098 vector long long vec_max (vector long long, vector long long);
14099 vector unsigned long long vec_max (vector unsigned long long,
14100 vector unsigned long long);
14102 vector long long vec_min (vector long long, vector long long);
14103 vector unsigned long long vec_min (vector unsigned long long,
14104 vector unsigned long long);
14106 vector long long vec_nand (vector long long, vector long long);
14107 vector long long vec_nand (vector bool long long, vector long long);
14108 vector long long vec_nand (vector long long, vector bool long long);
14109 vector unsigned long long vec_nand (vector unsigned long long,
14110 vector unsigned long long);
14111 vector unsigned long long vec_nand (vector bool long long,
14112 vector unsigned long long);
14113 vector unsigned long long vec_nand (vector unsigned long long,
14114 vector bool long long);
14115 vector int vec_nand (vector int, vector int);
14116 vector int vec_nand (vector bool int, vector int);
14117 vector int vec_nand (vector int, vector bool int);
14118 vector unsigned int vec_nand (vector unsigned int, vector unsigned int);
14119 vector unsigned int vec_nand (vector bool unsigned int,
14120 vector unsigned int);
14121 vector unsigned int vec_nand (vector unsigned int,
14122 vector bool unsigned int);
14123 vector short vec_nand (vector short, vector short);
14124 vector short vec_nand (vector bool short, vector short);
14125 vector short vec_nand (vector short, vector bool short);
14126 vector unsigned short vec_nand (vector unsigned short, vector unsigned short);
14127 vector unsigned short vec_nand (vector bool unsigned short,
14128 vector unsigned short);
14129 vector unsigned short vec_nand (vector unsigned short,
14130 vector bool unsigned short);
14131 vector signed char vec_nand (vector signed char, vector signed char);
14132 vector signed char vec_nand (vector bool signed char, vector signed char);
14133 vector signed char vec_nand (vector signed char, vector bool signed char);
14134 vector unsigned char vec_nand (vector unsigned char, vector unsigned char);
14135 vector unsigned char vec_nand (vector bool unsigned char, vector unsigned char);
14136 vector unsigned char vec_nand (vector unsigned char, vector bool unsigned char);
14138 vector long long vec_orc (vector long long, vector long long);
14139 vector long long vec_orc (vector bool long long, vector long long);
14140 vector long long vec_orc (vector long long, vector bool long long);
14141 vector unsigned long long vec_orc (vector unsigned long long,
14142 vector unsigned long long);
14143 vector unsigned long long vec_orc (vector bool long long,
14144 vector unsigned long long);
14145 vector unsigned long long vec_orc (vector unsigned long long,
14146 vector bool long long);
14147 vector int vec_orc (vector int, vector int);
14148 vector int vec_orc (vector bool int, vector int);
14149 vector int vec_orc (vector int, vector bool int);
14150 vector unsigned int vec_orc (vector unsigned int, vector unsigned int);
14151 vector unsigned int vec_orc (vector bool unsigned int,
14152 vector unsigned int);
14153 vector unsigned int vec_orc (vector unsigned int,
14154 vector bool unsigned int);
14155 vector short vec_orc (vector short, vector short);
14156 vector short vec_orc (vector bool short, vector short);
14157 vector short vec_orc (vector short, vector bool short);
14158 vector unsigned short vec_orc (vector unsigned short, vector unsigned short);
14159 vector unsigned short vec_orc (vector bool unsigned short,
14160 vector unsigned short);
14161 vector unsigned short vec_orc (vector unsigned short,
14162 vector bool unsigned short);
14163 vector signed char vec_orc (vector signed char, vector signed char);
14164 vector signed char vec_orc (vector bool signed char, vector signed char);
14165 vector signed char vec_orc (vector signed char, vector bool signed char);
14166 vector unsigned char vec_orc (vector unsigned char, vector unsigned char);
14167 vector unsigned char vec_orc (vector bool unsigned char, vector unsigned char);
14168 vector unsigned char vec_orc (vector unsigned char, vector bool unsigned char);
14170 vector int vec_pack (vector long long, vector long long);
14171 vector unsigned int vec_pack (vector unsigned long long,
14172 vector unsigned long long);
14173 vector bool int vec_pack (vector bool long long, vector bool long long);
14175 vector int vec_packs (vector long long, vector long long);
14176 vector unsigned int vec_packs (vector unsigned long long,
14177 vector unsigned long long);
14179 vector unsigned int vec_packsu (vector long long, vector long long);
14181 vector long long vec_rl (vector long long,
14182 vector unsigned long long);
14183 vector long long vec_rl (vector unsigned long long,
14184 vector unsigned long long);
14186 vector long long vec_sl (vector long long, vector unsigned long long);
14187 vector long long vec_sl (vector unsigned long long,
14188 vector unsigned long long);
14190 vector long long vec_sr (vector long long, vector unsigned long long);
14191 vector unsigned long long char vec_sr (vector unsigned long long,
14192 vector unsigned long long);
14194 vector long long vec_sra (vector long long, vector unsigned long long);
14195 vector unsigned long long vec_sra (vector unsigned long long,
14196 vector unsigned long long);
14198 vector long long vec_sub (vector long long, vector long long);
14199 vector unsigned long long vec_sub (vector unsigned long long,
14200 vector unsigned long long);
14202 vector long long vec_unpackh (vector int);
14203 vector unsigned long long vec_unpackh (vector unsigned int);
14205 vector long long vec_unpackl (vector int);
14206 vector unsigned long long vec_unpackl (vector unsigned int);
14208 vector long long vec_vaddudm (vector long long, vector long long);
14209 vector long long vec_vaddudm (vector bool long long, vector long long);
14210 vector long long vec_vaddudm (vector long long, vector bool long long);
14211 vector unsigned long long vec_vaddudm (vector unsigned long long,
14212 vector unsigned long long);
14213 vector unsigned long long vec_vaddudm (vector bool unsigned long long,
14214 vector unsigned long long);
14215 vector unsigned long long vec_vaddudm (vector unsigned long long,
14216 vector bool unsigned long long);
14218 vector long long vec_vclz (vector long long);
14219 vector unsigned long long vec_vclz (vector unsigned long long);
14220 vector int vec_vclz (vector int);
14221 vector unsigned int vec_vclz (vector int);
14222 vector short vec_vclz (vector short);
14223 vector unsigned short vec_vclz (vector unsigned short);
14224 vector signed char vec_vclz (vector signed char);
14225 vector unsigned char vec_vclz (vector unsigned char);
14227 vector signed char vec_vclzb (vector signed char);
14228 vector unsigned char vec_vclzb (vector unsigned char);
14230 vector long long vec_vclzd (vector long long);
14231 vector unsigned long long vec_vclzd (vector unsigned long long);
14233 vector short vec_vclzh (vector short);
14234 vector unsigned short vec_vclzh (vector unsigned short);
14236 vector int vec_vclzw (vector int);
14237 vector unsigned int vec_vclzw (vector int);
14239 vector long long vec_vmaxsd (vector long long, vector long long);
14241 vector unsigned long long vec_vmaxud (vector unsigned long long,
14242 unsigned vector long long);
14244 vector long long vec_vminsd (vector long long, vector long long);
14246 vector unsigned long long vec_vminud (vector long long,
14249 vector int vec_vpksdss (vector long long, vector long long);
14250 vector unsigned int vec_vpksdss (vector long long, vector long long);
14252 vector unsigned int vec_vpkudus (vector unsigned long long,
14253 vector unsigned long long);
14255 vector int vec_vpkudum (vector long long, vector long long);
14256 vector unsigned int vec_vpkudum (vector unsigned long long,
14257 vector unsigned long long);
14258 vector bool int vec_vpkudum (vector bool long long, vector bool long long);
14260 vector long long vec_vpopcnt (vector long long);
14261 vector unsigned long long vec_vpopcnt (vector unsigned long long);
14262 vector int vec_vpopcnt (vector int);
14263 vector unsigned int vec_vpopcnt (vector int);
14264 vector short vec_vpopcnt (vector short);
14265 vector unsigned short vec_vpopcnt (vector unsigned short);
14266 vector signed char vec_vpopcnt (vector signed char);
14267 vector unsigned char vec_vpopcnt (vector unsigned char);
14269 vector signed char vec_vpopcntb (vector signed char);
14270 vector unsigned char vec_vpopcntb (vector unsigned char);
14272 vector long long vec_vpopcntd (vector long long);
14273 vector unsigned long long vec_vpopcntd (vector unsigned long long);
14275 vector short vec_vpopcnth (vector short);
14276 vector unsigned short vec_vpopcnth (vector unsigned short);
14278 vector int vec_vpopcntw (vector int);
14279 vector unsigned int vec_vpopcntw (vector int);
14281 vector long long vec_vrld (vector long long, vector unsigned long long);
14282 vector unsigned long long vec_vrld (vector unsigned long long,
14283 vector unsigned long long);
14285 vector long long vec_vsld (vector long long, vector unsigned long long);
14286 vector long long vec_vsld (vector unsigned long long,
14287 vector unsigned long long);
14289 vector long long vec_vsrad (vector long long, vector unsigned long long);
14290 vector unsigned long long vec_vsrad (vector unsigned long long,
14291 vector unsigned long long);
14293 vector long long vec_vsrd (vector long long, vector unsigned long long);
14294 vector unsigned long long char vec_vsrd (vector unsigned long long,
14295 vector unsigned long long);
14297 vector long long vec_vsubudm (vector long long, vector long long);
14298 vector long long vec_vsubudm (vector bool long long, vector long long);
14299 vector long long vec_vsubudm (vector long long, vector bool long long);
14300 vector unsigned long long vec_vsubudm (vector unsigned long long,
14301 vector unsigned long long);
14302 vector unsigned long long vec_vsubudm (vector bool long long,
14303 vector unsigned long long);
14304 vector unsigned long long vec_vsubudm (vector unsigned long long,
14305 vector bool long long);
14307 vector long long vec_vupkhsw (vector int);
14308 vector unsigned long long vec_vupkhsw (vector unsigned int);
14310 vector long long vec_vupklsw (vector int);
14311 vector unsigned long long vec_vupklsw (vector int);
14314 If the cryptographic instructions are enabled (@option{-mcrypto} or
14315 @option{-mcpu=power8}), the following builtins are enabled.
14318 vector unsigned long long __builtin_crypto_vsbox (vector unsigned long long);
14320 vector unsigned long long __builtin_crypto_vcipher (vector unsigned long long,
14321 vector unsigned long long);
14323 vector unsigned long long __builtin_crypto_vcipherlast
14324 (vector unsigned long long,
14325 vector unsigned long long);
14327 vector unsigned long long __builtin_crypto_vncipher (vector unsigned long long,
14328 vector unsigned long long);
14330 vector unsigned long long __builtin_crypto_vncipherlast
14331 (vector unsigned long long,
14332 vector unsigned long long);
14334 vector unsigned char __builtin_crypto_vpermxor (vector unsigned char,
14335 vector unsigned char,
14336 vector unsigned char);
14338 vector unsigned short __builtin_crypto_vpermxor (vector unsigned short,
14339 vector unsigned short,
14340 vector unsigned short);
14342 vector unsigned int __builtin_crypto_vpermxor (vector unsigned int,
14343 vector unsigned int,
14344 vector unsigned int);
14346 vector unsigned long long __builtin_crypto_vpermxor (vector unsigned long long,
14347 vector unsigned long long,
14348 vector unsigned long long);
14350 vector unsigned char __builtin_crypto_vpmsumb (vector unsigned char,
14351 vector unsigned char);
14353 vector unsigned short __builtin_crypto_vpmsumb (vector unsigned short,
14354 vector unsigned short);
14356 vector unsigned int __builtin_crypto_vpmsumb (vector unsigned int,
14357 vector unsigned int);
14359 vector unsigned long long __builtin_crypto_vpmsumb (vector unsigned long long,
14360 vector unsigned long long);
14362 vector unsigned long long __builtin_crypto_vshasigmad
14363 (vector unsigned long long, int, int);
14365 vector unsigned int __builtin_crypto_vshasigmaw (vector unsigned int,
14369 The second argument to the @var{__builtin_crypto_vshasigmad} and
14370 @var{__builtin_crypto_vshasigmaw} builtin functions must be a constant
14371 integer that is 0 or 1. The third argument to these builtin functions
14372 must be a constant integer in the range of 0 to 15.
14374 @node RX Built-in Functions
14375 @subsection RX Built-in Functions
14376 GCC supports some of the RX instructions which cannot be expressed in
14377 the C programming language via the use of built-in functions. The
14378 following functions are supported:
14380 @deftypefn {Built-in Function} void __builtin_rx_brk (void)
14381 Generates the @code{brk} machine instruction.
14384 @deftypefn {Built-in Function} void __builtin_rx_clrpsw (int)
14385 Generates the @code{clrpsw} machine instruction to clear the specified
14386 bit in the processor status word.
14389 @deftypefn {Built-in Function} void __builtin_rx_int (int)
14390 Generates the @code{int} machine instruction to generate an interrupt
14391 with the specified value.
14394 @deftypefn {Built-in Function} void __builtin_rx_machi (int, int)
14395 Generates the @code{machi} machine instruction to add the result of
14396 multiplying the top 16 bits of the two arguments into the
14400 @deftypefn {Built-in Function} void __builtin_rx_maclo (int, int)
14401 Generates the @code{maclo} machine instruction to add the result of
14402 multiplying the bottom 16 bits of the two arguments into the
14406 @deftypefn {Built-in Function} void __builtin_rx_mulhi (int, int)
14407 Generates the @code{mulhi} machine instruction to place the result of
14408 multiplying the top 16 bits of the two arguments into the
14412 @deftypefn {Built-in Function} void __builtin_rx_mullo (int, int)
14413 Generates the @code{mullo} machine instruction to place the result of
14414 multiplying the bottom 16 bits of the two arguments into the
14418 @deftypefn {Built-in Function} int __builtin_rx_mvfachi (void)
14419 Generates the @code{mvfachi} machine instruction to read the top
14420 32 bits of the accumulator.
14423 @deftypefn {Built-in Function} int __builtin_rx_mvfacmi (void)
14424 Generates the @code{mvfacmi} machine instruction to read the middle
14425 32 bits of the accumulator.
14428 @deftypefn {Built-in Function} int __builtin_rx_mvfc (int)
14429 Generates the @code{mvfc} machine instruction which reads the control
14430 register specified in its argument and returns its value.
14433 @deftypefn {Built-in Function} void __builtin_rx_mvtachi (int)
14434 Generates the @code{mvtachi} machine instruction to set the top
14435 32 bits of the accumulator.
14438 @deftypefn {Built-in Function} void __builtin_rx_mvtaclo (int)
14439 Generates the @code{mvtaclo} machine instruction to set the bottom
14440 32 bits of the accumulator.
14443 @deftypefn {Built-in Function} void __builtin_rx_mvtc (int reg, int val)
14444 Generates the @code{mvtc} machine instruction which sets control
14445 register number @code{reg} to @code{val}.
14448 @deftypefn {Built-in Function} void __builtin_rx_mvtipl (int)
14449 Generates the @code{mvtipl} machine instruction set the interrupt
14453 @deftypefn {Built-in Function} void __builtin_rx_racw (int)
14454 Generates the @code{racw} machine instruction to round the accumulator
14455 according to the specified mode.
14458 @deftypefn {Built-in Function} int __builtin_rx_revw (int)
14459 Generates the @code{revw} machine instruction which swaps the bytes in
14460 the argument so that bits 0--7 now occupy bits 8--15 and vice versa,
14461 and also bits 16--23 occupy bits 24--31 and vice versa.
14464 @deftypefn {Built-in Function} void __builtin_rx_rmpa (void)
14465 Generates the @code{rmpa} machine instruction which initiates a
14466 repeated multiply and accumulate sequence.
14469 @deftypefn {Built-in Function} void __builtin_rx_round (float)
14470 Generates the @code{round} machine instruction which returns the
14471 floating-point argument rounded according to the current rounding mode
14472 set in the floating-point status word register.
14475 @deftypefn {Built-in Function} int __builtin_rx_sat (int)
14476 Generates the @code{sat} machine instruction which returns the
14477 saturated value of the argument.
14480 @deftypefn {Built-in Function} void __builtin_rx_setpsw (int)
14481 Generates the @code{setpsw} machine instruction to set the specified
14482 bit in the processor status word.
14485 @deftypefn {Built-in Function} void __builtin_rx_wait (void)
14486 Generates the @code{wait} machine instruction.
14489 @node S/390 System z Built-in Functions
14490 @subsection S/390 System z Built-in Functions
14491 @deftypefn {Built-in Function} int __builtin_tbegin (void*)
14492 Generates the @code{tbegin} machine instruction starting a
14493 non-constraint hardware transaction. If the parameter is non-NULL the
14494 memory area is used to store the transaction diagnostic buffer and
14495 will be passed as first operand to @code{tbegin}. This buffer can be
14496 defined using the @code{struct __htm_tdb} C struct defined in
14497 @code{htmintrin.h} and must reside on a double-word boundary. The
14498 second tbegin operand is set to @code{0xff0c}. This enables
14499 save/restore of all GPRs and disables aborts for FPR and AR
14500 manipulations inside the transaction body. The condition code set by
14501 the tbegin instruction is returned as integer value. The tbegin
14502 instruction by definition overwrites the content of all FPRs. The
14503 compiler will generate code which saves and restores the FPRs. For
14504 soft-float code it is recommended to used the @code{*_nofloat}
14505 variant. In order to prevent a TDB from being written it is required
14506 to pass an constant zero value as parameter. Passing the zero value
14507 through a variable is not sufficient. Although modifications of
14508 access registers inside the transaction will not trigger an
14509 transaction abort it is not supported to actually modify them. Access
14510 registers do not get saved when entering a transaction. They will have
14511 undefined state when reaching the abort code.
14514 Macros for the possible return codes of tbegin are defined in the
14515 @code{htmintrin.h} header file:
14518 @item _HTM_TBEGIN_STARTED
14519 @code{tbegin} has been executed as part of normal processing. The
14520 transaction body is supposed to be executed.
14521 @item _HTM_TBEGIN_INDETERMINATE
14522 The transaction was aborted due to an indeterminate condition which
14523 might be persistent.
14524 @item _HTM_TBEGIN_TRANSIENT
14525 The transaction aborted due to a transient failure. The transaction
14526 should be re-executed in that case.
14527 @item _HTM_TBEGIN_PERSISTENT
14528 The transaction aborted due to a persistent failure. Re-execution
14529 under same circumstances will not be productive.
14532 @defmac _HTM_FIRST_USER_ABORT_CODE
14533 The @code{_HTM_FIRST_USER_ABORT_CODE} defined in @code{htmintrin.h}
14534 specifies the first abort code which can be used for
14535 @code{__builtin_tabort}. Values below this threshold are reserved for
14539 @deftp {Data type} {struct __htm_tdb}
14540 The @code{struct __htm_tdb} defined in @code{htmintrin.h} describes
14541 the structure of the transaction diagnostic block as specified in the
14542 Principles of Operation manual chapter 5-91.
14545 @deftypefn {Built-in Function} int __builtin_tbegin_nofloat (void*)
14546 Same as @code{__builtin_tbegin} but without FPR saves and restores.
14547 Using this variant in code making use of FPRs will leave the FPRs in
14548 undefined state when entering the transaction abort handler code.
14551 @deftypefn {Built-in Function} int __builtin_tbegin_retry (void*, int)
14552 In addition to @code{__builtin_tbegin} a loop for transient failures
14553 is generated. If tbegin returns a condition code of 2 the transaction
14554 will be retried as often as specified in the second argument. The
14555 perform processor assist instruction is used to tell the CPU about the
14556 number of fails so far.
14559 @deftypefn {Built-in Function} int __builtin_tbegin_retry_nofloat (void*, int)
14560 Same as @code{__builtin_tbegin_retry} but without FPR saves and
14561 restores. Using this variant in code making use of FPRs will leave
14562 the FPRs in undefined state when entering the transaction abort
14566 @deftypefn {Built-in Function} void __builtin_tbeginc (void)
14567 Generates the @code{tbeginc} machine instruction starting a constraint
14568 hardware transaction. The second operand is set to @code{0xff08}.
14571 @deftypefn {Built-in Function} int __builtin_tend (void)
14572 Generates the @code{tend} machine instruction finishing a transaction
14573 and making the changes visible to other threads. The condition code
14574 generated by tend is returned as integer value.
14577 @deftypefn {Built-in Function} void __builtin_tabort (int)
14578 Generates the @code{tabort} machine instruction with the specified
14579 abort code. Abort codes from 0 through 255 are reserved and will
14580 result in an error message.
14583 @deftypefn {Built-in Function} void __builtin_tx_assist (int)
14584 Generates the @code{ppa rX,rY,1} machine instruction. Where the
14585 integer parameter is loaded into rX and a value of zero is loaded into
14586 rY. The integer parameter specifies the number of times the
14587 transaction repeatedly aborted.
14590 @deftypefn {Built-in Function} int __builtin_tx_nesting_depth (void)
14591 Generates the @code{etnd} machine instruction. The current nesting
14592 depth is returned as integer value. For a nesting depth of 0 the code
14593 is not executed as part of an transaction.
14596 @deftypefn {Built-in Function} void __builtin_non_tx_store (unsigned long long *, unsigned long long)
14598 Generates the @code{ntstg} machine instruction. The second argument
14599 is written to the first arguments location. The store operation will
14600 not be rolled-back in case of an transaction abort.
14603 @node SH Built-in Functions
14604 @subsection SH Built-in Functions
14605 The following built-in functions are supported on the SH1, SH2, SH3 and SH4
14606 families of processors:
14608 @deftypefn {Built-in Function} {void} __builtin_set_thread_pointer (void *@var{ptr})
14609 Sets the @samp{GBR} register to the specified value @var{ptr}. This is usually
14610 used by system code that manages threads and execution contexts. The compiler
14611 normally does not generate code that modifies the contents of @samp{GBR} and
14612 thus the value is preserved across function calls. Changing the @samp{GBR}
14613 value in user code must be done with caution, since the compiler might use
14614 @samp{GBR} in order to access thread local variables.
14618 @deftypefn {Built-in Function} {void *} __builtin_thread_pointer (void)
14619 Returns the value that is currently set in the @samp{GBR} register.
14620 Memory loads and stores that use the thread pointer as a base address are
14621 turned into @samp{GBR} based displacement loads and stores, if possible.
14629 int get_tcb_value (void)
14631 // Generate @samp{mov.l @@(8,gbr),r0} instruction
14632 return ((my_tcb*)__builtin_thread_pointer ())->c;
14638 @node SPARC VIS Built-in Functions
14639 @subsection SPARC VIS Built-in Functions
14641 GCC supports SIMD operations on the SPARC using both the generic vector
14642 extensions (@pxref{Vector Extensions}) as well as built-in functions for
14643 the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis}
14644 switch, the VIS extension is exposed as the following built-in functions:
14647 typedef int v1si __attribute__ ((vector_size (4)));
14648 typedef int v2si __attribute__ ((vector_size (8)));
14649 typedef short v4hi __attribute__ ((vector_size (8)));
14650 typedef short v2hi __attribute__ ((vector_size (4)));
14651 typedef unsigned char v8qi __attribute__ ((vector_size (8)));
14652 typedef unsigned char v4qi __attribute__ ((vector_size (4)));
14654 void __builtin_vis_write_gsr (int64_t);
14655 int64_t __builtin_vis_read_gsr (void);
14657 void * __builtin_vis_alignaddr (void *, long);
14658 void * __builtin_vis_alignaddrl (void *, long);
14659 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
14660 v2si __builtin_vis_faligndatav2si (v2si, v2si);
14661 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
14662 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
14664 v4hi __builtin_vis_fexpand (v4qi);
14666 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
14667 v4hi __builtin_vis_fmul8x16au (v4qi, v2hi);
14668 v4hi __builtin_vis_fmul8x16al (v4qi, v2hi);
14669 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
14670 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
14671 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
14672 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
14674 v4qi __builtin_vis_fpack16 (v4hi);
14675 v8qi __builtin_vis_fpack32 (v2si, v8qi);
14676 v2hi __builtin_vis_fpackfix (v2si);
14677 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
14679 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
14681 long __builtin_vis_edge8 (void *, void *);
14682 long __builtin_vis_edge8l (void *, void *);
14683 long __builtin_vis_edge16 (void *, void *);
14684 long __builtin_vis_edge16l (void *, void *);
14685 long __builtin_vis_edge32 (void *, void *);
14686 long __builtin_vis_edge32l (void *, void *);
14688 long __builtin_vis_fcmple16 (v4hi, v4hi);
14689 long __builtin_vis_fcmple32 (v2si, v2si);
14690 long __builtin_vis_fcmpne16 (v4hi, v4hi);
14691 long __builtin_vis_fcmpne32 (v2si, v2si);
14692 long __builtin_vis_fcmpgt16 (v4hi, v4hi);
14693 long __builtin_vis_fcmpgt32 (v2si, v2si);
14694 long __builtin_vis_fcmpeq16 (v4hi, v4hi);
14695 long __builtin_vis_fcmpeq32 (v2si, v2si);
14697 v4hi __builtin_vis_fpadd16 (v4hi, v4hi);
14698 v2hi __builtin_vis_fpadd16s (v2hi, v2hi);
14699 v2si __builtin_vis_fpadd32 (v2si, v2si);
14700 v1si __builtin_vis_fpadd32s (v1si, v1si);
14701 v4hi __builtin_vis_fpsub16 (v4hi, v4hi);
14702 v2hi __builtin_vis_fpsub16s (v2hi, v2hi);
14703 v2si __builtin_vis_fpsub32 (v2si, v2si);
14704 v1si __builtin_vis_fpsub32s (v1si, v1si);
14706 long __builtin_vis_array8 (long, long);
14707 long __builtin_vis_array16 (long, long);
14708 long __builtin_vis_array32 (long, long);
14711 When you use the @option{-mvis2} switch, the VIS version 2.0 built-in
14712 functions also become available:
14715 long __builtin_vis_bmask (long, long);
14716 int64_t __builtin_vis_bshuffledi (int64_t, int64_t);
14717 v2si __builtin_vis_bshufflev2si (v2si, v2si);
14718 v4hi __builtin_vis_bshufflev2si (v4hi, v4hi);
14719 v8qi __builtin_vis_bshufflev2si (v8qi, v8qi);
14721 long __builtin_vis_edge8n (void *, void *);
14722 long __builtin_vis_edge8ln (void *, void *);
14723 long __builtin_vis_edge16n (void *, void *);
14724 long __builtin_vis_edge16ln (void *, void *);
14725 long __builtin_vis_edge32n (void *, void *);
14726 long __builtin_vis_edge32ln (void *, void *);
14729 When you use the @option{-mvis3} switch, the VIS version 3.0 built-in
14730 functions also become available:
14733 void __builtin_vis_cmask8 (long);
14734 void __builtin_vis_cmask16 (long);
14735 void __builtin_vis_cmask32 (long);
14737 v4hi __builtin_vis_fchksm16 (v4hi, v4hi);
14739 v4hi __builtin_vis_fsll16 (v4hi, v4hi);
14740 v4hi __builtin_vis_fslas16 (v4hi, v4hi);
14741 v4hi __builtin_vis_fsrl16 (v4hi, v4hi);
14742 v4hi __builtin_vis_fsra16 (v4hi, v4hi);
14743 v2si __builtin_vis_fsll16 (v2si, v2si);
14744 v2si __builtin_vis_fslas16 (v2si, v2si);
14745 v2si __builtin_vis_fsrl16 (v2si, v2si);
14746 v2si __builtin_vis_fsra16 (v2si, v2si);
14748 long __builtin_vis_pdistn (v8qi, v8qi);
14750 v4hi __builtin_vis_fmean16 (v4hi, v4hi);
14752 int64_t __builtin_vis_fpadd64 (int64_t, int64_t);
14753 int64_t __builtin_vis_fpsub64 (int64_t, int64_t);
14755 v4hi __builtin_vis_fpadds16 (v4hi, v4hi);
14756 v2hi __builtin_vis_fpadds16s (v2hi, v2hi);
14757 v4hi __builtin_vis_fpsubs16 (v4hi, v4hi);
14758 v2hi __builtin_vis_fpsubs16s (v2hi, v2hi);
14759 v2si __builtin_vis_fpadds32 (v2si, v2si);
14760 v1si __builtin_vis_fpadds32s (v1si, v1si);
14761 v2si __builtin_vis_fpsubs32 (v2si, v2si);
14762 v1si __builtin_vis_fpsubs32s (v1si, v1si);
14764 long __builtin_vis_fucmple8 (v8qi, v8qi);
14765 long __builtin_vis_fucmpne8 (v8qi, v8qi);
14766 long __builtin_vis_fucmpgt8 (v8qi, v8qi);
14767 long __builtin_vis_fucmpeq8 (v8qi, v8qi);
14769 float __builtin_vis_fhadds (float, float);
14770 double __builtin_vis_fhaddd (double, double);
14771 float __builtin_vis_fhsubs (float, float);
14772 double __builtin_vis_fhsubd (double, double);
14773 float __builtin_vis_fnhadds (float, float);
14774 double __builtin_vis_fnhaddd (double, double);
14776 int64_t __builtin_vis_umulxhi (int64_t, int64_t);
14777 int64_t __builtin_vis_xmulx (int64_t, int64_t);
14778 int64_t __builtin_vis_xmulxhi (int64_t, int64_t);
14781 @node SPU Built-in Functions
14782 @subsection SPU Built-in Functions
14784 GCC provides extensions for the SPU processor as described in the
14785 Sony/Toshiba/IBM SPU Language Extensions Specification, which can be
14786 found at @uref{http://cell.scei.co.jp/} or
14787 @uref{http://www.ibm.com/developerworks/power/cell/}. GCC's
14788 implementation differs in several ways.
14793 The optional extension of specifying vector constants in parentheses is
14797 A vector initializer requires no cast if the vector constant is of the
14798 same type as the variable it is initializing.
14801 If @code{signed} or @code{unsigned} is omitted, the signedness of the
14802 vector type is the default signedness of the base type. The default
14803 varies depending on the operating system, so a portable program should
14804 always specify the signedness.
14807 By default, the keyword @code{__vector} is added. The macro
14808 @code{vector} is defined in @code{<spu_intrinsics.h>} and can be
14812 GCC allows using a @code{typedef} name as the type specifier for a
14816 For C, overloaded functions are implemented with macros so the following
14820 spu_add ((vector signed int)@{1, 2, 3, 4@}, foo);
14824 Since @code{spu_add} is a macro, the vector constant in the example
14825 is treated as four separate arguments. Wrap the entire argument in
14826 parentheses for this to work.
14829 The extended version of @code{__builtin_expect} is not supported.
14833 @emph{Note:} Only the interface described in the aforementioned
14834 specification is supported. Internally, GCC uses built-in functions to
14835 implement the required functionality, but these are not supported and
14836 are subject to change without notice.
14838 @node TI C6X Built-in Functions
14839 @subsection TI C6X Built-in Functions
14841 GCC provides intrinsics to access certain instructions of the TI C6X
14842 processors. These intrinsics, listed below, are available after
14843 inclusion of the @code{c6x_intrinsics.h} header file. They map directly
14844 to C6X instructions.
14848 int _sadd (int, int)
14849 int _ssub (int, int)
14850 int _sadd2 (int, int)
14851 int _ssub2 (int, int)
14852 long long _mpy2 (int, int)
14853 long long _smpy2 (int, int)
14854 int _add4 (int, int)
14855 int _sub4 (int, int)
14856 int _saddu4 (int, int)
14858 int _smpy (int, int)
14859 int _smpyh (int, int)
14860 int _smpyhl (int, int)
14861 int _smpylh (int, int)
14863 int _sshl (int, int)
14864 int _subc (int, int)
14866 int _avg2 (int, int)
14867 int _avgu4 (int, int)
14869 int _clrr (int, int)
14870 int _extr (int, int)
14871 int _extru (int, int)
14877 @node TILE-Gx Built-in Functions
14878 @subsection TILE-Gx Built-in Functions
14880 GCC provides intrinsics to access every instruction of the TILE-Gx
14881 processor. The intrinsics are of the form:
14885 unsigned long long __insn_@var{op} (...)
14889 Where @var{op} is the name of the instruction. Refer to the ISA manual
14890 for the complete list of instructions.
14892 GCC also provides intrinsics to directly access the network registers.
14893 The intrinsics are:
14897 unsigned long long __tile_idn0_receive (void)
14898 unsigned long long __tile_idn1_receive (void)
14899 unsigned long long __tile_udn0_receive (void)
14900 unsigned long long __tile_udn1_receive (void)
14901 unsigned long long __tile_udn2_receive (void)
14902 unsigned long long __tile_udn3_receive (void)
14903 void __tile_idn_send (unsigned long long)
14904 void __tile_udn_send (unsigned long long)
14908 The intrinsic @code{void __tile_network_barrier (void)} is used to
14909 guarantee that no network operations before it are reordered with
14912 @node TILEPro Built-in Functions
14913 @subsection TILEPro Built-in Functions
14915 GCC provides intrinsics to access every instruction of the TILEPro
14916 processor. The intrinsics are of the form:
14920 unsigned __insn_@var{op} (...)
14925 where @var{op} is the name of the instruction. Refer to the ISA manual
14926 for the complete list of instructions.
14928 GCC also provides intrinsics to directly access the network registers.
14929 The intrinsics are:
14933 unsigned __tile_idn0_receive (void)
14934 unsigned __tile_idn1_receive (void)
14935 unsigned __tile_sn_receive (void)
14936 unsigned __tile_udn0_receive (void)
14937 unsigned __tile_udn1_receive (void)
14938 unsigned __tile_udn2_receive (void)
14939 unsigned __tile_udn3_receive (void)
14940 void __tile_idn_send (unsigned)
14941 void __tile_sn_send (unsigned)
14942 void __tile_udn_send (unsigned)
14946 The intrinsic @code{void __tile_network_barrier (void)} is used to
14947 guarantee that no network operations before it are reordered with
14950 @node Target Format Checks
14951 @section Format Checks Specific to Particular Target Machines
14953 For some target machines, GCC supports additional options to the
14955 (@pxref{Function Attributes,,Declaring Attributes of Functions}).
14958 * Solaris Format Checks::
14959 * Darwin Format Checks::
14962 @node Solaris Format Checks
14963 @subsection Solaris Format Checks
14965 Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format
14966 check. @code{cmn_err} accepts a subset of the standard @code{printf}
14967 conversions, and the two-argument @code{%b} conversion for displaying
14968 bit-fields. See the Solaris man page for @code{cmn_err} for more information.
14970 @node Darwin Format Checks
14971 @subsection Darwin Format Checks
14973 Darwin targets support the @code{CFString} (or @code{__CFString__}) in the format
14974 attribute context. Declarations made with such attribution are parsed for correct syntax
14975 and format argument types. However, parsing of the format string itself is currently undefined
14976 and is not carried out by this version of the compiler.
14978 Additionally, @code{CFStringRefs} (defined by the @code{CoreFoundation} headers) may
14979 also be used as format arguments. Note that the relevant headers are only likely to be
14980 available on Darwin (OSX) installations. On such installations, the XCode and system
14981 documentation provide descriptions of @code{CFString}, @code{CFStringRefs} and
14982 associated functions.
14985 @section Pragmas Accepted by GCC
14987 @cindex @code{#pragma}
14989 GCC supports several types of pragmas, primarily in order to compile
14990 code originally written for other compilers. Note that in general
14991 we do not recommend the use of pragmas; @xref{Function Attributes},
14992 for further explanation.
14998 * RS/6000 and PowerPC Pragmas::
15000 * Solaris Pragmas::
15001 * Symbol-Renaming Pragmas::
15002 * Structure-Packing Pragmas::
15004 * Diagnostic Pragmas::
15005 * Visibility Pragmas::
15006 * Push/Pop Macro Pragmas::
15007 * Function Specific Option Pragmas::
15011 @subsection ARM Pragmas
15013 The ARM target defines pragmas for controlling the default addition of
15014 @code{long_call} and @code{short_call} attributes to functions.
15015 @xref{Function Attributes}, for information about the effects of these
15020 @cindex pragma, long_calls
15021 Set all subsequent functions to have the @code{long_call} attribute.
15023 @item no_long_calls
15024 @cindex pragma, no_long_calls
15025 Set all subsequent functions to have the @code{short_call} attribute.
15027 @item long_calls_off
15028 @cindex pragma, long_calls_off
15029 Do not affect the @code{long_call} or @code{short_call} attributes of
15030 subsequent functions.
15034 @subsection M32C Pragmas
15037 @item GCC memregs @var{number}
15038 @cindex pragma, memregs
15039 Overrides the command-line option @code{-memregs=} for the current
15040 file. Use with care! This pragma must be before any function in the
15041 file, and mixing different memregs values in different objects may
15042 make them incompatible. This pragma is useful when a
15043 performance-critical function uses a memreg for temporary values,
15044 as it may allow you to reduce the number of memregs used.
15046 @item ADDRESS @var{name} @var{address}
15047 @cindex pragma, address
15048 For any declared symbols matching @var{name}, this does three things
15049 to that symbol: it forces the symbol to be located at the given
15050 address (a number), it forces the symbol to be volatile, and it
15051 changes the symbol's scope to be static. This pragma exists for
15052 compatibility with other compilers, but note that the common
15053 @code{1234H} numeric syntax is not supported (use @code{0x1234}
15057 #pragma ADDRESS port3 0x103
15064 @subsection MeP Pragmas
15068 @item custom io_volatile (on|off)
15069 @cindex pragma, custom io_volatile
15070 Overrides the command-line option @code{-mio-volatile} for the current
15071 file. Note that for compatibility with future GCC releases, this
15072 option should only be used once before any @code{io} variables in each
15075 @item GCC coprocessor available @var{registers}
15076 @cindex pragma, coprocessor available
15077 Specifies which coprocessor registers are available to the register
15078 allocator. @var{registers} may be a single register, register range
15079 separated by ellipses, or comma-separated list of those. Example:
15082 #pragma GCC coprocessor available $c0...$c10, $c28
15085 @item GCC coprocessor call_saved @var{registers}
15086 @cindex pragma, coprocessor call_saved
15087 Specifies which coprocessor registers are to be saved and restored by
15088 any function using them. @var{registers} may be a single register,
15089 register range separated by ellipses, or comma-separated list of
15093 #pragma GCC coprocessor call_saved $c4...$c6, $c31
15096 @item GCC coprocessor subclass '(A|B|C|D)' = @var{registers}
15097 @cindex pragma, coprocessor subclass
15098 Creates and defines a register class. These register classes can be
15099 used by inline @code{asm} constructs. @var{registers} may be a single
15100 register, register range separated by ellipses, or comma-separated
15101 list of those. Example:
15104 #pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6
15106 asm ("cpfoo %0" : "=B" (x));
15109 @item GCC disinterrupt @var{name} , @var{name} @dots{}
15110 @cindex pragma, disinterrupt
15111 For the named functions, the compiler adds code to disable interrupts
15112 for the duration of those functions. If any functions so named
15113 are not encountered in the source, a warning is emitted that the pragma is
15114 not used. Examples:
15117 #pragma disinterrupt foo
15118 #pragma disinterrupt bar, grill
15119 int foo () @{ @dots{} @}
15122 @item GCC call @var{name} , @var{name} @dots{}
15123 @cindex pragma, call
15124 For the named functions, the compiler always uses a register-indirect
15125 call model when calling the named functions. Examples:
15134 @node RS/6000 and PowerPC Pragmas
15135 @subsection RS/6000 and PowerPC Pragmas
15137 The RS/6000 and PowerPC targets define one pragma for controlling
15138 whether or not the @code{longcall} attribute is added to function
15139 declarations by default. This pragma overrides the @option{-mlongcall}
15140 option, but not the @code{longcall} and @code{shortcall} attributes.
15141 @xref{RS/6000 and PowerPC Options}, for more information about when long
15142 calls are and are not necessary.
15146 @cindex pragma, longcall
15147 Apply the @code{longcall} attribute to all subsequent function
15151 Do not apply the @code{longcall} attribute to subsequent function
15155 @c Describe h8300 pragmas here.
15156 @c Describe sh pragmas here.
15157 @c Describe v850 pragmas here.
15159 @node Darwin Pragmas
15160 @subsection Darwin Pragmas
15162 The following pragmas are available for all architectures running the
15163 Darwin operating system. These are useful for compatibility with other
15167 @item mark @var{tokens}@dots{}
15168 @cindex pragma, mark
15169 This pragma is accepted, but has no effect.
15171 @item options align=@var{alignment}
15172 @cindex pragma, options align
15173 This pragma sets the alignment of fields in structures. The values of
15174 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
15175 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
15176 properly; to restore the previous setting, use @code{reset} for the
15179 @item segment @var{tokens}@dots{}
15180 @cindex pragma, segment
15181 This pragma is accepted, but has no effect.
15183 @item unused (@var{var} [, @var{var}]@dots{})
15184 @cindex pragma, unused
15185 This pragma declares variables to be possibly unused. GCC does not
15186 produce warnings for the listed variables. The effect is similar to
15187 that of the @code{unused} attribute, except that this pragma may appear
15188 anywhere within the variables' scopes.
15191 @node Solaris Pragmas
15192 @subsection Solaris Pragmas
15194 The Solaris target supports @code{#pragma redefine_extname}
15195 (@pxref{Symbol-Renaming Pragmas}). It also supports additional
15196 @code{#pragma} directives for compatibility with the system compiler.
15199 @item align @var{alignment} (@var{variable} [, @var{variable}]...)
15200 @cindex pragma, align
15202 Increase the minimum alignment of each @var{variable} to @var{alignment}.
15203 This is the same as GCC's @code{aligned} attribute @pxref{Variable
15204 Attributes}). Macro expansion occurs on the arguments to this pragma
15205 when compiling C and Objective-C@. It does not currently occur when
15206 compiling C++, but this is a bug which may be fixed in a future
15209 @item fini (@var{function} [, @var{function}]...)
15210 @cindex pragma, fini
15212 This pragma causes each listed @var{function} to be called after
15213 main, or during shared module unloading, by adding a call to the
15214 @code{.fini} section.
15216 @item init (@var{function} [, @var{function}]...)
15217 @cindex pragma, init
15219 This pragma causes each listed @var{function} to be called during
15220 initialization (before @code{main}) or during shared module loading, by
15221 adding a call to the @code{.init} section.
15225 @node Symbol-Renaming Pragmas
15226 @subsection Symbol-Renaming Pragmas
15228 For compatibility with the Solaris system headers, GCC
15229 supports two @code{#pragma} directives that change the name used in
15230 assembly for a given declaration. To get this effect
15231 on all platforms supported by GCC, use the asm labels extension (@pxref{Asm
15235 @item redefine_extname @var{oldname} @var{newname}
15236 @cindex pragma, redefine_extname
15238 This pragma gives the C function @var{oldname} the assembly symbol
15239 @var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME}
15240 is defined if this pragma is available (currently on all platforms).
15243 This pragma and the asm labels extension interact in a complicated
15244 manner. Here are some corner cases you may want to be aware of.
15247 @item Both pragmas silently apply only to declarations with external
15248 linkage. Asm labels do not have this restriction.
15250 @item In C++, both pragmas silently apply only to declarations with
15251 ``C'' linkage. Again, asm labels do not have this restriction.
15253 @item If any of the three ways of changing the assembly name of a
15254 declaration is applied to a declaration whose assembly name has
15255 already been determined (either by a previous use of one of these
15256 features, or because the compiler needed the assembly name in order to
15257 generate code), and the new name is different, a warning issues and
15258 the name does not change.
15260 @item The @var{oldname} used by @code{#pragma redefine_extname} is
15261 always the C-language name.
15264 @node Structure-Packing Pragmas
15265 @subsection Structure-Packing Pragmas
15267 For compatibility with Microsoft Windows compilers, GCC supports a
15268 set of @code{#pragma} directives that change the maximum alignment of
15269 members of structures (other than zero-width bit-fields), unions, and
15270 classes subsequently defined. The @var{n} value below always is required
15271 to be a small power of two and specifies the new alignment in bytes.
15274 @item @code{#pragma pack(@var{n})} simply sets the new alignment.
15275 @item @code{#pragma pack()} sets the alignment to the one that was in
15276 effect when compilation started (see also command-line option
15277 @option{-fpack-struct[=@var{n}]} @pxref{Code Gen Options}).
15278 @item @code{#pragma pack(push[,@var{n}])} pushes the current alignment
15279 setting on an internal stack and then optionally sets the new alignment.
15280 @item @code{#pragma pack(pop)} restores the alignment setting to the one
15281 saved at the top of the internal stack (and removes that stack entry).
15282 Note that @code{#pragma pack([@var{n}])} does not influence this internal
15283 stack; thus it is possible to have @code{#pragma pack(push)} followed by
15284 multiple @code{#pragma pack(@var{n})} instances and finalized by a single
15285 @code{#pragma pack(pop)}.
15288 Some targets, e.g.@: i386 and PowerPC, support the @code{ms_struct}
15289 @code{#pragma} which lays out a structure as the documented
15290 @code{__attribute__ ((ms_struct))}.
15292 @item @code{#pragma ms_struct on} turns on the layout for structures
15294 @item @code{#pragma ms_struct off} turns off the layout for structures
15296 @item @code{#pragma ms_struct reset} goes back to the default layout.
15300 @subsection Weak Pragmas
15302 For compatibility with SVR4, GCC supports a set of @code{#pragma}
15303 directives for declaring symbols to be weak, and defining weak
15307 @item #pragma weak @var{symbol}
15308 @cindex pragma, weak
15309 This pragma declares @var{symbol} to be weak, as if the declaration
15310 had the attribute of the same name. The pragma may appear before
15311 or after the declaration of @var{symbol}. It is not an error for
15312 @var{symbol} to never be defined at all.
15314 @item #pragma weak @var{symbol1} = @var{symbol2}
15315 This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}.
15316 It is an error if @var{symbol2} is not defined in the current
15320 @node Diagnostic Pragmas
15321 @subsection Diagnostic Pragmas
15323 GCC allows the user to selectively enable or disable certain types of
15324 diagnostics, and change the kind of the diagnostic. For example, a
15325 project's policy might require that all sources compile with
15326 @option{-Werror} but certain files might have exceptions allowing
15327 specific types of warnings. Or, a project might selectively enable
15328 diagnostics and treat them as errors depending on which preprocessor
15329 macros are defined.
15332 @item #pragma GCC diagnostic @var{kind} @var{option}
15333 @cindex pragma, diagnostic
15335 Modifies the disposition of a diagnostic. Note that not all
15336 diagnostics are modifiable; at the moment only warnings (normally
15337 controlled by @samp{-W@dots{}}) can be controlled, and not all of them.
15338 Use @option{-fdiagnostics-show-option} to determine which diagnostics
15339 are controllable and which option controls them.
15341 @var{kind} is @samp{error} to treat this diagnostic as an error,
15342 @samp{warning} to treat it like a warning (even if @option{-Werror} is
15343 in effect), or @samp{ignored} if the diagnostic is to be ignored.
15344 @var{option} is a double quoted string that matches the command-line
15348 #pragma GCC diagnostic warning "-Wformat"
15349 #pragma GCC diagnostic error "-Wformat"
15350 #pragma GCC diagnostic ignored "-Wformat"
15353 Note that these pragmas override any command-line options. GCC keeps
15354 track of the location of each pragma, and issues diagnostics according
15355 to the state as of that point in the source file. Thus, pragmas occurring
15356 after a line do not affect diagnostics caused by that line.
15358 @item #pragma GCC diagnostic push
15359 @itemx #pragma GCC diagnostic pop
15361 Causes GCC to remember the state of the diagnostics as of each
15362 @code{push}, and restore to that point at each @code{pop}. If a
15363 @code{pop} has no matching @code{push}, the command-line options are
15367 #pragma GCC diagnostic error "-Wuninitialized"
15368 foo(a); /* error is given for this one */
15369 #pragma GCC diagnostic push
15370 #pragma GCC diagnostic ignored "-Wuninitialized"
15371 foo(b); /* no diagnostic for this one */
15372 #pragma GCC diagnostic pop
15373 foo(c); /* error is given for this one */
15374 #pragma GCC diagnostic pop
15375 foo(d); /* depends on command-line options */
15380 GCC also offers a simple mechanism for printing messages during
15384 @item #pragma message @var{string}
15385 @cindex pragma, diagnostic
15387 Prints @var{string} as a compiler message on compilation. The message
15388 is informational only, and is neither a compilation warning nor an error.
15391 #pragma message "Compiling " __FILE__ "..."
15394 @var{string} may be parenthesized, and is printed with location
15395 information. For example,
15398 #define DO_PRAGMA(x) _Pragma (#x)
15399 #define TODO(x) DO_PRAGMA(message ("TODO - " #x))
15401 TODO(Remember to fix this)
15405 prints @samp{/tmp/file.c:4: note: #pragma message:
15406 TODO - Remember to fix this}.
15410 @node Visibility Pragmas
15411 @subsection Visibility Pragmas
15414 @item #pragma GCC visibility push(@var{visibility})
15415 @itemx #pragma GCC visibility pop
15416 @cindex pragma, visibility
15418 This pragma allows the user to set the visibility for multiple
15419 declarations without having to give each a visibility attribute
15420 @xref{Function Attributes}, for more information about visibility and
15421 the attribute syntax.
15423 In C++, @samp{#pragma GCC visibility} affects only namespace-scope
15424 declarations. Class members and template specializations are not
15425 affected; if you want to override the visibility for a particular
15426 member or instantiation, you must use an attribute.
15431 @node Push/Pop Macro Pragmas
15432 @subsection Push/Pop Macro Pragmas
15434 For compatibility with Microsoft Windows compilers, GCC supports
15435 @samp{#pragma push_macro(@var{"macro_name"})}
15436 and @samp{#pragma pop_macro(@var{"macro_name"})}.
15439 @item #pragma push_macro(@var{"macro_name"})
15440 @cindex pragma, push_macro
15441 This pragma saves the value of the macro named as @var{macro_name} to
15442 the top of the stack for this macro.
15444 @item #pragma pop_macro(@var{"macro_name"})
15445 @cindex pragma, pop_macro
15446 This pragma sets the value of the macro named as @var{macro_name} to
15447 the value on top of the stack for this macro. If the stack for
15448 @var{macro_name} is empty, the value of the macro remains unchanged.
15455 #pragma push_macro("X")
15458 #pragma pop_macro("X")
15463 In this example, the definition of X as 1 is saved by @code{#pragma
15464 push_macro} and restored by @code{#pragma pop_macro}.
15466 @node Function Specific Option Pragmas
15467 @subsection Function Specific Option Pragmas
15470 @item #pragma GCC target (@var{"string"}...)
15471 @cindex pragma GCC target
15473 This pragma allows you to set target specific options for functions
15474 defined later in the source file. One or more strings can be
15475 specified. Each function that is defined after this point is as
15476 if @code{attribute((target("STRING")))} was specified for that
15477 function. The parenthesis around the options is optional.
15478 @xref{Function Attributes}, for more information about the
15479 @code{target} attribute and the attribute syntax.
15481 The @code{#pragma GCC target} attribute is not implemented in GCC versions earlier
15482 than 4.4 for the i386/x86_64 and 4.6 for the PowerPC back ends. At
15483 present, it is not implemented for other back ends.
15487 @item #pragma GCC optimize (@var{"string"}...)
15488 @cindex pragma GCC optimize
15490 This pragma allows you to set global optimization options for functions
15491 defined later in the source file. One or more strings can be
15492 specified. Each function that is defined after this point is as
15493 if @code{attribute((optimize("STRING")))} was specified for that
15494 function. The parenthesis around the options is optional.
15495 @xref{Function Attributes}, for more information about the
15496 @code{optimize} attribute and the attribute syntax.
15498 The @samp{#pragma GCC optimize} pragma is not implemented in GCC
15499 versions earlier than 4.4.
15503 @item #pragma GCC push_options
15504 @itemx #pragma GCC pop_options
15505 @cindex pragma GCC push_options
15506 @cindex pragma GCC pop_options
15508 These pragmas maintain a stack of the current target and optimization
15509 options. It is intended for include files where you temporarily want
15510 to switch to using a different @samp{#pragma GCC target} or
15511 @samp{#pragma GCC optimize} and then to pop back to the previous
15514 The @samp{#pragma GCC push_options} and @samp{#pragma GCC pop_options}
15515 pragmas are not implemented in GCC versions earlier than 4.4.
15519 @item #pragma GCC reset_options
15520 @cindex pragma GCC reset_options
15522 This pragma clears the current @code{#pragma GCC target} and
15523 @code{#pragma GCC optimize} to use the default switches as specified
15524 on the command line.
15526 The @samp{#pragma GCC reset_options} pragma is not implemented in GCC
15527 versions earlier than 4.4.
15530 @node Unnamed Fields
15531 @section Unnamed struct/union fields within structs/unions
15532 @cindex @code{struct}
15533 @cindex @code{union}
15535 As permitted by ISO C11 and for compatibility with other compilers,
15536 GCC allows you to define
15537 a structure or union that contains, as fields, structures and unions
15538 without names. For example:
15552 In this example, you are able to access members of the unnamed
15553 union with code like @samp{foo.b}. Note that only unnamed structs and
15554 unions are allowed, you may not have, for example, an unnamed
15557 You must never create such structures that cause ambiguous field definitions.
15558 For example, in this structure:
15570 it is ambiguous which @code{a} is being referred to with @samp{foo.a}.
15571 The compiler gives errors for such constructs.
15573 @opindex fms-extensions
15574 Unless @option{-fms-extensions} is used, the unnamed field must be a
15575 structure or union definition without a tag (for example, @samp{struct
15576 @{ int a; @};}). If @option{-fms-extensions} is used, the field may
15577 also be a definition with a tag such as @samp{struct foo @{ int a;
15578 @};}, a reference to a previously defined structure or union such as
15579 @samp{struct foo;}, or a reference to a @code{typedef} name for a
15580 previously defined structure or union type.
15582 @opindex fplan9-extensions
15583 The option @option{-fplan9-extensions} enables
15584 @option{-fms-extensions} as well as two other extensions. First, a
15585 pointer to a structure is automatically converted to a pointer to an
15586 anonymous field for assignments and function calls. For example:
15589 struct s1 @{ int a; @};
15590 struct s2 @{ struct s1; @};
15591 extern void f1 (struct s1 *);
15592 void f2 (struct s2 *p) @{ f1 (p); @}
15596 In the call to @code{f1} inside @code{f2}, the pointer @code{p} is
15597 converted into a pointer to the anonymous field.
15599 Second, when the type of an anonymous field is a @code{typedef} for a
15600 @code{struct} or @code{union}, code may refer to the field using the
15601 name of the @code{typedef}.
15604 typedef struct @{ int a; @} s1;
15605 struct s2 @{ s1; @};
15606 s1 f1 (struct s2 *p) @{ return p->s1; @}
15609 These usages are only permitted when they are not ambiguous.
15612 @section Thread-Local Storage
15613 @cindex Thread-Local Storage
15614 @cindex @acronym{TLS}
15615 @cindex @code{__thread}
15617 Thread-local storage (@acronym{TLS}) is a mechanism by which variables
15618 are allocated such that there is one instance of the variable per extant
15619 thread. The runtime model GCC uses to implement this originates
15620 in the IA-64 processor-specific ABI, but has since been migrated
15621 to other processors as well. It requires significant support from
15622 the linker (@command{ld}), dynamic linker (@command{ld.so}), and
15623 system libraries (@file{libc.so} and @file{libpthread.so}), so it
15624 is not available everywhere.
15626 At the user level, the extension is visible with a new storage
15627 class keyword: @code{__thread}. For example:
15631 extern __thread struct state s;
15632 static __thread char *p;
15635 The @code{__thread} specifier may be used alone, with the @code{extern}
15636 or @code{static} specifiers, but with no other storage class specifier.
15637 When used with @code{extern} or @code{static}, @code{__thread} must appear
15638 immediately after the other storage class specifier.
15640 The @code{__thread} specifier may be applied to any global, file-scoped
15641 static, function-scoped static, or static data member of a class. It may
15642 not be applied to block-scoped automatic or non-static data member.
15644 When the address-of operator is applied to a thread-local variable, it is
15645 evaluated at run time and returns the address of the current thread's
15646 instance of that variable. An address so obtained may be used by any
15647 thread. When a thread terminates, any pointers to thread-local variables
15648 in that thread become invalid.
15650 No static initialization may refer to the address of a thread-local variable.
15652 In C++, if an initializer is present for a thread-local variable, it must
15653 be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++
15656 See @uref{http://www.akkadia.org/drepper/tls.pdf,
15657 ELF Handling For Thread-Local Storage} for a detailed explanation of
15658 the four thread-local storage addressing models, and how the runtime
15659 is expected to function.
15662 * C99 Thread-Local Edits::
15663 * C++98 Thread-Local Edits::
15666 @node C99 Thread-Local Edits
15667 @subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage
15669 The following are a set of changes to ISO/IEC 9899:1999 (aka C99)
15670 that document the exact semantics of the language extension.
15674 @cite{5.1.2 Execution environments}
15676 Add new text after paragraph 1
15679 Within either execution environment, a @dfn{thread} is a flow of
15680 control within a program. It is implementation defined whether
15681 or not there may be more than one thread associated with a program.
15682 It is implementation defined how threads beyond the first are
15683 created, the name and type of the function called at thread
15684 startup, and how threads may be terminated. However, objects
15685 with thread storage duration shall be initialized before thread
15690 @cite{6.2.4 Storage durations of objects}
15692 Add new text before paragraph 3
15695 An object whose identifier is declared with the storage-class
15696 specifier @w{@code{__thread}} has @dfn{thread storage duration}.
15697 Its lifetime is the entire execution of the thread, and its
15698 stored value is initialized only once, prior to thread startup.
15702 @cite{6.4.1 Keywords}
15704 Add @code{__thread}.
15707 @cite{6.7.1 Storage-class specifiers}
15709 Add @code{__thread} to the list of storage class specifiers in
15712 Change paragraph 2 to
15715 With the exception of @code{__thread}, at most one storage-class
15716 specifier may be given [@dots{}]. The @code{__thread} specifier may
15717 be used alone, or immediately following @code{extern} or
15721 Add new text after paragraph 6
15724 The declaration of an identifier for a variable that has
15725 block scope that specifies @code{__thread} shall also
15726 specify either @code{extern} or @code{static}.
15728 The @code{__thread} specifier shall be used only with
15733 @node C++98 Thread-Local Edits
15734 @subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage
15736 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
15737 that document the exact semantics of the language extension.
15741 @b{[intro.execution]}
15743 New text after paragraph 4
15746 A @dfn{thread} is a flow of control within the abstract machine.
15747 It is implementation defined whether or not there may be more than
15751 New text after paragraph 7
15754 It is unspecified whether additional action must be taken to
15755 ensure when and whether side effects are visible to other threads.
15761 Add @code{__thread}.
15764 @b{[basic.start.main]}
15766 Add after paragraph 5
15769 The thread that begins execution at the @code{main} function is called
15770 the @dfn{main thread}. It is implementation defined how functions
15771 beginning threads other than the main thread are designated or typed.
15772 A function so designated, as well as the @code{main} function, is called
15773 a @dfn{thread startup function}. It is implementation defined what
15774 happens if a thread startup function returns. It is implementation
15775 defined what happens to other threads when any thread calls @code{exit}.
15779 @b{[basic.start.init]}
15781 Add after paragraph 4
15784 The storage for an object of thread storage duration shall be
15785 statically initialized before the first statement of the thread startup
15786 function. An object of thread storage duration shall not require
15787 dynamic initialization.
15791 @b{[basic.start.term]}
15793 Add after paragraph 3
15796 The type of an object with thread storage duration shall not have a
15797 non-trivial destructor, nor shall it be an array type whose elements
15798 (directly or indirectly) have non-trivial destructors.
15804 Add ``thread storage duration'' to the list in paragraph 1.
15809 Thread, static, and automatic storage durations are associated with
15810 objects introduced by declarations [@dots{}].
15813 Add @code{__thread} to the list of specifiers in paragraph 3.
15816 @b{[basic.stc.thread]}
15818 New section before @b{[basic.stc.static]}
15821 The keyword @code{__thread} applied to a non-local object gives the
15822 object thread storage duration.
15824 A local variable or class data member declared both @code{static}
15825 and @code{__thread} gives the variable or member thread storage
15830 @b{[basic.stc.static]}
15835 All objects that have neither thread storage duration, dynamic
15836 storage duration nor are local [@dots{}].
15842 Add @code{__thread} to the list in paragraph 1.
15847 With the exception of @code{__thread}, at most one
15848 @var{storage-class-specifier} shall appear in a given
15849 @var{decl-specifier-seq}. The @code{__thread} specifier may
15850 be used alone, or immediately following the @code{extern} or
15851 @code{static} specifiers. [@dots{}]
15854 Add after paragraph 5
15857 The @code{__thread} specifier can be applied only to the names of objects
15858 and to anonymous unions.
15864 Add after paragraph 6
15867 Non-@code{static} members shall not be @code{__thread}.
15871 @node Binary constants
15872 @section Binary constants using the @samp{0b} prefix
15873 @cindex Binary constants using the @samp{0b} prefix
15875 Integer constants can be written as binary constants, consisting of a
15876 sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or
15877 @samp{0B}. This is particularly useful in environments that operate a
15878 lot on the bit level (like microcontrollers).
15880 The following statements are identical:
15889 The type of these constants follows the same rules as for octal or
15890 hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL}
15893 @node C++ Extensions
15894 @chapter Extensions to the C++ Language
15895 @cindex extensions, C++ language
15896 @cindex C++ language extensions
15898 The GNU compiler provides these extensions to the C++ language (and you
15899 can also use most of the C language extensions in your C++ programs). If you
15900 want to write code that checks whether these features are available, you can
15901 test for the GNU compiler the same way as for C programs: check for a
15902 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
15903 test specifically for GNU C++ (@pxref{Common Predefined Macros,,
15904 Predefined Macros,cpp,The GNU C Preprocessor}).
15907 * C++ Volatiles:: What constitutes an access to a volatile object.
15908 * Restricted Pointers:: C99 restricted pointers and references.
15909 * Vague Linkage:: Where G++ puts inlines, vtables and such.
15910 * C++ Interface:: You can use a single C++ header file for both
15911 declarations and definitions.
15912 * Template Instantiation:: Methods for ensuring that exactly one copy of
15913 each needed template instantiation is emitted.
15914 * Bound member functions:: You can extract a function pointer to the
15915 method denoted by a @samp{->*} or @samp{.*} expression.
15916 * C++ Attributes:: Variable, function, and type attributes for C++ only.
15917 * Function Multiversioning:: Declaring multiple function versions.
15918 * Namespace Association:: Strong using-directives for namespace association.
15919 * Type Traits:: Compiler support for type traits
15920 * Java Exceptions:: Tweaking exception handling to work with Java.
15921 * Deprecated Features:: Things will disappear from G++.
15922 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
15925 @node C++ Volatiles
15926 @section When is a Volatile C++ Object Accessed?
15927 @cindex accessing volatiles
15928 @cindex volatile read
15929 @cindex volatile write
15930 @cindex volatile access
15932 The C++ standard differs from the C standard in its treatment of
15933 volatile objects. It fails to specify what constitutes a volatile
15934 access, except to say that C++ should behave in a similar manner to C
15935 with respect to volatiles, where possible. However, the different
15936 lvalueness of expressions between C and C++ complicate the behavior.
15937 G++ behaves the same as GCC for volatile access, @xref{C
15938 Extensions,,Volatiles}, for a description of GCC's behavior.
15940 The C and C++ language specifications differ when an object is
15941 accessed in a void context:
15944 volatile int *src = @var{somevalue};
15948 The C++ standard specifies that such expressions do not undergo lvalue
15949 to rvalue conversion, and that the type of the dereferenced object may
15950 be incomplete. The C++ standard does not specify explicitly that it
15951 is lvalue to rvalue conversion that is responsible for causing an
15952 access. There is reason to believe that it is, because otherwise
15953 certain simple expressions become undefined. However, because it
15954 would surprise most programmers, G++ treats dereferencing a pointer to
15955 volatile object of complete type as GCC would do for an equivalent
15956 type in C@. When the object has incomplete type, G++ issues a
15957 warning; if you wish to force an error, you must force a conversion to
15958 rvalue with, for instance, a static cast.
15960 When using a reference to volatile, G++ does not treat equivalent
15961 expressions as accesses to volatiles, but instead issues a warning that
15962 no volatile is accessed. The rationale for this is that otherwise it
15963 becomes difficult to determine where volatile access occur, and not
15964 possible to ignore the return value from functions returning volatile
15965 references. Again, if you wish to force a read, cast the reference to
15968 G++ implements the same behavior as GCC does when assigning to a
15969 volatile object---there is no reread of the assigned-to object, the
15970 assigned rvalue is reused. Note that in C++ assignment expressions
15971 are lvalues, and if used as an lvalue, the volatile object is
15972 referred to. For instance, @var{vref} refers to @var{vobj}, as
15973 expected, in the following example:
15977 volatile int &vref = vobj = @var{something};
15980 @node Restricted Pointers
15981 @section Restricting Pointer Aliasing
15982 @cindex restricted pointers
15983 @cindex restricted references
15984 @cindex restricted this pointer
15986 As with the C front end, G++ understands the C99 feature of restricted pointers,
15987 specified with the @code{__restrict__}, or @code{__restrict} type
15988 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
15989 language flag, @code{restrict} is not a keyword in C++.
15991 In addition to allowing restricted pointers, you can specify restricted
15992 references, which indicate that the reference is not aliased in the local
15996 void fn (int *__restrict__ rptr, int &__restrict__ rref)
16003 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
16004 @var{rref} refers to a (different) unaliased integer.
16006 You may also specify whether a member function's @var{this} pointer is
16007 unaliased by using @code{__restrict__} as a member function qualifier.
16010 void T::fn () __restrict__
16017 Within the body of @code{T::fn}, @var{this} has the effective
16018 definition @code{T *__restrict__ const this}. Notice that the
16019 interpretation of a @code{__restrict__} member function qualifier is
16020 different to that of @code{const} or @code{volatile} qualifier, in that it
16021 is applied to the pointer rather than the object. This is consistent with
16022 other compilers that implement restricted pointers.
16024 As with all outermost parameter qualifiers, @code{__restrict__} is
16025 ignored in function definition matching. This means you only need to
16026 specify @code{__restrict__} in a function definition, rather than
16027 in a function prototype as well.
16029 @node Vague Linkage
16030 @section Vague Linkage
16031 @cindex vague linkage
16033 There are several constructs in C++ that require space in the object
16034 file but are not clearly tied to a single translation unit. We say that
16035 these constructs have ``vague linkage''. Typically such constructs are
16036 emitted wherever they are needed, though sometimes we can be more
16040 @item Inline Functions
16041 Inline functions are typically defined in a header file which can be
16042 included in many different compilations. Hopefully they can usually be
16043 inlined, but sometimes an out-of-line copy is necessary, if the address
16044 of the function is taken or if inlining fails. In general, we emit an
16045 out-of-line copy in all translation units where one is needed. As an
16046 exception, we only emit inline virtual functions with the vtable, since
16047 it always requires a copy.
16049 Local static variables and string constants used in an inline function
16050 are also considered to have vague linkage, since they must be shared
16051 between all inlined and out-of-line instances of the function.
16055 C++ virtual functions are implemented in most compilers using a lookup
16056 table, known as a vtable. The vtable contains pointers to the virtual
16057 functions provided by a class, and each object of the class contains a
16058 pointer to its vtable (or vtables, in some multiple-inheritance
16059 situations). If the class declares any non-inline, non-pure virtual
16060 functions, the first one is chosen as the ``key method'' for the class,
16061 and the vtable is only emitted in the translation unit where the key
16064 @emph{Note:} If the chosen key method is later defined as inline, the
16065 vtable is still emitted in every translation unit that defines it.
16066 Make sure that any inline virtuals are declared inline in the class
16067 body, even if they are not defined there.
16069 @item @code{type_info} objects
16070 @cindex @code{type_info}
16072 C++ requires information about types to be written out in order to
16073 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
16074 For polymorphic classes (classes with virtual functions), the @samp{type_info}
16075 object is written out along with the vtable so that @samp{dynamic_cast}
16076 can determine the dynamic type of a class object at run time. For all
16077 other types, we write out the @samp{type_info} object when it is used: when
16078 applying @samp{typeid} to an expression, throwing an object, or
16079 referring to a type in a catch clause or exception specification.
16081 @item Template Instantiations
16082 Most everything in this section also applies to template instantiations,
16083 but there are other options as well.
16084 @xref{Template Instantiation,,Where's the Template?}.
16088 When used with GNU ld version 2.8 or later on an ELF system such as
16089 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
16090 these constructs will be discarded at link time. This is known as
16093 On targets that don't support COMDAT, but do support weak symbols, GCC
16094 uses them. This way one copy overrides all the others, but
16095 the unused copies still take up space in the executable.
16097 For targets that do not support either COMDAT or weak symbols,
16098 most entities with vague linkage are emitted as local symbols to
16099 avoid duplicate definition errors from the linker. This does not happen
16100 for local statics in inlines, however, as having multiple copies
16101 almost certainly breaks things.
16103 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
16104 another way to control placement of these constructs.
16106 @node C++ Interface
16107 @section #pragma interface and implementation
16109 @cindex interface and implementation headers, C++
16110 @cindex C++ interface and implementation headers
16111 @cindex pragmas, interface and implementation
16113 @code{#pragma interface} and @code{#pragma implementation} provide the
16114 user with a way of explicitly directing the compiler to emit entities
16115 with vague linkage (and debugging information) in a particular
16118 @emph{Note:} As of GCC 2.7.2, these @code{#pragma}s are not useful in
16119 most cases, because of COMDAT support and the ``key method'' heuristic
16120 mentioned in @ref{Vague Linkage}. Using them can actually cause your
16121 program to grow due to unnecessary out-of-line copies of inline
16122 functions. Currently (3.4) the only benefit of these
16123 @code{#pragma}s is reduced duplication of debugging information, and
16124 that should be addressed soon on DWARF 2 targets with the use of
16128 @item #pragma interface
16129 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
16130 @kindex #pragma interface
16131 Use this directive in @emph{header files} that define object classes, to save
16132 space in most of the object files that use those classes. Normally,
16133 local copies of certain information (backup copies of inline member
16134 functions, debugging information, and the internal tables that implement
16135 virtual functions) must be kept in each object file that includes class
16136 definitions. You can use this pragma to avoid such duplication. When a
16137 header file containing @samp{#pragma interface} is included in a
16138 compilation, this auxiliary information is not generated (unless
16139 the main input source file itself uses @samp{#pragma implementation}).
16140 Instead, the object files contain references to be resolved at link
16143 The second form of this directive is useful for the case where you have
16144 multiple headers with the same name in different directories. If you
16145 use this form, you must specify the same string to @samp{#pragma
16148 @item #pragma implementation
16149 @itemx #pragma implementation "@var{objects}.h"
16150 @kindex #pragma implementation
16151 Use this pragma in a @emph{main input file}, when you want full output from
16152 included header files to be generated (and made globally visible). The
16153 included header file, in turn, should use @samp{#pragma interface}.
16154 Backup copies of inline member functions, debugging information, and the
16155 internal tables used to implement virtual functions are all generated in
16156 implementation files.
16158 @cindex implied @code{#pragma implementation}
16159 @cindex @code{#pragma implementation}, implied
16160 @cindex naming convention, implementation headers
16161 If you use @samp{#pragma implementation} with no argument, it applies to
16162 an include file with the same basename@footnote{A file's @dfn{basename}
16163 is the name stripped of all leading path information and of trailing
16164 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
16165 file. For example, in @file{allclass.cc}, giving just
16166 @samp{#pragma implementation}
16167 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
16169 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
16170 an implementation file whenever you would include it from
16171 @file{allclass.cc} even if you never specified @samp{#pragma
16172 implementation}. This was deemed to be more trouble than it was worth,
16173 however, and disabled.
16175 Use the string argument if you want a single implementation file to
16176 include code from multiple header files. (You must also use
16177 @samp{#include} to include the header file; @samp{#pragma
16178 implementation} only specifies how to use the file---it doesn't actually
16181 There is no way to split up the contents of a single header file into
16182 multiple implementation files.
16185 @cindex inlining and C++ pragmas
16186 @cindex C++ pragmas, effect on inlining
16187 @cindex pragmas in C++, effect on inlining
16188 @samp{#pragma implementation} and @samp{#pragma interface} also have an
16189 effect on function inlining.
16191 If you define a class in a header file marked with @samp{#pragma
16192 interface}, the effect on an inline function defined in that class is
16193 similar to an explicit @code{extern} declaration---the compiler emits
16194 no code at all to define an independent version of the function. Its
16195 definition is used only for inlining with its callers.
16197 @opindex fno-implement-inlines
16198 Conversely, when you include the same header file in a main source file
16199 that declares it as @samp{#pragma implementation}, the compiler emits
16200 code for the function itself; this defines a version of the function
16201 that can be found via pointers (or by callers compiled without
16202 inlining). If all calls to the function can be inlined, you can avoid
16203 emitting the function by compiling with @option{-fno-implement-inlines}.
16204 If any calls are not inlined, you will get linker errors.
16206 @node Template Instantiation
16207 @section Where's the Template?
16208 @cindex template instantiation
16210 C++ templates are the first language feature to require more
16211 intelligence from the environment than one usually finds on a UNIX
16212 system. Somehow the compiler and linker have to make sure that each
16213 template instance occurs exactly once in the executable if it is needed,
16214 and not at all otherwise. There are two basic approaches to this
16215 problem, which are referred to as the Borland model and the Cfront model.
16218 @item Borland model
16219 Borland C++ solved the template instantiation problem by adding the code
16220 equivalent of common blocks to their linker; the compiler emits template
16221 instances in each translation unit that uses them, and the linker
16222 collapses them together. The advantage of this model is that the linker
16223 only has to consider the object files themselves; there is no external
16224 complexity to worry about. This disadvantage is that compilation time
16225 is increased because the template code is being compiled repeatedly.
16226 Code written for this model tends to include definitions of all
16227 templates in the header file, since they must be seen to be
16231 The AT&T C++ translator, Cfront, solved the template instantiation
16232 problem by creating the notion of a template repository, an
16233 automatically maintained place where template instances are stored. A
16234 more modern version of the repository works as follows: As individual
16235 object files are built, the compiler places any template definitions and
16236 instantiations encountered in the repository. At link time, the link
16237 wrapper adds in the objects in the repository and compiles any needed
16238 instances that were not previously emitted. The advantages of this
16239 model are more optimal compilation speed and the ability to use the
16240 system linker; to implement the Borland model a compiler vendor also
16241 needs to replace the linker. The disadvantages are vastly increased
16242 complexity, and thus potential for error; for some code this can be
16243 just as transparent, but in practice it can been very difficult to build
16244 multiple programs in one directory and one program in multiple
16245 directories. Code written for this model tends to separate definitions
16246 of non-inline member templates into a separate file, which should be
16247 compiled separately.
16250 When used with GNU ld version 2.8 or later on an ELF system such as
16251 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
16252 Borland model. On other systems, G++ implements neither automatic
16255 You have the following options for dealing with template instantiations:
16260 Compile your template-using code with @option{-frepo}. The compiler
16261 generates files with the extension @samp{.rpo} listing all of the
16262 template instantiations used in the corresponding object files that
16263 could be instantiated there; the link wrapper, @samp{collect2},
16264 then updates the @samp{.rpo} files to tell the compiler where to place
16265 those instantiations and rebuild any affected object files. The
16266 link-time overhead is negligible after the first pass, as the compiler
16267 continues to place the instantiations in the same files.
16269 This is your best option for application code written for the Borland
16270 model, as it just works. Code written for the Cfront model
16271 needs to be modified so that the template definitions are available at
16272 one or more points of instantiation; usually this is as simple as adding
16273 @code{#include <tmethods.cc>} to the end of each template header.
16275 For library code, if you want the library to provide all of the template
16276 instantiations it needs, just try to link all of its object files
16277 together; the link will fail, but cause the instantiations to be
16278 generated as a side effect. Be warned, however, that this may cause
16279 conflicts if multiple libraries try to provide the same instantiations.
16280 For greater control, use explicit instantiation as described in the next
16284 @opindex fno-implicit-templates
16285 Compile your code with @option{-fno-implicit-templates} to disable the
16286 implicit generation of template instances, and explicitly instantiate
16287 all the ones you use. This approach requires more knowledge of exactly
16288 which instances you need than do the others, but it's less
16289 mysterious and allows greater control. You can scatter the explicit
16290 instantiations throughout your program, perhaps putting them in the
16291 translation units where the instances are used or the translation units
16292 that define the templates themselves; you can put all of the explicit
16293 instantiations you need into one big file; or you can create small files
16300 template class Foo<int>;
16301 template ostream& operator <<
16302 (ostream&, const Foo<int>&);
16306 for each of the instances you need, and create a template instantiation
16307 library from those.
16309 If you are using Cfront-model code, you can probably get away with not
16310 using @option{-fno-implicit-templates} when compiling files that don't
16311 @samp{#include} the member template definitions.
16313 If you use one big file to do the instantiations, you may want to
16314 compile it without @option{-fno-implicit-templates} so you get all of the
16315 instances required by your explicit instantiations (but not by any
16316 other files) without having to specify them as well.
16318 The ISO C++ 2011 standard allows forward declaration of explicit
16319 instantiations (with @code{extern}). G++ supports explicit instantiation
16320 declarations in C++98 mode and has extended the template instantiation
16321 syntax to support instantiation of the compiler support data for a
16322 template class (i.e.@: the vtable) without instantiating any of its
16323 members (with @code{inline}), and instantiation of only the static data
16324 members of a template class, without the support data or member
16325 functions (with (@code{static}):
16328 extern template int max (int, int);
16329 inline template class Foo<int>;
16330 static template class Foo<int>;
16334 Do nothing. Pretend G++ does implement automatic instantiation
16335 management. Code written for the Borland model works fine, but
16336 each translation unit contains instances of each of the templates it
16337 uses. In a large program, this can lead to an unacceptable amount of code
16341 @node Bound member functions
16342 @section Extracting the function pointer from a bound pointer to member function
16344 @cindex pointer to member function
16345 @cindex bound pointer to member function
16347 In C++, pointer to member functions (PMFs) are implemented using a wide
16348 pointer of sorts to handle all the possible call mechanisms; the PMF
16349 needs to store information about how to adjust the @samp{this} pointer,
16350 and if the function pointed to is virtual, where to find the vtable, and
16351 where in the vtable to look for the member function. If you are using
16352 PMFs in an inner loop, you should really reconsider that decision. If
16353 that is not an option, you can extract the pointer to the function that
16354 would be called for a given object/PMF pair and call it directly inside
16355 the inner loop, to save a bit of time.
16357 Note that you still pay the penalty for the call through a
16358 function pointer; on most modern architectures, such a call defeats the
16359 branch prediction features of the CPU@. This is also true of normal
16360 virtual function calls.
16362 The syntax for this extension is
16366 extern int (A::*fp)();
16367 typedef int (*fptr)(A *);
16369 fptr p = (fptr)(a.*fp);
16372 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
16373 no object is needed to obtain the address of the function. They can be
16374 converted to function pointers directly:
16377 fptr p1 = (fptr)(&A::foo);
16380 @opindex Wno-pmf-conversions
16381 You must specify @option{-Wno-pmf-conversions} to use this extension.
16383 @node C++ Attributes
16384 @section C++-Specific Variable, Function, and Type Attributes
16386 Some attributes only make sense for C++ programs.
16389 @item abi_tag ("@var{tag}", ...)
16390 @cindex @code{abi_tag} attribute
16391 The @code{abi_tag} attribute can be applied to a function or class
16392 declaration. It modifies the mangled name of the function or class to
16393 incorporate the tag name, in order to distinguish the function or
16394 class from an earlier version with a different ABI; perhaps the class
16395 has changed size, or the function has a different return type that is
16396 not encoded in the mangled name.
16398 The argument can be a list of strings of arbitrary length. The
16399 strings are sorted on output, so the order of the list is
16402 A redeclaration of a function or class must not add new ABI tags,
16403 since doing so would change the mangled name.
16405 The @option{-Wabi-tag} flag enables a warning about a class which does
16406 not have all the ABI tags used by its subobjects and virtual functions; for users with code
16407 that needs to coexist with an earlier ABI, using this option can help
16408 to find all affected types that need to be tagged.
16410 @item init_priority (@var{priority})
16411 @cindex @code{init_priority} attribute
16414 In Standard C++, objects defined at namespace scope are guaranteed to be
16415 initialized in an order in strict accordance with that of their definitions
16416 @emph{in a given translation unit}. No guarantee is made for initializations
16417 across translation units. However, GNU C++ allows users to control the
16418 order of initialization of objects defined at namespace scope with the
16419 @code{init_priority} attribute by specifying a relative @var{priority},
16420 a constant integral expression currently bounded between 101 and 65535
16421 inclusive. Lower numbers indicate a higher priority.
16423 In the following example, @code{A} would normally be created before
16424 @code{B}, but the @code{init_priority} attribute reverses that order:
16427 Some_Class A __attribute__ ((init_priority (2000)));
16428 Some_Class B __attribute__ ((init_priority (543)));
16432 Note that the particular values of @var{priority} do not matter; only their
16435 @item java_interface
16436 @cindex @code{java_interface} attribute
16438 This type attribute informs C++ that the class is a Java interface. It may
16439 only be applied to classes declared within an @code{extern "Java"} block.
16440 Calls to methods declared in this interface are dispatched using GCJ's
16441 interface table mechanism, instead of regular virtual table dispatch.
16444 @cindex @code{warn_unused} attribute
16446 For C++ types with non-trivial constructors and/or destructors it is
16447 impossible for the compiler to determine whether a variable of this
16448 type is truly unused if it is not referenced. This type attribute
16449 informs the compiler that variables of this type should be warned
16450 about if they appear to be unused, just like variables of fundamental
16453 This attribute is appropriate for types which just represent a value,
16454 such as @code{std::string}; it is not appropriate for types which
16455 control a resource, such as @code{std::mutex}.
16457 This attribute is also accepted in C, but it is unnecessary because C
16458 does not have constructors or destructors.
16462 See also @ref{Namespace Association}.
16464 @node Function Multiversioning
16465 @section Function Multiversioning
16466 @cindex function versions
16468 With the GNU C++ front end, for target i386, you may specify multiple
16469 versions of a function, where each function is specialized for a
16470 specific target feature. At runtime, the appropriate version of the
16471 function is automatically executed depending on the characteristics of
16472 the execution platform. Here is an example.
16475 __attribute__ ((target ("default")))
16478 // The default version of foo.
16482 __attribute__ ((target ("sse4.2")))
16485 // foo version for SSE4.2
16489 __attribute__ ((target ("arch=atom")))
16492 // foo version for the Intel ATOM processor
16496 __attribute__ ((target ("arch=amdfam10")))
16499 // foo version for the AMD Family 0x10 processors.
16506 assert ((*p) () == foo ());
16511 In the above example, four versions of function foo are created. The
16512 first version of foo with the target attribute "default" is the default
16513 version. This version gets executed when no other target specific
16514 version qualifies for execution on a particular platform. A new version
16515 of foo is created by using the same function signature but with a
16516 different target string. Function foo is called or a pointer to it is
16517 taken just like a regular function. GCC takes care of doing the
16518 dispatching to call the right version at runtime. Refer to the
16519 @uref{http://gcc.gnu.org/wiki/FunctionMultiVersioning, GCC wiki on
16520 Function Multiversioning} for more details.
16522 @node Namespace Association
16523 @section Namespace Association
16525 @strong{Caution:} The semantics of this extension are equivalent
16526 to C++ 2011 inline namespaces. Users should use inline namespaces
16527 instead as this extension will be removed in future versions of G++.
16529 A using-directive with @code{__attribute ((strong))} is stronger
16530 than a normal using-directive in two ways:
16534 Templates from the used namespace can be specialized and explicitly
16535 instantiated as though they were members of the using namespace.
16538 The using namespace is considered an associated namespace of all
16539 templates in the used namespace for purposes of argument-dependent
16543 The used namespace must be nested within the using namespace so that
16544 normal unqualified lookup works properly.
16546 This is useful for composing a namespace transparently from
16547 implementation namespaces. For example:
16552 template <class T> struct A @{ @};
16554 using namespace debug __attribute ((__strong__));
16555 template <> struct A<int> @{ @}; // @r{OK to specialize}
16557 template <class T> void f (A<T>);
16562 f (std::A<float>()); // @r{lookup finds} std::f
16568 @section Type Traits
16570 The C++ front end implements syntactic extensions that allow
16571 compile-time determination of
16572 various characteristics of a type (or of a
16576 @item __has_nothrow_assign (type)
16577 If @code{type} is const qualified or is a reference type then the trait is
16578 false. Otherwise if @code{__has_trivial_assign (type)} is true then the trait
16579 is true, else if @code{type} is a cv class or union type with copy assignment
16580 operators that are known not to throw an exception then the trait is true,
16581 else it is false. Requires: @code{type} shall be a complete type,
16582 (possibly cv-qualified) @code{void}, or an array of unknown bound.
16584 @item __has_nothrow_copy (type)
16585 If @code{__has_trivial_copy (type)} is true then the trait is true, else if
16586 @code{type} is a cv class or union type with copy constructors that
16587 are known not to throw an exception then the trait is true, else it is false.
16588 Requires: @code{type} shall be a complete type, (possibly cv-qualified)
16589 @code{void}, or an array of unknown bound.
16591 @item __has_nothrow_constructor (type)
16592 If @code{__has_trivial_constructor (type)} is true then the trait is
16593 true, else if @code{type} is a cv class or union type (or array
16594 thereof) with a default constructor that is known not to throw an
16595 exception then the trait is true, else it is false. Requires:
16596 @code{type} shall be a complete type, (possibly cv-qualified)
16597 @code{void}, or an array of unknown bound.
16599 @item __has_trivial_assign (type)
16600 If @code{type} is const qualified or is a reference type then the trait is
16601 false. Otherwise if @code{__is_pod (type)} is true then the trait is
16602 true, else if @code{type} is a cv class or union type with a trivial
16603 copy assignment ([class.copy]) then the trait is true, else it is
16604 false. Requires: @code{type} shall be a complete type, (possibly
16605 cv-qualified) @code{void}, or an array of unknown bound.
16607 @item __has_trivial_copy (type)
16608 If @code{__is_pod (type)} is true or @code{type} is a reference type
16609 then the trait is true, else if @code{type} is a cv class or union type
16610 with a trivial copy constructor ([class.copy]) then the trait
16611 is true, else it is false. Requires: @code{type} shall be a complete
16612 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
16614 @item __has_trivial_constructor (type)
16615 If @code{__is_pod (type)} is true then the trait is true, else if
16616 @code{type} is a cv class or union type (or array thereof) with a
16617 trivial default constructor ([class.ctor]) then the trait is true,
16618 else it is false. Requires: @code{type} shall be a complete
16619 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
16621 @item __has_trivial_destructor (type)
16622 If @code{__is_pod (type)} is true or @code{type} is a reference type then
16623 the trait is true, else if @code{type} is a cv class or union type (or
16624 array thereof) with a trivial destructor ([class.dtor]) then the trait
16625 is true, else it is false. Requires: @code{type} shall be a complete
16626 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
16628 @item __has_virtual_destructor (type)
16629 If @code{type} is a class type with a virtual destructor
16630 ([class.dtor]) then the trait is true, else it is false. Requires:
16631 @code{type} shall be a complete type, (possibly cv-qualified)
16632 @code{void}, or an array of unknown bound.
16634 @item __is_abstract (type)
16635 If @code{type} is an abstract class ([class.abstract]) then the trait
16636 is true, else it is false. Requires: @code{type} shall be a complete
16637 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
16639 @item __is_base_of (base_type, derived_type)
16640 If @code{base_type} is a base class of @code{derived_type}
16641 ([class.derived]) then the trait is true, otherwise it is false.
16642 Top-level cv qualifications of @code{base_type} and
16643 @code{derived_type} are ignored. For the purposes of this trait, a
16644 class type is considered is own base. Requires: if @code{__is_class
16645 (base_type)} and @code{__is_class (derived_type)} are true and
16646 @code{base_type} and @code{derived_type} are not the same type
16647 (disregarding cv-qualifiers), @code{derived_type} shall be a complete
16648 type. Diagnostic is produced if this requirement is not met.
16650 @item __is_class (type)
16651 If @code{type} is a cv class type, and not a union type
16652 ([basic.compound]) the trait is true, else it is false.
16654 @item __is_empty (type)
16655 If @code{__is_class (type)} is false then the trait is false.
16656 Otherwise @code{type} is considered empty if and only if: @code{type}
16657 has no non-static data members, or all non-static data members, if
16658 any, are bit-fields of length 0, and @code{type} has no virtual
16659 members, and @code{type} has no virtual base classes, and @code{type}
16660 has no base classes @code{base_type} for which
16661 @code{__is_empty (base_type)} is false. Requires: @code{type} shall
16662 be a complete type, (possibly cv-qualified) @code{void}, or an array
16665 @item __is_enum (type)
16666 If @code{type} is a cv enumeration type ([basic.compound]) the trait is
16667 true, else it is false.
16669 @item __is_literal_type (type)
16670 If @code{type} is a literal type ([basic.types]) the trait is
16671 true, else it is false. Requires: @code{type} shall be a complete type,
16672 (possibly cv-qualified) @code{void}, or an array of unknown bound.
16674 @item __is_pod (type)
16675 If @code{type} is a cv POD type ([basic.types]) then the trait is true,
16676 else it is false. Requires: @code{type} shall be a complete type,
16677 (possibly cv-qualified) @code{void}, or an array of unknown bound.
16679 @item __is_polymorphic (type)
16680 If @code{type} is a polymorphic class ([class.virtual]) then the trait
16681 is true, else it is false. Requires: @code{type} shall be a complete
16682 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
16684 @item __is_standard_layout (type)
16685 If @code{type} is a standard-layout type ([basic.types]) the trait is
16686 true, else it is false. Requires: @code{type} shall be a complete
16687 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
16689 @item __is_trivial (type)
16690 If @code{type} is a trivial type ([basic.types]) the trait is
16691 true, else it is false. Requires: @code{type} shall be a complete
16692 type, (possibly cv-qualified) @code{void}, or an array of unknown bound.
16694 @item __is_union (type)
16695 If @code{type} is a cv union type ([basic.compound]) the trait is
16696 true, else it is false.
16698 @item __underlying_type (type)
16699 The underlying type of @code{type}. Requires: @code{type} shall be
16700 an enumeration type ([dcl.enum]).
16704 @node Java Exceptions
16705 @section Java Exceptions
16707 The Java language uses a slightly different exception handling model
16708 from C++. Normally, GNU C++ automatically detects when you are
16709 writing C++ code that uses Java exceptions, and handle them
16710 appropriately. However, if C++ code only needs to execute destructors
16711 when Java exceptions are thrown through it, GCC guesses incorrectly.
16712 Sample problematic code is:
16715 struct S @{ ~S(); @};
16716 extern void bar(); // @r{is written in Java, and may throw exceptions}
16725 The usual effect of an incorrect guess is a link failure, complaining of
16726 a missing routine called @samp{__gxx_personality_v0}.
16728 You can inform the compiler that Java exceptions are to be used in a
16729 translation unit, irrespective of what it might think, by writing
16730 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
16731 @samp{#pragma} must appear before any functions that throw or catch
16732 exceptions, or run destructors when exceptions are thrown through them.
16734 You cannot mix Java and C++ exceptions in the same translation unit. It
16735 is believed to be safe to throw a C++ exception from one file through
16736 another file compiled for the Java exception model, or vice versa, but
16737 there may be bugs in this area.
16739 @node Deprecated Features
16740 @section Deprecated Features
16742 In the past, the GNU C++ compiler was extended to experiment with new
16743 features, at a time when the C++ language was still evolving. Now that
16744 the C++ standard is complete, some of those features are superseded by
16745 superior alternatives. Using the old features might cause a warning in
16746 some cases that the feature will be dropped in the future. In other
16747 cases, the feature might be gone already.
16749 While the list below is not exhaustive, it documents some of the options
16750 that are now deprecated:
16753 @item -fexternal-templates
16754 @itemx -falt-external-templates
16755 These are two of the many ways for G++ to implement template
16756 instantiation. @xref{Template Instantiation}. The C++ standard clearly
16757 defines how template definitions have to be organized across
16758 implementation units. G++ has an implicit instantiation mechanism that
16759 should work just fine for standard-conforming code.
16761 @item -fstrict-prototype
16762 @itemx -fno-strict-prototype
16763 Previously it was possible to use an empty prototype parameter list to
16764 indicate an unspecified number of parameters (like C), rather than no
16765 parameters, as C++ demands. This feature has been removed, except where
16766 it is required for backwards compatibility. @xref{Backwards Compatibility}.
16769 G++ allows a virtual function returning @samp{void *} to be overridden
16770 by one returning a different pointer type. This extension to the
16771 covariant return type rules is now deprecated and will be removed from a
16774 The G++ minimum and maximum operators (@samp{<?} and @samp{>?}) and
16775 their compound forms (@samp{<?=}) and @samp{>?=}) have been deprecated
16776 and are now removed from G++. Code using these operators should be
16777 modified to use @code{std::min} and @code{std::max} instead.
16779 The named return value extension has been deprecated, and is now
16782 The use of initializer lists with new expressions has been deprecated,
16783 and is now removed from G++.
16785 Floating and complex non-type template parameters have been deprecated,
16786 and are now removed from G++.
16788 The implicit typename extension has been deprecated and is now
16791 The use of default arguments in function pointers, function typedefs
16792 and other places where they are not permitted by the standard is
16793 deprecated and will be removed from a future version of G++.
16795 G++ allows floating-point literals to appear in integral constant expressions,
16796 e.g.@: @samp{ enum E @{ e = int(2.2 * 3.7) @} }
16797 This extension is deprecated and will be removed from a future version.
16799 G++ allows static data members of const floating-point type to be declared
16800 with an initializer in a class definition. The standard only allows
16801 initializers for static members of const integral types and const
16802 enumeration types so this extension has been deprecated and will be removed
16803 from a future version.
16805 @node Backwards Compatibility
16806 @section Backwards Compatibility
16807 @cindex Backwards Compatibility
16808 @cindex ARM [Annotated C++ Reference Manual]
16810 Now that there is a definitive ISO standard C++, G++ has a specification
16811 to adhere to. The C++ language evolved over time, and features that
16812 used to be acceptable in previous drafts of the standard, such as the ARM
16813 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
16814 compilation of C++ written to such drafts, G++ contains some backwards
16815 compatibilities. @emph{All such backwards compatibility features are
16816 liable to disappear in future versions of G++.} They should be considered
16817 deprecated. @xref{Deprecated Features}.
16821 If a variable is declared at for scope, it used to remain in scope until
16822 the end of the scope that contained the for statement (rather than just
16823 within the for scope). G++ retains this, but issues a warning, if such a
16824 variable is accessed outside the for scope.
16826 @item Implicit C language
16827 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
16828 scope to set the language. On such systems, all header files are
16829 implicitly scoped inside a C language scope. Also, an empty prototype
16830 @code{()} is treated as an unspecified number of arguments, rather
16831 than no arguments, as C++ demands.